JP5072449B2 - Medical image processing apparatus and medical image processing method - Google Patents

Description

  The present invention relates to a medical image processing apparatus and a medical image processing method for processing image data of a medical image, and more particularly to a technique for aligning two medical images (image data thereof).

  Conventionally, for a certain subject, by comparing a plurality of medical images with different imaging timings, it is possible to grasp the progress of a disease, the state of healing, etc., and to examine a treatment policy (for example, (See Patent Document 1). In such comparative interpretation, in order to compare slice images (CT images or the like) at substantially the same position of the subject, it is necessary to perform alignment between images having different imaging timings. In recent years, with the advent of devices capable of capturing 3D medical images, it has become necessary to align 3D medical images.

  As registration methods between images, an iterative registration method using image correlation and a registration method using landmarks (called “feature points”) are generally known. It has been.

  For example, Non-Patent Document 1 discloses an iterative registration method using image correlation. This method is an iterative alignment that uses image correlation such as correlation coefficient. This method is called “rigid registration” in which the image is aligned without deformation, and the image is deformed. Are classified into a method called “non-rigid registration”. In this specification, these two methods will be collectively referred to as “registration”.

  On the other hand, an alignment method using landmarks is disclosed in Non-Patent Document 2, for example. This approach inherently has a shorter processing time compared to an iterative registration method using image correlation. That is, once a landmark is specified, a linear optimization method can be applied to align the landmarks that are paired, and it is not necessary to use an iterative method. Image alignment can be executed in a processing time of about a second or less.

JP 2005-334219 A Jeongtae Kim, Jeffery A. Fessler, “Intensity-Based Image Registration Usable Robust Correlation Coefficients”, IEEE TRANSACTIONS ON MEDICAL IMAGEING, VOL. 23, NO. 11, NOVEMBER 2004, pp. 1430-1444 Hidenori Shikata, 7 others, “Vessel branch point position detection algorithm for non-rigid registration of lungs”, IEICE Transactions, Vol. J85-D-II, no. 10, pp. 1613-1623, 2002

  However, the conventional positioning method as described above has the following problems.

  First, in the iterative registration method using image correlation, several tens of local regions are set for each image to be compared, image correlation is calculated for the corresponding local region pair, and Processing such as translation, rotation, scaling (size change), and deformation of the image is repeatedly performed so that the image correlation becomes high. This method is effective to some extent for a two-dimensional image with a small amount of data, but for a three-dimensional image with a much larger amount of data, it takes a long time to process, and is not practical at this stage.

  For example, Non-Patent Document 1 describes that the calculation time of the image correlation of a local region of a 10 × 10 pixel two-dimensional image is about 0.1 seconds. When similar processing is applied to a local region of 10 × 10 × 10 voxels of a three-dimensional image, it is assumed that it takes a time of approximately 10 × 0.1 = 1 second. Therefore, for example, when 50 pairs of local regions are set in a three-dimensional image, a time of about 1 × 50 = 50 seconds is required for one calculation.

  A known optimization method is used as an image alignment algorithm, but it is known that the number of iterations of calculation increases as the number of variables increases. For example, in non-rigid registration of a three-dimensional image, it is necessary to optimize about several tens of variables, and for this purpose, iterative calculation is required at least about 100 times. Therefore, it is estimated that it takes about 50 × 100 = 5000 seconds (about 83 minutes) to align the three-dimensional image. In this case, even if a technique for speeding up the processing is additionally applied, it is expected that it is difficult to achieve a practical processing time.

  On the other hand, the alignment method using landmarks has a merit that the processing time is short as described above, but has a difficulty in alignment accuracy. That is, for example, when aligning an image including a pulmonary blood vessel, as described in Non-Patent Document 2, in order to perform alignment using a landmark with high accuracy, the position of a branch point is highly accurate. Decisions and accurate graphical representations of blood vessels are required. This is due to the fact that the specific accuracy of the corresponding landmarks is reflected in the alignment accuracy for the landmark groups independently extracted from the two images to be compared.

  In addition, it is practically very difficult to accurately obtain a graph representation of an object that is stretched in a complicated manner such as a pulmonary blood vessel. This is also a big obstacle to realizing the accuracy improvement of the alignment method using landmarks. Particularly in the medical field, it is difficult to think that the method using landmarks is highly useful because accuracy must be emphasized in order to avoid medical errors.

  The present invention has been made to solve the above-described problems, and a medical image processing apparatus and medical device capable of performing alignment with high accuracy while shortening the time required for alignment of medical images. An object is to provide an image processing method.

In order to achieve the above object, the invention described in claim 1 is a medical image processing apparatus for aligning volume data of two medical images, and sets landmarks of medical images based on the respective volume data And a plurality of pairs of two landmarks of the same medical image are formed for each volume data based on the landmarks set by the setting means, and the landmarks are set for each of the plurality of pairs. calculating a distance between a generating means for generating distance information including the distance as the positional relationship information, based on the positional relationship information, and distance between pairs of landmark of one of the medical images, other medical Compare the distances between the landmark pairs in the image, and tentatively associate the pairs with each other based on the comparison result. Add a landmark not included in the pair to form a set of three or more landmarks, and based on the distance or angle determined using the landmarks included in the pair, between the two medical images Calculating an evaluation value indicating the degree of correspondence of the set at, removing some of the landmarks based on the evaluation value, and association means for associating the landmarks of the medical image remaining after the exclusion with each other; Alignment means for aligning two volume data based on the association of the landmarks.

The invention according to claim 14 is a medical image processing method for aligning volume data of two medical images, wherein landmarks of the medical image are set based on each volume data, and each volume data A plurality of pairs of two landmarks of the same medical image are formed based on the set landmarks, a distance between the landmarks is calculated for each of the plurality of pairs, and a distance including the distance Information is generated as positional relationship information, and based on the positional relationship information, the distance between the landmark pair of one medical image is compared with the distance between the landmark pair of the other medical image; Based on the comparison result, pairs are provisionally associated with each other, and for each of the provisional pairs, three or more labels are added by adding landmarks not included in the pair. An evaluation value indicating the degree of correspondence between the two medical images is calculated based on the distance or angle obtained using the landmark included in the set, and the evaluation value Based on this , some of the landmarks are excluded, the landmarks of the two medical images remaining after the exclusion are associated with each other, and the two volume data are aligned based on the association of the landmarks It is characterized by that.

According to the invention described in claim 1 or claim 14, to exclude some of the landmarks based on the positional relationship of the landmarks in a medical image, to perform alignment of the volume data by using the remaining landmarks Therefore, it is possible to realize highly accurate alignment while shortening the time required for alignment.

  An example of an embodiment of a medical image processing apparatus and a medical image processing method according to the present invention will be described in detail with reference to the drawings. In this embodiment, alignment of image data of a medical image obtained by imaging a subject's lung will be described. In general, a medical image obtained by imaging a lung shows a lung parenchyma, bronchi, pulmonary blood vessels (pulmonary artery, pulmonary vein), and the like. In this embodiment, the image data may be identified with the medical image displayed based on the image data.

[Overview]
The present invention is for suitably performing image alignment when performing comparative interpretation of medical images acquired by a medical image diagnostic apparatus. In comparative interpretation, the progress of a disease and the treatment effect are grasped by comparing the latest captured image (latest image) with an image captured in the past (past image). When comparing slice images of the subject, it is necessary to display the latest image and the past image at (substantially) the same slice position of the subject.

  It should be noted that technically, the relationship between the shooting times of the images to be compared is not related to the realization of the alignment method, so in this specification, one image is referred to as a “reference image” and the other image as “target. It will be referred to as “image”, and a method for aligning both will be described.

  FIG. 7 shows an example of the flow of processing according to the present invention. First, landmarks are extracted from the past image and the latest image (S1, S2). As this landmark, for example, a branch point of a pulmonary blood vessel is extracted. An example of a technique for specifying a branch point of a blood vessel is disclosed in Non-Patent Document 2 described above. Note that the order of the landmarks to be extracted may be indefinite between the past image and the latest image, and the number of landmarks may not match. A landmark is a mark (feature point) artificially set for a characteristic part in a medical image, and is a reference for performing alignment of two medical images (image data thereof). It will be.

  Next, the correspondence between the landmarks extracted from the past image and the latest image is determined (S3). In this embodiment, the correspondence between the landmark of the past image and the landmark of the latest image is determined using information on the distance between the landmarks (or information on the angle formed by the landmark). Although details will be described later, such association enables image alignment even when there is an extraction error in the landmark or even when the branching structure of the blood vessel cannot be specified accurately. This is one of the features of this embodiment.

  In the subsequent conversion parameter determination (I), the linear optimization method is applied to the associated landmarks, and the coordinate conversion parameters between these landmarks are determined (S4). Hereinafter, an example of this process will be described. Assume that the landmark coordinates of the latest image are (r1, r2, r3), the landmark coordinates of the past image are (t1, t2, t3), and these landmarks are associated with each other. These landmarks have the following relationship.

[Equation 1]
t1-r1 = a1 * r1 + b1 * r2 + c1 * r3 + d1
t2-r2 = a2 * r1 + b2 * r2 + c2 * r3 + d2
t3-r3 = a3 * r1 + b3 * r2 + c3 * r3 + d3

  Here, twelve coefficients ai, bi, ci, di (i = 1, 2, 3) are conversion parameters to be obtained. Since the number of equations is 3 and the number of unknowns is 12, this simultaneous equation cannot be solved uniquely as it is, but if the coordinates of four or more landmarks are known, Twelve conversion parameters can be obtained by using a known linear optimization method such as a least square method.

  If some of the landmark groups extracted in steps S1 and S2 are very close to each other, an association error may occur in step S3. If an association error occurs, in step S4, the alignment result of the landmark that is associated in error is reflected in the alignment of the entire image (volume data of the image), and the accuracy deteriorates.

  Therefore, in this embodiment, in the determination of the conversion parameter (II), the position of the image (volume data) is obtained by removing or reducing the influence of the landmark having a large residual after alignment and obtaining the conversion parameter again. The alignment accuracy is improved (S5).

  Using the result of such alignment, for the past image and the latest image, tomographic images at (almost) the same slice position are displayed in parallel (S6, S7). The interpretation doctor performs comparative interpretation of the past and latest tomographic images displayed in parallel.

[Configuration and operation]
The configuration and operation of the medical image processing apparatus according to the present invention will be described. A medical image processing apparatus 1 shown in FIG. 1 is connected to a medical image database 200 that stores image data of medical images generated by the medical image diagnostic apparatus 1000. The medical image processing apparatus 1, the medical image diagnostic apparatus 1000, and the medical image database 2000 are communicably connected via an arbitrary network such as a LAN (Local Area Network).

  The medical image diagnostic apparatus 1000 is an apparatus that generates image data of a medical image based on a signal that reflects the internal form of a subject (not shown). As a specific example of the medical image diagnostic apparatus 1000, X Line CT (Computed Tomography) apparatus, X-ray diagnostic apparatus, MRI (Magnetic Resonance Imaging) apparatus, PET (Positron Emission Computed Tomography), SPECT (Single Photon EmulationTone diagnosis apparatus) The medical image diagnostic apparatus 1000 is particularly an apparatus that can generate image data (volume data) of a three-dimensional medical image of a subject.

  Here, the volume data is image data having three-dimensional voxels as pixels. Each voxel is assigned pixel data such as luminance, shading, and color. The volume data is sometimes called voxel data. When displaying an image based on the volume data, rendering processing such as volume rendering or MIP (Maximum Intensity Projection) is performed to convert the three-dimensional volume data into a two-dimensional image for image display. Convert to data.

  The medical image database 2000 is, for example, a database of a PACS (Picture Archiving and Communication System) that is a system for managing image data of various medical images, or a database of an electronic medical record system that manages an electronic medical record to which medical images are attached. Composed.

  It is also possible to construct a network so that image data can be directly input from the medical image diagnostic apparatus 1000 to the medical image processing apparatus 1.

  As shown in FIG. 1, the medical image processing apparatus 1 includes an extraction processing unit 2, a positional relationship information generation unit 3, a landmark association unit 4, an image registration unit 5, a tomographic image generation unit 6, a display unit 7, and an operation. The unit 8 is configured to be included. Hereinafter, each of these units will be described in detail.

(Display section, operation section)
First, the display unit 7 and the operation unit 8 will be described. The display unit 7 and the operation unit 8 are included in the user interface of the medical image processing apparatus 1. The display unit 7 is a display device that displays a medical image acquired by the medical image diagnostic apparatus 1000. For example, a plurality of medical images (for example, the latest image and the past image) are displayed in parallel on the display unit 7. The display unit 7 functions as an example of the “display unit” of the present invention.

  The operation unit 8 is an operation device (input device) used for various data input operations and operation request input operations for the medical image processing apparatus 1. The user operates the operation unit 8 to input, for example, a request for acquiring image data from the medical image database 2000 or a request for displaying a desired medical image on the display unit 7.

  The acquisition request for image data is performed by, for example, inputting a patient ID of a patient to be interpreted and operating a soft key for requesting acquisition of image data.

  The display request for a desired medical image includes, for example, a display request for a tomographic image at a desired slice position (cross-sectional position) of the subject. On the interpretation screen displayed on the display unit 7, a screen area for designating a slice position is provided. In this screen area, for example, an input space for designating a slice position with a numerical value (coordinates), a soft key for designating a slice position on a scanogram, and a plurality of tomographic images in order according to the slice direction A soft key for performing an image turning function to be displayed on the screen is provided. The user designates the slice position by performing a predetermined operation on the image area. The display unit 7 displays a tomographic image at the designated slice position.

[Extraction processing unit]
The block diagram shown in FIG. 2 represents an example of the configuration of the extraction processing unit 2. The extraction processing unit 2 performs processing for setting a landmark in a medical image, and corresponds to an example of a “setting unit” according to the present invention. The extraction processing unit 2 includes an image region extraction unit 21, a seed setting unit 22, a landmark extraction unit 23, and a landmark extraction prohibition unit 24. Hereinafter, each of these units 21 to 24 will be described in detail.

(Image area extraction unit)
The image area extraction unit 21 performs processing for extracting an image area corresponding to a specific part from volume data of a medical image input from the medical image database 2000.

  In this embodiment, processing is performed on a medical image obtained by imaging a lung. This medical image includes an image region corresponding to a portion having a branch structure, such as an image region corresponding to a bronchus or an image region corresponding to a pulmonary blood vessel. The image region extracting unit 21 includes a bronchus extracting unit 211 that extracts an image region corresponding to a bronchus (sometimes referred to as a bronchial region), and an image region that is a candidate for an image region corresponding to a pulmonary blood vessel (pulmonary blood vessel). A pulmonary blood vessel extraction unit 212 that performs extraction of a candidate region).

  First, extraction of the bronchial region will be described. The pixel values (CT values, etc.) of the image region corresponding to the lung parenchyma (sometimes referred to as the lung parenchyma region) are not uniform and have values close to the pixel value of the bronchial region. Therefore, when the bronchial region is extracted using a single threshold, if the threshold is set to a relatively low value, only the image region of the thick part of the bronchus is extracted. When the threshold is set to a relatively high value, the lung parenchyma region adjacent to the bronchial region may be extracted (this is referred to as “overflow” in this specification). In order to extract the image region corresponding to the distal part of the bronchus as far as possible (the tip of the branched bronchus) without causing overflow, it is necessary to adjust the threshold value for each patient. Therefore, in this embodiment, the bronchial region is extracted using local threshold processing (local threshold processing).

  The bronchus extraction unit 211 includes a local threshold processing unit 211a and an image region separation processing unit 211b.

  The local threshold processing unit 211a extracts a bronchial region by local threshold processing. More specifically, the local threshold processing unit 211a first obtains a threshold curved surface based on medical image data (volume data or tomographic image data). For example, the threshold curved surface is obtained by adding a predetermined constant to a result obtained by performing morphological operation processing such as erosion of several voxels on the gray scale (gray scale: monochrome) image data. Can be acquired.

  Subsequently, the local threshold processing unit 211a performs threshold processing on the pixel values of the image data using the acquired threshold curved surface. That is, for each pixel, the pixel value is compared with the value of the threshold curved surface, and an image region that is a candidate for a bronchial region is extracted based on the magnitude relationship.

  The extraction result by such local threshold processing can extract the bronchi in the distal part relatively far compared to the case of using a single threshold, while extracting regions other than the bronchi (pulmonary parenchymal region, etc.). There is a possibility that.

  In order to eliminate such inconvenience, the image region separation processing unit 211b performs a process of separating the bronchial region extracted by the local threshold processing unit 211a and the region around the bronchial region by morphological calculation. As this morphological operation, for example, morphological operation processing for a binary image is applied, and erosion that acts to reduce the number of pixels of the image or processing similar thereto can be used. Thereby, a thin line portion or the like in the medical image is removed, and the bronchial region and the other region are separated.

  In the medical image obtained by the image region separation processing unit 211b, the bronchial region and other regions are separated, but the bronchial region has been thinned by the morphological calculation. It is not preferable to extract the region.

  Therefore, the bronchi extraction unit 211 extracts an image region other than the bronchial region separated by the image region separation processing unit 211b (an image region around the bronchial region), and dilation is performed on the image region other than the bronchial region. After performing a morphological operation such as (dilation) processing to make the image slightly thicker, the image region other than the bronchial region is removed from the image region extracted by the local threshold processing unit 211a. Thereby, overflow from the bronchial region to the surrounding region is also removed. Finally, a connected region of the image region corresponding to the trachea is detected, and this connected region is extracted to extract a target bronchial region.

(Pulmonary blood vessel extraction part)
Since image regions corresponding to the pulmonary artery and pulmonary vein have pixel values that are significantly different from those of the lung parenchymal region, they can be extracted to a certain extent by a normal threshold processing. However, the bronchial wall and the pulmonary blood vessels have close pixel values, and it is difficult to separate them in the hilar region, and because many thin blood vessels are extracted, so many branch points (landmarks) are obtained. Problem arises. Therefore, the pulmonary blood vessel extraction unit 212 extracts pulmonary blood vessels by the following threshold processing in order to deal with such a problem.

  The pulmonary blood vessel extraction unit 212 first extracts an image region corresponding to the lung. Next, a morphological operation such as dilation of 1 to 5 voxels is performed on the extracted image region to remove fine holes due to thin blood vessels or the like. Subsequently, for example, a morphological operation such as erosion of 5 to 30 voxels is performed, and an image corresponding to the hilar portion (a portion near the entrance where the trachea enters the lungs, a portion near the bronchus of both lungs) in the image region. Remove region. The image obtained in this way becomes an image region (pulmonary blood vessel extraction target region) from which pulmonary blood vessels are extracted.

  Furthermore, the pulmonary blood vessel extraction unit 212 performs a threshold process on the pulmonary blood vessel extraction target region, so that, for example, (−300) to (−700) HU (Hounsfield Unit; a unit of CT value) or more. Extract the image area. Next, for example, morphological operations such as erosion of 1 to 5 voxels are performed to remove thin blood vessels. Subsequently, a common region between the image region from which the thin blood vessel has been removed and the pulmonary blood vessel extraction target region is extracted. Through the processing as described above, the pulmonary blood vessel extraction unit 212 obtains image regions (pulmonary blood vessel candidate regions) that are candidates for image regions corresponding to the pulmonary blood vessels.

(Seed setting part)
The seed setting unit 22 performs a process of setting a position as a starting point (seed) for extracting a pulmonary blood vessel landmark (branch point) in the pulmonary blood vessel candidate region extracted by the pulmonary blood vessel extraction unit 212.

  More specifically, the seed setting unit 22 first performs a morphological operation such as erosion on the pulmonary blood vessel extraction target region, extracts an internal region of the pulmonary blood vessel extraction target region, and extracts the internal region to the lung It removes from the blood vessel extraction target region (calculates the difference between the pulmonary blood vessel extraction target region and the internal region), and acquires the boundary region of the pulmonary blood vessel extraction target region. Then, by extracting a common region between the boundary region and the pulmonary blood vessel candidate region, an image region (pulmonary blood vessel end region) corresponding to the end of the pulmonary blood vessel is acquired.

  The seed setting unit 22 is a seed for extracting a pulmonary blood vessel landmark at an arbitrary position in the pulmonary blood vessel end region, for example, at the center of gravity of each connected region of the image region or in the entire pulmonary blood vessel end region. Set.

  The seed setting unit 22 performs a process of setting a seed at the most root position of the trachea in the bronchial region extracted by the bronchial extraction unit 211 in the same manner as the pulmonary blood vessel.

(Landmark extraction part)
The landmark extraction unit 23 executes a process of extracting a landmark (branch point) of a part having a branch structure of the subject. The landmark extraction unit 23 performs the following two landmark extraction processes: (1) Landmark extraction process in the bronchial region extracted by the bronchial extraction unit 211; (2) Seed set by the seed setting unit 22 Based landmark extraction process in pulmonary blood vessel region. Since these two landmark extraction processes are essentially the same, here, the landmark extraction process in the bronchial region will be described in detail.

  By the way, thinning processing is widely used as a landmark extraction of a target having a branch structure. However, the thinning method has a problem in that a thin “whisker” occurs in a portion that is not a branch point due to small unevenness on the surface of the object, and the branch point cannot be extracted with high accuracy. To solve this problem, it is necessary to remove the “beard” from the thinned image. However, since the thickness of the bronchi and pulmonary blood vessels changes depending on the region, a good standard for removing the “beard” is set. It is difficult to solve this problem (for example, refer to the following document for the above contents: Patent Document 2 mentioned above; Toyofumi Saito, two others, “Using Euclidean distance conversion Sequential algorithms for thinning and thinning of 3D digital images and their properties ", IEICE Transactions D-11, Vol. J79-D-II, no. 10, pp. 1675-1685, 1996. Tatsuhiro Yonekura, 3 others, “About 1-element erasability and figure contraction in 3D digital images”, IEICE Transactions, Vol. J65-D, no. 12, pp. 1543-1549, 1982).

  Therefore, in this embodiment, a method is adopted in which a region in the target image region (bronchial region, pulmonary blood vessel region) having a branch structure is expanded stepwise and branch points are sequentially extracted. This method is described in Patent Document 2 described above. In this embodiment, processing for preventing the occurrence of overflow is performed simultaneously with the landmark extraction (landmark extraction prohibition unit 24: described later).

  First, the landmark extraction unit 23 expands the voxel layer by layer starting from the seed (root position of the trachea) set by the seed setting unit 22 in the bronchial region, and is expanded (added) at each stage. The common area (cross section) is acquired. When the common region (cross section) is acquired while expanding the voxels in this way, the connection region (or the cross section of the common region) of the common region is separated into a plurality when passing through the branch point.

  The landmark extraction unit 23 detects a branch point by detecting a change in the number of connected regions and cross sections. Then, for example, the position of the center of gravity or the center position of the voxel is calculated at the stage immediately before that stage and set as the position of the branch point.

(Landmark extraction prohibition part)
The landmark extraction prohibition unit 24 executes processing for preventing occurrence of overflow in parallel with the landmark extraction processing by the landmark extraction unit 23.

  It should be noted that processing for removing overflow is also performed when the bronchial region is extracted by the bronchial extraction unit 211. However, it is desirable to remove the overflow that was missed during the bronchus extraction by performing again the process of preventing the occurrence of overflow during the landmark extraction.

  The landmark extraction prohibition unit 24 is configured such that, at each stage where the landmark extraction unit 23 expands the voxel, the extracted common area or cross section has a predetermined size than the common area or cross section extracted in the previous stage. Judge whether it has become larger. This process calculates the size of the common region (or its cross section) at each stage, compares the size at that stage with the size at the previous stage, and determines whether the change in size is greater than or equal to a predetermined value. This is done by judging.

  When it is determined that the change in size is less than the predetermined value, the landmark extraction prohibition unit 24 instructs the landmark extraction unit 23 to continue the landmark extraction process for the trachea or blood vessel branch. On the other hand, if it is determined that the change in size is greater than or equal to a predetermined value, the end of landmark extraction processing at and after that stage is instructed for the trachea or blood vessel branch. As a result, the processing by the landmark extraction unit 23 ends up to the stage immediately before the stage. By such processing, it is possible to detect a portion where the common area or the cross section suddenly increases (that is, a portion where overflow occurs) and prohibit the subsequent landmark extraction processing.

  Note that the processing by the landmark extraction prohibition unit 24 need not be executed at each stage of voxel expansion. For example, when a branch point is searched for in a region where the trachea or blood vessel has a certain thickness, such as when a voxel is expanded by several stages from the seed, it is not necessary to perform the landmark extraction prohibition process. For this purpose, for example, the size of the common region or the cross section may be calculated, and the size comparison process with the previous stage may be performed only when the size is equal to or smaller than a predetermined size.

  As described above, it is one of the features of this embodiment that the landmark extraction process and the overflow removal process are executed in parallel. This parallel processing is executed as follows, for example. The landmark (branch point) is extracted by the above-described method in which the voxel is expanded step by step starting from the seed. At this time, in each expansion stage, identification information (label) for distinguishing the layer is given to each voxel of the layer added in the stage. In addition, a flag is assigned to each voxel of the common region or cross section, or voxels representing the common region or cross section, and information (size flag) indicating the volume of the common region or the area (size) of the cross section. Give. By referring to such labels and flags, the landmark extraction process and the overflow removal process can be performed in parallel.

[Positional information generator]
The positional relationship information generation unit 3 generates positional relationship information indicating the positional relationship of the landmarks extracted by the landmark extraction unit 23, and functions as an example of the “generation unit” of the present invention.

  The positional relationship information is information used for positioning the reference image used for comparative interpretation and the target image, and is information for associating landmarks of both images. As this positional relationship information, for example, a distance list (distance information) including information about the distance between two landmarks, or an angle list (including information about angles of polygons formed by three (or more) landmarks) Angle information). Here, the case where the distance list is used will be described, and the case where the angle list is used will be described later in the [Modification] section.

  In order to generate a distance list, the positional relationship information generation unit 3 forms a plurality of pairs of two landmarks based on the landmarks extracted by the landmark extraction unit 23, and each of the pairs Calculate the distance between two landmarks. Then, a distance list is generated by associating information for identifying each landmark (or information for identifying each pair) with the distance between each pair.

  The positional relationship information generation unit 3 generates a distance list for each of the reference image and the target image. The distance list of the reference image includes a distance Dr (i1, i2) between two landmarks ri1 and ri2 included in each of a plurality of pairs formed based on the landmark extracted from the reference image. ) Is collected list information. Further, the distance list of the target image includes a distance Dt (j1) between two landmarks tj1 and tj2 included in each of a plurality of pairs formed based on the landmark extracted from the target image. , J2). Each distance list is generated, for example, by arranging landmark pairs in order from the smallest distance between landmarks.

  Here, as for the distance between landmarks, a predetermined number of landmarks included in the extracted landmarks are selected rather than calculating all combinations (pairs) that select two from all the extracted landmarks. It is desirable to use it for processing. Although details will be described later, in this embodiment, new landmarks are sequentially added to each pair of two landmarks, and three sets of landmarks, four sets,... The reference image and the target image are associated with each other while increasing the points. Therefore, the number of operations required for the process of increasing the number of points is approximately the product of the number of landmark pairs set first and the number of landmarks used for the process. For example, when about 100 pairs are considered for about 100 landmarks, about 10,000 calculations are performed. Note that the number of operations decreases as new landmarks are added. Therefore, by selecting a predetermined number (for example, 100) of landmarks from the extracted landmarks and selecting them for processing, it is possible to perform the target processing in a realistic processing time.

  Further, in the normal image registration method, even if the subject part is deformed, it is not so large. Even if the deformation looks large, it can often be regarded as a small deformation locally. In addition, changes in the size of a subject's body itself are generally greatly expanded or reduced within a few months or years, except in special cases such as when the subject is a child. It is not considered.

  On the other hand, in image alignment, a large displacement of parallel movement or rotation may be required. When searching for the correspondence between points using the coordinate values of the points (landmarks) as in the past, it is necessary to repeat translation and rotation many times in order to evaluate the correspondence. Invited to increase.

  In this embodiment, by using a distance between two points or an angle formed by a plurality of points, the correspondence between the landmarks of the reference image and the target image is evaluated without considering parallel movement or rotation, thereby processing Reduce time. This utilizes the fact that the distance and angle are not changed by translation or rotation. Note that the angle does not change with respect to a change in size (enlargement or reduction), and therefore can be suitably applied even in a case where the size of the site of the subject changes.

[Landmark mapping part]
The landmark association unit 4 performs processing for associating the landmarks of the reference image and the target image with each other based on the positional relationship information generated by the positional relationship information generation unit 3. It functions as an example.

  As shown in FIG. 3, the landmark association unit 4 includes a line segment association unit 41, a polygon association unit 42, a landmark selection unit 43, an evaluation value setting unit 44, and an association evaluation processing unit 45. Is provided. The processes executed by these units 41 to 45 will be described later.

  The landmark of the reference image is represented by ri (i = 1 to M), and the landmark of the target image is represented by tj (j = 1 to N). These are landmarks that are (optionally) subjected to processing (described above). Note that the number of landmarks (M, N) in each image may or may not be equal.

  The landmark associating unit 4 calculates an evaluation value fij indicating the degree (possibility) of correspondence between the landmarks ri and tj (described later), and performs, for example, the following evaluation processing (stages 1 to 3). By executing, it is determined whether or not these landmarks ri and tj correspond. This evaluation process is executed by the association evaluation processing unit 45.

(Stage 1)
i = i0 is fixed. For j = 1 to N, the total sum Si0 of the evaluation values fi0j is calculated. It is determined whether or not there is j = J0 having an evaluation value fi0j0 that exceeds the product of a preset constant Cf and the total sum Si0 (that is, whether or not j0 exists such that fi0j0> Cf × Si0). To do).

(Stage 2)
j = j0 is fixed. For i = 1 to M, the total sum Sj0 of the evaluation values fij0 is calculated. It is determined whether there is i = i1 having an evaluation value fi1j0 exceeding the product of the constant Cf and the sum Sj0 (that is, it is determined whether i1 exists such that fi1j0> Cf × Sj0).

(Stage 3)
Whether or not i0 and i1 are equal when it is determined in step 1 that j0 satisfying fi0j0> Cf × Si0 exists and in step 2 it is determined that i1 satisfying fi1j0> Cf × Sj0 exists to decide. When i0 = i1, it is determined that the landmark ri0 of the reference image corresponds to the landmark tj0 of the target image, and these are associated with each other. On the other hand, when i0 ≠ i1, it is determined that these landmarks do not correspond.

  For example, by setting constant Cf> 0.5, it is possible to exclude a case where two or more landmarks tj are associated with one landmark ri and vice versa.

  It is also possible to eliminate correspondence duplication by performing the following processing. First, among the landmarks tj corresponding to one landmark ri0, the one having the maximum evaluation value fi0j is defined as tj0. Similarly, the landmark tj corresponding to one landmark tj0 having the largest evaluation value fij0 is defined as ri1. Then, it is determined whether i0 = i1. When i0 = i1, it is determined that the landmark i0 of the reference image corresponds to the landmark j0 of the target image.

  As described above, if the evaluation value fij is configured, the landmarks can be easily associated with each other. Hereinafter, an example of a method for configuring the evaluation value fij will be described. This evaluation value construction process goes through the steps of landmark pair (line segment) association processing → polygon association processing → landmark selection processing → evaluation value setting processing.

(Landmark pair matching process)
The process of associating landmark pairs (line segments) is executed by the line segment association unit 41 shown in FIG. The line segment association unit 41 compares the distance between the landmark pair of the reference image and the distance between the landmark pair of the target image based on the distance list generated by the positional relationship information generation unit 3. , Calculate the difference between those distances. Then, a pair whose distance difference is smaller than a predetermined value is provisionally associated.

  For example, the distance between the landmark pair <ri1, ri2> shown in the distance list of the reference image is Dr (ri1, ri2), and the distance between the landmark pair <tj1, tj2> shown in the distance list of the target image Is Dt (tj1, tj2), the line segment association unit 41 determines the difference ΔD = | Dr (ri1, ri2) −Dt (tj1, tj2) between the distances Dr (ri1, ri2) and Dt (tj1, tj2). ) | Is calculated. Then, it is determined whether or not the distance difference ΔD is equal to or smaller than a predetermined constant Cd2. When ΔD <Cd2, the line segment association unit 41 provisionally associates the landmark pair <ri1, ri2> of the reference image with the landmark pair <tj1, tj2> of the target image.

  Here, a pair formed by two landmarks is expressed by <., ..> (a group of three or more landmarks described later is also expressed in a similar manner).

  Note that the association processing described above is performed on the distance list of the target image and the predetermined number of pairs from the smallest distance Dr (ri1, ri2) among the pairs (line segments) included in the distance list of the reference image. It is desirable to execute only a predetermined number of pairs (in order to avoid an increase in processing time) from among the included pairs (line segments) having a small distance Dt (tj1, tj2). In this association process, duplicate association may be performed. For example, one or more landmark pairs of the target image can be associated with one landmark pair of the reference image.

(Polygon matching process)
Polygon (a set of three or more landmarks) is associated by the polygon associating unit 42. The polygon associating unit 42 adds three landmarks that are not included in each of the reference image landmark pair and the target image landmark pair associated with each other by the line segment associating unit 41. Form a (provisional) set of (and above) landmarks.

  This process will be described more specifically. The polygon associating unit 42 first determines the distance between each landmark ri1, ri2 of the reference image landmark pair <ri1, ri2> and another landmark ri3 (i3 ≠ i1, i2) of the reference image. Dr (ri1, ri3) and Dr (ri2, ri3) are calculated. Similarly, a distance Dt (tj1, tj3) between each landmark tj1, tj2 of the landmark pair <tj1, tj2> of the target image and another landmark tj3 (j3 ≠ j1, j2) of the target image, Dt (tj2, tj3) is calculated.

  Next, the polygon associating unit 42 determines the lengths Dr (ri1, ri2), Dr () of the three sides of the triangle formed by the (provisional) set of three landmarks <ri1, ri2, ri3> of the reference image. ri1, ri3), Dr (ri2, ri3) and a (provisional) set of three landmarks of the target image <tj1, tj2, tj3>, the lengths Dt (ti1, ti2) of the three sides of the triangle It is determined whether the distance difference between Dt (tj1, tj3) and Dt (tj2, tj3) is equal to or smaller than a predetermined value Cd3.

  This processing is performed by, for example, distances Dr (ri1, ri2), Dr (ri1, ri3), Dr (ri2, ri3) on the reference image side, and distances Dt (ti1, ti2), Dt (tj1, tj3) on the target image side. ) And Dt (tj2, tj3) are performed by determining whether or not the distance difference of all combinations is equal to or less than a predetermined value Cd3. That is, it is determined whether or not the following equation holds.

[Equation 2]
max {| Dr (ri1, ri2) -Dt (ti1, ti2) |,
| Dr (ri1, ri2) −Dt (tj1, tj3) |,
| Dr (ri1, ri2) −Dt (tj2, tj3) |,
| Dr (ri1, ri3) -Dt (ti1, ti2) |,
| Dr (ri1, ri3) -Dt (tj1, tj3) |,
| Dr (ri1, ri3) −Dt (tj2, tj3) |,
| Dr (ri2, ri3) -Dt (ti1, ti2) |,
| Dr (ri2, ri3) −Dt (tj1, tj3) |,
| Dr (ri2, ri3) −Dt (tj2, tj3) |} ≦ Cd3

  Here, max {·} represents an operation that employs the maximum value of a plurality of values in parentheses. The fact that this equation is satisfied indicates that the difference between the length of an arbitrary side of the triangle on the reference image side and the length of an arbitrary side of the triangle on the target image side is equal to or less than a predetermined value Cd3. This is because the distance between any two of the three landmark sets <ri1, ri2, ri3> on the reference image side and the three landmark sets <tj1, tj2, tj3> on the target image side The error with the distance between any two of them is equivalent to being equal to or less than a predetermined value Cd3.

  The polygon associating unit 42 searches for a pair of such a reference image landmark set <ri1, ri2, ri3> and a target image landmark set <tj1, tj2, tj3>. Corresponding pairs of landmarks (triangles).

  Similarly, the polygon association unit 42 adds other landmarks ri4 and tj4 to the associated triangle pairs <ri1, ri2, ri3>, <tj1, tj2, tj3>, respectively, A set of landmarks (rectangles) <ri1, ri2, ri3, ri4>, <tj1, tj2, tj3, tj4> are formed. Further, it is determined whether the difference between the length of an arbitrary side of the square on the reference image side and the length of the arbitrary side of the square on the target image side is equal to or less than a predetermined value Cd4. Then, a rectangular pair that satisfies this criterion is searched and associated with each other.

  Such a process is repeated until a pair of L landmarks (L square) is associated. Here, L is an arbitrary integer of 3 or more. However, considering the processing time, it is not desirable to apply a large value as L. In practice, L = 3, 4, and 5 are often sufficient.

(Landmark selection process)
The landmark selection unit 43 uses a pair of landmarks (a pair of landmarks) used for calculating the evaluation value fij based on the L-angle pair on the reference image side and the target image side associated by the polygon association unit 42. ) Is selected. For this purpose, the landmark selection unit 43 executes the following processing (steps 1 to 4), for example, and selects a landmark suitable for calculating the evaluation value fij.

(Stage 1)
First, the length of a line segment connecting two vertices (that is, the distance between two landmarks) is calculated for each of the L-angles on the reference image side and the target image side. This calculation process is performed for all combinations of L landmarks forming an L-gon (all combinations when selecting two from L).

  As a specific example of this processing, a case will be described in which a square <ri1, ri2, ri3, ri4> on the reference image side is associated with a square <tj1, tj2, tj3, tj4> on the target image side. The landmark selection unit 43 sets the distances Dr (ri1, ri2), Dr (ri1, ri3), Dr (ri1, ri4), Dr (ri2, ri3) for the square <ri1, ri2, ri3, ri4> on the reference image side. ), Dr (ri2, ri4), and Dr (ri3, ri4), respectively. Further, for the squares <tj1, tj2, tj3, tj4> on the target image side, distances Dt (ti1, ti2), Dt (tj1, tj3), Dt (tj1, tj4), Dt (tj2, tj3), Dt (tj2) , Tj4) and Dt (tj3, tj4) are calculated. Note that distances that have already been calculated, for example, Dr (ri1, ri2), Dr (ri1, ri3), Dr (ri2, ri3), Dt (ti1, ti2), Dt (tj1, tj3), Dt (tj2, tj3) ) And the like need not be calculated again here, and the previous calculation result may be referred to.

(Stage 2)
Next, the distance (the length of the line segment) between the landmarks forming the L-angle on the reference image side obtained in step 1 and the distance (the length of the line segment) between the landmarks forming the L-angle on the target image side. It is determined whether or not the difference exceeds the predetermined value Cdf. This process is executed for all combinations of the line segment on the reference image side and the line segment on the target image side, for example.

  One point is added to each landmark at both ends of the line segment for which the distance is determined to exceed the predetermined value Cdf. In this process, for example, a counter is provided for each of L landmarks forming an L-shaped square, and when the length of a line segment having the landmark as an end point exceeds a predetermined value Cdf, the value of the counter is incremented by +1. Can be done.

  By such processing, a large score is given to a landmark having a poor correspondence between the landmark of the reference image and the landmark of the target image with the landmark of the other image.

(Stage 3)
Subsequently, a process of excluding some landmarks having a large number of points given in Step 2 is performed. At this time, a point threshold value can be set in advance, and landmarks larger than this threshold value can be excluded. Alternatively, the number of landmarks to be excluded may be set in advance, and the number of landmarks may be excluded in order from the landmark having the highest score. As a result, only the landmark having a good correspondence between the landmark of the reference image and the landmark of the target image with the landmark of the other image remains.

(Stage 4)
Next, the same processing as in steps 2 and 3 is performed on the landmarks remaining after the processing in step 3. At this time, the predetermined value Cdf ′ serving as a criterion for exclusion is set to a value smaller than the predetermined value Cdf in step 2. Such a process is repeated until, for example, all remaining landmarks have zero scores.

  Thereby, the distance difference between an arbitrary line segment formed by all landmarks remaining on the reference image side and an arbitrary line segment formed by all landmarks remaining on the target image side becomes a predetermined value or less. That is, only a landmark having a very good correspondence with the landmark of the other image is left.

  Note that it is not necessary to perform the landmark exclusion process until the number of points of all landmarks becomes 0, and it is generally sufficient to leave only the landmarks whose assigned points are within a predetermined range. Here, it is desirable that the predetermined range of the score has 0 as the lower limit, and it is not desirable to use a large value as the upper limit (a landmark having a poor correspondence with the landmark in the other image remains. To end up).

  The landmark selection unit 43 selects a landmark for calculating the evaluation value fij through the above processing (steps 1 to 4).

(Evaluation value setting process)
Based on the landmark selected by the landmark selection unit 43, the evaluation value setting unit 44 calculates an evaluation value fij indicating the degree (possibility) that the landmark of the reference image corresponds to the landmark of the target image. Perform the setting process.

  As the evaluation value fij, for example, the number of landmarks on the reference image side and the target image side remaining by the exclusion process of the landmark selection unit 43 can be set.

  The association evaluation processing unit 45 uses the evaluation value fij set by the evaluation value setting unit 44 to check whether each landmark ri of the reference image and each landmark tj of the target image correspond to each other. (A specific example of this determination process has been described above).

  Note that the landmark pairs selected by the landmark selection unit 43 may be associated with each other as they are. That is, the setting of the evaluation value fij and the evaluation process of the association are not performed, and the pair of the landmark of the reference image and the landmark of the target image remaining after the exclusion process by the landmark selection unit 43 is finally associated The subsequent processing may be performed as a landmark to be performed.

(Image alignment part)
The image alignment unit 5 performs alignment processing of the two images (volume data thereof) based on the reference image and the landmark of the target image associated with each other by the landmark association unit 4. This corresponds to an example of the “positioning means”.

  The image alignment unit 5 includes a conversion parameter calculation unit 51 and a conversion parameter correction unit 52. The conversion parameter calculation unit 51 is a parameter for coordinate conversion for aligning one position of the reference image and the target image (volume data thereof) with the other position based on the landmark associated by the landmark association unit 4. And functions as an example of the “conversion parameter calculation means” of the present invention.

  The conversion parameter correction unit 52 performs processing for correcting the calculated conversion parameter, and functions as an example of the “conversion parameter correction unit” of the present invention. Further, the conversion parameter correction unit 52 selects several landmarks from the landmarks used in the calculation processing based on the conversion parameter calculated by the conversion parameter calculation unit 51, and calculates a new landmark. Acts like

(Conversion parameter calculator)
In the reference image (volume data thereof), a three-dimensional coordinate system (ri, rj, rk) is defined in advance (this three-dimensional coordinate system may be referred to as a “reference coordinate system”). Further, a three-dimensional coordinate system (ti, tj, tk) is defined in advance for the target image (volume data thereof) (this three-dimensional coordinate system may be referred to as a “target coordinate system”). The conversion parameter calculation unit 51 calculates a coordinate conversion parameter (coordinate conversion formula) for converting the coordinates (ri, rj, rk) of the reference coordinate system into the coordinates (ti, tj, tk) of the target coordinate system.

  Note that parameters for converting the target coordinate system to the reference coordinate system may be calculated, and parameters for converting both the reference coordinate system and the target coordinate system to another three-dimensional coordinate system are calculated. You may do it. That is, the conversion parameter calculation unit 51 calculates a parameter that relatively converts the reference coordinate system and the target coordinate system and shifts to the same three-dimensional coordinate system.

  The conversion parameter calculation unit 51 calculates the conversion parameter using, for example, a linear optimization method similar to the conventional one.

  As an example, first, the difference ti-ri of the coordinate ti of the target coordinate system with respect to the coordinate ri of the reference coordinate system is expressed by a parametric function (parametric function), for example, a cubic polynomial (each coefficient is undetermined). The difference tj−rj of the coordinate tj of the target coordinate system with respect to the coordinate rj of the target coordinate system is expressed by a cubic polynomial (each coefficient is undetermined), and the difference tk−rk of the coordinate t × of the target coordinate system with respect to the coordinate rk of the reference coordinate system is 3 Represented by a second-order polynomial (each coefficient is undetermined). By determining the coefficients (parameters) of these cubic polynomials, a coordinate conversion formula (third order) for converting the coordinates (ri, rj, rk) of the reference coordinate system into the coordinates (ti, tj, tk) of the target coordinate system. (Represented by a polynomial).

  An example of a method for calculating the coefficients of these cubic polynomials will be described. For the landmark r of the reference image and the landmark t of the target image associated by the landmark associating unit 4, the coordinates of the landmark r in the reference coordinate system are (ri, rj, rk), and the target of the landmark t. The coordinates in the coordinate system are (ti, tj, tk).

  At this time, the transformation of the coordinates (ri, rj, rk) of the landmark r into the coordinates (ti, tj, tk) of the landmark t satisfies the above-mentioned three cubic polynomials. Such a relationship is established for each of the landmarks associated with the landmark associating unit 4 (assuming that P landmarks are associated with each other), and therefore, a 3P-order simultaneous linear equation can be obtained by coupling them together. When this simultaneous equation is expressed by a matrix, the transformation matrix is a 3P × 60th order matrix.

  As is well known, the minimum norm solution of the 3P-order simultaneous linear equations can be expressed using the transformation matrix. In addition, each unknown is an independent variable, and assuming the standard deviation thereof, the minimum norm solution of the above simultaneous equations can be obtained by assuming the standard deviation of the measurement noise. Thereby, a coordinate conversion parameter (coordinate conversion formula) for aligning one position of the reference image and the target image (volume data thereof) with the other position is obtained.

(Conversion parameter correction unit)
Some degree of error intervenes in the landmark associating process by the landmark associating unit 4 and the calculating process by the conversion parameter calculating unit 51. For example, when different landmarks exist at very close positions, there is a possibility that an incorrect landmark is associated with the landmark. In such a case, the conversion parameter calculation unit 51 described above displays the wrongly associated landmark and other correctly associated landmarks even though the wrong landmark is associated. Since the conversion parameters are calculated with the same treatment, the alignment accuracy between the reference image and the target image is reflected in the alignment result of the reference image and the target image, and the alignment accuracy deteriorates.

  Therefore, in this embodiment, in order to reduce or eliminate such an adverse effect, the conversion parameter correction unit 52 operates so as to perform correction and recalculation of the conversion parameter. As a result, it is possible to realize high-accuracy image alignment that effectively uses information of landmarks that are correctly associated.

  Hereinafter, a specific example of such processing by the conversion parameter correction unit 52 will be described. Three specific examples will be described below, but the medical image processing apparatus according to the present invention only needs to include at least one of these specific examples. When two or more are provided, the user may be configured to select a desired process as appropriate, or the apparatus may be configured to automatically select and apply one.

(Specific example 1)
The first specific example calculates a residual when the coordinate transformation of the landmark is performed based on the conversion parameter obtained by the conversion parameter calculation unit 51, and excludes a landmark whose residual is a predetermined value or more. Then, the conversion parameter is calculated again using only the remaining landmarks.

  An example of this process will be described in more detail. The conversion parameter correction unit 52 uses the conversion parameters obtained by the conversion parameter calculation unit 51 to convert the coordinates (ri, rj, rk) of each landmark r of the reference image into the coordinates (ti ′, tj ′) of the target coordinate system. , Tk ′).

  Next, the coordinates (ti ′, tj ′, tk ′) obtained by this conversion and the coordinates (ti, tj, tk) of the landmark t of the target image associated with the landmark r in the target coordinate system. The difference Δ (r, t) is calculated. This difference Δ (r, t) is calculated by, for example, calculating the equation Δ (r, t) = {(ti′−ti) ^ 2 + (tj′−tj) ^ 2 + (tk′−tk) ^ 2} ^ (1 / 2). The arithmetic expression of the difference Δ (r, t) is not limited to this, and for example, the difference Δ (r, t) = max {ti′−ti, tj′−tj, tk′−tk}. It is possible to use an arbitrary arithmetic expression that represents the residual of coordinate conversion by a conversion parameter.

  Subsequently, for each of the landmark pairs r and t, the difference Δ (r, t) is compared with a predetermined value δ1, and a landmark pair satisfying the condition Δ (r, t) <δ1 is searched. Then, a new conversion parameter is calculated using only the landmark pair that satisfies the condition. This calculation process is performed by the same process as the conversion parameter calculation unit 51.

  It is desirable to repeat such processing while sequentially reducing the threshold value in the above conditions. That is, threshold values δ1> δ2> δ3>... Are set in advance, the threshold value δ1 is used as described above in the first process, the threshold value δ2 is used in the second process, and the third In the second process, the threshold value δ3 is used, and a new conversion parameter is repeatedly obtained as follows. Thereby, a conversion parameter capable of performing alignment with higher accuracy can be acquired.

  In the above processing, only a pair of landmarks having a large residual is excluded. However, a distance list is created again using coordinate coordinates after coordinate conversion, and landmarks are associated again. It is desirable to remove the landmarks that are improperly associated with the results (those with large errors). Thereby, a conversion parameter capable of performing alignment with higher accuracy can be acquired. In the landmark association process again, it is desirable to set the values Cd and Cf, which are the determination criteria for association, to smaller values to increase the accuracy of association.

(Specific example 2)
The second specific example uses the weighting function (objective function) corresponding to the difference Δ (r, t) obtained in the same manner as the specific example 1 to correct the conversion parameter by weighting each landmark pair. To do.

  FIG. 5 shows an example of this weighting function. In the weighting function F shown in the figure, when the value of the difference Δ (r, t) is small (when the residual due to the conversion parameter is small), the value of the gradient (slope) F′Δ (r, t) is large. The nonlinear function is such that the value of the gradient F′Δ (r, t) gradually decreases as the value of the difference Δ (r, t) increases. The weighting function F has, for example, a gradient F′Δ (r, t) = 1 when the value of the difference Δ (r, t) is small.

  The conversion parameter correction unit 52 corrects the conversion parameter by a non-linear optimization method based on an iterative solution based on the weighting function F. Although this process is based on an iterative solution method, since the weighting function F is a simple function, the calculation can be performed at a high speed and the processing time is not increased.

  The weighting function F has a large gradient F′Δ (r, t) when the difference Δ (r, t) is small, and a gradient F′Δ (r, t) when the difference Δ (r, t) is large. Therefore, in the optimization process, the influence of the landmark pair having a small difference Δ (r, t) on the alignment increases, and the influence of the landmark pair having a large difference Δ (r, t). Becomes smaller. In particular, when the number of landmark pairs having a large difference Δ (r, t) is sufficiently smaller than the number of landmark pairs having a small difference Δ (r, t), the difference Δ (r, t) is reduced. The effects of large landmark pairs are almost ignored. As a result, even if there is a pair of landmarks having a large difference Δ (r, t), the influence on the alignment result is reduced (ignored), so that highly accurate alignment is possible.

  Note that the weighting function used in this specific example is not limited to the above-described one, and an arbitrary function can be used. However, a function that increases the influence on alignment by a pair of landmarks having a small difference Δ (r, t) and reduces the influence on alignment by a pair of landmarks having a large difference Δ (r, t). It is necessary to adopt.

(Specific example 3)
The third specific example is a modification of specific example 2. This specific example is characterized in that the weighting function (objective function) is changed in the iterative process of the nonlinear optimization method based on the iterative solution described in specific example 2.

  FIG. 6 shows an example of a weighting function used in this specific example. In the figure, two weighting functions F1 and F2 are shown. The weighting function F1 is used in the initial iteration process of the nonlinear optimization method by the iterative solution, and the weighting function F2 is used in the latter iteration process. The weighting function F2 is set so that the gradient F′Δ (r, t) becomes gradually smaller at the value of the difference Δ (r, t) smaller than the weighting function F1.

  The conversion parameter correction unit 52 performs processing using the weighting function F1 until the number of iterations reaches a predetermined number of times, and switches to the weighting function F2 when the predetermined number of times is reached. Note that the residual may be calculated for each iteration, and the weighting function F2 may be switched when the residual becomes a predetermined value or less.

  By applying such a configuration, the following merits can be enjoyed. The weighting function F2 acts so as to reduce the influence on the alignment even with respect to the pair of landmarks whose difference Δ (r, t) is somewhat smaller than the weighting function F1, so that the weighting function F2 from the beginning of the iteration is used. It seems that F2 may be used. However, if the weighting function F2 is used from the beginning, the number of landmark pairs that have a large influence on alignment (that is, landmark pairs with a very small difference Δ (r, t)) is extremely small, and the position There is a possibility that the error of the alignment result may be increased.

  Therefore, in the initial iteration, the transformation parameter is suitably corrected using a pair of landmarks (a relatively large number) having a certain difference Δ (r, t) using the weighting function F1. Then, by changing to the weighting function F2 when the number of iterations progresses, a conversion parameter capable of highly accurate alignment is obtained.

  In the example shown in FIG. 6, two weighting functions F1 and F2 are switched and used, but the number of weighting functions to be used may be three or more. It is also possible to apply a weighting function that changes continuously according to the number of iterations.

[Tomographic image generator]
The tomographic image generation unit 6 performs processing for generating image data of a tomographic image at a cross-section at the same position by alignment based on the reference image and target image volume data aligned by the image alignment unit 5. This is performed and corresponds to an example of the “tomographic image generating means” of the present invention. The generation processing of the tomographic image data based on the volume data can be performed by, for example, MPR (Multi-Planar Reconstruction).

  The processing of the tomographic image generation unit 6 will be described more specifically. The image alignment unit 5 aligns the reference image and the target image by performing the above-described processing. For example, the coordinates of the volume data of the reference image expressed in the reference coordinate system and the coordinates of the volume data of the target image expressed in the target coordinate system are associated by coordinate conversion using conversion parameters.

  When the user operates the operation unit 8 to specify the slice position of the reference image (described above), the tomographic image generation unit 6 sets the image data of the tomographic image of the reference image at this slice position based on the volume data of the reference image. Generate. Further, the tomographic image generation unit 6 calculates the slice position of the target image corresponding to the designated slice position based on the alignment result (conversion parameter) by the image alignment unit 5. Then, based on the volume data of the target image, image data of the tomographic image at the calculated slice position is generated. Image data generated for each of the reference image and the target image is sent to the display unit 7.

  Note that when the medical image processing apparatus 1 stores image data of a tomographic image at a certain slice position in advance, when the slice position is designated, the image data of the tomographic image corresponding to the designated slice position is obtained. Instead of generating, it is possible to read out the image data of the tomographic image corresponding to the slice position and send it to the display unit 7.

  The display unit 7 receives the image data from the tomographic image generation unit 6 and displays the tomographic image of the reference image and the tomographic image of the target image in parallel. As described above, these tomographic images are based on the alignment result by the image alignment unit 5 and are (substantially) tomographic images at the same slice position.

[Hardware configuration]
A hardware configuration of the medical image processing apparatus 1 will be described. The medical image processing apparatus 1 includes a general computer device. The medical image processing apparatus 1 includes a microprocessor, a volatile storage device, a nonvolatile storage device, a user interface, a communication interface, and the like.

  The microprocessor includes a CPU (Central Processing Unit), an MPU (Micro Processing Unit), and the like. The volatile storage device includes a RAM (Random Access Memory) or the like.

  The nonvolatile storage device is configured by a ROM (Read Only Memory), a hard disk drive, or the like. The hard disk drive stores in advance a computer program that causes the computer device to execute the processing described above. The hard disk drive stores medical image data and patient-related data (such as medical chart information).

  The user interface has a display device and an operation device (input device). The display device includes an arbitrary display device such as an LCD (Liquid Crystal Display) or a CRT (Cathode Ray Tube) display. The operation device is configured by an arbitrary device such as a mouse, a keyboard, a trackball, or a control panel. Note that a display device and an operation device can be integrally configured by applying a touch panel type LCD or the like.

  The communication interface includes a network adapter such as a LAN card. It is also possible to configure so that data communication can be performed via the Internet by installing a modem.

  The extraction processing unit 2, the positional relationship information generation unit 3, the landmark association unit 4, the image registration unit 5, and the tomographic image generation unit 6 shown in FIG. 1 and the like each include a microprocessor that executes the computer program. Consists of.

  The display unit 7 includes a display device, and the operation unit 8 includes an operation device (input device).

  The overall control of the medical image processing apparatus 1 for executing the above-described processing is performed by the execution microprocessor executing the computer program. For example, the microprocessor performs control for displaying various screens and images on the display device (display unit 7). Further, the microprocessor controls each part of the apparatus based on an operation signal (signal output corresponding to the operation) from the operation device (operation unit 8).

[Function and effect]
According to the medical image processing apparatus 1 described above, the following actions and effects can be mainly achieved.

  First, the medical image processing apparatus 1 is configured to operate as follows: (1) The extraction processing unit 2 displays landmarks in an image for each of the volume data of the reference image and the target image. (2) The positional relationship information generation unit 3 generates a distance list (positional relationship information) indicating the positional relationship of the extracted landmarks for each of these two volume data; (3) Landmark correspondence The attaching unit 4 excludes some of the landmarks of the reference image and the target image based on the generated positional relationship information, and associates the remaining landmarks of the reference image and the target image with each other; (4 ) Based on the landmark associated with the landmark associating unit 4 by the image alignment unit 5 Aligning the two volume data.

  This image alignment process employs a registration method using landmarks instead of the conventional iterative registration method, which takes a long time to process, and analyzes the position of the landmarks themselves to associate the landmarks with each other. Instead, some landmarks are excluded based on the positional relationship of landmarks shown in the distance list (distance between landmarks), and the remaining landmarks are associated with each other. Here, the distance between the landmarks is an invariable amount due to translation and rotation of the image.

  Here, processing executed by the medical image processing apparatus 1 will be described with reference to FIGS. 8 and 9. Here, an alignment process between the past image GP shown in FIG. 8A and the latest image GN shown in FIG. 8B will be described.

  First, the medical image processing apparatus 1 extracts the landmarks of the past image GP and the latest image GN, respectively. Numbers 1 to 7 surrounded by a circle shown in FIG. 8A represent seven landmarks extracted from the past image GP. Also, numerals 1 to 8 surrounded by a square shown in FIG. 8B represent eight landmarks extracted from the latest image GN.

  Next, the medical image processing apparatus 1 generates a distance list representing the positional relationship between the landmarks 1 to 7 of the past image GP and the landmarks 1 to 8 of the latest image GN.

  Subsequently, the medical image processing apparatus 1 excludes landmarks based on the distance list. In this example, the landmark 7 of the past image GP is excluded, and the landmarks 7 and 8 of the latest image GN are excluded. Thereby, the remaining landmarks are the landmarks 1 to 6 of the past image GP and the landmarks 1 to 6 of the latest image GN.

  FIG. 9 shows the positional relationship between the remaining landmarks of both the images GP and GN. In this example, the landmarks 1, 2,..., 6 of the past image GP and the landmarks 1, 2,.

  The medical image processing apparatus 1 determines the conversion parameters so that the positions of the landmarks 1 to 6 of the past image GP are matched with the positions of the landmarks 1 to 6 of the corresponding latest image GN. , GN is aligned. Note that the conversion parameters may be determined so that the positions of the landmarks 1 to 6 of the latest image GN are aligned with the positions of the landmarks 1 to 6 of the latest image GN.

  In this way, it is possible to reduce the processing time by performing association after removing some landmarks using the distance list. In particular, by excluding landmarks, it is possible to perform image alignment by simple threshold processing, and the processing time can be shortened. Further, since it is not necessary to translate or rotate the landmark as in the prior art, the processing time can be shortened as compared with the prior art. Further, as described above, the image alignment accuracy is also good. Further, even if there is an error in landmark extraction, the error is removed at the stage of landmark association, so that there is an effect that image alignment can be performed with high accuracy and speed.

  The medical image processing apparatus 1 is configured to operate as follows: (1) The extraction processing unit 2 extracts landmarks in the image for each of the volume data of the reference image and the target image. (2) The landmark associating unit 4 associates the landmark extracted by the extraction processing unit 2 between these two medical images; (3) The conversion parameter calculating unit 51 is configured as a landmark associating unit. 4 is used to calculate a coordinate conversion parameter for matching one position of these two volume data to the other position based on the landmark associated with 4; (4) the conversion parameter correction unit 52 is calculated The distance between the associated landmarks in the two volume data aligned based on the parameters Calculated, and corrects the parameter in response to the computed distance; is (5) image registration unit 5, based on the corrected parameters, to position the two volume data.

  According to such an image alignment process, an alignment method using landmarks is employed to avoid an increase in processing time, and the parameters are corrected based on the distance between the landmarks. Since the volume data of the reference image and the target image is aligned using, image alignment can be performed with good accuracy.

  Further, the medical image processing apparatus 1 is configured to operate as follows: (1) a reference image acquired by the extraction processing unit 2 for a portion (pulmonary blood vessel) of a subject having a branch structure; For each volume data of the target image, a landmark in the image is extracted. At this time, the extraction processing unit 2 (1a) extracts an image area corresponding to a part having a branch structure for each of these two volume data, and (1b) extracts an image for each of these two images. Extracting a common area between the boundary area of the partial area of the image including at least a part of the area and the extracted image area, and (1c) setting the position in the extracted common area as a seed, (1d) A landmark in an image area corresponding to a branch position of a part having a branch structure is extracted starting from the seed; (2) the landmark association unit 4 associates the landmarks extracted for these two medical images with each other. (3) The image alignment unit 5 aligns these two volume data based on the associated landmarks.

  According to such an image alignment process, when extracting a landmark of a part having a branch structure, a seed as a starting point of the extraction process can be automatically set, so that an increase in processing time can be prevented. It is possible to achieve image alignment and image alignment with good accuracy.

[Modification]
The content described above is only a specific configuration example for carrying out the present invention, and arbitrary modifications can be appropriately made.

[Modification 1]
First, as mentioned in the description of the above embodiment, as positional relationship information indicating the positional relationship of landmarks extracted from a medical image, an angle list including angles of vertices of a polygon composed of three or more landmarks ( A modification using (angle information) will be described.

  Here, an angle list including the angles of the vertices of a triangle composed of three landmarks is taken as a specific example. Based on the landmarks extracted by the landmark extraction unit 23, the positional relationship information generation unit 3 forms a plurality of sets of three landmarks. More specifically, first, two landmarks with a short distance between each other are selected and their pairs are formed. At this time, pairs may be formed in order from the shortest distance, or a pair whose distance is a predetermined value or less may be formed. Next, the positional relationship information generation unit 3 adds a new landmark to each pair to form a set of three landmarks, as with the objective form association unit 42 of the above embodiment.

  Subsequently, the positional relationship information generation unit 3 calculates the angle of the vertex of the triangle formed by the three landmarks in each group. Then, information for identifying each landmark (or information for identifying each set) is associated with the calculated angle, and an angle list is generated for each of the reference image and the target image.

  Next, landmark association processing using an angle list will be described. The triangle on the reference image side (a set of three landmarks) is <ri1, ri2, ri3>, and the triangle on the target image side is <tj1, tj2, tj3>. Further, the angle of the vertex ri1 of the triangle <ri1, ri2, ri3> is θr (ri1), the angle of the vertex ri2 is θr (ri2), and the angle of the vertex ri3 is θr (ri3). Further, the angle of the vertex tj1 of the triangle <tj1, tj2, tj3> is θt (tj1), the angle of the vertex tj2 is θt (tj2), and the angle of the vertex tj3 is θt (tj3).

  The landmark associating unit 4 sets the angles on the reference image side to θr (ri1), θr (ri2), θr (ri3), and the angles θt (tj1), θt (tj2), θt (tj3) on the target image side. Are determined to be equal to or less than a predetermined value, and triangles with the difference equal to or less than the predetermined value are associated with each other (that is, the association is performed by excluding triangles with the difference exceeding the predetermined value). At this time, the angle difference is calculated for all combinations of the angles of both images, and it is determined whether the angle difference of all the combinations is equal to or less than a predetermined value.

  Here, if necessary, as in the above-described embodiment, a landmark is added to form an L-gon, and the same determination process as described above is performed for the angle of the vertex to associate the L-gons with each other. The following processing is the same as in the above embodiment.

[Modification 2]
In the above embodiment, the alignment of the two medical images is performed using the branch point of the bronchus or the pulmonary blood vessel as a landmark. In this modification, the landmark is determined using the shape of the boundary portion (contour) of the lung. The method to do is explained.

  An example of the configuration of a medical image processing apparatus according to this modification is shown in FIG. The medical image processing apparatus 100 shown in the figure includes a specific part extraction processing unit 110, a landmark setting unit 120, a landmark association unit 130, an image alignment unit 140, a tomographic image generation unit 150, a display unit 160, and an operation unit 170. It has. The tomographic image generation unit 150, the display unit 160, and the operation unit 170 have the same configurations as the tomographic image generation unit 6, the display unit 7, and the operation unit 8 of the above-described embodiment, respectively.

(Specific part extraction processing unit)
The specific part extraction processing unit 110 performs processing for extracting an image region corresponding to the specific part of the subject for each of the reference image and the target image, and functions as an example of the “specific part extraction unit” of the present invention. . The specific part extraction processing unit 110 extracts an image region corresponding to the lungs of the subject as a specific part. In this modification, it is not necessary to extract small irregularities in the hilar portion in detail, and it is sufficient to extract the lung boundary shape (contour shape).

  For this purpose, the specific area extraction processing unit 110 first performs threshold processing on a medical image (reference image, target image) including a lung image area, and extracts an image area of −400 HU or less, for example. Next, a label (identification information) is assigned to each connected region (connected component) of the extracted image region using, for example, a known method (connected component labeling or the like).

  Subsequently, the number of voxels in each connected area is counted, and the connected area having the maximum number of voxels is selected from the connected areas that do not include any of the eight corners of the volume data of the medical image. Here, the reason why the connected region including any one of the eight corners of the volume data is excluded is to avoid selecting this air region when the number of voxels in the air region outside the body of the subject is maximum. It is. As described above, the region (voxel) corresponding to the lung is extracted from the medical image (volume data thereof).

(Landmark setting part)
The landmark setting unit 120 performs processing for setting the landmark of the specific part based on the image region corresponding to the specific part extracted by the specific part extraction processing unit 110. It functions as an example of “means”.

  The landmark setting unit 120 sets the landmarks located at the upper position, the left position, the right position, the front position, and the rear position of the image area (lung area) corresponding to the lung extracted as the specific part. Works. In the present invention, it is sufficient to set a landmark located at at least one of these positions. Hereinafter, the landmark setting processing of these positions will be described.

  First, the process for setting the landmark at the upper position will be described. The landmark setting unit 120 first detects the side position of the extracted lung region in order to set the landmark at the upper position. The side position means the left end position of the left lung and / or the right end position of the right lung. Here, since the landmarks in the upper position are set in both lungs, both the left end position of the left lung and the right end position of the right lung are detected.

  The left end position of the left lung is detected, for example, among the voxels of the image area (left lung area) corresponding to the extracted left lung, the horizontal coordinate value is the maximum (or minimum: depends on the direction of the coordinate axis in the horizontal direction Do this by searching for things. The right end position of the right lung is detected in the same manner.

  Next, the landmark setting unit 120 detects the position corresponding to the lung apex by tracing upward (in the direction of the head) within the lung region starting from the detected side position. An example in which the position of the apex of the left lung is detected will be specifically described. First, it is determined whether or not a voxel that is one above (in the head direction) the detected voxel at the left end position of the left lung is included in the left lung region. If it is included, it is determined whether the upper left voxel is included in the left lung region. On the other hand, when the voxel one position above the left end position is not included in the left lung region, it is determined whether one right voxel is included in the left lung region. Such determination processing is repeated until it is determined that neither the upper voxel nor the right voxel is included in the left lung region. Thereby, a position corresponding to the apex of the left lung is detected.

  Subsequently, the landmark setting unit 120 sets the landmark at the upper position at a position that is a predetermined distance below the position corresponding to the detected lung apex. As the predetermined distance, for example, the distance from the apex of the lung to the uppermost slice position where the area of the slice plane in the front-rear and left-right directions of the lung is equal to or greater than a predetermined value (such as 700 square millimeters) is employed. As a method for obtaining the slice position, for example, the area of the slice plane is calculated at a predetermined interval downward from the apex of the lung, and the position where the area first becomes a predetermined value or more is set as the slice position.

  The landmark setting unit 120 sets the landmark at the upper position, for example, at the center of gravity position of the cross section of the lung at the slice position (set for each of the left lung and the right lung). This landmark may be referred to as a subpulmonary landmark.

  It is also conceivable to detect the apex of the lung by analyzing the image region corresponding to the lung from the top (head direction) to the bottom (foot direction). However, since the trachea is also extracted together with the extraction process by the specific part extraction processing unit 110, the trachea is detected before reaching the lung apex, and thus the lung apex may not be extracted effectively. . In this modified example, since the lung apex is extracted by tracing the lung region upward from the side position as described above, there is an advantage that such a situation can be avoided.

  Next, the landmark setting process for the left and right positions will be described. Therefore, the landmark setting unit 120 detects the lower end positions of the left and right lungs. The lower end position is obtained by searching for the voxel located at the lower end with reference to the coordinates of the respective voxels in the left lung region and the right lung region. Note that a landmark (a landmark at the lower position) may be set at the lower end position.

  The landmark setting unit 120 obtains a slice position that internally divides the sub-pulmonary landmark (slice position thereof) and the lower end position by a predetermined ratio (for example, 8: 2). Then, at this slice position, the left end position of the left lung region and the right end position of the right lung region are detected. This detection process is performed by referring to the voxel coordinates of the left lung region and the right lung region at the slice position.

  Further, the landmark setting unit 120 sets a left position landmark and a right position landmark based on the detected left end position of the left lung and right end position of the right lung. These landmarks can be set by any method based on the left end position of the left lung and the right end position of the right lung. For example, the positions of these landmarks in the left-right direction can be set on the left line and the right line among three lines (or planes) that divide the lung region into four in the left-right direction. The position in the front-rear direction can also be set in the same manner (for example, it can be set at a position on a line or plane that divides the lung region in the front-rear direction).

  Finally, the landmark setting process for the front and rear positions will be described. These landmarks are set based on, for example, landmarks at the left position and the right position. For this purpose, the landmark setting unit 120 detects the front end position and the rear end position in the slice position of the internal division position. Then, based on the set left and right landmarks and the front and rear end positions of the slice position, the front and rear position landmarks are set. The setting method is arbitrary. For example, the midpoint of the left and right landmarks is obtained, and a straight line extending in the forward direction from the midpoint to the line connecting the left and right landmarks is considered. Then, a landmark at the previous position is set at a position advanced from the midpoint along the straight line by half the length in the front-rear direction of the lung. The landmark at the rear position can be set similarly.

(Landmark mapping part)
Landmark association processing by the landmark association unit 130 is easy. In other words, in this modified example, each landmark is given meaning (upper position, right position, left position, front position, rear position). This meaning is given by adding identification information such as a flag when setting each landmark. In accordance with this meaning, the landmark associating unit 130 associates the landmark of the reference image with the landmark of the target image (for example, landmarks in the upper position of the left lung are associated with each other). Note that some landmarks may be excluded and the remaining landmarks may be associated as in the above embodiment.

(Image alignment part)
The image alignment unit 140 uses, for example, the reference image side landmark coordinates (reference coordinate system coordinates) for the landmark pairs associated by the landmark association unit 130, and the target image side landmark coordinates. A conversion parameter to be converted into (coordinates of the target coordinate system) is calculated. A known method can be used for this calculation.

  According to this modification, instead of using the conventional iterative alignment method with a long processing time, an alignment method using landmarks is adopted, and each landmark is made meaningful, thereby making it short. Image alignment can be performed in processing time. In addition, since the number of landmarks to be considered is very small, image alignment can be performed in a very short time. In addition, since the landmarks that have been given meaning are associated with each other, there is no mistake in association, and it is possible to perform image alignment with high accuracy.

[Medical image processing method]
A medical image processing method according to the present invention will be described. The medical image processing apparatus according to the above embodiment is an example of an apparatus for realizing this medical image processing method.

  A first medical image processing method according to the present invention is a medical image processing method for aligning volume data of two medical images, and includes the following processes: (1) Medical based on each volume data (2) generating positional relationship information indicating the positional relationship of landmarks for each volume data; (3) excluding some of the landmarks based on the positional relationship information; The landmarks of the two medical images remaining after the exclusion are associated with each other; (4) The two volume data are aligned based on the association of the landmarks.

  According to such a medical image processing method, as described in the above embodiment, it is possible to shorten the processing time and to perform image alignment with high accuracy.

  A second medical image processing method according to the present invention is a medical image processing method for aligning volume data of two medical images, and includes the following processes: (1) Medical based on each volume data (2) Associate two medical image landmarks with each other: (3) To match the position of one volume data with the position of the other volume data based on the association of the landmarks (4) Calculate the distance between the associated landmarks after alignment based on the parameter; (5) Correct the parameter according to the distance; (6) Correction Based on the set parameters, the two volume data are aligned.

  According to such a medical image processing method, an increase in processing time can be avoided by using landmarks. In addition, since the parameters are corrected based on the distance between the landmarks and the images are aligned using the corrected parameters, the images can be aligned with high accuracy.

  A third medical image processing method according to the present invention is a medical image processing method for aligning volume data of two medical images, and includes the following processing: (1) For each medical image, the subject Image regions corresponding to the left and right lungs are extracted; (2) the position of the apex and the lower end of each lung in the image region is detected; and (3) the position of the apex and the lower end of the lung in the image region Detecting the left end position of the left lung image region and the right end position of the right lung image region at a cross-sectional position that internally divides the position by a predetermined ratio; (4) apex position, lower end position, left end position And a landmark of each medical image is set based on the right end position; (5) the landmarks of the two medical images are associated with each other; (6) the two volume data are aligned based on the association of the landmarks. Do.

  According to such a medical image processing method, it is possible to perform image alignment in a short processing time by using landmarks and making each landmark meaningful. In addition, since the number of landmarks to be considered is very small, image alignment can be performed in a very short time. In addition, since the landmarks that have been given meanings are associated with each other, there is no mistake in association, and image alignment can be performed with high accuracy.

1 is a schematic block diagram showing an example of the overall configuration of an embodiment of a medical image processing apparatus according to the present invention. It is a schematic block diagram showing an example of a structure of the extraction process part of embodiment of the medical image processing apparatus concerning this invention. It is a schematic block diagram showing an example of a structure of the landmark matching part of embodiment of the medical image processing apparatus which concerns on this invention. It is a schematic block diagram showing an example of a structure of the image position alignment part of embodiment of the medical image processing apparatus concerning this invention. It is a graph for demonstrating an example of the process by the image position alignment part of embodiment of the medical image processing apparatus concerning this invention. It is a graph for demonstrating an example of the process by the image position alignment part of embodiment of the medical image processing apparatus concerning this invention. It is the schematic showing an example of the flow of a process by embodiment of the medical image processing apparatus concerning this invention. It is the schematic for demonstrating the process by embodiment of the medical image processing apparatus concerning this invention. It is the schematic for demonstrating the process by embodiment of the medical image processing apparatus concerning this invention. It is a schematic block diagram showing an example of the whole structure of the modification of the medical image processing apparatus which concerns on this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1,100 Medical image processing apparatus 2 Extraction processing part 21 Image area extraction part 211 Bronchial extraction part 211a Local threshold value processing part 211b Image area separation processing part 212 Pulmonary blood vessel extraction part 22 Seed setting part 23 Landmark extraction part 24 Landmark extraction prohibition Unit 3 positional relationship information generation unit 4, 130 landmark association unit 41 line segment association unit 42 polygon association unit 43 landmark selection unit 44 evaluation value setting unit 45 association evaluation processing unit 5, 140 image registration unit 51 Conversion Parameter Calculation Unit 52 Conversion Parameter Correction Unit 6, 150 Tomographic Image Generation Unit 7, 160 Display Unit 8, 170 Operation Units F, F1, F2 Weighting Function (Objective Function)
110 Specific part extraction processing unit 120 Landmark setting unit

Claims (14)

A medical image processing apparatus for aligning volume data of two medical images,
Setting means for setting a landmark of a medical image based on each volume data;
For each volume data, a plurality of pairs of two landmarks of the same medical image are formed based on the landmarks set by the setting means, and the distance between the landmarks is calculated for each of the plurality of pairs. Generating means for generating distance information including the distance as positional relationship information;
Based on the positional relationship information, the distance between the pair of landmarks of one medical image is compared with the distance between the pair of landmarks of the other medical image, and the pair is determined based on the comparison result. Temporarily correlate, for each of the provisional pairs, add a landmark not included in the pair to form a set of three or more landmarks, and use the landmarks included in the set An evaluation value indicating the degree of correspondence of the set between two medical images is calculated based on the measured distance or angle, and some of the landmarks are excluded based on the evaluation value and remain after the exclusion Association means for associating landmarks of medical images with each other;
Alignment means for aligning two volume data based on the association of the landmarks;
Comprising
A medical image processing apparatus. The generating means includes
Based on the landmarks set by the setting means, a plurality of sets of three or more landmarks of the same medical image are formed,
For each of the plurality of sets, calculate an angle of the vertex of a polygon having three or more landmarks as vertices;
Generating angle information including the angle as the positional relationship information;
The medical image processing apparatus according to claim 1. The association means includes
Based on the positional relationship information, an evaluation value indicating the degree of correspondence between the landmarks set for the two medical images is calculated,
Based on the evaluation value, the landmark is associated between two medical images.
The medical image processing apparatus according to claim 1. The association means includes
When one landmark not included in the temporary pair is added to form a set of three landmarks, a distance between the one landmark and each landmark of the pair is calculated,
When the difference between these distances is less than or equal to a predetermined value, the group of two medical images is provisionally associated with each other,
Calculating the evaluation value based on the provisional set;
The medical image processing apparatus according to claim 1 . The association means includes
By sequentially adding landmarks whose difference in distance is equal to or less than a predetermined value to the temporary pair, the set of three or more landmarks is temporarily associated with each other,
Calculating the evaluation value based on the provisional set;
The medical image processing apparatus according to claim 1 . The association means includes
Calculating a distance between two landmarks included in the set of landmarks;
Assigning a score to a landmark in which the difference in distance between the two medical images exceeds a predetermined value, and calculating the evaluation value based on the landmark whose score is included in the predetermined range;
The medical image processing apparatus according to claim 1 . The evaluation value is the number of landmarks whose score is included in a predetermined range.
The medical image processing apparatus according to claim 6 . The alignment means includes
Based on the landmarks correlated by the correlating means, a coordinate conversion parameter for adjusting the position of one volume data to the position of the other volume data is calculated,
Calculating the distance between the associated landmarks after alignment based on the parameters;
Based on only landmarks whose distance is less than or equal to a predetermined value, the new parameter is calculated.
The medical image processing apparatus according to claim 1. The alignment means includes
Based on the landmarks correlated by the correlating means, a coordinate conversion parameter for adjusting the position of one volume data to the position of the other volume data is calculated,
Aligning two volume data based on a predetermined continuous function based on the parameters,
The medical image processing apparatus according to claim 1. The predetermined continuous function is a smooth function;
The medical image processing apparatus according to claim 9 . The smooth function is a polynomial;
The medical image processing apparatus according to claim 10 . The alignment means calculates a solution of the polynomial by a linear optimization method, and aligns two volume data based on the solution.
The medical image processing apparatus according to claim 11 . Based on the two volume data aligned by the alignment means, tomographic image generation means for respectively generating image data of tomographic images at the same cross section by the alignment;
Display means for displaying two tomographic images based on the image data;
Further comprising
The medical image processing apparatus according to claim 1. A medical image processing method for aligning volume data of two medical images,
Set landmarks for medical images based on each volume data,
For each volume data, a plurality of pairs of two landmarks of the same medical image are formed based on the set landmarks, and a distance between the landmarks is calculated for each of the plurality of pairs, Generate distance information including distance as positional relationship information,
Based on the positional relationship information, the distance between the pair of landmarks of one medical image is compared with the distance between the pair of landmarks of the other medical image, and the pair is determined based on the comparison result. Temporarily correlate, for each of the provisional pairs, add a landmark not included in the pair to form a set of three or more landmarks, and use the landmarks included in the set An evaluation value indicating the degree of correspondence of the set between two medical images is calculated based on the measured distance or angle, and some of the landmarks are excluded based on the evaluation value and remain after the exclusion The two medical image landmarks are associated with each other,
Based on the association of the landmarks, the two volume data are aligned.
A medical image processing method characterized by the above.

Images