JP4972477B2 - Medical image processing device - Google Patents

Description

  The present invention relates to a medical image processing apparatus, and more particularly to a medical image processing apparatus that performs feature region extraction in a cardiac cine image or a time-resolved angiographic image.

  The MRI apparatus measures NMR signals generated by nuclear spins constituting a subject, particularly a human tissue, and images the form and function of the head, abdomen, limbs, etc. two-dimensionally or three-dimensionally. Device. In imaging, the NMR signal is given a phase encoding and a frequency encoding that differ depending on the gradient magnetic field. The measured NMR signal is subjected to two-dimensional Fourier transformation or three-dimensional Fourier transformation to reconstruct an image.

  In cardiac imaging and angiography using an MRI apparatus, time series images are acquired by continuously imaging the same part at a high speed in order to obtain information such as movement and temporal change of the subject.

  In the case of cardiac imaging, a pulse sequence is synchronized with an ECG (electrocardiogram), and one heartbeat (RR interval) is divided into a plurality of time phases (phases) to capture time-series images. In this way, an imaging method in which one heartbeat is divided into a plurality of time phases and a heart motion cycle is captured is generally referred to as “cardiac cine imaging”.

  On the other hand, in the case of angiography, high-speed imaging is continuously performed in parallel with intravenous injection of a contrast medium such as gadolinium (Gd), and the state of angiography is captured as a time-series image. Typical imaging methods used for angiography include time-resolved angiography (TRCE-MRA, Time-Resolved Contrast-Enhanced Magnetic Resonance Angiography).

  The acquired cardiac cine images and time-resolved angiographic images are converted into moving images by being continuously displayed, and the motion, function, etc. of the imaging target can be evaluated. For example, in the case of cardiac function analysis, the cardiac function parameters can be calculated from the thickness and movement of the myocardium by tracing the cardiac wall position in each cardiac time phase (phase). In the case of angiography, hemangiomas and arteriovenous malformations can be depicted.

  Filtering, noise reduction, feature region extraction, etc. for time-series images include a method of spatially processing each image of the time-series image and processing for Fourier data that is Fourier-transformed in the time direction. There is a way to do. Here, the Fourier data is a frequency spectrum and a phase obtained when a luminance change at each pixel of the time-series image is extracted in the time direction and Fourier-transformed in the time direction.

  As an example of processing for Fourier data, Patent Document 1 discloses a temporal mismatch between the application of a stimulus to a subject and the corresponding brain activity in brain function measurement (functional MRI, fMRI). Has been proposed to improve the extraction accuracy of the activated site by detecting the phase difference of Fourier data.

Patent Documents 2 and 3 propose a method for improving a signal-to-noise ratio (S / N) of a time-series image by removing a high-frequency noise component by filtering a high-frequency spectrum in Fourier data. .
JP 11-235323 A Japanese Patent Laid-Open No. 7-23926 JP-A-5-305065

  However, Patent Document 1 proposes a method of extracting a region using Fourier data obtained by performing Fourier transform in the time direction in the field of fMRI. However, similarly to fMRI, cardiac cine using time-series images is proposed. In the field of images and TRCE-MRA, a method for processing Fourier data has not been realized.

  Further, since the method of Patent Document 1 is a method for extracting an activation region of the brain, the luminance change to be extracted is about several percent, but in the case of a cardiac cine image or TRCE-MRA, the luminance change is several. Therefore, when the method described in Patent Document 1 is applied to cardiac cine images and TRCE-MRA, a new region extraction method different from the method described in Patent Document 1 is required.

  In Patent Documents 2 and 3, image quality can be improved by appropriately removing image noise by providing a frequency filter, but feature regions such as organ boundaries and blood vessel regions are not extracted.

  The present invention has been made in view of the above circumstances, and by automatically determining a desired region, contour, and the like from a time-series image, the burden on the operator is reduced and the quantitative analysis of functional analysis of organs and the like is performed. An object is to provide a medical image processing apparatus capable of improving reproducibility.

  In order to solve the above-described problem, the medical image processing apparatus according to claim 1, an acquisition unit that acquires a plurality of images captured continuously, and a signal for each pixel of the acquired plurality of images. Frequency spectrum calculation means for calculating the frequency spectrum of each pixel by performing a Fourier transform in the time direction, and a plurality of signals having a predetermined temporal change, respectively, by performing a Fourier transform in the time direction. Comparing the frequency spectrum for each pixel calculated by the frequency spectrum calculating means with the template set by the template setting means, and setting the frequency spectrum as a plurality of templates Pixels whose correlation between the frequency spectrum and the template for each frequency is a certain value or more are extracted as boundary candidate points Means out, is characterized by comprising a display means for displaying the identifiably displayed image and the extracted pixel and other pixels in the extraction means.

  According to the medical image processing apparatus of claim 1, a frequency spectrum obtained by performing Fourier transform on each pixel of a plurality of images (time-series images) taken in succession in a time direction, and a predetermined time A frequency spectrum (template) obtained by Fourier-transforming a signal having a local change is compared, and a pixel whose correlation between the frequency spectrum obtained for each pixel and the template is a certain value or more is extracted. And the image which displayed the extracted image and the other image so that identification was possible is displayed. Thereby, a pixel in which a signal change of a time-series image changes with a predetermined time can be easily identified from other pixels. Also, by performing Fourier transform in the time axis direction for each pixel, the temporal change of each pixel can be represented by one frequency spectrum, so that the amount of calculation can be reduced.

  The medical image processing apparatus according to claim 2, in the imaging apparatus according to claim 1, identification image creation means for creating an identification image in which the pixels extracted by the extraction means and other pixels are displayed in a distinguishable manner. And an identification image created by the identification image creation means, and a display image creation means for creating a display image in which the plurality of images are superimposed, the display means being the display image creation means The created display image is displayed.

  According to the medical image processing apparatus according to claim 2, an identification image in which the pixel extracted by the extraction unit and other pixels are displayed in a distinguishable manner is created, and the identification image and the time-series image are superimposed on each other. An image is created and the display image is displayed on the display means. Thereby, a pixel in which a signal change of a time-series image changes with a predetermined time can be easily identified from other pixels.

  The medical image processing apparatus according to claim 3 is the medical image processing apparatus according to claim 1 or 2, wherein the extraction unit has a correlation with a template of one of the plurality of templates of a certain value or more. It is characterized in that a boundary candidate point for each time is extracted by performing processing for extracting a pixel having a certain frequency spectrum as a boundary candidate point at a predetermined time for all of the plurality of templates.

  According to the medical image processing apparatus according to claim 3, a pixel having a frequency spectrum whose correlation with a predetermined template is equal to or greater than a predetermined value among a plurality of set templates is extracted as a boundary candidate point at a predetermined time. To do. By performing this processing on all the set templates, boundary candidate points by time are extracted. Thereby, the change of a desired area | region, an outline, etc. can be automatically extracted from a time series image. In addition, when the start of signal change is the same and the duration of the signal change is different, the frequency spectrum changes, that is, when the start of the signal change is the same and the duration of the signal change is long, the frequency spectrum changes. By using the fact that the amplitude at frequency 0 becomes larger and the frequency width becomes shorter, a plurality of templates corresponding to a plurality of desired signal changes having the same signal change start time and different signal change durations are set. A pixel having a frequency spectrum equal to that of the template, that is, a pixel having the same signal change start time and a different signal change duration can be extracted from the signal that is the basis of the template.

  A medical image processing apparatus according to a fourth aspect is the medical image processing apparatus according to any one of the first to third aspects, wherein a correlation between a boundary candidate point extracted by the extraction unit and a circular function or an elliptic function is obtained. A boundary line acquisition unit that acquires a boundary line by connecting boundary candidate points having high correlation, and the identification image generation unit includes the pixel extracted by the extraction unit, the boundary line, and another pixel. It is characterized in that an identification image displayed so as to be identifiable is created.

  According to the medical image processing apparatus of the fourth aspect, the pixel line extracted by the extracting unit is correlated with the circular function or the elliptic function, and the boundary line is acquired by connecting the pixels having high correlation. Then, the boundary lines and the extracted pixels and other images are displayed so as to be distinguishable. This makes it possible to automatically extract the myocardial boundary (intracardial wall, extracardiac wall) that can be approximated by a circular function or an elliptic function included in the extracted pixel.

  According to the present invention, by automatically determining a desired region, contour, etc. from a time-series image, it is possible to reduce the burden on the operator and improve the quantitative / reproducibility of functional analysis of organs and the like. .

  The best mode for carrying out the present invention will be described below with reference to the accompanying drawings.

<First Embodiment>
In the first embodiment, a myocardial contour is extracted from a time-series image in a cardiac cine image taken by a magnetic resonance imaging apparatus (MRI apparatus) which is one of medical imaging apparatuses.

  FIG. 1 is an external view of a magnetic resonance imaging apparatus (hereinafter referred to as an MRI apparatus) 1 to which the present invention is applied. The MRI apparatus 1 mainly includes a magnet 101 that generates a static magnetic field, a bed 103 on which the subject 102 is placed, an RF coil that irradiates the subject 102 with a high-frequency magnetic field (hereinafter referred to as RF) and detects an echo signal (transmission of a high-frequency magnetic field). And gradient signal coils 105 and 106 for generating a slice selection, phase encoding, or frequency encoding gradient magnetic field in any of the X, Y, and Z directions, respectively. 107, an RF power source 108 for supplying power to the RF coil 104, a gradient magnetic field power source 109, 110, 111 for supplying current to each of the gradient magnetic field coils 105, 106, 107, an RF power source 108, and a synthesizer. 112, sends instructions to peripheral devices such as modulator 113, amplifier 114, receiver 115, etc. A sequencer 116 that controls the operation, a storage medium 117 that stores data such as imaging conditions, a computer 118 that performs image reconstruction by referring to echo signals input from the receiver 115 and data in the storage medium 117, And a display 119 for displaying a result of image reconstruction performed by the computer 118.

  In FIG. 1, the RF coil is shown as a transmitter / receiver for the sake of simplicity, but a device on which each of a transmitting coil and a receiving coil is mounted is common. As for the receiving coil, a plurality of receiving coils may be used in parallel.

  An operation procedure when imaging the subject 102 using the MRI apparatus 1 shown in FIG. 1 will be described.

  In accordance with the imaging conditions specified by the operator, the sequencer 116 transmits a command to the gradient magnetic field power sources 109, 110, and 111 in accordance with a predetermined pulse sequence, and the gradient magnetic field coils 105, 106, and 107 change the gradient magnetic field in each direction. generate. At the same time, the sequencer 116 transmits an instruction to the synthesizer 112 and the modulation device 113 to generate an RF waveform, generates an RF pulse amplified by the RF power supply 108 from the RF coil 104, and irradiates the subject 102.

  The echo signal generated from the subject 102 is received by the RF coil 104, amplified by the amplifier 114, and A / D converted and detected by the receiver 115. Since the center frequency used as a reference for detection is stored in the storage medium 117 as a value measured in advance, it is read out by the sequencer 116 and set in the receiver 115. The detected echo signal is sent to the computer 118 for image reconstruction processing. Results such as image reconstruction are displayed on the display 119.

  Next, the relationship between the frequency spectrum (amplitude spectrum and phase spectrum) at each pixel obtained by Fourier-transforming each pixel of the time-series image acquired by the MRI apparatus 1 in the time direction and the signal change will be described. To do.

  FIG. 2 is a graph showing the result of performing Fourier transform on four signal changes (FIGS. 2 (1) to (4)) having different signal start and duration. In these four graphs showing changes in signal, (a) is a graph showing changes in signal strength over time, the vertical axis shows signal strength, and the horizontal axis shows time t. (B) is a graph obtained by deriving a frequency spectrum by Fourier-transforming (a) and displaying the amplitude spectrum in a graph. The vertical axis represents amplitude and the horizontal axis represents frequency. (C) is a graph obtained by deriving a frequency spectrum by performing Fourier transform on (a) and displaying the phase spectrum in a graph, where the vertical axis represents the phase and the horizontal axis represents the frequency.

  2 (1) and FIG. 2 (2) have the same signal change start time, and FIG. 2 (2) has a longer signal change duration than FIG. 2 (1). 2 (3) and FIG. 2 (4) have the same signal change start time, and FIG. 2 (4) has a longer signal change duration than FIG. 2 (3). As a result, it can be seen that when the signal change start time is the same and the signal change duration is long, the amplitude near the frequency 0 in the frequency spectrum increases. In addition, when half the amplitude at frequency 0 is set as a threshold value and a frequency width (band width ΔB) having a value equal to or greater than the threshold value is compared, it can be seen that the longer the signal change duration is, the narrower the bandwidth ΔB is.

  2 (1) and FIG. 2 (3) have the same signal change duration, and FIG. 2 (3) shows a case where the start of the signal change is later than in FIG. 2 (1). 2 (2) and 2 (4) have the same signal change duration, and FIG. 2 (4) shows a case where the start of the signal change is later than in FIG. 2 (2). Thus, it can be seen that when the duration of the signal change is the same and the start of the signal change is delayed, the phase difference Δθ between the i-th and i + 1-th frequency components increases.

  Next, the flow of processing for extracting a myocardial contour from a cardiac cine image (time series image) will be described.

Cardiac cine imaging with an MRI apparatus is an imaging method performed to measure cardiac functions such as cardiac wall motion and ejection fraction, using an SSFP (steady-state free precession) sequence. Often filmed. An image obtained by this cardiac cine imaging is called a cardiac cine image, regardless of whether it is a two-dimensional image or a three-dimensional image. Since the image obtained by the SSFP sequence has a contrast depending on T2 ^* / T1 of the tissue, the blood in the heart lumen is extracted with a high signal, and the myocardium is rendered with a lower signal than the heart lumen. Therefore, when extracting the outline of the myocardium from the cardiac cine image, first, the boundary between the heart lumen and the myocardium (inner heart wall) with clear contrast is extracted, and then the outer heart wall is extracted. Hereinafter, a case where a time-series image when the heart contracts is acquired will be described as an example.

  FIG. 3 shows a relationship between part of the time-series image and the signal intensity at times t0 and t1 in the time-series image. As described above, the signal intensity of the heart lumen (blood) is a high signal, the signal intensity of the myocardium (muscle) is a medium signal (between a high signal and a low signal), and the signal intensity of air (for example, the lung) is a low signal. It is drawn in. Therefore, as shown in FIG. 3B, in the time-series image, the location that is always a cardiac lumen region (point A) is always a high signal, and the initial location is the cardiac lumen region and then becomes the myocardial region. (Point B) changes from a high signal to a medium signal, and the location (Point C) that initially passes through the myocardial region and then becomes the air region changes from a high signal to a medium signal and low signal. However, a location (point D) that is initially a myocardial region and then becomes an air region changes from a medium signal to a low signal, and a location that is always an air region (point E) is always a low signal.

  Thereby, if the signal change at the point A and the point E where the signal does not change with time is Fourier-transformed in the time direction, a similar frequency spectrum is obtained. Further, when the signal changes at points B, C, and D where the signal changes with time are Fourier-transformed in the time direction, different frequency spectra are obtained.

  FIG. 4 is a flowchart showing a flow of processing for extracting a myocardial contour from a time-series image. The computer 118 operates according to this flowchart. The following process is started after the computer 118 acquires a time-series image of the subject 102.

  First, a frequency spectrum, that is, an amplitude spectrum and a phase spectrum are calculated by Fourier-transforming the luminance of each pixel of the acquired time-series image in the time direction (step S10). Note that the frequency spectrum obtained by Fourier transform in the time direction corresponds to the luminance change of each pixel. The process of step S10 is performed on all the pixels constituting the time series image.

  Next, a desired luminance change is set as a template function g (t) (step S11), and a frequency spectrum (amplitude spectrum and phase spectrum) is calculated by performing Fourier transform in the time direction, and the calculated amplitude A spectrum is defined as a template (see step S12, FIG. 5). In this embodiment, only the amplitude spectrum is used, and the phase spectrum is not used.

  The template function g (t) is a function that lowers the signal at a predetermined time. As shown in FIG. 5B, a function that lowers the signal at time ts + Δt (Δt can be set to positive or negative) is used as a template function. Set. In this embodiment, Δt, that is, a plurality of template functions having different times for changing the signal intensity are set.

The number of templates necessary to extract a rectangular wave-shaped signal change like the signal changes at points B and D in FIG. 3B is equal to or less than the number of frames constituting the time-series image. For example, if a time-series image is composed of 20 images, 20 or less template functions with different Δt can be set. Further, when the signal intensity changes from a high signal to a medium signal and a low signal as in the signal change at the point C in FIG. 3B, the signal change can be extracted by _N C _two templates.

  As a simple method, the template is set to be equal to or less than the number of frames of the time-series image, and only a rectangular wave signal change is extracted. The signal change of the coordinates for which the signal change could not be extracted by this template may be obtained by interpolation from surrounding coordinate points so as to maintain the continuity of the movement of the myocardium.

  Since the image contrast, that is, the signal intensity of high signal, medium signal, and low signal, can be theoretically calculated from the imaging parameters (repetition time TR, echo time TE, flip angle FA, etc.) of the SSFP sequence, the template function is automatically selected. Can be determined.

  A heart region (boundary candidate point) is extracted by applying a high-pass filter (low frequency filter) to remove static components such as the rib cage and other organs from the time-series image (step S13). The bandwidth of the high-pass filter is set so that it is sufficiently narrower than the heart rate [Hz], that is, the time when the signal intensity is weakened in the template function is sufficiently shorter than the half cycle of the heart rate. Pixels whose amplitude spectrum has been eliminated by this filtering process are excluded from the subsequent processes. As a result, it is possible to prevent erroneous extraction of regions other than the myocardium, and to reduce the subsequent processing load (calculation amount). In this embodiment, the heart region is extracted by applying a low-frequency filter. However, a band-pass filter (band-pass filter) may be applied to simultaneously remove high-frequency noise components.

  Next, heart wall candidate points for each time phase are extracted from the extracted heart region by comparing the amplitude spectrum calculated in step S10 with the template calculated in step S11 (step 14).

  That is, the amplitude spectrum of each pixel obtained in step S10 is compared with the template obtained in step S12 based on the template function whose signal becomes low at the predetermined time set in step S11. Pixels that are included and whose correlation between the amplitude spectrum of the pixel and the template is equal to or greater than a certain value are extracted as heart wall candidate points in a predetermined time phase. By performing the above process on all the templates set in step S11, heart wall candidate points for each time phase are extracted.

  For example, as shown in FIG. 6, an amplitude spectrum at each pixel of the boundary candidate point is compared with a template calculated from a template function whose signal becomes low at time t0, and a pixel whose amplitude spectrum matches the template is found. A pixel whose signal becomes low at time t0, that is, a heart wall candidate point at time t0 is extracted. The same processing is performed on the template functions whose signals become low at times t1 and t2, respectively, so that heart wall candidate points at times t1 and t2 are extracted. Thereby, the heart wall candidate point in each time phase of time t0, t1, and t2 is extracted.

  From the heart wall candidate points in each time phase obtained in step S13, the position (pixel) of the heart wall (the left heart inner wall, the right heart inner wall, and the outer heart wall) is determined (step S15). Since the cardiac wall candidate points are determined using the contrast of the myocardial boundary, the endocardial wall, the epicardial wall, epicardial fat exhibiting a high signal similar to the endocardial wall, and the like are included. Therefore, it is necessary to separate the inner and outer walls from the candidate heart wall points.

  First, a method for determining the inner wall and the outer wall of the left heart will be described with reference to FIG.

  In the cardiac wall candidate points in each time phase obtained in step S13, the inner and inner walls of the left and right ventricles and the fat outer wall are extracted simultaneously. For this reason, the position of the endocardial wall cannot be determined simply by connecting the heart wall candidate points. Therefore, by using the fact that the left ventricle looks circular in the cardiac short-axis image (the heart image in a direction substantially perpendicular to the body axis), the cardiac wall candidate points in each time phase obtained in step S13 are used. And a pixel having a large correlation with the circular function are extracted as the left heart inner wall in each time phase (step S701). For example, the center of the FOV (Field of view, imaging region) is used as a reference point, and the number of heart wall candidate points that coincide with the circle is counted while gradually increasing the radius. Then, a circle including the largest number of heart wall candidate points is determined as a circular function at that time phase, and a heart wall candidate point that matches the circle function is extracted as a left heart inner wall candidate point. As a result, pixels having a large correlation with the circular function are extracted at the heart wall candidate points in each time phase. As the reference point, the center of the FOV may be automatically set as described above, or the point that is predicted to be the center of the heart lumen is set by the operator using the pointing device 17 or the like. The center may be determined. In the above left heart inner wall extraction method, the center of the FOV is used as a reference point and the circular function is obtained while gradually increasing the radius. However, the present invention is not limited to this, and a plurality of circular functions are set in advance. You may make it select the circular function with the highest correlation from among them. Further, since the change between the time phases is not a problem size, the circular function before the one time phase may be referred to.

  The heart wall candidate points that are not extracted as the left heart inner wall candidate points in step S701 are stored and used in the subsequent processing.

  And the left heart inner wall candidate point in each time phase extracted in step S701 is connected, and the left heart inner wall is determined by performing the smoothing process (step S702).

  When the left heart inner wall in each time phase is determined, a pixel having a large correlation between the heart wall candidate point and the elliptic function in each time phase obtained in step S13 is extracted as an epicardial wall candidate point (step S703). At this time, accuracy is improved by referring to the heart wall candidate point deviated from the left heart inner wall candidate point in step S701 and taking in information on the boundary position between the myocardium and epicardial fat. Note that the elliptic function is determined in the same manner as in step S701, with the point predicted to be the center of the FOV or the center of the heart lumen as the reference point, and the vertical and horizontal axes are gradually expanded to match the ellipse. The number of heart wall candidate points to be counted and the ellipse with the largest number of matching heart wall candidate points may be determined as the elliptic function at that time phase, or a plurality of elliptic functions may be set in advance, An elliptic function having the highest correlation among them may be selected. Further, since the change between the time phases is not a problem size, the elliptic function before the time phase may be referred to.

  The extracardiac wall is determined by connecting and smoothing the pixels thus extracted (step S704).

  Then, it is determined whether the inner wall of the left heart determined in step S702 intersects or contacts the outer wall of the heart determined in step S704 (step S705).

  Since the inner and outer walls are located inside and outside the myocardium, they do not cross or touch each other. Therefore, when it is determined in step S705 that the inner wall of the left heart and the outer wall of the heart intersect or contact, the boundary point that intersects or contacts and the connection line adjacent thereto are deleted (step S706), and step S704 is performed again. The determination process of the epicardial wall is performed.

  If it is determined in step S705 that the inner wall of the left heart and the outer wall of the heart do not intersect or contact each other, the process is terminated.

  Next, a method for determining the inner wall of the right heart will be described with reference to FIG.

  At the heart wall candidate points for each time phase extracted in step S14, the pixels determined as the left heart inner wall and the outer heart wall are removed, and the remaining pixels are extracted as right heart inner wall candidate points (step S801). .

  The right heart inner wall is determined by connecting and smoothing the right heart inner wall candidate points extracted in step S801 (step S802).

  Then, it is determined whether the left heart inner wall determined in step S702 intersects or contacts the right heart inner wall determined in step S802 (step S803).

  If it is determined in step S803 that the inner wall of the left heart and the inner wall of the right heart intersect or contact each other, the connection line adjacent to the intersecting or contacting boundary point is deleted (step S804), and the right in step S802 again. An intracardiac wall determination process is performed.

  If it is determined in step S803 that the inner wall of the left heart and the inner wall of the right heart do not intersect or contact each other, the process proceeds to step S805.

  Next, it is determined whether the outer wall determined in step S704 intersects or contacts the inner wall of the right heart determined in step S802 (step S805).

  If it is determined in step S803 that the outer wall of the heart and the inner wall of the right heart intersect or contact, the connecting line adjacent to the boundary point that intersects or contacts is deleted (step S806), and the right heart in step S802 again. An inner wall determination process is performed.

  If it is determined in step S805 that the outer heart wall and the inner right wall do not intersect or contact, the processing ends.

  Thereby, the myocardial boundary (heart wall) in each time phase can be automatically extracted. Further, by approximating the myocardial boundary (heart wall) with a circular function or an elliptic function, the myocardial boundary (heart wall) can be determined more accurately.

  In the method for determining the left heart inner wall and the outer heart wall, the short heart image is used. However, the present invention is not limited to this, and the heart long axis image (heart image in a direction substantially parallel to the body axis). May be used. In this case, since the left ventricle appears to be elliptical, an elliptical function may be used instead of the circular function in the above left heart inner wall determination method.

  When the heart wall position is determined, an identification image in which the heart wall determined in step S15 and the time-series image are superimposed is created (step S16) and displayed on the display 119 (step S17). In this case, the heart wall and other portions can be visually separated by highlighting the heart wall (changing the display color, surrounding with a color line, etc.). Therefore, the identification image can be used for functional analysis of the subject. In this case, processing for inputting the position of the myocardium to the cardiac function analysis processing may be performed. This makes it easier to use the function for analyzing the function of the subject.

  According to the present embodiment, by using a frequency spectrum obtained by Fourier transform of a temporal change of a signal, in particular, an amplitude spectrum, a pixel in which a signal undergoes a predetermined temporal change in a time-series image, and other A pixel can be easily identified. In addition, by applying a high-pass filter after performing the Fourier transform, regions other than the desired region such as the heart region can be eliminated, so that the amount of calculation can be reduced.

  Further, according to the present embodiment, by extracting pixels having a frequency spectrum equal to that of a plurality of templates, it is possible to automatically extract a desired region such as a heart wall, a change in contour, etc. from a time-series image. it can.

  In the present embodiment, pixels whose amplitude spectrum obtained by Fourier transforming a time-series image and a plurality of templates are extracted as heart wall candidate points. However, one type of preset template is used. Standard by setting the frequency spectrum obtained by Fourier transform of the function as a template and multiplying the frequency spectrum obtained by Fourier transform of the time change of the signal of the predetermined pixel of the time series image by the predetermined coefficient If the converted frequency spectrum matches the template, the pixel may be extracted as a heart wall candidate point.

  Hereinafter, a case where the amplitude spectrum is normalized will be described. As shown in FIG. 9, coefficients B / B1 and B / B2 are set as coefficients in the horizontal axis direction so that the frequency widths B1 and B2 of the amplitude spectrum match the frequency width B of the template. The coefficients A / A1 and A / A2 are set such that the amplitude widths A1 and A2 of the frequency spectrum coincide with the amplitude width A of the template, and these coefficients are added to the amplitude spectrum obtained by Fourier transforming the time series image. Is applied to normalize the amplitude spectrum obtained by Fourier transform of the time-series image. Then, the normalized amplitude spectrum is compared with the template, and a pixel having an amplitude spectrum in which the normalized amplitude spectrum matches the template is set as a heart wall candidate point. Also, pixels that are set as heart wall candidate points and that have the same coefficient used for normalization are extracted as heart wall candidate points having the same temporal change, that is, a predetermined time phase. The

<Second Embodiment>
In the MRI apparatus 1 according to the first embodiment, a myocardial contour is extracted from a time-series image in a cardiac cine image photographed by the MRI apparatus 1, but an area of the image photographed by the MRI apparatus 1 is used. Or the method of extracting an outline is not limited to this.

  The MRI apparatus 1 of the present embodiment performs separation of arteries and veins, rendering hemangiomas, arteriovenous malformations (lesion sites), and the like from a time-series image obtained by angiography with TRCE-MRA. In the figure, the same portions as those of the first embodiment are denoted by the same reference numerals, and description thereof is omitted.

  TRCE-MRA is a technique in which high-speed imaging is continuously performed in parallel with intravenous injection of a contrast agent such as Gd, and the state of angiography is captured as a time-series image. The time-series images acquired in this way are divided into a time phase until the injected contrast agent reaches the imaging region, a time phase that reaches the artery of the imaging region, and a time phase that reaches the vein of the imaging region. As shown in FIG. 10, there is a difference in the time for the contrast medium to reach between the artery and the vein, that is, the contrast medium reaches the vein after the contrast medium reaches the artery. When the present invention is applied to TRCE-MRA, it is possible to separate an artery and a vein in a moving image using the difference in luminance change of the artery and vein.

  FIG. 11 is a flowchart showing a flow of processing for separating an artery and a vein in a time-series image and extracting a lesion site. The computer 118 operates according to this flowchart. The following process is started after the computer 118 acquires a time-series image of the subject 102.

  First, a frequency spectrum, that is, an amplitude spectrum and a phase spectrum are calculated by Fourier-transforming the luminance of each pixel of the acquired time-series image in the time direction (step S20). Note that the frequency spectrum obtained by Fourier transform in the time direction corresponds to the luminance change of each pixel. The process of step S20 is performed on all the pixels constituting the time series image.

  An arterial region and a venous region are extracted by applying a high-pass filter (low-frequency filter) to remove a region where there is no luminance change due to a contrast agent from the time-series image (step S21). Pixels whose frequency spectrum has been eliminated by this filtering process are excluded from the subsequent processes. As a result, it is possible to erase a region where there is no change in luminance due to the contrast agent, and to reduce the subsequent processing load (calculation amount).

  Next, the preset arterial luminance change a (t) and venous luminance change v (t) are set as template functions, and the frequency spectrum obtained by Fourier transforming the template function in the time direction is set. Among them, the amplitude spectra are set as template amplitude spectra A (ω) and V (ω), respectively, and the phase spectra obtained by Fourier transforming the template function in the time direction are set as template phase spectra φA and φV, respectively (step) S22).

  The frequency spectrum obtained by Fourier-transforming the luminance of each pixel in the time-series image in the time direction has the same amplitude spectrum even if there is a Δt offset at the start of the signal change if the signal change is the same. .

  On the other hand, the phase spectrum obtained by Fourier transforming the luminance of each pixel of the time-series image in the time direction is only 2πnΔt if there is an offset of Δt at the start of the signal change even if the signal change is the same. The inclination increases. The slope of the phase spectrum difference is obtained by fitting the phase spectrum difference with a linear function with respect to n. The slope of this phase spectrum difference is set as a phase spectrum difference Δφ (Δφ = 2πΔt). Therefore, when the time difference (brightness peak time, start of signal change, etc.) between the arterial luminance change a (t) and the venous luminance change v (t) as shown in FIG. 10 is Δt, The phase spectrum difference between the template phase spectrum φA of the artery and the template phase spectrum φV of the vein is Δφ.

  From the above, in this embodiment, the template amplitude spectra A (ω) and V (ω) are considered to be the same, but the template phase spectra φA and φV are different, and the template phase spectrum φA of the artery and the template phase spectrum φV of the vein are different. Has a phase spectrum difference of Δφ.

  The time difference Δt of the luminance peak between the arterial luminance change a (t) and the venous luminance change v (t) varies depending on the region to be imaged. This is because the vein is imaged after the artery is first imaged by the contrast agent, and once passes through the end of the body, the time difference Δt becomes smaller as the end of the body is approached. The phase spectrum difference Δφ is expressed by a relational expression of Δt = Δφ / 2π, and the phase spectrum difference Δφ and the time offset Δt have a one-to-one correspondence. Therefore, a template function (brightness change a of the artery a) (T), the luminance change v (t) of the vein needs to be set in advance.

  In order to improve the extraction accuracy, the template function is set by the user specifying the artery and vein at a desired position in the time-series image as a point or a region, and setting the brightness change at that position as the template function. You can also. It is also possible to determine a change in the brightness of the arteries and veins from the signal intensity of echo signals continuously acquired at the time of imaging, and to set a template function based on this.

  When the template amplitude spectra A (ω) and V (ω) and the template phase spectra φA and φV are set in step S22, the amplitude spectrum calculated in step S20 and the template amplitude spectrum A (ω) calculated in step S22. , V (ω) and between the phase spectrum calculated in step S20 and the template phase spectrums φA and φV calculated in step S22 (step S23), and calculated in step S20. The regions where the amplitude spectrum and the template amplitude spectra A (ω) and V (ω) are equal and the correlation between the phase spectrum calculated in step S20 and the template phase spectra φA and φV are large are the arterial region and the vein region, respectively. Extracted (step S24).

  Pixels in which the amplitude spectrum and phase spectrum coincide with the template amplitude spectra A (ω) and V (ω) and the template phase spectra φA and φV have the same luminance as the template functions a (t) and v (t). It can be judged that it is changing. The amplitude spectrum matches the template amplitude spectra A (ω) and V (ω), but the phase spectrum is different from the template phase spectra φA and φV. Is considered to be offset (there is a time delay).

  As a method for extracting a region where the correlation between the phase spectrum calculated in step S20 and the template phase spectrums φA and φV is large, a case where the phase spectrum matches the template phase spectra φA and φV, a phase spectrum and a template are used. The case where the difference between the phase spectra φA and φV is a predetermined value or less is extracted. Here, the allowable error as the phase spectrum difference Δφ is equivalent to an allowable error with respect to Δt (for example, one frame or less of a moving image). That is, when the total number of frames is N frames, the order of error that can be allowed as the phase spectrum difference Δφ is about 1 / N. For example, in the case of a 20-frame moving image, the error order is approximately 5% (= 1/20), and the phase spectrum difference Δφ is 5% or less between the phase spectrum and the template phase spectra φA and φV. Extraction may be performed as a case where the difference is equal to or less than a predetermined value.

  An identification image in which the arterial region and vein region for each time phase extracted in step S24 and the time-series image are superimposed is created (step S25) and displayed on the display 15 (step S26). In this case, the arterial region and the venous region are highlighted (colored lines that can be visually separated (the arteries are red, the veins are blue, etc.), etc.) Are visually separated. Further, by observing the highlighted arterial region and venous region in the identification image, lesions such as hemangiomas and arteriovenous malformations can be easily found. Also, compare the contours of the arterial region and vein region highlighted in the identification image with the contours of the normal arterial region and vein region created using the contours of the arterial region and vein region highlighted in the identification image. Alternatively, lesions such as hemangiomas and arteriovenous malformations may be automatically depicted. This makes it easier to use the function for analyzing the function of the subject.

  According to the present embodiment, when the start of signal change is delayed, the difference in phase becomes large, so that the amplitude spectrum for each pixel matches the template, and the phase spectrum and phase spectrum for each pixel match. Automatically extract pixels whose difference from the template is equal to or less than a predetermined value, that is, pixels whose start of signal change is different from the signal that is the basis of the template and phase spectrum template, and whose signal change duration is the same. Can do.

  In the present embodiment, since the injected contrast agent is divided into a time phase until reaching the imaging site, a time phase reaching the artery of the imaging site, and a time phase reaching the vein of the imaging site, Although the arterial luminance change a (t) and the venous luminance change v (t) are set one by one, the time phases are subdivided according to the flow of the contrast agent, and the arterial luminance changes a (t ) And a plurality of vein luminance changes v (t) may be prepared. In this case, since a plurality of template amplitude spectra A (ω), V (ω) and template phase spectra φA, φV are set, template amplitude spectra A (ω), V (ω) and templates for each time phase are set. The above processing is performed for each of the phase spectra φA and φV.

1 is a schematic diagram showing an overall configuration of a first embodiment of an MRI apparatus to which the present invention is applied. It is explanatory drawing which shows the amplitude spectrum and phase spectrum which are obtained when a predetermined signal is Fourier-transformed in the time direction. It is explanatory drawing explaining the relationship between the motion of the heart of 1st Embodiment of the said MRI apparatus, and the signal of a time series image. It is a flowchart which shows the flow of a process of 1st Embodiment of the said MRI apparatus. It is explanatory drawing explaining the relationship between the template function and spectrum of 1st Embodiment of the said MRI apparatus. It is explanatory drawing explaining the method of extracting the heart wall candidate point for every time phase of 1st Embodiment of the said MRI apparatus. It is explanatory drawing explaining the method to determine the left heart inner wall and the heart outer wall of 1st Embodiment of the said MRI apparatus. It is explanatory drawing explaining the method to determine the inner wall of the right heart of 1st Embodiment of the said MRI apparatus. It is explanatory drawing explaining the method of normalizing the frequency spectrum of 1st Embodiment of the said MRI apparatus. It is explanatory drawing explaining the relationship between the signal and phase spectrum of 2nd Embodiment of the MRI apparatus with which this invention was applied. It is a flowchart which shows the flow of a process of 2nd Embodiment of the said MRI apparatus.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1: MRI apparatus, 101: Static magnetic field generating magnet, 102: Subject, 103: Bed, 104: High frequency magnetic field coil, 105: X direction gradient magnetic field coil, 106: Y direction gradient magnetic field coil, 107: Z direction gradient magnetic field coil , 108: high-frequency magnetic field power supply, 109: X-direction gradient magnetic field coil, 110: Y-direction gradient magnetic field coil, 111: Z-direction gradient magnetic field coil, 112: synthesizer, 113: modulator, 114: amplifier, 115: receiver, 116 : Sequencer, 117: Storage medium, 118: Computer, 119: Display

Claims (4)

Acquisition means for acquiring a plurality of images photographed in succession;
A frequency spectrum calculating means for calculating a frequency spectrum for each pixel by performing a Fourier transform in a time direction of a temporal change in a signal for each pixel of the acquired plurality of images;
Template setting means for setting, as a plurality of templates, frequency spectra obtained by performing a Fourier transform on each of a plurality of signals having a predetermined temporal change in the time direction;
The frequency spectrum for each pixel calculated by the frequency spectrum calculation means is compared with the template set by the template setting means, and a pixel whose correlation between the frequency spectrum for each pixel and the template is equal to or greater than a certain value is obtained. Extraction means for extracting as boundary candidate points;
Display means for displaying an image in which the pixels extracted by the extraction means and other pixels are displayed in an identifiable manner;
A medical image processing apparatus comprising: An identification image creation means for creating an identification image in which the pixels extracted by the extraction means and other pixels are displayed in a distinguishable manner;
An identification image created by the identification image creating means, and a display image creating means for creating a display image in which the plurality of images are superimposed,
The medical image processing apparatus according to claim 1, wherein the display unit displays the display image created by the display image creation unit.   The extraction means is configured to extract a pixel having a frequency spectrum whose correlation with one type of template among the plurality of templates is a predetermined value or more as a boundary candidate point at a predetermined time. The medical image processing apparatus according to claim 1, wherein a boundary candidate point for each time is extracted by performing the operation on the medical image processing apparatus. A boundary line acquisition means for obtaining a boundary line by correlating a boundary candidate point extracted by the extraction means and a circular function or an elliptic function, and connecting boundary candidate points having high correlation,
The said identification image preparation means produces the identification image which displayed the pixel extracted by the said extraction means, the said boundary line, and another pixel so that identification was possible. The medical image processing apparatus described.

Images