JP2008161266A - Image processor, computer program and image processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a novel art for improving visibility of an affected part even if the quality of an image acquired in an operation is low such as an open MR image. <P>SOLUTION: This image processor 1 is provided with a means for positioning a high-definition image including the affected part image and a low-quality image including the affected part image by a rigid body positioning method, and a means for deforming and converting the affected part L2 in the high-definition image and the affected part L1 in the low-quality image; and deforms and positions the high-definition image into the low-quality image. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an image processing apparatus, a computer program, and an image processing method, and particularly to a technique for assisting surgery by visualizing an invisible affected part.

  In recent years, due to significant progress in medical equipment and computer technology, high-definition images related to the internal structure of the human body have been obtained. In particular, visual guidance can be used to expand visual functions and visualize, measure, and judge invisible areas. As a result, it is becoming possible to improve the diagnosis and treatment accuracy of doctors and reduce the burden on patients.

For example, in liver tumor microwave coagulation therapy using vertical open MR (Magnetic Resonance), minimally invasive surgical treatment is possible in which treatment and surgery are performed while observing the tumor in real time with images obtained from open MR .
In other words, unlike ultrasound-guided therapy in liver tumor microwave coagulation, the needle is inserted into the tumor while observing the changing form and temperature of the tumor without being affected by air or bone. Thermocoagulation treatment can be performed with microwaves.
A. Carrillo., JL Duerk., JS Lewin and DL Wilson, Semiautomatic 3-D Image Registration as Applied to Interventional MRI Liver Cancer Treatment.IEEE Trans. On Med.Img.Vol.19, No.3.pp: 175- 185,2000

However, since an open MR apparatus has a lower magnetic field than a general MR apparatus, the SN ratio of a captured MR image is low, and only a low-quality image can be obtained. In addition, since the open MR image has low image quality, the tumor portion may not be clearly visible.
In Non-Patent Document 1, simple rigid body (not deformed) alignment is performed. However, since a soft organ such as the liver is deformed, the alignment accuracy is not sufficient.

  Therefore, an object of the present invention is to provide a new technique for improving the visibility of an affected area even when the image quality obtained during an operation like an open MR image is low.

[Image processing device]
An image processing apparatus according to the present invention includes means for aligning a high-quality image including an affected area image and a low-quality image including an affected area image by a rigid registration method, and an affected area in a high-quality image and an affected area in a low-quality image. And means for performing deformation conversion, and aligning deformation positions of the high-quality image and the low-quality image.
According to the present invention, since the means for transforming and transforming the affected part is provided, a high-quality image and a low-quality image can be accurately associated even with a soft affected part such as the liver.

The means for aligning by the rigid alignment method preferably performs from low-resolution image alignment to high-resolution alignment by gradually increasing the image resolution.
Further, it is preferable that the means for performing transformation conversion is performed by gradually increasing the image resolution from transformation conversion at a low image resolution to transformation conversion at a high resolution.
Furthermore, it is preferable that the means for performing transformation conversion performs the transformation conversion at a low transformation grid resolution to the transformation transformation at a high transformation grid resolution by gradually increasing the grid resolution.
As described above, the calculation can be efficiently performed by gradually increasing the resolution.

  It is preferable that the deformation conversion means obtains a parameter for deformation conversion using the rigid body alignment parameters obtained by the rigid body alignment method. By performing transformation transformation after performing overall alignment by the rigid alignment method, it is possible to reliably associate a high-quality image and a low-quality image.

[Computer program]
A computer program according to the present invention is for causing a computer to function as the image processing apparatus.

[Image processing method]
In the image processing method according to the present invention, the image processing apparatus aligns a high-quality image including an affected area image captured before surgery and a low-quality image including an affected area image captured during surgery by a rigid alignment method. Then, the affected part in the high-quality image and the affected part in the low-quality image are deformed and converted to align the high-quality image and the low-quality image.

  According to the present invention, deformation position alignment is performed, so that it is possible to cope with an affected part that is easily deformed, such as a liver.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 shows a surgery support system having an image processing apparatus 1 of the present invention. This system is a system that supports (navigates) surgery for a liver tumor, and includes an open MR device 2 and a display device 3 that displays surgical navigation images in addition to the image processing device 1.

  The image processing apparatus 1 is configured by installing an image processing computer program in a computer. Each function of the image processing apparatus 1 described below is realized by executing the computer program by a computer.

The open MR apparatus 2 can take a real-time image of an affected area inside a human body during surgery, but can obtain only a low-quality image compared to a normal MR apparatus because the magnetic field is low. An MR image taken by the open MR apparatus 2 during the operation is given to the image processing apparatus 1.
In the image processing apparatus 1, a navigation image is generated by combining the MR image and a CT image taken before surgery. The navigation image is displayed on the display device 3.
The doctor observes the navigation image and observes the form of the tumor, inserts a surgical instrument such as a needle into the body, and performs treatment such as liver tumor microwave coagulation without laparotomy.

[Generation of navigation image (alignment image)]
FIG. 2 shows a navigation image generation process performed in real time during the operation. When an MR image (MR volume data) is input from the open MR apparatus 2 to the image processing apparatus 1 during the operation (step S101), the image processing apparatus 1 stores the MR image and a storage device (not shown) of the image processing apparatus 1. From the CT image stored in (CT volume data), conversion parameters for deforming and aligning the MR image and the CT image are generated in real time (step S102).

  Subsequently, based on the generated parameters, the image processing apparatus 1 adjusts the affected part image in the CT image so that the form of the affected part (liver) in the CT image matches the form of the affected part (liver) in the MR image. Deformation alignment is performed in real time (step S103). That is, a navigation image in which the affected part (liver) image of the CT image subjected to deformation positioning is combined with the MR image is displayed in real time during the operation (step S104).

  In this navigation image, the liver in the affected part image in the high-quality CT image and the liver tumor contained therein appear in a state where necessary deformation has been applied to the corresponding position in the MR image. Therefore, even if the liver tumor is not shown in the low-quality MR image, the doctor performs the minimally invasive surgical treatment such as liver tumor microwave coagulation therapy while confirming the liver tumor on the navigation image. be able to.

  FIG. 3A shows an example of the MR image (cross section) of the human body H including the liver L1 portion. FIG. 4 shows an example of CT images (cross sections) of the same cross section for the same human body H. The MR image (low-quality image) has a lower image quality than the CT image (high-quality image), and the liver tumor LC that appears in the high-quality CT image does not appear in the MR image.

  As is apparent from FIG. 4 where only the liver L1 and L2 portions are extracted, not only the position of the liver but also the shape of the liver is different between the MR image and the CT image. This is because the liver is soft and easily deformed. Therefore, even if the same liver (affected part; organ) is targeted between images taken at different times and places, the simple rigid body alignment method cannot be used for successful alignment.

For example, when there are many parts to be markers like the brain and there is not much deformation, it is easy to cope with the alignment by the position correspondence between the markers.
However, since the liver is an organ with a simple form, there are few parts that can serve as markers for alignment, and the liver is easily deformed, so it is difficult to adopt a method that performs alignment by matching the positions of the markers. .

[Conversion parameter generation processing for deformation alignment]
Hereinafter, conversion parameter generation processing for deformation positioning for an easily affected part such as a liver will be described with reference to FIG.

First, the image processing apparatus 1 extracts liver portions L1 and L2 that are affected portions to be aligned from a CT image (CT volume data) and an MR image (MR volume data) obtained by photographing the inside of a human body ( Steps S1a and S1b). That is, the liver portions L1 and L2 images (volume data) shown in FIGS. 4A and 4B are extracted from the images H1 and H2 shown in FIGS.
Further, the image processing apparatus 1 calculates the center of gravity to be used for alignment from the shape of the extracted liver portion (step S1c).

Subsequently, the image processing apparatus 1 generates a low-resolution reduced image from the extracted liver partial volume data L1 and L2 using a Gaussian pyramid (steps S2a and S2b).
Further, after obtaining the center of gravity in step S1c, the image processing apparatus 1 performs a calculation for the rotation center for alignment and initial conversion in alignment (step S2c).

  Subsequently, the image processing apparatus 1 determines whether or not the overall alignment between the liver portions L1 and L2 is necessary (step S3). If the positions of the liver portions L1 and L2 are shifted, step S4 is performed. Alignment processing (rigid body alignment) up to step S8 is performed. If rigid body alignment is not necessary, the process immediately proceeds to the deformation conversion process after step S10.

[Steps S4 to S8: Rigid body alignment processing]
The rigid body alignment process is repeated from the low resolution image generated in steps S2a and S2b until the resolution is gradually increased until the resolution of the original image (Fine Image Resolution) is reached (step S4). By performing processing from a low resolution, it is possible to perform computation efficiently.

In the rigid body alignment process, a rigid body conversion process that performs rotation, parallel movement, and scale conversion (enlargement / reduction) of the object (liver portions L1, L2) is performed (step S5). In rigid body transformation, alignment is performed assuming that the object to be transformed is a rigid body (an object that does not deform). The rigid body transformation is represented by a rigid body transformation function T _Global and a rigid body transformation parameter (rotation amount / parallel movement amount / enlargement / reduction amount) U _Global given to the function. The alignment is performed by optimizing the rigid body transformation parameter U _Global .

Specifically, the image processing apparatus 1 includes the mutual information amount between the liver portion L1 of the MR image and the transformed image when the liver portion L2 image of the CT image is rigidly transformed with the current rigid transformation parameter U _Global. (Normalized Mutual Information) is calculated, and alignment evaluation and alignment are performed (step S6). _Further, the optimization of the rigid body transformation parameter U _Globaln is performed by a gradient method (Gradient _Decent Optimizer) (step S7).

For a certain resolution, the above steps S5 to S7 are repeated several times (step S8), and when the value of the rigid body transformation parameter U _Global converges, the loop of parameter calculation processing at that resolution ends (step S8).
Then, the image processing apparatus 1 increases the resolution of the image (step S4), and repeats the processing loop of steps S5 to S8 again. The first rigid transformation parameters _{U Global} when the resolution is raised, the rigid transformation parameters _{U Global} obtained in the previous resolution (lower resolution) is used.
When the resolution is increased to the resolution of the original image (Finest Image Resolution), the rigid registration process (Global Registration) ends (Step S4).

[Steps S10 to S16: Deformation conversion processing of liver]
When the rigid body alignment is completed, the obtained rigid body transformation parameter U _Global is given to the liver transformation transformation process (step S9).
Also in the deformation conversion process, starting from an image obtained by converting the low-resolution image generated in steps S2a and S2b with the rigid body conversion parameter U _Global , the resolution is gradually increased to the resolution of the original image (Finest Image Resolution). Is repeated (step S10).

In the deformation conversion process, the object (liver portions L1, L2) is deformed by an FFD (Free-Form Deformation) method (step S12). In FFD, a deformation operation is performed by operating a control curve represented by a B-Spline curve at a control point. The transformation conversion is expressed by a transformation transformation function T _Local and a transformation transformation parameter U _Local given to this function. The deformation conversion parameter U _Local here is the position of the control point after deformation of the liver portion represented by the FDD model.
The transformation conversion is performed by optimizing the transformation transformation parameter U _Local .

Specifically, the image processing apparatus 1 includes the mutual information amount between the liver portion L1 of the MR image and the converted image obtained by performing FFD conversion on the liver portion L2 image of the CT image with the current deformation conversion parameter U _Local. (Normalized Mutual Information) is calculated, and deformation alignment is evaluated and deformation alignment is performed (step S13). The deformation transformation parameter U _Local is optimized by the L-BFGS-B method (step S14).

In the present embodiment, for an image with a certain resolution, optimization of the deformation transformation parameter U _Local is repeatedly performed for a plurality of lattice resolutions from low to high FFD deformation lattice resolution (step S11).
That is, for an image of a certain resolution, starting from a low grid resolution, steps S12 to S15 are repeated, and when the value of the transformation conversion parameter U _Local converges, the parameter calculation process (steps S12 to S14) at that grid resolution is completed. (Step S15).
Then, the image processing apparatus 1 increases the lattice resolution (step S11) and repeats steps S12 to S15) again. The first deformation transformation parameters U _Local when grid resolution is raised, the modified conversion parameters U _Local determined by a straight line grating resolution (lower grid resolution) _is used.
When the lattice resolution is sufficiently high (Finest FFD Grid Resolution), the parameter calculation processing loop is terminated at the image resolution (step S11).

When the parameter calculation process at a certain image resolution is completed, the image processing apparatus 1 increases the image resolution (step S10) and repeats steps S11 to S15 again. The first deformation transformation parameters _{U Local} when the image resolution is raised, the modified conversion parameters _{U Local} obtained in the previous resolution (lower resolution) is used.
When the image resolution is increased to the resolution of the original image (Fine Image Resolution), the deformation conversion process (Local Registration) ends (step S10), and the deformation conversion parameter U _Local is obtained (step S16).

A navigation image is generated from the MR image and the CT image using the parameters U _Global and U _Local obtained as described above.

FIG. 6 shows a navigation image obtained by synthesizing the CT image obtained by aligning the deformation position with the MR image. The liver tumor LC that was not visible in the MR image is visible in the navigation image of FIG.
As described above, when the volume deformation alignment of the open MR image and CT image of the liver was performed, the alignment could be performed with an alignment accuracy of 1.5 mm. The accuracy of 1.5 mm was sufficient for practical use as a surgery support system.
On the other hand, when only the rigid body alignment was performed, the accuracy was 3.5 mm, which was not sufficient for practical use.

  In addition, this invention is not limited to the said embodiment, A various deformation | transformation is possible.

It is a figure which shows the whole structure of a surgery assistance system. It is a flowchart which shows a navigation image generation process. (A) shows an open MR image, and (b) shows a CT image. FIG. 3A shows an image of the liver portion cut out from the image in FIG. 3A, and FIG. 3B shows an image of the liver portion cut out from the image in FIG. It is a flowchart which shows a conversion parameter production | generation process.

Explanation of symbols

  1 Image processing device

Claims (7)

Means for aligning a high-quality image including an affected area image and a low-quality image including an affected area image by a rigid alignment method;
Means for transforming the affected area in the high-quality image and the affected area in the low-quality image;
With
An image processing apparatus characterized by aligning deformation positions of a high-quality image and a low-quality image.   2. The image processing according to claim 1, wherein the means for aligning by the rigid body aligning method performs the process from the alignment at a low image resolution to the alignment at a high resolution by gradually increasing the image resolution. apparatus.   3. The image processing apparatus according to claim 1 or 2, wherein the means for performing transformation conversion performs from transformation transformation at a low image resolution to transformation transformation at a high resolution by gradually increasing the image resolution.   4. The image according to claim 2, wherein the deformation conversion means performs the deformation conversion at a low deformation lattice resolution to the deformation conversion at a high deformation lattice resolution by gradually increasing the lattice resolution. Processing equipment.   5. The deformation transformation unit obtains a transformation transformation parameter using a rigid body alignment parameter obtained by the rigid body registration method. The image processing apparatus described.   A computer program for causing a computer to function as the image processing apparatus according to claim 1.   The image processing device aligns the high-quality image including the affected area image taken before the operation and the low-quality image including the affected area image taken during the operation by the rigid body alignment method, and the affected area in the high-quality image and the low image quality are aligned. An image processing method characterized by deforming and transforming an affected area in an image to perform deformation alignment of a high-quality image and a low-quality image.

Images