JP2015156894A - Medical image processor, medical object area extraction method thereof and medical object area extraction processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To extract a two-dimensional or three-dimensional object area with a simple operation in a short period of time from a plurality of two-dimensional tomographic images acquired by a medical image diagnostic device.SOLUTION: A control part 10 controls operations of a preprocessing part 20 and an image processing part 30, the preprocessing part 20 receives a designation input by a manual operation of an input part 40, acquires data of an object area and the approximate size thereof at the same time, and performs preprocessing for acquiring data of a non-object area, and the image processing part 30 generates a histogram of luminance data of a pixel of the object area and a histogram of luminance data of a pixel of the non-object area from the data of the object area and data showing the approximate size, and the data of the non-object area and data showing its approximate size acquired by the preprocessing part 20, approximates the respective histograms to a normal distribution, determines a threshold of area division by a Bays rule by using an approximate normal distribution of the object area and an approximate normal distribution of the non-object area, and extracts the object area and the non-object area.

Description

  The present invention relates to a medical image processing apparatus that extracts a target area with high accuracy from image data acquired by a medical image diagnostic apparatus, a medical target area extraction method, and a medical target area extraction processing program.

  In recent years, it has been acquired by medical diagnostic imaging equipment such as X-ray computer tomography (CT), magnetic resonance imaging (MRI), and positron emission tomography (PET). A lesion is found from a reconstructed two-dimensional tomographic image or three-dimensional image, and the state of the lesion is observed to diagnose the presence or absence of disease and the progress.

  Images picked up by these devices are expressed by density information (pixel density information) regarding a certain point in the image. From the density information, attention is paid to a target area, for example, an organ or the like, and it can be used for diagnosis of the presence or absence of a lesion. .

  Since a large amount of images can be acquired by a medical image diagnostic apparatus, a great deal of labor is required for a doctor to make a tumor diagnosis. Such an increase in labor for diagnosis may lead to oversight or misdiagnosis, and therefore it is desired to reduce it as much as possible. Therefore, a technique (CAD: Computer Aided Detection) for automatically extracting a target area such as a tumor and an apparatus equipped with the technique have been proposed (for example, see Patent Document 1).

In addition, the following various methods (1) to (8) have been proposed as a segmentation technique for extracting a target region from a grayscale image such as a CT image.
(1) A method for separating a region by determining a suitable binarization threshold value from a luminance (density) histogram of an image.
(2) A method of extracting a boundary region using a gradient (gradient) of image brightness (density), a Laplacian, and a Laplacian zero crossing point.
(3) A dynamic contour extraction method (Snake method) that sequentially optimizes the boundary line initially determined by some method.
(4) A method of determining a boundary by sequentially determining pixels so that a starting point and an ending point are determined, and a predetermined evaluation function between them is small (or large).
(5) Reach point search for extending the tentacles in each direction from each pixel in the target area and determining the boundary from the distribution state of the points (reach points) whose brightness (density) value difference from the starting pixel is equal to or greater than a threshold value Law.
(6) A method of acquiring images having different contrasts with the passage of time (time phase) after the injection of the contrast agent, and utilizing density characteristics for each time phase, that is, multi-time correlation density characteristics.
(7) Tumor region extraction method using wavelet transform and smoothing as seen in lung cancer region extraction.
(8) A method in which a deformable three (2) dimensional curved surface is analytically determined to obtain an intended image, and the shape of the target region is obtained under the assumption of elastic deformation or the like.

JP 2012-245085 A

  By the way, in the above method (1), it is not easy to determine a suitable binarization threshold when separation of histograms between different regions is not good.

  In the method (2), the luminance gradient and the Laplacian value are large as noise other than the boundary, and are not easily distinguished from the boundary.

  In the method (3), there are many parameters for determining the desired shape and sequential deformation, and it is difficult to determine a value suitable for the target image, and the robustness (robustness) is low between different images.

  In the method (4), a common evaluation function cannot be determined between images. In addition, the result depends greatly on the selection of the start and end points.

  Further, in the method (5), since the number of times of processing is proportional to the square of the number of target pixels, the processing time varies greatly for each image (particularly in three dimensions). Moreover, it is not easy to determine a general-purpose threshold value.

  Further, in the above method (6), it is not easy to align the images in the abnormal correlation, and the error greatly affects the detection accuracy.

  In the method (7), the selection of the wavelet suitable for the target areas having different sizes and shapes and the determination of the degree of smoothing have not been confirmed.

  Further, in the method (8), it is not easy to set a deformation parameter and a deformation method suitable for the target image.

  The inventors of the present invention try to determine various parameters such as a threshold value when extracting a target area for all of the conventional methods (1) to (8) as described above. The contrast between the target area and the non-target area is usually very different, and it is necessary to ensure the robustness of the process to those changes. Have difficulty.

  In particular, when a target area is determined in three dimensions, three-dimensional data is often directly processed, but generally the processing program becomes complicated and the processing time increases.

  In view of the above, the object of the present invention is to provide an X-ray computed tomography (CT) apparatus, a nuclear magnetic resonance imaging (MRI) apparatus, a positron emission tomography (PET). Medical image processing that makes it possible to extract a two-dimensional or three-dimensional target region in a short time from a plurality of three-dimensional tomographic images acquired by a medical image diagnostic apparatus such as a Positron Emission Tomography device. The present invention provides an apparatus, a medical target area extraction method, and a medical target area extraction processing program.

  Other objects of the present invention and specific advantages obtained by the present invention will become more apparent from the description of embodiments described below.

  In the present invention, when a target region (for example, liver cell carcinoma in the liver) is automatically extracted from a plurality of two-dimensional tomographic images acquired by a medical image diagnostic apparatus, a two-dimensional or three-dimensional target region is extracted. , Accepting input by simple manual operation, obtain the approximate size at the same time as the data of the target area, perform pre-processing to acquire the data of non-target area, and use the data acquired by the above pre-processing Image processing for automatically extracting a two-dimensional or three-dimensional target region. In the above image processing, approximation processing of density distribution to normal distribution, statistical region extraction processing by Bayes rule, target region correction processing by global and local statistical tests, and two-dimensional image for three-dimensionalization The two-region extraction processing for separating the target region is performed by repeating the above-described processing in the region where the already extracted region is limited to the enlarged / reduced lower intermediate region.

  That is, the present invention is a medical image processing apparatus that accepts a designation input by manual operation, acquires a rough size at the same time as target area data, and performs preprocessing for acquiring non-target area data. A pre-processing unit that performs the processing, and an image processing unit that extracts a two-dimensional or three-dimensional target region using the data acquired by the pre-processing unit. The image processing unit has a function of accepting a designation input by manual operation for designating a size, and the image processing unit obtains the data of the target area acquired by the pre-processing unit, the data indicating the approximate size, and the data of the non-target area And a histogram of the luminance data of the pixels in the target area and the histogram of the luminance data of the pixels in the non-target area are generated from the data indicating the approximate size, and each histogram is normalized. Using the normal distribution approximation processing function that approximates the cloth, the approximate normal distribution of the target region obtained by the normal distribution approximation processing function, and the approximate normal distribution of the non-target region, the threshold value for region division according to the Bayes rule is determined, It has a region extraction function for extracting a target region and a non-target region.

  In the medical image processing apparatus according to the present invention, the pre-processing unit may have an image luminance adjustment processing function that keeps the luminance range of the processing target image to be acquired within a predetermined luminance range.

  The image processing unit uses only the periphery of the target area that has been enlarged as the detected target area as a test execution area, and performs raster scanning of pixels in the test execution area to obtain a plurality of pixels around the target area. A global target area correction process in which a Z test is performed using an average value and a population mean and standard deviation obtained as a population within an area that has already been determined as a target area, and the target area is corrected based on the test result. And only the periphery of the target area that has already been expanded is set as the test execution area, and the test target pixel obtained by searching only the pixels in the test execution area and a plurality of (n) pixels around it Z test is performed using the average value and the average value and standard deviation of multiple (m> n) pixels around the test target pixel only in the area already determined as the target area as the population mean and population standard deviation. , Its inspection It is possible to have a target area correction processing function for performing local target area correction processing for correcting the target area based on the fixed result.

  Furthermore, in the medical image processing apparatus according to the present invention, the image processing unit repeats the repeated processing on the two-dimensional image for three-dimensionalizing the extracted target region into an intermediate region obtained by enlarging or reducing the extracted region. It is possible to have a continuous two-region extraction processing function for separating a target region within a limited region.

  The present invention is a medical target region extraction method that receives a designation input by manual operation, acquires a rough size at the same time as target region data, and performs preprocessing for acquiring non-target region data A pre-processing step, and an image processing step for extracting a two-dimensional or three-dimensional target region using the data acquired in the pre-processing step. In the pre-processing step, an outline of the region of the image to be processed is included. Accepting a designation input by manual operation for designating the size, and in the image processing step, the data of the target area acquired by the preprocessing step, the data indicating the approximate size, the data of the non-target area, and the approximate From the data indicating the size, a histogram of luminance data of pixels in the target area and a histogram of luminance data of pixels in the non-target area are generated. Using the normal distribution approximation processing function that approximates each histogram to a normal distribution, and the approximate normal distribution of the target region and the approximate normal distribution of the non-target region obtained by the normal distribution approximation processing function, the threshold of region division according to the Bayes rule is set. It is characterized by determining and extracting a target area and a non-target area.

  In the medical target region extraction method according to the present invention, in the preprocessing step, an image luminance adjustment process may be performed in which the luminance range of the processing target image to be acquired falls within a predetermined luminance range.

  Further, in the medical target region extraction method according to the present invention, in the image processing step, only the periphery of the target region that has already been detected is set as the test execution region, and the pixels in the test execution region are raster scanned. Z-test is performed using the average value of the plurality of pixels around the target area obtained above and the population mean and standard deviation obtained as a population within the area already determined as the target area. Based on the global target area correction process that corrects the target area based on the above, and only the surrounding area of the target area that has already been expanded is set as the test execution area, and only the pixels in the test execution area are searched for above The Z test is performed using the average value of a plurality of pixels around the target area and the population mean and standard deviation obtained as a population within the area already determined as the target area. It can be made to perform a local object area correction processing for correcting the target region Zui.

  Further, in the medical target region extraction method according to the present invention, in the image processing step, an intermediate region obtained by enlarging / reducing an already extracted region is obtained by repeating a process on a two-dimensional image for three-dimensionalizing the extracted target region. It is possible to perform a continuous two-region extraction process for separating the target region within a region limited to the above.

  The present invention is a medical target area extraction processing program executed by a computer mounted on a medical image processing apparatus, and the computer accepts a designation input by manual operation, and simultaneously approximates the size of the target area data. Functions as a pre-processing unit that performs pre-processing for acquiring non-target region data and an image processing unit that extracts a two-dimensional or three-dimensional target region using the data acquired by the pre-processing unit The preprocessing unit has a function of accepting a designation input by a manual operation for designating an approximate size of a region of the image to be processed, and the image processing unit is configured to acquire the target region acquired by the preprocessing unit. Data of the target area, and data indicating the approximate size and non-target area data and data indicating the approximate size. A normal distribution approximation processing function that generates a histogram of the luminance data of the pixels of the gram and the non-target region and approximates each histogram to a normal distribution, and the approximate normal distribution and non-target region of the target region obtained by the normal distribution approximation processing function The approximate normal distribution is used to determine a threshold value for region division according to the Bayes rule, and a region extraction function for extracting a target region and a non-target region is provided.

  The medical target region extraction processing program according to the present invention causes the computer to function as the pre-processing unit having an image luminance adjustment processing function for keeping the luminance range of an image to be processed within a predetermined luminance range. it can.

  Further, the medical target area extraction processing program according to the present invention provides the above object obtained by raster-scanning pixels in the test execution area only with the target area surrounding the target area that has already been detected as the test execution area. A Z test is performed using the average value of a plurality of pixels around the area, and the population mean and standard deviation obtained as a population within the area that has already been determined as the target area, and the target area is determined based on the test result. A global target area correction process to be corrected and a plurality of areas around the target area obtained by searching only the pixels in the test execution area, with only the target area surrounding the already detected target area enlarged as the test execution area A Z test is performed using the average value of the individual pixels and the population mean and population standard deviation obtained as the population within the area already determined as the target area, and the target area is corrected based on the test result. It can be made to function the computer as the image processing unit having the target area correction processing function of performing a local object area correction processing that.

  Furthermore, the medical target region extraction processing program according to the present invention is configured so that an iterative process on a two-dimensional image for three-dimensionalizing an extracted target region is limited to an intermediate region obtained by enlarging or reducing the extracted region. The computer can be made to function as the image processing unit having a continuous two-region extraction processing function for separating the target regions.

  According to the present invention, a designation input by a simple manual operation is accepted, and the approximate size is simultaneously obtained from the plurality of two-dimensional tomographic images acquired by the medical image diagnostic apparatus, and the non-purpose By performing preprocessing for acquiring region data and performing image processing for automatically extracting a two-dimensional or three-dimensional target region using the data acquired by the preprocessing, a desired target region can be quickly obtained. Can be extracted.

  In the above image processing, approximation processing of density distribution to normal distribution, statistical region extraction processing by Bayes rule, target region correction processing by global and local statistical tests, and two-dimensional image for three-dimensionalization The desired target area is accurately extracted in a short time by performing the continuous two-area extraction process that repeats the process in the area where the already extracted area is limited to the middle area that is enlarged or reduced and separates the target area. be able to.

  Therefore, a plurality of images acquired by medical diagnostic imaging apparatuses such as an X-ray computer tomography (CT) apparatus, a nuclear magnetic resonance imaging (MRI) apparatus, and a positron emission tomography (PET) apparatus. Medical image processing apparatus, medical target region extraction method and medical target region extraction processing program capable of extracting a two-dimensional or three-dimensional target region in a short time from a two-dimensional tomographic image Can be provided.

It is a block diagram which shows the structure of the medical image processing apparatus to which this invention is applied. It is a flowchart which shows the processing content of the pre-processing part in the said medical image processing apparatus. It is a flowchart which shows the content of the target image selection process in the said pre-processing part. It is a figure which shows the display content by the display part in the said target image selection process, (A) shows an initial screen, (B) has shown the display content in the process image determination input reception process. It is a flowchart which shows the content of the area | region and data acquisition process in the said pre-processing part. It is a figure which shows the display content by the display part in the said area | region and data acquisition process, (A) shows the display content in the target area designation | designated input reception process, (B) is in the non-target area designation | designated input reception process. Is displayed. It is a flowchart which shows the processing content of the image process part in the said medical image processing apparatus. The image processing unit each approximate a normal distribution _H Hcc created from Hcc data and Liver data in _a diagram showing the _{H Liver,} (A) is Hcc data and Liver utilization image brightness adjusting processing in the preprocessing unit has been performed Each approximate normal distribution created from the data is shown, and (B) shows each approximate normal distribution created from the Hcc data and the Liver data not subjected to the use image luminance adjustment processing. It is a flowchart which shows the content of the threshold value determination and area | region extraction process in the said image processing part. It is a figure which shows an example of the histogram produced in the said threshold value determination / area | region extraction process. It is a flowchart which shows the content of the area | region extraction process on the YZ plane in the said image process part. It is a figure which shows the display content by the display part in the area | region extraction process on the said YZ surface, (A) shows a target candidate area | region (Hcc1), (B) extracts the maximum area in the said target candidate area | region (Hcc1). (C) shows an image in which the maximum area in the target candidate area (Hcc1) is extracted, and (D) shows a non-target candidate area (Live1), and (E) shows ( A non-target candidate area including the target candidate area obtained by adding (C) to D) is shown. (F) shows a binarized version of (E), and (G) shows a final non-target area. It is a flowchart which shows the content of the extraction process of the magnitude | size of the X direction on the XZ plane in the said image process part. It is a figure which shows the display content by the display part in the extraction process of the magnitude | size of the X direction on the said XZ surface, (A) shows the selection reception screen of the target area (Hcc), (B) is selected An image of the target area (Hcc) is shown. It is a figure which shows the display content by the display part in the extraction process of the magnitude | size of the X direction on the said XZ surface, (A) shows the designation input reception screen of length, (B) shows the measurement screen of length. Show. It is a flowchart which shows the content of the target area extraction process on XY plane in the said image process part. It is a figure which shows the display content by the display part in the said target area extraction process on the said XY surface, (A) shows the extraction screen of a target area (Hcc) and a non-target candidate area (Live), (B) is extraction. The target screen is shown, (C) shows the detection screen of the maximum area of the target area (Hcc), and (D) shows the area (Liver for Data) setting screen. It is a figure which shows an example of the histogram used for the target area extraction process on the said XY plane. It is a figure which shows the display content by the display part in the target area extraction process on the said XY surface, (A) shows the extracted target area superimposed on the original image, (B) was extracted. Although the non-target region is shown, the target region is also shown in a superimposed manner in order to show the relationship with the target region. It is a flowchart which shows the content of the target area correction process in the said image processing part. It is a flowchart which shows the content of the global objective area | region extraction correction | amendment process in the said objective area | region correction process. It is a figure which shows the display content by the display part in the said global target area correction process, (A) shows the target area (Hcc) which has already been detected, (B) shows the test execution area which expanded the target area (Hcc). Show. It is a flowchart which shows the content of the local target area | region extraction correction | amendment process in the said target area correction process. It is a figure which shows the display content by the display part in the said local target area | region extraction correction | amendment process, (A) shows the target area (Hcc) already detected, (B) is the test | inspection execution area which expanded the target area (Hcc). Is shown. It is a figure which shows the display content by the display part in the said local target area correction process. It is a figure which shows the display content by the display part in the said local target area correction process. It is a flowchart which shows the content of the 2 area | region continuous extraction process on different Z in the said image process part. It is a figure which shows the display content by the display part in the said 2 area | region continuous extraction process on said different Z, (A) shows an object candidate area | region (Hcc), (B) shows a non-object candidate area | region (Liver). Yes. It is a figure which shows an example of the histogram used for the 2 area | region continuous extraction process on the said different Z. It is a flowchart which shows the content of the new target area search range determination process in the said image process part. It is a figure which shows the display content by the display part in the said new destination area search range determination process, (A) shows an intermediate area ( _IHcc ), (B) shows the execution screen of an area _| region separation process, (C) is A new target area (Hcc _NEW ) is shown, and (D) shows the extraction result. It is a flowchart which shows the content of the search end determination process of the Z up-down direction in the said new destination area search range determination process. It is a flowchart which shows the content of the three-dimensional display process in the said image process part. It is a figure which shows the continuous extraction result of the Z direction used for the said three-dimensional display process, (A) shows an example of the continuous extraction result of a different Z upper direction, (B) is a Z continuous down extraction result. It is a figure which shows an example. It is a figure which shows the display result by the said three-dimensional display process, The three-dimensional display screen is shown and the two-dimensional display screen is shown. It is a figure which uses for description of extraction of a feature-value, (A) shows the two-dimensional abdominal CT image before operation, (B) has shown the extracted liver cancer. It is a figure which shows the case where the area | region filling process of the cancer dent used by two-dimensional shape evaluation is applied to three dimensions, (A) shows the image which the dent area | region was filled, and (B) shows the image before filling. , (C) shows a recessed area. It is a figure where it uses for description about extraction of the convex part field in the above-mentioned medical image processing device, (A) shows a horizontal section picture, (B) shows the state which laminated and constituted the cancer field extracted automatically, ( C) shows a three-dimensional image of visceral cancer. It is a figure with which it uses for description about the radius calculation of the inscribed circle by a midpoint method, (A) shows an object figure, (B) shows a structural element (circle), (C) has shown the shrinkage | contraction result. . It is a figure used for description of the extraction process of a convex part area | region, (A) shows an object figure, (B) shows the drawn inscribed circle, (C) shows the extracted convex part area | region. . It is a figure which uses for the description of extraction processing by making the difference with the three-dimensional image of cancer into a convex part area | region, (A) shows an object figure, (B) shows a structural element (sphere), (C) shows 2 pixels. A shaved convex region is shown. It is a figure which shows the discrimination | determination result classified into 2 classes (vp0 / vp1) by the method of extracting one by making the presence or absence of portal vein invasion (vp) into a correct answer (learning pattern). It is a figure which shows the survival curve created by the Kaplan-Meier method for every class of the discrimination result. It is a figure which shows the naked eye classification prescribed | regulated by the "primary liver cancer handling rule." It is a figure which shows the image of a pathological specimen photograph. It is a figure which shows the classification | category by the color of a cancer area | region, (a) is a white cancer, (c) is a green cancer (a lot of bile), (d) is a pink cancer, (e) is a black cancer, (f ) Indicates the bleeding area. It is a figure which shows a liver cancer area | region, (a) is a red liver cancer area | region, (b) of FIG. 47 is the whitened area | region where cirrhosis has occurred, (c) has shown the wrinkle of the liver. It is a figure which shows another area | region, (a) is a scale, (b) is a blue base of a background, (c) has shown reflection of light. It is a figure used for description of edge preservation | save smoothing process, (a) shows the image before edge preservation | save smoothing process application, (b) has shown the image after edge preservation | save smoothing process application. It is a figure which shows the result of having performed the discriminant analysis which makes unknown data the class with the highest probability by the Mahalanobis generalized distance. Discriminant analysis of cirrhosis is performed with two variables, the first feature amount Rate _A (ratio of the liver region around the cirrhosis region and the cancer region) and the second feature amount Rate _B (ratio of the liver region including cancer and the cancer region). It is a figure which shows the result. Discrimination of pink cancer by bivariate of second feature amount Rate _B (ratio of liver region including cancer and cancer region) and (third feature amount Rate _A ratio of cancer region around pink cancer region) It is a figure which shows the result of having analyzed. It is a figure where it uses for discrimination | determination of a black cancer, bleeding, and the reflection area of light.

  Hereinafter, the best mode for carrying out the present invention will be described in detail with reference to the drawings.

  The present invention is applied to, for example, a medical image processing apparatus 100 configured as shown in the block diagram of FIG.

  The medical image processing apparatus 100 includes a control unit 10, a preprocessing unit 20, an image processing unit 30, an input unit 40, a storage unit 50, a display unit 60, and the like. The medical image processing apparatus 100 is a computer system in which a predetermined processing program is installed, and the control unit 10 controls operations of the preprocessing unit 20 and the image processing unit 30 according to the processing program. It has become.

  The medical image processing apparatus 100 is a medical image processing apparatus that extracts a two-dimensional or three-dimensional target region from a plurality of two-dimensional tomographic images acquired by a medical image diagnostic apparatus, and includes a plurality of medical image diagnostic apparatuses acquired by the medical image diagnostic apparatus. In the present embodiment, a large number of CT images acquired by, for example, an X-ray computer tomography (CT) apparatus are stored in advance in the storage unit. It is assumed that the data is stored in the memory 50.

  As shown in the flowchart of FIG. 2, the control unit 10 controls the operation of the preprocessing unit 20 in accordance with the processing program so that the preprocessing unit 20 in the medical image processing apparatus 100 is processed. The unit 20 performs target image selection processing (S1), region and data acquisition processing (S2).

  That is, the pre-processing unit 20 has a target image selection function, a region, and a data acquisition function.

  In the target image selection process (S1) in the pre-processing unit 20, as shown in the flowchart of FIG. 3, an initial screen display process (S11), a phase selection input reception process (S12), and a processed image determination input reception process (S13). )I do.

  In the initial screen display process (S11), as shown in FIG. 4A, a three-section image of a YZ (sagittal) section, an XY (horizontal) section, and an XZ (head amount) section read from the storage unit 50. A1, A2, and A3 are simultaneously displayed on the display unit 60 together with attribute information A4 such as a patient number, pixel size, and image slice thickness.

  In the phase selection input receiving process (S12), an operation input by an input device such as a mouse or a keyboard of the input unit 40 is received, and an imaging condition of a different phase is determined by a selection operation input by a selection button displayed together with the attribute information A4. The “phase” of the condition suitable for achieving the purpose is selected from the image below.

  In the processed image determination input reception process (S13), an operation input by an input device such as a mouse or a keyboard of the input unit 40 is received, and a dotted line position (cursor) or +/− buttons for selecting an image in the display screen by the display unit 60. By the operation of A5, as shown in FIG. 4B, an optimal image is selected as a processing target. On the display screen by the display unit 60, the selected position information A6 is displayed.

  When the pre-processing unit 20 receives a manual operation input from the input unit 40 and selects an optimum image as a processing target in the target image selection process (S1), the pre-processing unit 20 performs a data acquisition process (S2) with the next region. .

  In the area and data acquisition process (S2) in the pre-processing unit 20, as shown in the flowchart of FIG. 5, the use image brightness adjustment process (S21), the target area designation input reception process (S22), and the target area image data acquisition Processing (S23), non-target region designation input acceptance processing (S24), and non-target region image data acquisition processing (S25) are performed.

  In the use image brightness adjustment process (S21), the brightness range of the image read from the storage unit 50 is adjusted and is stored within the brightness 0 to 255 level.

  In the target area designation input reception process (S22), an operation input by an input device such as a mouse or a keyboard of the input unit 40 is received, a cross section to be processed is selected, and the display unit 60 performs processing as shown in FIG. On the YZ plane displayed on the display screen, two diagonal points are selected inside the target area (Hcc) so that the line segment connecting the two points is a diagonal line substantially including the target area (Hcc). To do. Then, as shown in FIG. 6B, the rectangle determined by the diagonal line L1 is appropriately enlarged (or reduced) to obtain a temporary target area (Hcc).

  In the target area image data acquisition process (S23), luminance value data of peripheral pixels on each XY plane along the diagonal line L1 is acquired as Hcc data.

  In the next non-target area designation input reception process (S24), an operation input by an input device such as a mouse or a keyboard of the input unit 40 is received and displayed on the display screen by the display unit 60 as shown in FIG. On the YZ plane, two diagonal points inside the non-target region (Liver) are selected.

  In the non-target area image data acquisition process (S25), luminance value data of peripheral pixels on each XY plane along the line segment L2 connecting the two points is acquired as Liver data.

  Next, when the target image selection process (S1) and the area and data acquisition process (S2) by the preprocessing unit 20 are completed, the control unit 10 controls the image processing by the image processing unit 30.

  That is, as shown in the flowchart of FIG. 7, the control unit 10 controls the operation of the image processing unit 30 in accordance with the processing program, and displays the histogram as the processing content by the image processing unit 30 in the medical image processing apparatus 100 is shown in FIG. Generation and normal distribution approximation processing (S3), threshold determination / region extraction processing (S4), target region correction processing (S5), continuous two-region extraction processing (S6), search end determination processing (S7), three-dimensional display processing (S8) is performed.

In the histogram generation and normal distribution approximation process (S3) by the image processing unit 30, the Hcc data and the Liver data obtained by the preprocessing unit 20, that is, peripherals on each XY plane along the diagonal line L1. Histograms, averages and variances of the luminance value data of the pixels and the luminance value data of the surrounding pixels on each XY plane along the line segment L2 are obtained, and the luminance distribution of the pixels in the target area (Hcc) is obtained therefrom. and creating an approximate normal distribution _{H Liver} of the luminance distribution of the pixels of the normal distribution _{H Hcc} and untargeted area of approximation (Liver), and displays on the display screen by the display unit 60.

Here, in the generation of the histogram and the normal distribution approximation process (S3) by the image processing unit 30, each approximate normal distribution is obtained from the Hcc data and the Liver data subjected to the use image luminance adjustment process (S21) in the preprocessing unit 20. since H _Hcc, creating a _{H Liver,} each approximated normal distribution _H Hcc _{created, H Liver,} for example, as shown in FIG. 8 (a), correctly entirely housed in the inner to the luminance 0 to 255 levels It is an approximation.

The above-using image brightness adjusting processing (S21) is performed not even Hcc data and the approximated normal distribution from Liver data _{H Hcc ', H Liver'} If you have created, for example, as shown in FIG. 8 (B), Since a part of the brightness falls within the range of 0 to 255 and is not approximated correctly, the robustness of the target area extraction in the subsequent threshold value determination / area extraction process (S4) is greatly affected.

  In the threshold value determination / region extraction process (S4) in the image processing unit 30, as shown in the flowchart of FIG. 9, the threshold value determination process based on the Bayes rule (S41), the region extraction process on the YZ (sagittal cut) plane (S42), size extraction process in the X direction on the XZ (head cut) plane (S43), area extraction process on the XY (horizontal cut) plane (S44), on the next XY (horizontal cut) plane Region extraction processing (S45), region extraction processing on the XY (horizontal section) surface (S46), size extraction processing in the Z direction on the XZ (head frame) surface (S47), XZ (head amount) Section extraction processing (S48) on the plane and area extraction processing (S49) on the XY (horizontal section) plane.

Threshold value determination process based on Bays rule (S41), the generation of the histogram and the normal distribution approximation process (S3) by two approximate normal distribution _H Hcc _obtained, using _{H Liver,} these two approximations normal distribution _{H Hcc} , find the intersection _{H Liver,} and binarization threshold at Bays law (iEqual).

  Here, in the area division processing based on the Bayes rule, for example, as shown in FIG. 10, when binarizing an image composed of luminances having different normal distributions, the intersection point is set as a binarization threshold value (iEqual), and more than that 1 and less than 0. In the region division processing based on this Bayes rule, in FIG. 10, an error (Err1) that originally belongs to the low luminance region but is determined to be high luminance and an error (Err2) that originally belongs to the high luminance region but is determined to be low luminance are shown. ) (ErrT) is logically minimum.

  In the area extraction process (S42) on the YZ (sagittal cut) plane, as shown in the flowchart of FIG. 11, candidate area extraction process (S421) of the target area (Hcc) and the non-target area (Live), the final target Outer region determination processing (S422) and rectangular region determination processing (S423) surrounding the target region are performed.

  In the candidate area extraction process (S421), the target candidate area (Hcc1) is obtained by the binarization threshold (iEqual) on the YZ (sagittal section) plane determined by the threshold value determination process (S41) based on the Bayes rule. ) And a non-target candidate region (Liver1).

  For example, an image obtained by extracting the maximum area in the target candidate region (Hcc1) shown in FIG. 12B from the image showing the target candidate region (Hcc1) shown in FIG. As shown in (H), a target area (Hcc2) that is binarized by filling the inside thereof is used.

  For example, an image obtained by extracting the maximum area in the target candidate region (Hcc1) shown in FIG. 12B from the image showing the target candidate region (Hcc1) shown in FIG. As shown in (C), a binarized target area (Hcc2) is formed by filling the inside.

  In the next final non-target area determination process (S422), for example, from the image showing the target candidate area (Hcc1) shown in FIG. 12A extracted by the candidate area extraction process (S421), (B ), The image obtained by extracting the maximum area in the target candidate area (Hcc1) is obtained, and as shown in FIG. 12C, the inside is filled into a binarized target area (Hcc2). In addition, for example, in the image showing the non-target candidate area (Liver1) shown in (D) of FIG. 12 extracted by the candidate area extraction processing (S421), the inside of the target candidate area (Hcc1) shown in (C) of FIG. By adding an image obtained by extracting the maximum area of the target area, as shown in FIG. 12E, a non-target candidate area that includes the target candidate area is determined, and as shown in FIG. Filling the inside After binarization Te, as shown in (G) in FIG. 12, not even in non-target regions in the target area, the final non-target region excluding the example noise or vascular regions.

  In the rectangular area determination process (S423), based on the binarized target area (Hcc2) shown in (C) of FIG. 12 determined by the candidate area extraction process (S421), (H) of FIG. As shown in FIG. 4, a rectangular area surrounding the target area (Hcc2) is determined.

  In addition, in the size extraction process in the X direction on the XZ (head slice) plane (S43), as shown in the flowchart of FIG. 13, selection input reception for receiving selection input of the target area (Hcc) on the XZ plane. Processing (S431), length designation input acceptance processing (S432) for accepting a designation input of the length in the X (horizontal) direction in the selected target area (Hcc), and X (in the selected target area (Hcc) Width information measurement processing (S433) for measuring width information in the (horizontal) direction and width information storage processing (S434) for storing width information in the X (horizontal) direction are performed.

  That is, in the selection input receiving process (S431), the image processing unit 30 receives an operation input from an input device such as a mouse or a keyboard of the input unit 40, and as shown in FIG. The selection of the target area (Hcc) by the pull-down menu P in the menu bar on the displayed screen is accepted, and the image of the selected target area (Hcc) is displayed on another window as shown in FIG. Show on

  In the next length designation input acceptance process (S432), an operation input by an input device such as a mouse or a keyboard of the input unit 40 is accepted, and as shown in FIG. 15A, within the selected target area (Hcc). A designation input of the length in the X (horizontal) direction is received, and a line segment L3 having the designated length is drawn.

  In the next width information measurement process (S433), as shown in FIG. 15B, the length of the line segment L3 designated by the length designation input reception process (S432) is measured, and the target area is obtained. Width information in the X (horizontal) direction of (Hcc) is obtained.

  Even when the line segment L3 is specified with a slight inclination in the length designation input receiving process (S432), in the width information measurement process (S433), the X (horizontal) direction of the target area (Hcc). Only the length of is measured.

  In the next width information storage process (S434), the length (width information) in the X (horizontal) direction of the target area (Hcc) obtained by the width information measurement process (S433) is extended to 120%. Then, the position information Xmin, Xmax, Ymin, Ymax, Zmin, Zmax indicating the size of the target area (Hcc) is stored with the both end positions as Xmin, Xmax.

  In the area extraction process (S44) on the next XY (horizontal section) plane, the area extraction process (S441), the area restriction frame process (S442), the area maximum area detection process (S443), and the non-target area setting process (S444), normal distribution approximation processing (S445), threshold value determination processing S446), and region extraction processing (S447).

  In the region extraction process (S441), as shown in FIG. 17A, a target region (Hcc) and a non-target candidate region (Liver) are extracted on the Z section of interest on the same basis as the YZ plane.

  In the next area limit frame process (S442), as shown in FIG. 17B, a rectangular frame indicating the size of the target area (Hcc) indicated by the position information Xmin, Xmax, Ymin, Ymax, Zmin, Zmax. Is expanded, and only the inside of the rectangular frame S near the target area (Hcc) is extracted.

  In the next area maximum area detection process (S443), the maximum area (XYHccMaxYZ) of the target area (Hcc) in the rectangular frame S is detected as shown in FIG.

  In the next non-target area setting process (S444), as shown in FIG. 17D, a new area (Liver for Data) for acquiring luminance data of the non-target area is set.

  In the above explanation, the threshold value determination process (S41) based on the Bayes rule shown in FIG. 9 is first applied to the area extraction process (S42) on the YZ (sagittal cut) plane performed on the YZ plane. After the completion of this process, the process proceeds to the normal distribution approximation process (S445) described later.

  On the other hand, since the threshold value determination process (S41) based on the Bayes rule shown in FIG. 9 can be adopted, the method of first performing the above-described region extraction process (S46) on the XY (horizontal section) plane can be adopted. In this case, the same process as the above-described area extraction process (S42) on the YZ (sagittal section) plane is performed, and the area extraction process on the XY (horizontal section) plane (S46) with the corresponding appropriate plane as a target. As implemented.

  As described above, when processing is first started from the XY plane, the size of the target area on the XY plane and the surrounding rectangular area, that is, the target areas in the X and Y directions are detected. The size of the target area in the Z direction is determined by the size extraction process in the Z direction on the (headframe) plane (S47). After this process is completed, the process proceeds to a normal distribution approximation process (S445) described later.

  On the other hand, since the threshold value determination process (S41) based on the Bayes rule shown in FIG. 9 can be adopted, first, the method of performing the area extraction process (S48) on the XZ (head frame) plane can be adopted. In this case, the same process as the area extraction process (S42) on the YZ (sagittal cut) plane described above is performed, and the area extraction process on the XZ (head cut) plane is performed for the corresponding appropriate plane. This is implemented as (S48).

  As described above, when processing is first started from the XZ plane, the size of the target area on the XZ plane and the surrounding rectangular area, that is, the target areas in the X and Z directions are detected. The size in the Y direction of the target region is determined by the size extraction process in the Y direction on the (horizontal section) plane (S49). After this process is completed, the process proceeds to the next normal distribution approximation process (S445).

In the next normal distribution approximation processing (S445), the region (Liver for Data) set by the non-target region setting processing (S444) and the target region (Hcc) detected by the region maximum area detection processing (S443). From the histogram of the new luminance data of the maximum area (XYHccMaxYZ), the approximate normal distribution H _Hcc of the luminance distribution of the pixel of the target area (Hcc) of the image to be processed next and the approximate normal of the luminance distribution of the pixel of the non-target area (Live) A distribution H _Liver is created.

In the next threshold value determination process S446, the approximate normal distribution _HHcc of the luminance distribution of the pixel in the target area (Hcc) and the approximate normality of the luminance distribution of the pixel in the non-target area (Live) created by the normal distribution approximation process (S445). with distribution _{H Liver,} as shown in FIG. 18, these two approximations normal distribution _{H Hcc,} obtain the intersection of _{H Liver,} region segmentation based on Bays law to binarization threshold (IEqual) at Bays law ( A process of determining a threshold (iEqual) of binarization is performed.

  Then, in the next area extraction process (S447), using the binarization threshold value (iEqual) determined in the threshold value determination process S446, as shown in FIG. 19, two new areas are extracted in the XY plane. I do.

  In the target area correction process (S5), the image processing unit 30 performs a global target area correction process (S51) and a local target area correction process (S52) as shown in the flowchart of FIG.

In the global target area correction process 51, as shown in the flowchart of FIG. 21, a test necessity determination process (S511), a test rejection criterion setting process (S512), a test execution setting process (S513), and a test area search process (S514). ), Population data acquisition processing (S515), verification execution processing (S516), target region correction processing (S517), and rectangular region determination processing (S511).

  First, in the necessity determination process (S511), the necessity of the test is determined based on the total error value based on the histogram. If the test is not required, the results before the test shall be used.

  In the next test rejection criterion setting process (S512), the test rejection criterion is set based on the total error obtained from the histograms of the two regions.

  In the next test execution setting process (S513), the target area (Hcc) is expanded as shown in FIG. 22 (B) using the target area (Hcc) already detected as shown in FIG. 22 (A). The test execution area (Act) is set only around the target area.

  In the next test area search process (S514), pixels in the test execution area (Act) set by the test execution setting process (S513) are sequentially searched by raster scanning.

  In the next population data acquisition process (S515), a process for obtaining a population mean and a population standard deviation is performed using the region already determined as the target region as a population.

  In the next test execution process (S516), a Z test is performed using the average value of n pixels around the test target pixel, the population average of the population, and the population standard deviation. Do what you consider.

  In the next target area correction process (S517), a process for correcting the target area is performed based on the test result obtained by the test execution process (S516).

  In the next rectangular area determination process (S511), a rectangular area surrounding the target area determined on the XY plane in a certain Z that has been subjected to the area correction process by the target area correction process (S517) is determined. Based on the target area, the target area is determined on each of the upper and lower Z XY planes. The target area at a certain Z determined in this way is set as the initial target area.

  In the local target area correction process 51, as shown in the flowchart of FIG. 23, a test necessity determination process (S521), a test rejection criterion setting process (S522), a test execution setting process (S523), and a test area search process (S524). Then, a verification target pixel determination process (S525), a population data acquisition process (S526), a test execution process (S527), a target area correction process (S528), and a rectangular area determination process (S529) are performed.

  First, in the necessity determination process (S521), the necessity of the test is determined based on the total error value based on the histogram. If the test is not required, the results before the test shall be used.

  In the next test rejection criterion setting process (S522), a test rejection criterion is set based on the total error obtained from the histograms of the two regions.

  In the next test execution setting process (S523), the target area (Hcc) is enlarged as shown in (B) of FIG. 24 using the target area (Hcc) already detected as shown in (A) of FIG. The test execution area (Act) is set only around the target area.

  In the next test area search process (S524), only the pixels in the test execution area (Act) set by the test execution setting process (S523) are sequentially searched.

  In the next verification target pixel determination process (S525), it is determined to determine a non-tested pixel in the test execution area (Act), process the pixel as a test target, and register it as a tested pixel after the process is completed. The test is performed for all the pixels in the test execution area (Act).

  In the next population data acquisition process (S526), the test target pixel and a plurality of (n) pixel data around the test target pixel and a plurality (m) around the test target pixel only in the area already determined as the target area. A process of determining the area for acquiring the data of (n> n) pixels as the population data for local verification is performed.

  In the next test execution process (S527), a Z test is performed using the average value of the n pixels around the test target pixel and the average value and standard deviation of the local test population data as the population mean and population standard deviation. If rejected, the pixel is processed as a target area.

  In the next target area correction process (S528), as shown in FIG. 25, a process for correcting the target area (Hcc) is performed based on the test result of the test execution process (S527).

  In the next rectangular area determination process (S529), as shown in FIG. 26, the rectangular area S surrounding the target area determined on the XY plane in a certain Z subjected to the area correction process by the target area correction process (S528). Based on the target area (Hcc) in the rectangular area S, the target area (Hcc) is determined on each of the upper and lower XY planes. The target area (Hcc) at a certain Z determined in this way is set as the initial target area.

  In the continuous two-region extraction process (S6), the image processing unit 30 performs the initial target region screen setting process (S61), the two-region continuous extraction process (S62), and the new Z as shown in the flowchart of FIG. The two-region continuous extraction process (S63), the new threshold adoption determination process (S64), the new target area search range determination process (S65), and the search end determination process (S66) are performed.

  In the initial target area screen setting process (S61), a process of setting the target area in the initial Z as the initial target area screen is performed.

  In the next two-region continuous extraction processing (S62), as shown in FIGS. 28A and 28B, the target candidate region (Hcc) and the non-target candidate region are continuously changed while changing the plane to be processed in the Z direction. A process of extracting two areas of (Liver) is performed.

  In the two-region continuous extraction process (S63) in the next new Z, the XY plane on the new Z is obtained using the two-region data on the old Z (initially the initial target region, and then the next previous target region). The process of newly extracting two regions, the target region (Hcc) and the non-target region (Liver), is performed.

In the next new threshold adoption feasibility determination process (S64), when the size of the old target area (converted diameter) is smaller than a certain value, the old threshold is used as it is. Otherwise, as shown in FIG. object candidate region (Hcc) and untargeted candidate region (Liver) of the two regions histograms approximate normal distribution _H Hcc _data, performs processing used to determine the new threshold value for binarization (iEqual) from _{H Liver.}

  In the next new target area search range determination process (S65), as shown in the flowchart of FIG. 30, intermediate area determination process (S651), area separation process (S652), new area addition process (S653), extraction result display process (S654) is performed.

In the intermediate area determination process (S651), as shown in FIG. _31A , a process of determining an intermediate area (I _Hcc ) obtained by contracting and expanding the old target area (Hcc _OLD ) is performed.

In the next area separation process (S652), as shown in FIG. _31B , the area separation process is performed according to the _Bayes rule only in the intermediate area (I _Hcc ) determined in the intermediate area determination process (S651).

In the next new area addition processing (S653), as shown in (C) of FIG. 31, the new and the area new region separated by the segmentation process (S652) in addition to the old object region _{(Hcc OLD)} Processing for setting the target area (Hcc _NEW ) is performed.

  Then, in the next extraction result display process (S654), as shown in FIG. 31D, a process of displaying the extraction results at different Z together with the original image is performed.

  In the search end determination process (S7), as shown in the flowchart of FIG. 32, a target area extraction process (S71), an end condition determination process (S72), and a target area extraction process (S73) are performed.

  In the target area extraction process (S71), a process of extracting a plurality of two-dimensional target images in the Z upward direction is performed.

  In the end condition determination process (S72), the end condition is determined based on the size (height) of the target image obtained on the YZ plane.

  In the target area extraction process (S73), a process of extracting a plurality of two-dimensional target images in the Z downward direction is performed.

  In the three-dimensional display process (S8), as shown in the flowchart of FIG. 33, a storage process (S81) and a display process (S82) into a three-dimensional matrix are performed.

In the storing process into the three-dimensional matrix (S81), a process of storing a plurality of two-dimensional target region extraction results in the three-dimensional matrix is performed.

  In the next display process (S82), a three-dimensional matrix in which a plurality of two-dimensional target region extraction results are stored by the storage process (S81), for example, Z on different Z as shown in FIG. As shown in FIGS. 35A and 35B, a three-dimensional matrix storing the continuous extraction results in the direction and the continuous extraction results in the lower Z direction at different Zs as shown in FIG. 34B is used. As described above, the processing for displaying the target area in three dimensions is performed.

  In the medical image processing apparatus 100 as described above, the control unit 10 controls the operation of the preprocessing unit 20 according to the processing program, and the preprocessing unit 20 performs the target image selection process (S1), the area and data acquisition process ( By performing S2), it receives a designation input by a simple manual operation, acquires the approximate size simultaneously with the data of the target area from a plurality of two-dimensional tomographic images acquired by the medical image diagnostic apparatus, The image processing unit 30 performs image processing for acquiring the target area data and automatically extracting the two-dimensional or three-dimensional target area using the data acquired by the pre-processing, thereby shortening the desired target area. Can be extracted in time. Then, the control unit 10 controls the operation of the image processing unit 30 according to the processing program, and approximates the density distribution to a normal distribution (S3), statistical region extraction processing by Bayes rule (S4), Target area correction processing (S5) based on global and local statistical tests, and iterative processing on a two-dimensional image for three-dimensionalization are performed within an area where the already extracted area is limited to an enlarged / reduced intermediate area By performing the continuous two-region extraction process (S6) for separating the regions, a desired target region can be accurately extracted in a short time.

  Various functions of the preprocessing unit 20 and the image processing unit 30 whose operations are controlled by the control unit 10 according to the processing program are that a computer mounted on the medical image processing apparatus executes a medical target region extraction processing program. X-ray computed tomography (CT) apparatus, nuclear magnetic resonance imaging (MRI) apparatus, positron emission tomography (PET) apparatus, etc. By installing the above-described medical target region extraction processing program in a medical image diagnostic apparatus, a two-dimensional or three-dimensional target region can be quickly converted from a plurality of two-dimensional tomographic images acquired by the medical image diagnostic apparatus with a simple operation. Can be extracted.

The image data of the two-dimensional or three-dimensional target area (Hcc) extracted by the medical image processing apparatus 100 as described above is, for example, a feature amount such as the outline of the target area (Hcc) for evaluation / classification of liver cancer. Is extracted.
Note that the preprocessing unit 20 in the medical image processing apparatus 100 accepts an input of a rectangular area specified by a diagonal line as a designation input by a manual operation that designates the approximate size of the area of the image to be processed. However, it is not limited to a rectangular region, but may be a region having another shape such as a polygon, an arbitrarily-shaped closed curved surface, or a circle.

  For example, the feature amount of the target area (Hcc) is extracted as follows.

The following describes the contents of the two submitted manuscripts for feature extraction.
Here, the inventors of the present invention, for example, have a depression feature at the boundary of the liver cancer as shown in FIG. 36B extracted from a preoperative two-dimensional abdominal CT image as shown in FIG. The result shows that it corresponds well with the later survival curve.

  However, since there are many irregularities that cannot be seen in one CT image, shape evaluation using a three-dimensional image of liver cancer is performed.

37 (A), (B), and (C) are diagrams showing a case in which the area filling process is applied to the cancer depression used in the two-dimensional shape evaluation in three dimensions, and (A) is the depression. An image in which the area is filled is shown, (B) an image before being filled is shown, and (C) is an indentation area. As shown in (A), (B), and (C) of FIG. 37, when the area filling process is applied to the three-dimensional area of the cancer depression used in the two-dimensional shape evaluation, the time calculation amount is O (m ² n ² ) (m: number of slices, n: number of contour pixels per image), which is not practical as a system used in an actual medical field.

  Therefore, here, the difference between the liver cancer three-dimensional image extracted from the preoperative abdominal CT image and the inscribed sphere is extracted as the convex region feature.

  That is, in this medical image processing apparatus 100, as shown in FIGS. 38A, 38B and 38C, a three-dimensional image of liver cancer formed by stacking cancer regions automatically extracted for each horizontal cross-sectional image. Is used to extract the convex region based on the difference between the inscribed sphere obtained from the three-dimensional image of cancer and the cancer. FIG. 38 is a diagram for explaining the extraction of the convex region in the medical image processing apparatus 100, where (A) shows a horizontal cross-sectional image, and (B) is configured by stacking automatically extracted cancer regions. The state is shown, and (C) shows a three-dimensional image of visceral cancer.

  Prior to the description of the convex region extraction based on the difference between the inscribed sphere and the cancer, the principle will be described in two dimensions (inscribed circle).

  The inscribed circle here is “the largest circle that can be drawn in a certain figure”. Therefore, there may be a plurality of inscribed circles for one figure.

  In calculating the radius of the inscribed circle using the midpoint method, the radius is increased if the circle scanned within the target figure using the midpoint method fits within the figure, and the stub is reduced if it does not fit within the figure. The value when converged is the radius of the inscribed circle. Whether or not it fits within the figure is determined by the presence or absence of the own pixel remaining in the image contracted by the structural element of the circle of the target image, as shown in FIGS. 39 (A), (B), and (C). It is a figure with which it uses for description about the radius calculation of the inscribed circle by a midpoint method, (A) shows an object figure, (B) shows a structural element (circle), (C) has shown the shrinkage | contraction result. .

The initial conditions of the central method are shown below.
r ₁ = 0
r ₂ = √ ((xSize / 2) ² + (ySize / 2) ² )
Tolerance: 1 pixel, r ₁ : minimum radius range, r ₂ : maximum radius range, xSize: horizontal width of image, ySize: vertical width of image

  Further, the center of the inscribed circle is obtained from the target figure by the following procedure.

That is,
(Procedure 1) Shrink a half-circular circle obtained by calculating the radius of the inscribed circle by the midpoint method
(Procedure 2) The remaining pixel is expanded by the circular structure element of (Procedure 1).

  However, there may be a plurality of inscribed circles for the quantization error of the image or one figure, and there are as many center candidate points as the number remaining in (Procedure 1). As shown in B) and (C), the center candidate point closest to the center of gravity of the target image is drawn as the center of the inscribed circle. FIG. 40 is a diagram for explaining the extraction of the convex region, where (A) shows the target graphic, (B) shows the drawn inscribed circle, and (C) shows the extracted convex region. Show.

  Then, an inscribed sphere is created by changing a symmetric figure from a two-dimensional image to a three-dimensional image, and changing a structural element used for contraction / expansion from a circle to a sphere, and FIGS. 41 (A), (B), ( As shown in C), the difference from the three-dimensional cancer image is extracted as a convex region. FIG. 41 is a diagram for explaining the extraction process using a difference from a three-dimensional image of cancer as a convex region, where (A) shows a target graphic, (B) shows a structural element (sphere), and (C ) Shows a convex region where two pixels have been removed.

  In consideration of the influence of the quantization error of the image, the convex region is an inscribed sphere + a region where two pixels are removed from the outside.

  Since the amount of time calculation in creating the inscribed sphere is O (mn) (m: the number of slices, n: the number of contour pixels per image), it is more practical than the filling process of the recessed area.

  When the clinical usefulness is verified according to the following verification steps, the verification result that the three-dimensional shape feature of liver cancer is useful for classifying clinical data is obtained.

Verification step 1: A three-dimensional shape feature amount is obtained from the extracted convex region of the cancer.
Verification step 2: Create a discriminant function using post-operative information as a learning pattern.
Verification step 3: An unknown pattern (27 cases) is separated by a discriminant function.
Verification step 4: A survival curve is created by the Kaplan-Meier method.
Verification step 5: Test for significant differences between survival curves.

  Here, portal vein invasion (vp) was used as post-operative information. Portal vein invasion (vp) is a phenomenon in which cancer enters the portal vein and grows in the opposite direction to normal blood flow, and is a medically useful prognostic factor that is known after surgery. .

 In addition, as feature amounts used for shape evaluation from the extracted three-dimensional convex region, feature amounts representing “cancer boundary complexity” and “cancer size” were used, respectively.

  In addition, the volume ratio between the cancer and the convex region represented by the following formula (1) was used as the feature quantity having the highest degree of separation that is well separated depending on whether or not portal vein invasion (vp) is present.

  Furthermore, a radius r corresponding to a sphere of cancer represented by the following formula (2) was used as a feature amount representing “the size of cancer”.

 Then, when all the 27 cases were classified into 2 classes (vp0 / vp1) by a single extraction method with the presence or absence of portal vein invasion (vp) as a correct answer (learning pattern), the discrimination results as shown in FIG. 42 were obtained. .

  The total discriminant probability is 74% (the discriminant probability of vp0 is 82% and the discriminant probability of vp1 is 75%), which is a sufficiently high value from a medical point of view.

  A survival curve was created for each class of discrimination results by the Kaplan-Meier method, and the significance between the classes shown in FIG. Therefore, it can be said that there is a significant difference between the classes.

  That is, the three-dimensional shape feature of liver cancer is useful for classifying clinical data.

  Therefore, an inscribed circle or an inscribed sphere can be approximated from the principal component analysis of the target image, and a difference area from the target boundary can be used as a feature amount. Further, instead of an inscribed circle or an inscribed sphere, an inscribed ellipse or an inscribed ellipsoid may be approximated.

  In the binary case, an approximate circle or an approximate ellipse obtained by principal component analysis may be obtained, and a difference area from the target boundary may be used as an uneven feature.

  It is also possible to obtain approximate circles, approximate ellipses, and concavo-convex features from the music coordinate display of the target image and frequency analysis thereof.

  That is, the approximate circle diameter and approximate circle center can be obtained from the direct current component and the fundamental frequency component. Further, the approximate ellipse and approximate ellipse center can be obtained from the direct current component, the fundamental frequency component, and the second harmonic component.

  Further, when the region of the difference between the approximate ellipse and the target boundary is an uneven feature, the uneven feature may be obtained from the curved coordinate display of the target boundary and its frequency analysis-high frequency component.

  In addition, an approximate circle, an approximate ellipse, and an uneven feature can be obtained from the Fourier descriptor display of the target image and its frequency analysis.

  In addition, as a method of obtaining the inscribed circle and circumscribed circle of the target image at a high speed and a method for extracting the concavo-convex features using them, a method for obtaining the inscribed circle and circumscribed circle of the target image at a high speed by morphological calculation, There are a method of obtaining a difference area from the target image boundary as an uneven feature, a method of obtaining a difference area between the circumscribed circle and the target image boundary as an uneven feature, and a method of using the two features together.

  In the case of 3D, it is assumed that the 3D image is composed of a set of a plurality of 2D images, and many 3D images seem to have such a structure. It may be used after being decomposed, and the decomposition is easy.

  For example, in the case of 3D, the features on each surface obtained by applying the above-described various methods on a plurality of 2D planes are handled in combination. Normal statistical quantities such as mean, median, and frequent values can be easily considered as composite features.

  Conventionally, doctors and other experts have a habit of searching for “maximum cleavage plane = surface with the largest area of the target area” in the target area, and performing analysis / diagnosis on that plane. However, for example, in CT images, a set of horizontal sections is generally provided and used without being three-dimensional, and the "maximum cleavage plane" under limited conditions must be used. A maximum split plane extraction method that increases the area of the target region in the three-dimensional image can be employed.

  The maximum area of the target area on the three cross-sections that make up a three-dimensional image, typically the horizontal cross-section, sagittal cross-section, and head-head cross-section, is detected and the largest one is extracted. .

  The longest axis of the target image in the three-dimensional image is extracted (for example, the longest line segment connecting all the boundary pixels may be obtained), and the maximum area around the axis is obtained. In addition, the maximum split surface created by the first principal component and the second principal component of the three-dimensional target image is used. All the features on the two-dimensional surface of the above 2 can be applied on the obtained maximum split surface.

  In the three-dimensional case, an approximate sphere or approximate ellipsoid can be obtained from the principal component analysis of the target image, and the difference area between the approximate sphere or approximate ellipsoid and the target boundary can be used as an uneven feature.

  In this case, all of the three-dimensional features that are adopted in two dimensions can be applied.

  In addition, as a method of obtaining the inscribed sphere and circumscribed sphere of the target image at high speed and a method for extracting the uneven features using them, a method of obtaining the inscribed sphere and circumscribed sphere of the target image at high speed by morphological calculation, A method for obtaining a difference area between the target image boundary as the uneven feature, a method for obtaining a difference area between the circumscribed sphere and the target image boundary as the uneven feature, and the like can be employed.

  Furthermore, in this medical image processing apparatus 100, a color image obtained by excision of a cancer after removal at the maximum diameter is captured, and the presence or absence of portal vein invasion (vp) is evaluated, thereby objectively determining the malignancy of the cancer. , Has a function to perform quantitative morphological classification.

  As one prognosis estimation method for patients with liver cancer, a method is known in which a part of the liver containing cancer is excised and morphologically classified according to the macroscopic classification shown in FIG. 44 defined in the “Primary Liver Cancer Handling Rules”. However, this medical image processing apparatus 100 analyzes the pathological specimen photograph of a color image (256 gradations for each of RGB) as shown in FIG. By evaluating the presence or absence of portal vein invasion (vp), objective and quantitative morphological classification of cancer malignancy was performed as follows.

  That is, since the color of cancer varies depending on the degree of differentiation (maturity of cancer cells) and the region involved, as shown in FIGS. 46 (a) to (f) based on the characteristics of cancer presented by a doctor, (A) White cancer, (c) Green cancer (bile is much), (d) Pink cancer, (e) Black cancer. In addition, as shown in FIG. 46 (f), the inside of the cancer may be bleeding, and this state is included in the cancer region and classified into six types.

  The liver cancer region is red as shown in FIG. 47 (a), but when liver cirrhosis occurs, it becomes white as shown in FIG. 47 (b). In addition, as shown in FIG. 47 (c), the color of the dark blue platform is reflected in the liver folds.

  As other regions, as shown in FIGS. 48A to 48C, there are (a) a scale, (b) a blue base in the background, and (c) light reflection. The scale shown in FIG. 48 (a) is black and white. Further, as shown in FIG. 48 (c), the reflection of light generated at the time of photographing is in the liver including cancer.

  49 (a), the boundary between the cancer and the liver region is ambiguous. Therefore, by performing the edge preserving smoothing process as the preprocessing, the cancer as shown in FIG. 49 (b) is obtained. Each region was discriminated by discriminant analysis.

  In order to classify the cancer area into the above six types according to the difference in the color of the cancer area, the brightness value of each area in the target image was investigated. As a result, each area has a distribution of different brightness values in the RGB three-dimensional space. Turned out to be. Therefore, when discriminant analysis was performed with unknown data as the class having the highest probability based on Mahalanobis's general distance, the results shown in FIG. 50 were obtained, and the cancer area could be classified by the difference in color.

 As described above, the cancer region can be discriminated based on the difference in color. However, since each region has an overlap in the RGB three-dimensional space and an error occurs, detailed classification is performed for each connected region of each region.

First, cirrhosis is similar in color to a white cancer region, but since there are many liver regions in the surrounding region, the ratio of the liver region and the cancer region around the cirrhosis region is defined as a first feature amount Rate _A. The first feature amount Rate _A is expressed by the following equation (3).

In addition, since there are cases where most white cancers are misidentified as cirrhosis due to contrast at the time of imaging, the ratio of the liver region containing cancer to the cancer region is defined as a second feature amount Rate _B. The second feature amount Rate _B is expressed by the following equation (4).

Then, as shown in FIG. 51, the cirrhosis discriminant analysis was performed with the two variables of the first feature value Rate _A and the second feature value Rate _B.

The pink cancer is similar to the cirrhosis region, but since there are many white cancer regions in the surrounding region, the ratio of the cancer region around the pink cancer region is defined as the third feature amount Rate _C. The third feature amount Rate _C is expressed by the following equation (5).

Then, as shown in FIG. 52, a pink cancer discriminant analysis was performed with two variables of the second feature amount Rate _B and the third feature amount Rate _A.

  In addition, as shown in FIG. 53, the surrounding area of the black cancer, bleeding, and light reflection area is surrounded by cancer, so search is performed in 8 directions for each pixel of the black cancer, bleeding, light reflection area. If the cancer is in all directions, the cancer region is determined.

  Thereafter, opening, closing, and hole filling processing were performed on the maximum area connected region with a 5 × 5 structural element.

  By the above processing, 192 cases out of all 350 cases, that is, 54.8% of cases were extracted with an overextraction rate of less than 10% and an unextraction rate of less than 10%.

  For other cases, the user gives candidate points on the contour to be corrected, corrects the path between them using the Normalized Cost (NC) method, which is a method of searching for the path between them, and when contour tracking, the concentration gradient The local cost is determined so as to take a path with a large distance to the end point.

  This was applied to a region where the overextraction rate or the unextraction rate was 10% or more to perform contour correction.

  And the presence or absence of portal vein invasion (vp) is discriminate | determined with respect to the cancer area | region extracted in this way using the complexity of a cancer area | region, and the magnitude | size of cancer.

 Portal vein invasion (vp) ranges from vp0 (no involvement) to vp4 depending on the location of the portal vein involved, and the greater the value, the more malignant. Since vp2 or higher can be confirmed with the naked eye on preoperative CT, vp0 and vp1 are discriminated.

As a shape feature that represents the overall complexity of cancer,
Shape feature value = 1-4πS / L ²
S: Cancer area L: Cancer perimeter L
The 1-circularity indicated by is used.

In addition, as a feature quantity obtained from an image, a feature quantity representing the size of a cancer that is used for the malignancy of cancer is calculated from the cancer area S to R = 2√ (S / π).
The equivalent circle diameter R shown in FIG.

  The result of evaluating the recognition rate by applying one of these features to the 319 cases of vp0 or vp1 was 73.3%, which is medically sufficiently high.

  That is, as described above, color features such as cancer and liver, features of surrounding regions, and extraction of cancer regions by boundary correction, and evaluation of the complexity of the contour shape and the size of the cancer, can be trusted medically. Recognition rate can be obtained.

DESCRIPTION OF SYMBOLS 10 Control part, 20 Pre-processing part, 30 Image processing part, 40 Input part, 50 Storage part, 60 Display part, 100 Medical image processing apparatus

Claims (13)

A pre-processing unit that accepts a manual input and obtains the approximate size at the same time as the data of the target area, and performs pre-processing for acquiring the data of the non-target area;
An image processing unit that extracts a two-dimensional or three-dimensional target region using the data acquired by the preprocessing unit,
The pre-processing unit has a function of receiving a designation input by a manual operation for designating an approximate size of a region of an image to be processed,
The image processing unit obtains the pixel of the target region from the target region data acquired by the preprocessing unit, the data indicating the approximate size, the data of the non-target region and the data indicating the approximate size. A histogram of luminance data and a luminance data histogram of pixels in the non-target region are generated, a normal distribution approximation processing function for approximating each histogram to a normal distribution, and an approximate normal distribution of the target region obtained by the normal distribution approximation processing function A medical image processing apparatus characterized by having a region extraction function for determining a threshold value for region division according to Bayes rule using an approximate normal distribution of a non-target region and extracting a target region and a non-target region.   The medical image processing apparatus according to claim 1, wherein the preprocessing unit has an image luminance adjustment processing function for keeping a luminance range of a processing target image to be acquired within a predetermined luminance range.   The image processing unit uses only the periphery of the target area that has been enlarged as the detected target area as a test execution area, and performs raster scanning of pixels in the test execution area to obtain a plurality of pixels around the target area. A global target area correction process in which a Z test is performed using an average value and a population mean and standard deviation obtained as a population within an area that has already been determined as a target area, and the target area is corrected based on the test result. And an average of a plurality of pixels around the target area obtained by searching only the pixels in the test execution area, only the periphery of the target area obtained by enlarging the target area that has already been detected, A Z-test is performed using a population mean and a standard deviation obtained as a population within an area that has already been determined as a target area, and a local target area correction process is performed to correct the target area based on the test result. The medical image processing apparatus according to any one of claims 1 or claim 2, characterized in that it has a specific area correction processing function.   The image processing unit uses only the periphery of the target area that has been enlarged as the detected target area as a test execution area, and performs raster scanning of pixels in the test execution area to obtain a plurality of pixels around the target area. A global target area correction process in which a Z test is performed using an average value and a population mean and standard deviation obtained as a population within an area that has already been determined as a target area, and the target area is corrected based on the test result. And only the periphery of the target area that has already been expanded is set as the test execution area, and the test target pixel obtained by searching only the pixels in the test execution area and a plurality of (n) pixels around it Z test is performed using the average value and the average value and standard deviation of multiple (m> n) pixels around the test target pixel only in the area already determined as the target area as the population mean and population standard deviation. , Its inspection 3. The medical image processing apparatus according to claim 1, further comprising a target area correction processing function for performing a local target area correction process for correcting a target area based on a predetermined result. 4. .   The image processing unit performs a repetitive process on a two-dimensional image for three-dimensionalizing an extracted target area in a region limited to an intermediate region obtained by enlarging or reducing the extracted region. The medical image processing apparatus according to claim 4, further comprising an automatic two-region extraction processing function. A pre-processing step for accepting a designation input by manual operation, obtaining a rough size at the same time as the data of the target area, and performing a pre-process for acquiring the data of the non-target area;
An image processing step of extracting a two-dimensional or three-dimensional target region using the data acquired in the preprocessing step,
In the pre-processing step, a manual input for specifying the approximate size of the region of the image to be processed is accepted,
In the image processing step, from the data of the target area acquired in the preprocessing step and the data indicating the approximate size and the data of the non-target area and the data indicating the approximate size, the pixel of the target area is determined. A histogram of luminance data and a luminance data histogram of pixels in the non-target region are generated, a normal distribution approximation processing function for approximating each histogram to a normal distribution, and an approximate normal distribution of the target region obtained by the normal distribution approximation processing function A medical target region extraction method characterized in that, by using an approximate normal distribution of a non-target region, a threshold value for region division according to the Bayes rule is determined to extract a target region and a non-target region.   6. The medical target region extracting method according to claim 5, wherein in the preprocessing step, an image luminance adjustment process is performed so that the luminance range of the processing target image to be acquired falls within a predetermined luminance range.   In the image processing step, only the periphery of the target area obtained by enlarging the already detected target area is set as a test execution area, and a plurality of pixels around the target area obtained by raster scanning the pixels in the test execution area are obtained. A global target area correction process in which a Z test is performed using an average value and a population mean and standard deviation obtained as a population within an area that has already been determined as a target area, and the target area is corrected based on the test result. And an average of a plurality of pixels around the target area obtained by searching only the pixels in the test execution area, only the periphery of the target area obtained by enlarging the target area that has already been detected, A local target area correction process for performing a Z test using a population mean and standard deviation obtained as a population within an area that has already been determined as a target area, and correcting the target area based on the test result Medical purposes region extraction method according to any one of claims 5 or claim 6, characterized in that to perform.   In the image processing step, iterative processing on the two-dimensional image for converting the extracted target area into a three-dimensional image is performed in a region limited to the intermediate region obtained by enlarging or reducing the extracted region. 8. The medical target region extraction method according to claim 7, wherein a target two region extraction process is performed. A medical target region extraction processing program executed by a computer mounted on a medical image processing apparatus,
The above computer
A pre-processing unit that receives a specification input by manual operation, acquires the approximate size at the same time as the data of the target area, and performs pre-processing for acquiring the data of the non-target area, and the data acquired by the pre-processing unit To function as an image processing unit that extracts a two-dimensional or three-dimensional target area using
The pre-processing unit has a function of receiving a designation input by a manual operation for designating an approximate size of a region of an image to be processed,
The image processing unit obtains the pixel of the target region from the target region data acquired by the preprocessing unit, the data indicating the approximate size, the data of the non-target region and the data indicating the approximate size. A histogram of luminance data and a luminance data histogram of pixels in the non-target region are generated, a normal distribution approximation processing function for approximating each histogram to a normal distribution, and an approximate normal distribution of the target region obtained by the normal distribution approximation processing function A medical target region extraction processing program characterized by having a region extraction function for extracting a target region and a non-target region by determining a threshold value for region division according to the Bayes rule using an approximate normal distribution of the non-target region.   10. The medical target region extraction processing program according to claim 9, wherein the computer is caused to function as the pre-processing unit having an image luminance adjustment processing function for keeping the luminance range of the processing target image to be acquired within a predetermined luminance range.   Only the target area around the target area that has already been expanded is set as the test execution area, and the average value of a plurality of pixels around the target area obtained by raster scanning the pixels in the test execution area, and the target A global target area correction process for correcting the target area based on the Z test using the population mean and the standard deviation obtained as a population within the area determined as the area, and the already detected area Only the area around the target area that is the enlarged target area is set as the test execution area, and the average value of the plurality of pixels around the target area obtained by searching only the pixels in the test execution area is already determined as the target area. A target area correction processor for performing a Z test using a population mean and a standard deviation obtained as a population within a predetermined area, and a local target area correction process for correcting the target area based on the test result The medical target area extraction processing program according to any one of claims 9 or claim 10, characterized in that for the computer to function as the image processing unit having a.   Continuous two-region extraction processing function that separates the target region by performing iterative processing on the two-dimensional image to make the extracted target region three-dimensional within the region limited to the intermediate region where the already extracted region is enlarged or reduced The medical target region extraction processing program according to claim 11, wherein the computer is caused to function as the image processing unit including: