Medical image processing method and medical image processing apparatus

Abstract

A medical image processing method comprises the following steps: a segmentation step of segmenting a medical image; a selection step of selecting an image for medical image correction from the medical image as a key image based on the difference between the segmentation result and its answer data; and a display step of displaying the key image.

Description

TECHNICAL FIELD

[0001] The present application relates to a medical image processing method and a medical image processing apparatus. BACKGROUND

[0002] A technique is known in which a biological tissue (for example, a human organ) is extracted from a medical image by performing a segmentation process on the medical image. However, sometimes the biological tissue cannot be correctly extracted. A user (an expert) performs a correction of a region in which the biological tissue is indicated for an image in which the segmentation has not been correctly performed. For example, in a case where a plurality of slice images (medical images) constituting a medical image group are obtained by a CT (Computed Tomography) apparatus and a segmentation process is performed on the plurality of slice images, a user (an expert) confirms a result of the segmentation, and corrects a region in which a biological tissue (for example, a human organ) is indicated for an image in which the segmentation has not been correctly performed. The user has to confirm the result of the segmentation of the plurality of slice images constituting the medical image group one by one, and find out a slice image that needs to be corrected. For such a work, the more the number of slice images that are the objects of the segmentation process, the more difficult it is to appropriately select a slice image that needs to be corrected, and the burden on the user becomes greater.

[0003] PROBLEMS TO BE SOLVED BY THE INVENTION

[0004] An object of the present application is to provide a medical image processing method and a medical image processing apparatus that can appropriately select a medical image that needs to be corrected with respect to a result of a segmentation, and thereby can reduce a burden on a user. SUMMARY

[0005] A medical image processing method according to one embodiment of the present application includes: a segmentation step of segmenting a medical image; a selection step of selecting an image for correction of the medical image as a key image based on a difference between a result of the segmentation and answer data thereof; and a display step of displaying the key image.

[0006] It is preferable that, in the segmentation step, a medical image group obtained by scanning a plurality of scanning positions of an organ region is segmented.

[0007] It is preferable that the medical image processing method further includes an interpolation process step of correcting a result of the segmentation of the medical image group by an interpolation process, and the interpolation process is performed based on the key image selected in the selection step.

[0008] Preferably, the selecting step has a difference value calculating step, a difference value is calculated and the calculated difference value is stored, the difference value corresponds to each of the scan positions of the medical image group, and indicates a difference between the result of the segmentation of the medical image at the plurality of scan positions of the medical image group and the answer data thereof, in the selecting step, a plurality of scan positions are selected as key positions from the scan positions of the medical image group according to the record of the difference values, and the medical image corresponding to the key positions is set as the key image.

[0009] Preferably, the interpolation processing step has a user correction step, the result of the segmentation of the key image is corrected by a user, and the answer data of the segmentation of the key image is obtained, in the interpolation processing step, the result of the segmentation of a non-key image other than the key image is corrected by using an interpolation algorithm using the key image in the medical image group and the answer data of the segmentation of the key image obtained in the user correction step, and the answer data of the segmentation of the non-key image is obtained.

[0010] Preferably, in the selecting step, a difference value graph indicating the relationship between the difference values and the scan positions is drawn based on the record of the difference values, a plurality of candidate positions are determined for each of the difference value graphs, and the key positions are selected based on all the determined candidate positions.

[0011] Preferably, in the selecting step, for each of the difference value graphs, a plurality of the scan positions of the medical image group are substituted into the difference value graph, the position with the maximum difference value among the scan positions of the medical image group and the position on the horizontal axis of all the straight lines perpendicular to the horizontal axis that satisfy the following condition are set as the candidate positions, the condition refers to the area of the figure enclosed by the straight line perpendicular to the horizontal axis at the position with the maximum difference value, the difference value graph, and the horizontal axis being a multiple of a pre-set area value.

[0012] Preferably, in the interpolation processing step, the result of the segmentation of the non-key image is corrected by the intensity gradient of the boundary of the target tissue in the answer data of the segmentation of the key image.

[0013] Preferably, the segmentation is performed by a neural network, and the neural network is trained by generating teacher data from each medical image and the answer data thereof.

[0014] Preferably, in the difference value calculating step, a value obtained by dividing the area of the different part of the result of the segmentation of the medical image and the answer data thereof by the total area is taken as the difference value of the image.

[0015] The organ type determining step is preferably further provided for determining a type of an organ to which the medical image set corresponds, and in the key position selecting step, the key position is selected from all of the candidate positions according to the type.

[0016] The medical image processing apparatus of one embodiment of the present application includes a dividing section that divides a medical image, a selecting section that selects an image for medical image correction as a key image from the medical image based on a difference between a result of the division and answer data thereof, and a display control section that causes the key image to be displayed. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a schematic view showing that a set of lung region scan images is generated by scanning a lung region.

[0018] Figure 2 is a block diagram showing the structure of the image dividing apparatus of Embodiment 1 of the present application.

[0019] Figure 3 is a flowchart showing the flow of the image dividing method of Embodiment 1 of the present application.

[0020] Figure 4 is a flowchart showing the flow included in step S20 of the image dividing method of Embodiment 1 of the present application.

[0021] Figure 5 is a schematic view showing an image displayed by the display apparatus in step S105.

[0022] Figure 6 is a schematic view showing an image displayed by the display apparatus in step S106.

[0023] Figure 7 is an explanatory view for explaining the interpolation process of Embodiment 1 of the present application.

[0024] Figure 8 is a schematic view showing a difference value graph of the image dividing method of Embodiment 1 of the present application.

[0025] Figure 9 is a view for explaining one example of the determination method of the candidate position of Embodiment 1 of the present application.

[0026] Figure 10 is a block diagram showing the structure of the image dividing apparatus of Embodiment 2 of the present application.

[0027] Figure 11 is a flowchart showing the flow of the image dividing method of Embodiment 2 of the present application.

[0028] Figure 12is a block diagram showing the structure of an image segmentation device according to Embodiment 3 of the present application.

[0029] Figure 13 is a flowchart showing the flow of a segmentation step of an image segmentation method according to Embodiment 3 of the present application. DETAILED DESCRIPTION

[0030] [Medical images and scan positions]

[0031] Next, medical images used in the present application will be described. In the segmentation method of the present application, medical images are segmented in units of groups. A group of medical images is, for example, a plurality of organ scan images obtained by scanning a plurality of scan positions at regular intervals in a region of an organ of a living body.

[0032] Figure 1 is a schematic diagram showing a plurality of scan positions in a lung region, and a plurality of lung region scan images obtained by scanning the lung region at the scan positions. In Figure 1 In (a) of FIG. 8, each horizontal line represents a scan position. By scanning each position, a plurality of scan images showing the cross-sectional structure of the lung region corresponding to each scan position are obtained. Figure 1 (b), (c), and (d) of FIG. 8 respectively show scan images corresponding to three different scan positions. In Figure 1 In FIG. 8, only 14 scan positions are marked, but the number of scan positions and the interval between scan positions can be freely set as needed.

[0033] In the image segmentation method of the present application, a plurality of groups of medical images are processed, each group of medical images being obtained by scanning a plurality of scan positions at regular intervals in the same organ region of a patient (which can be the same patient or different patients). The number and interval of scan positions in each group of medical images can be different.

[0034] [Pre-segmentation]

[0035] In a semi-automatic segmentation method of medical images, a medical image is first pre-segmented using AI technology or a specialized segmentation algorithm, or the like. However, the accuracy of pre-segmentation is generally not sufficient to meet the accuracy requirements of medical image segmentation, and therefore, after pre-segmentation, the pre-segmentation result of the medical image needs to be further corrected.

[0036] A pre-segmentation method using a machine learning model is a highly efficient and accurate pre-segmentation method. In the case of pre-segmentation using a machine learning model, a medical image is input to a trained machine learning model, and information showing all pixels belonging to a target tissue in the image is output by the operation of the machine learning model, thereby completing the pre-segmentation of the image. The machine learning model includes various types of neural networks such as ordinary neural networks and convolutional neural networks.

[0037] <Embodiment 1>

[0038] The image segmentation method of the present embodiment is an image segmentation method in which a semi-automatic segmentation method is improved. In the present embodiment, a pre-segmentation result and a correction segmentation result (answer data) obtained by correcting the pre-segmentation result are stored for each image, and the pre-segmentation result and the correction segmentation result of each image are respectively recorded with information indicating all pixels of the image corresponding to a target tissue. In the following description, a case in which a medical image is segmented with the lung as the target tissue is described as an example, but is not limited thereto.

[0039] Next, Embodiment 1 of the present application will be described with reference to the drawings.

[0040] Figure 2 is a block diagram showing the structure of the image segmentation apparatus 10. The image segmentation apparatus 10 has a control section 110, a display control section 120, an input processing section 130, a storage section 140, a communication section 150, and a bus not shown.

[0041] In the image segmentation apparatus 10, the control section 110, the display control section 120, the input processing section 130, the storage section 140, and the communication section 150 are connected to the bus, and exchange information with each other via the bus. In addition, the image segmentation apparatus 10 is connected to a display apparatus and an input apparatus as external apparatuses via the communication section 150.

[0042] The control section 110 has a segmentation section 111, a difference value calculation section 112, a key position selection section 113, and an interpolation processing section 114, and causes the image segmentation apparatus 10 to execute an image segmentation method by executing a program stored in a program storage section 143 included in the storage section 140.

[0043] The display control section 120 controls an image displayed on a display apparatus connected to the image segmentation apparatus 10 according to an instruction from the control section 110. The display control section 120 is, for example, a graphics processor or the like. The display apparatus is, for example, a liquid crystal display or a plasma display or the like.

[0044] The input processing section 130 receives an input from an input apparatus, and converts input information from the input apparatus into information readable by a computer. The input processing section 130 is, for example, a communication interface such as a USB interface, and the input apparatus is, for example, a mouse, a keyboard, a touch panel, or the like.

[0045] The storage section 140 has an image storage section 141 that stores medical images, a segmentation storage section 142 that stores segmentation information of the medical images, a program storage section 143 that stores programs, a machine learning model storage section 144 that stores a machine learning model that has been trained, and a difference value storage section 145 that stores difference values associated with scan positions for each set of medical images.

[0046] The communication section 150 connects the image segmentation apparatus 10 and display apparatuses and input apparatuses that are external apparatuses by wired or wireless means.

[0047] The segmentation section 111 performs pre-segmentation of each set of medical images, for example, by inputting the set of medical images to the machine learning model, and stores the pre-segmentation results to the segmentation storage section 142. The manner of pre-segmentation is not limited thereto.

[0048] The difference value calculation section 112 calculates a difference value corresponding to the scan position of each set of medical images from the pre-segmentation results of the set of medical images and the corrected segmentation results obtained by correcting the pre-segmentation results, and stores the calculated difference value in association with the scan position to the difference value storage section 145. The difference value indicates the difference between the pre-segmentation results and the corrected segmentation results of each image corresponding to each scan position of a set of images.

[0049] The key position selection section 113 selects a plurality of key positions from the plurality of scan positions of each set of medical images based on the calculated difference values. The medical image corresponding to the key position is referred to as a key image, and the key image is used to correct non-key images in the interpolation process described later. The key position selection section 113 has a candidate position determination section that generates a difference value graph for the input difference values and determines candidate positions that are candidates for the plurality of key positions from each scan position based on the difference value graph. Details of the candidate positions are described later.

[0050] The interpolation processing section 114 corrects the pre-segmentation results of the remaining images in a set of images by an interpolation algorithm based on a part of the images in the set of images and their corrected segmentation results to obtain the corrected segmentation results of the remaining images.

[0051] Figure 3 is a flowchart showing the flow of the image segmentation method of the present embodiment. Hereinafter, the image segmentation method of the present embodiment will be described with reference to Figure 3 The image segmentation method of the present embodiment will be described.

[0052] In step S10, the control section 110 executes the program stored in the program storage section 143, starts the execution of the image segmentation method, reads the machine learning model stored in the machine learning model storage section 144, and causes the segmentation section 111 to load the machine learning model.

[0053] Step S20 includes steps S101 to S112, and detailed information about steps S101 to S112 will be described later. In step S20, a set of medical images is segmented.

[0054] In step S30, it is determined whether all of the medical image data has been processed, and if the determination is YES, the processing is ended, otherwise the processing returns to step S20.

[0055] Hereinafter, the details of the processing of step S20 will be described with reference to Figure 4 The details of the processing of step S20 will be described.

[0056] In step S101, the control section 110 reads a set of medical images from the image storage section 141, and inputs the set of medical images to the segmentation section 111. The set of medical images is, for example, a set of data composed of a plurality of medical images obtained by scanning a lung region.

[0057] In step S102, the segmentation section 111 performs pre-segmentation on the set of medical images and stores the pre-segmentation result in the segmentation storage section 142. Specifically, the segmentation section 111 inputs the set of medical images to the loaded machine learning model, thereby obtaining a pre-segmentation result calculated by the machine learning model.

[0058] In step S103, the control section 110 determines whether this execution is the first execution, and if the determination is YES, the processing proceeds to step S104, otherwise the processing proceeds to step S110.

[0059] In step S104, an expert (user) checks all of the images in the set of medical images and their pre-segmentation results, selects a plurality of images having poor segmentation accuracy and appropriate positional intervals therebetween as key images, and records the scan positions (key positions) corresponding to the plurality of key images, the key images being used to correct non-key images in the interpolation processing described later. The method of selecting the key images is not unique, and in general, the lower the pre-segmentation accuracy of the set of medical images, the more key images are needed for the interpolation processing described later, but considering that the more key images there are, the more manual segmentation is needed, the number of key images should be limited as much as possible. In addition, regarding the intervals between the key positions corresponding to the key images, in the case of low pre-segmentation accuracy, if the distance between two key positions is too long, it is not easy to correct the pre-segmentation results of other non-key images between the two key positions by the interpolation algorithm described later. However, if the distance is too short, the user needs to correct frequently, and therefore the positions of the key positions should be set evenly according to the actual situation.

[0060] In step S105, the display control section 120 causes the display device to display an organ map of the segmentation target (in this embodiment, the lung), and label the key positions on the corresponding positions of the organ map.Figure 5 is a schematic diagram representing the image displayed by the display device in step S105, in which Figure 5 the organ map of the lung is shown, and the key positions are marked on the organ map with horizontal lines.

[0061] In step S106, the pre-segmentation result of the key image is corrected by the user. Figure 6 is a schematic diagram representing the image displayed by the display device in step S106. As Figure 6 shown, if the user clicks on a certain key position marked on the organ map, the display control section 120 causes the display device to highlight the key position and display the current segmentation result of the image (key image) corresponding to the position, and then the user corrects the segmentation result of the key image according to the display of the display device by the input device, and the input processing section 130 receives the user input from the input device and updates the segmentation result of the key image according to the user input.

[0062] In step S107, the interpolation processing section 114 corrects the pre-segmentation result of the non-key images other than the key images in the current set of medical images according to the key images and the corrected segmentation result of the key images by an interpolation algorithm to obtain the corrected segmentation result of the non-key images, and stores the corrected segmentation result of the non-key images to the segmentation storage section 142.

[0063] Reference is made to Figure 7 to explain the interpolation processing of the present embodiment. Figure 7 is an explanatory diagram for explaining the interpolation processing of the present embodiment, in which Figure 7 (a) represents the true situation of the target tissue of the two key images (the uppermost and the lowermost) and the non-key images sandwiched between the two key images, (b), (c) columns respectively represent the segmentation situations of each image before and after the interpolation processing, with the correctly segmented part and the incorrectly segmented part of the target tissue represented by dark color and light color respectively. In Figure 7 the (b) column, for the segmentation result of the two key images at the uppermost and the lowermost, all the parts belonging to the target tissue have been segmented out since the correction by the user. For the two non-key images in the middle, only a part of the region belonging to the target tissue has been correctly segmented out in the pre-segmentation, so there are both dark color part and light color part. In Figure 7 the (c) column, after the interpolation processing of the segmentation result of the non-key images in the middle by the segmentation result of the key images on both sides, for the two non-key images in the middle, all the parts belonging to the target tissue have been segmented out, so the segmentation result only contains dark color part.

[0064] Next, the interpolation algorithm used in the present embodiment will be described. In the interpolation algorithm of the present embodiment, the intensity gradient of the boundary of the target tissue shown by the corrected segmentation result of the key image is used to find the correct boundary of the target tissue of the non-key image. In medical images, the boundary between tissues generally has a specific intensity gradient because each tissue exhibits a different intensity. In the interpolation algorithm of the present embodiment, first, the boundary of the target tissue in each key image is accurately found from the corrected segmentation result of each key image, and the intensity gradient of the boundary is calculated. Then, for all the other non-key images between each key image and the next key image, the pre-segmented boundary of the target tissue of each non-key image is found from the pre-segmentation result of the non-key image, and a boundary having an intensity gradient that matches the intensity gradient of the boundary of the target tissue of the key images at both ends of the non-key image is found within a certain range around the pre-segmented boundary of the target tissue to replace the pre-segmented boundary. Thereafter, the pixels included in the target tissue are determined by the corrected boundary of the target tissue, and the pre-segmentation result is corrected.

[0065] In addition, in addition to the interpolation algorithm using the intensity gradient of the boundary of the tissue, a shape interpolation algorithm can be used to correct the pre-segmentation result of the non-key image between each key image and the next key image based on the shape of the target tissue in each key image and the next key image.

[0066] In step S108, the expert determines whether the corrected segmentation result of each medical image after the interpolation process meets the desired accuracy, and if the determination is "Yes", the process proceeds to step S109, otherwise the process returns to step S106. The determination of step S108 is not necessary, and the determination of S108 can be omitted, and the process can directly proceed to step S109 after step S107 is completed.

[0067] In step S109, the control unit 110 stores the pre-segmentation result and the corrected segmentation result of all the images in the segmentation storage unit 142 of the storage unit 140.

[0068] In step S103, if the control section 110 determines that the image segmentation method according to the present embodiment is not executed for the first time, that is, in the case where the determination is "No", the process proceeds to step S110. In step S110, the difference value calculating section 112 calculates the difference value for each scan position based on the pre-segmentation result and the corrected segmentation result of the previous set of medical images stored in the segmentation storage section 142, and stores the calculated difference value in association with the scan position in the difference value storage section 145. Since the difference value of the previous set of medical images is calculated every time step S110 is executed, the difference values of each set of medical images up to the present are stored in the difference value storage section 145. In the present embodiment, the difference value at each scan position is obtained by dividing the area of the portion where the segmentation classification of the pre-segmentation result and the classification to which the corrected segmentation result belongs differ, by the total area, where the area of the portion where the segmentation classification of the pre-segmentation result and the classification to which the corrected segmentation result belongs differ is equal to the number of pixels in which the segmentation classification of the pre-segmentation result and the classification to which the corrected segmentation result belongs differ, and the total area is equal to the total number of pixels of the target tissue of the medical image. The method of setting the difference value is not limited to this, and another index indicating the difference between the pre-segmentation result and the corrected segmentation result can be used as the difference value.

[0069] In step S111, the candidate position determining section of the key position selecting section 113 plots a difference value graph for each set of medical images based on the difference values of all sets of medical images stored in the difference value storage section 145, and determines a plurality of candidate positions for each difference value graph. Figure 8 is a schematic view of a difference value graph indicating the image segmentation method according to Embodiment 1, where the horizontal axis indicates the continuous positions of the target tissue, that is, the lung, of the present embodiment, and the vertical axis indicates the difference value. First, the candidate position determining section reads the scan positions and the difference values associated with the scan positions of each set of medical images up to the present from the difference value storage section 145, and for each set of medical images, marks points corresponding to each scan position and the difference value of the set of medical images on the difference value graph, and then, for example, performs curve fitting based on these points to obtain a difference value curve. Then, the candidate position determining section determines a plurality of candidate positions for each difference value graph, and records these candidate positions. The difference value graph indicates the segmentation accuracy of the pre-segmentation for each position within a certain scan interval, and in the present embodiment, the difference value graph indicates the segmentation accuracy of the pre-segmentation for each position within the lung region, and the greater the difference value for a position, the lower the pre-segmentation accuracy corresponding to the position.

[0070] Next, the method of determining the candidate positions according to the present embodiment will be described, but this method is merely an example, and the method of determining the candidate positions is not limited to this method. The candidate positions are a plurality of positions suitable as the key positions of the scan interval, which are presumed from the difference value graph. Figure 9is a diagram for explaining one example of the candidate position determination method of the present embodiment. In the candidate position determination method of the present embodiment, for a certain difference value graph, first, the scan position of the present set of medical images being processed is substituted into the difference value graph, the difference value of each scan position of the present set of images on the difference value graph is calculated, and the scan position having the largest difference value is set as a candidate position. In Figure 9 , the position having the largest difference value is position pi, and therefore position pi is set as a candidate position. Then, positions on the horizontal axis where all straight lines that satisfy the following conditions are located are set as candidate positions, the straight lines being perpendicular to the horizontal axis and being a multiple of a predetermined area value of the area of the figure enclosed by the straight line perpendicular to the horizontal axis at the position having the largest difference value, the difference value graph, and the horizontal axis. Specifically, first, a straight line perpendicular to the horizontal axis is sought on the left side of position pi, such that the area of the figure enclosed by the straight line, the straight line 11 perpendicular to the horizontal axis at position pi, the difference value graph, and the horizontal axis is a predetermined area value, and the position where the sought straight line is located is also set as a candidate position, and Figure 9 , the area of the figure enclosed by the straight line 12, the straight line 11, the difference value graph, and the horizontal axis is the predetermined area value, and therefore position p2 corresponding to the straight line 12 is set as a candidate position. Thereafter, a straight line perpendicular to the horizontal axis is sought on the left side of the newly sought position p2, such that the area of the figure enclosed by the straight line, the difference value graph, and the horizontal axis is the predetermined area value, and the position where the straight line is located is set as a candidate position, and this process is repeated until a new candidate position cannot be sought. Thereafter, the same process is performed on the right side of the position having the largest difference value. After all candidate positions on the left side and the right side are sought, the candidate positions are recorded. The above-described predetermined area value is set according to the needs of the user, and if the user requires high segmentation accuracy, a lower area value can be set, and if the user requires low segmentation accuracy, a higher area value can be set. However, the lower the area value is set, the more images the user needs to manually correct.

[0071] In step S112, the key position selection section 113 selects a plurality of key positions from all the candidate positions determined in step S111 for each difference value graph. Since each candidate position can be different from the scan position of the present set of medical images being processed, it is necessary to associate each candidate position with each scan position of the present set of medical images being processed. Specifically, the key position selection section 113 associates all the candidate positions determined in step S111 with the scan position closest to the candidate position among each scan position of the present set of medical images being processed. After associating all the candidate positions with the scan positions, the probability of each scan position being selected as a candidate position is calculated based on the number of times each scan position is selected as a candidate position and the number of sets of medical images processed so far, and a scan position whose probability exceeds a threshold value set in advance is selected as a key position. In the present embodiment, for example, the threshold value is set to 0.6, and a scan position whose probability of being selected as a candidate position in the segmentation of past medical images exceeds 60% is selected as a key position.

[0072] The machine learning model or algorithm generally used for medical image segmentation outputs similar segmentation results for similar images (e.g., a plurality of medical images obtained by scanning a plurality of scan positions within a certain scan interval), and thus images with poor segmentation accuracy often correspond to the same position. In the present application, since the same machine learning model is used to pre-segment a plurality of sets of images obtained by scanning a plurality of scan positions within a certain scan interval, the positions corresponding to the images suitable for being used as key images, i.e., the positions suitable for being used as key positions, in each set of images are similar. By statistically processing the positions suitable for being used as key positions in each set of past images, it is possible to accurately select the positions suitable for being used as key positions for each set of images in the future.

[0073] According to the image segmentation method of the present embodiment, since the key position can be automatically and accurately selected to prompt the key image, the process of manually selecting the key image by the user is omitted, the work burden of the user can be reduced, and the work efficiency of image segmentation can be improved.

[0074] In addition, in the past, since the user could only select based on experience according to visual observation when selecting the key image, accurate calculation could not be performed, and thus an inappropriate medical image could be selected as the key image. According to the image segmentation method and the image segmentation apparatus of the present application, the key position is selected based on the difference value of each set of past medical images, and thus the key image can be accurately selected.

[0075] <Embodiment 2>

[0076] Next, Embodiment 2 of the present application will be described with reference to the drawings. In the present embodiment, the same reference numerals are assigned to portions common or equivalent to those of Embodiment 1, and repetitive description will be omitted.

[0077] In the present embodiment, a neural network is used as the machine learning model, and for example, a general neural network or a convolutional neural network can be used.

[0078] Figure 10 is a block diagram showing the structure of the image segmentation apparatus 10B of Embodiment 2 of the present application. The image segmentation apparatus 10B differs from the image segmentation apparatus 10 of Embodiment 1 in that the control section 110 further has a training section 115.

[0079] The training section 115 trains the machine learning model according to the result of segmenting the image. For example, the training section 115 trains the neural network stored in the machine learning model storage section 144 using teacher data. The teacher data is constituted by a plurality of medical images and their segmentation answer data. The training method used by the training section 115 can be, for example, a method such as batch gradient descent or stochastic gradient descent.

[0080] Figure 11 is a flowchart showing the flow of the image segmentation method of the present embodiment. Next, the image segmentation method of Embodiment 2 will be described with reference to Figure 11 The image segmentation method of the present embodiment will be described. The image segmentation method of Embodiment 2 further has a step S40B and a step S50B compared to Embodiment 1.

[0081] In the step S40B, the control section 110 determines whether the number of medical images of the preset group has been processed, and in the case of determining "Yes", proceeds to the step S50B, and otherwise returns to the step S20.

[0082] In the step S50B, the training section 115 generates teacher data from the medical images segmented so far and their corrected segmentation results, trains the neural network stored in the machine learning model storage section 144, and thereafter initializes the number of times of execution of the image segmentation method of the present embodiment to 0.

[0083] According to the image segmentation method of the present embodiment, the neural network is trained (updated) after the medical images of the preset group are segmented, and therefore the neural network that segments the medical images is trained while the medical images are segmented. If the pre-segmentation accuracy of the neural network improves, the number of selected key images decreases, and therefore the efficiency of segmentation can be further improved.

[0084] <Embodiment 3>

[0085] Next, Embodiment 3 of the present application will be described with reference to the drawings. In the present embodiment, the same reference numerals are assigned to portions common or equivalent to those of Embodiment 1, and repetitive description will be omitted.

[0086] In the present embodiment, each set of medical images is obtained by scanning a lung region, and the medical images are segmented with the lung as a segmentation target. Since the lung of each patient is different, for example, there are lungs deformed by external force or diseased lungs in addition to normal adult lungs, and so on, the scan images of the lung are classified into a plurality of kinds.

[0087] Figure 12 is a block diagram showing the structure of the image segmentation apparatus 10C of Embodiment 3 of the present application. The image segmentation apparatus 10C differs from the image segmentation apparatus 10 of Embodiment 1 in that it has a difference value storage section 145C instead of the difference value storage section 145, and the control section 110 further has an organ kind determination section 116.

[0088] The difference value storage section 145C stores, for each kind, the difference values of each set of medical images in association with the scan positions.

[0089] The organ kind determination section 116 determines the kind of the organ to which the set of medical images corresponds.

[0090] The image segmentation method of Embodiment 3 differs from Embodiment 1 in that Embodiment 3 has S111C, S112C instead of S111, S112, and further has S113C. Hereinafter, repetitive description will be omitted for the steps common to Embodiment 1.

[0091] Figure 13 is a flowchart showing the flow of the segmentation step of the image segmentation method of Embodiment 3.

[0092] In step S113C, the organ kind determination section 116 determines the kind of the organ to which the set of medical images corresponds.

[0093] In step S111C, the candidate position determination section of the key position selection section 113 plots each difference value curve from the difference values of each set of medical images of the kind same as the organ to which the set of medical images corresponds, and determines a plurality of candidate positions for each difference value curve.

[0094] In step S112C, the key position selection section 113 selects a plurality of key positions from the candidate positions determined in step S111 for the difference value curves of each set of medical images of the kind same as the organ to which the set of medical images corresponds.

[0095] According to the image segmentation method of the present embodiment, since the key positions are selected for each organ classification, the selection accuracy of the key positions can be further improved.

[0096] Several embodiments of the present application have been described, but these embodiments are presented by way of example only, and are not intended to limit the scope of the application. These new embodiments can be implemented in other various forms, and various omissions, substitutions, modifications, and the like can be made within the scope of the gist of the application. These embodiments and variations thereof are included in the scope or gist of the application, and are included in the scope of the application and its equivalents described in the claims.

Claims

1. A medical image processing method comprising the following steps: a segmentation step of segmenting a medical image group comprising a plurality of medical images and outputting a first result of the segmentation; A selection step of selecting a key image suitable as a correction target for the first result from the plurality of medical images based on a difference value between a second result of the segmentation performed in the past and a third result, which is answer data obtained by user correction of the second result; as well as The display step is to display the key image.

2. The medical image processing method according to claim 1, wherein: In the segmentation step, the medical image group is segmented, and the medical image group is obtained by scanning multiple scanning positions of the organ region.

3. The medical image processing method according to claim 2, wherein: The method further comprises an interpolation processing step of correcting the first segmentation result of the medical image group by interpolation processing. The interpolation process is performed based on the key image selected in the selection step.

4. The medical image processing method according to claim 3, wherein: The selecting step includes a difference value calculating step of calculating and storing difference values, wherein the difference values ​​correspond to the scanning positions of the medical image group, respectively, and represent the difference between the second result and the third result of the medical images at the plurality of scanning positions of the medical image group. In the selecting step, a plurality of scanning positions are selected from the scanning positions of the medical image group as key positions based on the records of the difference values, and the medical images corresponding to the key positions are set as the key images.

5. The medical image processing method according to claim 4, wherein: The interpolation processing step includes a user correction step of obtaining answer data of the segmentation of the key image by having the user correct the segmentation result of the key image. In the interpolation processing step, the segmentation results of non-key images other than the key image are corrected by using the interpolation algorithm of the key image in the medical image group and the segmented answer data of the key image obtained in the user correction step, and the segmented answer data of the non-key image is obtained.

6. The medical image processing method according to claim 5, wherein: In the selection step, a difference value curve diagram representing the relationship between the difference value and the scanning position is drawn based on the records of the difference value, multiple candidate positions are determined for each of the difference value curve diagrams, and the key position is selected based on all the determined candidate positions.

7. The medical image processing method according to claim 6, wherein: In the selection step, for each of the difference value curve graphs, the multiple scanning positions of the medical image group are substituted into the difference value curve, and the position with the largest difference value among the scanning positions of the medical image group and the positions on the horizontal axis where all straight lines perpendicular to the horizontal axis that meet the following conditions are located are set as the candidate positions, where the condition is that the area of ​​the figure enclosed by the straight line perpendicular to the horizontal axis at the position with the largest difference value, the difference value curve, and the horizontal axis is a multiple of a pre-set area value.

8. The medical image processing method according to any one of claims 5 to 7, wherein: In the interpolation processing step, the segmentation result of the non-key image is interpolated using the intensity gradient of the boundary of the target tissue in the segmented answer data of the key image.

9. The medical image processing method according to any one of claims 5 to 7, wherein: Segmentation is performed using a neural network, The neural network is trained by generating teacher data using each medical image and its segmented answer data.

10. The medical image processing method according to any one of claims 5 to 7, wherein: In the difference value calculation step, a value obtained by dividing the area of ​​the different parts of the second result and the third result of the medical image by the total area is used as the difference value of the image.

11. The medical image processing method according to claim 6 or 7, wherein: The method further comprises an organ type determination step for determining the type of organ corresponding to the medical image group. In the selecting step, the key position is selected based on all the candidate positions of the category.

12. A medical image processing apparatus, comprising: a segmentation unit, which segments a medical image group including a plurality of medical images and outputs a first result of the segmentation; a selection unit for selecting, from the plurality of medical images, a key image suitable as a correction target for the first result based on a difference value between a second result of the segmentation performed in the past and a third result, which is answer data obtained by user correction of the second result; as well as The display control unit causes the key image to be displayed.

Images