JP2012011109A - Medical image processor and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve comprehension of a detecting function for an abnormal shade pick on a medical image processor and program.SOLUTION: The medical image processor displays a sample image for a CAD processing (Step S22), and a CAD mark is displayed on a position of the abnormal shade pick on the sample image (Step S24). When operations to search similar cases are done (Step S27; YES), processing for searching the similar cases (Step S28) is done. Specifically, the position of the abnormal shade pick corresponding to the sample image, information on authenticity of the CAD mark, and the medical images that correspond to comments on the CAD mark are selected from a teaching file DB, and a predetermined number of medical images are selected from the selected medical images based on an amount of image feature of the sample image and that of the selected medical image in order of higher similarity. The CAD mark is applied on the position of the abnormal shade pick on the medical image selected as a similar case (Step S29).

Description

  The present invention relates to a medical image processing apparatus and a program.

  In recent years, a computer-aided diagnosis (CAD) function has been used in diagnosis of medical images obtained by radiographing a patient. This function is a function that analyzes a medical image, adds a mark to a region that seems to be a lesion, displays it on a display, and helps a doctor interpret. For breast images, abnormal shadow candidates that may be lesions such as tumors or microcalcifications are detected.

  In general, various medical devices are used in medical institutions. When a new medical device is introduced, it is necessary to learn how to handle the medical device each time. Therefore, educational software provides training on how to handle medical devices, performs an understanding test to check the understanding of the educational content, and issues a password that allows the user to use the medical device on the condition that the test passes. A medical device handling education system has been proposed (see Patent Document 1).

JP 2004-157177 A

  However, it has been difficult to learn how to use each function of CAD as compared to a general medical device. For example, it is desirable to understand each function of CAD by an actual case. In order to use CAD effectively, not only true positives (TP: True Positive) but also false positives (FP: False Positive) and false negatives (FN: False Negative) are understood as examples of abnormal shadows. There is a need to. Further, since the detection sensitivity of CAD can be set, it is necessary to be able to cope with the use of CAD at each sensitivity.

  The present invention has been made in view of the above problems in the prior art, and an object of the present invention is to improve the understanding of the abnormal shadow candidate detection function.

  In order to solve the above-described problem, the invention according to claim 1 associates the sample image prepared for each target of the abnormal shadow candidate detection function with the positional information of the abnormal shadow candidate detected from the sample image. First storage means stored and stored, second storage means for storing each of a plurality of medical images and positional information of abnormal shadow candidates detected from the respective medical images in association with each other, Operation means for selecting any one of the sample images stored in the first storage means, and the position information of the selected sample image and the abnormal shadow candidate corresponding to the sample image in the first storage Read out from the means, the selected sample image and a mark indicating an abnormal shadow candidate corresponding to the sample image are displayed on the display means, and the second image is stored in the second storage means. One or a plurality of medical images having similar detection results of the selected sample image and abnormal shadow candidates are extracted from the selected medical images, and correspond to the extracted one or more medical images and the medical images. And a control unit that displays a mark indicating an abnormal shadow candidate on the display unit.

  According to a second aspect of the present invention, in the medical image processing apparatus according to the first aspect, the control unit uses the positional information of the abnormal shadow candidate corresponding to the selected sample image and the image feature amount of the sample image. Based on the plurality of medical images stored in the second storage unit, one or a plurality of medical images having similar detection results of the selected sample image and abnormal shadow candidates are extracted.

  According to a third aspect of the present invention, in the medical image processing apparatus according to the first or second aspect, each of the abnormal shadow candidates detected from the sample image is further true positive in the first storage unit. Information indicating whether each of the abnormal shadow candidates detected from the respective medical images is true positive or false positive is stored in the second storage means. When the abnormal shadow candidate detected from the selected sample image is true positive, the control means stores a plurality of medical images stored in the second storage means. Of these, when only abnormal shadow candidates are true positives, and the abnormal shadow candidates detected from the selected sample images are false positives, a plurality of medical items stored in the second storage means Image Abnormal shadow candidate as a target only those false positive detection results of the selected sample image and the abnormal shadow candidate and extracts one or a plurality of medical images are similar.

  According to a fourth aspect of the present invention, in the medical image processing apparatus according to any one of the first to third aspects, the control unit is configured to use the medical image as a plurality of medical images stored in the second storage unit. Add information in the facility where the image processing device is installed.

  In the invention according to claim 5, the computer stores the sample image prepared for each target of the abnormal shadow candidate detection function and the positional information of the abnormal shadow candidate detected from the sample image in association with each other. Stored in the first storage means, the second storage means in which each of the plurality of medical images and the positional information of the abnormal shadow candidates detected from the respective medical images are stored in association with each other. The operation means for selecting any one of the sample images, the selected sample image, and the position information of the abnormal shadow candidate corresponding to the sample image are read from the first storage means, and the selected A sample image and a mark indicating an abnormal shadow candidate corresponding to the sample image are displayed on the display unit, and a plurality of medical images stored in the second storage unit are displayed. Then, one or a plurality of medical images having similar detection results of the selected sample image and abnormal shadow candidates are extracted, and the extracted one or more medical images and abnormal shadow candidates corresponding to the medical images are indicated. It is a program for functioning as control means for displaying a mark on the display means.

  According to a sixth aspect of the present invention, in the program according to the fifth aspect, the control means is based on position information of an abnormal shadow candidate corresponding to the selected sample image and an image feature amount of the sample image. From the plurality of medical images stored in the second storage means, one or a plurality of medical images having similar detection results of the selected sample image and abnormal shadow candidates are extracted.

  According to a seventh aspect of the present invention, in the program according to the fifth or sixth aspect, each of the abnormal shadow candidates detected from the sample image is further true positive or false positive in the first storage means. Information indicating whether or not each of the abnormal shadow candidates detected from each of the medical images is true positive or false positive is stored in the second storage means. And when the abnormal shadow candidate detected from the selected sample image is true positive, the control means stores the abnormal shadow among the plurality of medical images stored in the second storage means. If only the candidate is true positive and the abnormal shadow candidate detected from the selected sample image is false positive, among the plurality of medical images stored in the second storage means Abnormal Targeting only those shadow candidate is false positives, the detection result of the selected sample image and the abnormal shadow candidate is to extract one or more medical images are similar.

  According to the first and fifth aspects of the present invention, one or a plurality of medical images having similar detection results of the sample image and the abnormal shadow candidate are extracted, so that the understanding of the abnormal shadow candidate detection function can be improved. .

  According to the second and sixth aspects of the present invention, one or a plurality of medical images having similar detection results of the sample image and the abnormal shadow candidate are extracted based on the positional information of the abnormal shadow candidate and the image feature amount of the sample image. can do.

  According to the third and seventh aspects of the invention, it is possible to distinguish between true positive and false positive and extract one or a plurality of medical images having similar detection results of the sample image and the abnormal shadow candidate. it can.

  According to the invention described in claim 4, information suitable for each facility can be added.

It is a system configuration diagram of a medical information system. It is a block diagram which shows the functional structure of a medical image processing apparatus. It is a flowchart which shows the teaching file creation process performed in a medical image processing apparatus. It is an example of the input screen of finding information and a schema image. It is an example of a TP / FP selection column. This is an example in which a mark is added at the position of FN. It is an example of the comment regarding a CAD mark (FP). It is a flowchart which shows the CAD training process performed in a medical image processing apparatus. (A) is an example of a sensitivity setting screen. (B) is an example of the detection result of an abnormal shadow candidate when high sensitivity is selected. (C) is an example of a detection result of an abnormal shadow candidate when low sensitivity is selected. This is an example in which a CAD mark is displayed on a sample image. It is a flowchart which shows a similar case search process. It is an example of the medical image extracted as a similar case. It is a flowchart which shows the understanding level confirmation test execution process performed in a medical image processing apparatus.

Hereinafter, an embodiment of a medical image processing apparatus according to the present invention will be described with reference to the drawings.
FIG. 1 shows a system configuration of the medical information system 100. As shown in FIG. 1, a medical information system 100 includes a medical image processing apparatus 10, a CAD processing apparatus 20, a medical image server 30, and an electronic medical record system 40 that perform data communication via a network N such as a LAN (Local Area Network). It is connected and configured. The DICOM (Digital Imaging and Communications in Medicine) standard is applied to the network N.

  The medical image processing apparatus 10 is configured by WS (WorkStation), and is used for interpretation of medical images, training of CAD functions for medical images, and the like.

  FIG. 2 shows a functional configuration of the medical image processing apparatus 10. As shown in FIG. 2, the medical image processing apparatus 10 includes a control unit 11, an operation unit 12, a display unit 13, a communication unit 14, and a storage unit 15, and each unit is connected by a bus 16.

  The control unit 11 includes a CPU (Central Processing Unit), a RAM (Random Access Memory), and the like, and comprehensively controls the processing operations of each unit of the medical image processing apparatus 10. Specifically, the CPU reads various processing programs stored in the storage unit 15 according to an operation signal input from the operation unit 12 or an instruction signal received by the communication unit 14, and is formed in the RAM. The work area is expanded and various processes are performed in cooperation with the program.

  The operation unit 12 includes a keyboard having cursor keys, numeric input keys, various function keys, and the like, and a pointing device such as a mouse, and controls operation signals input by keyboard operations or mouse operations. 11 is output.

  The display unit 13 is configured by an LCD (Liquid Crystal Display) and displays various screens based on display data input from the control unit 11.

  The communication unit 14 is an interface that transmits and receives data to and from external devices such as the CAD processing device 20, the medical image server 30, and the electronic medical record system 40.

  The storage unit 15 is configured by a hard disk, a nonvolatile memory, or the like, and stores various data. Specifically, the storage unit 15 includes a teaching file creation processing program 151, a CAD training processing program 152, an understanding level confirmation test execution processing program 153, a sample image DB (DataBase) 154, a teaching file DB 155, a test image DB 156, and the like. Is remembered.

  The sample image DB 154 is a database that stores sample images prepared in advance for each target (imaging region / detection purpose) of the abnormal shadow candidate detection function. In the sample image DB 154, for each sample image, position information of abnormal shadow candidates detected from the sample image, CAD mark authenticity information (TP / FP), comments regarding the CAD mark, and the like are stored in association with each other. Has been. The true / false information of the CAD mark is information indicating whether each of the abnormal shadow candidates is true positive (TP) or false positive (FP). Note that the position information of the abnormal shadow candidate, the authenticity information of the CAD mark, and the comment related to the CAD mark are stored for each CAD sensitivity (high sensitivity / medium sensitivity / low sensitivity).

  The teaching file DB 155 is a database that stores medical images searched in a CAD training process (see FIG. 8) described later. The teaching file DB 155 includes, for each medical image, basic patient information (patient ID, age, imaging information, etc.), image feature amount information, positional information of abnormal shadow candidates detected from the medical image, and CAD mark truth. False information (TP / FP), comments about CAD marks, schema images, findings information, FN (false negative) position information, and the like are stored in association with each other. Note that the position information of the abnormal shadow candidate, the authenticity information of the CAD mark, the comment regarding the CAD mark, and the position information of the FN are stored for each CAD sensitivity.

  The image feature amount information is information including image feature amounts such as image density, grayscale gradient, and line structure concentration. The image density is an average or variance of pixel values obtained from the density histogram. The light / dark gradient indicates the gradient of density change of each pixel. For example, a gradient vector is calculated by a filter typified by an iris filter, and the shape of the gray value is grasped based on whether the gradient vector is directed to one point or whether the gray value is uniform as a whole. The degree of concentration of the line structure is a numerical value indicating whether a certain line structure is concentrated at one point or facing the same direction.

  The FN position information is information indicating the position of an abnormal shadow (false negative) that has not been detected by CAD for a medical image.

  In addition to medical images prepared in advance, information in a facility such as a hospital where the medical image processing apparatus 10 is installed can be added to the teaching file DB 155.

  The test image DB 156 is a database that stores test images used in an understanding level confirmation test execution process (see FIG. 13) described later. The test image DB 156 corresponds to each test image with position information of abnormal shadow candidates detected from the test image, CAD mark authenticity information (TP / FP), schema image, finding information, and the like. Attached and memorized.

  The control unit 11 reads from the sample image DB 154 the sample image selected by the user's operation from the operation unit 12 and the position information of the abnormal shadow candidate corresponding to the sample image, and stores the selected sample image and the sample image in the selected sample image. A mark (CAD mark) indicating a corresponding abnormal shadow candidate is displayed on the display unit 13.

  In addition, when adding a mark to the position of the abnormal shadow candidate, a circular mark surrounding the abnormal shadow candidate area may be added, or the mark may be added so as to illustrate the outline (boundary line) of the abnormal shadow candidate area. You may add (CAD outline), and you may add a mark to the representative point (for example, gravity center etc.) of an abnormal shadow candidate area | region.

  The control unit 11 extracts one or a plurality of medical images having similar detection results of the selected sample image and the abnormal shadow candidate from the plurality of medical images stored in the teaching file DB 155. Specifically, the control unit 11 selects from among a plurality of medical images stored in the teaching file DB 155 based on the position information of the abnormal shadow candidate corresponding to the selected sample image and the image feature amount of the sample image. Then, one or a plurality of medical images having similar detection results of the selected sample image and the abnormal shadow candidate are extracted.

  At this time, if the abnormal shadow candidate detected from the selected sample image is true positive, the control unit 11 is true positive among the plurality of medical images stored in the teaching file DB 155. When only a certain object is targeted and the abnormal shadow candidate detected from the selected sample image is false positive, the abnormal shadow candidate is false positive among a plurality of medical images stored in the teaching file DB 155. One or a plurality of medical images having similar detection results of the selected sample image and the abnormal shadow candidate are extracted for only a subject.

  The control unit 11 causes the display unit 13 to display one or more extracted medical images and a mark (CAD mark) indicating an abnormal shadow candidate corresponding to the medical image.

  The control unit 11 adds information in the facility where the medical image processing apparatus 10 is installed as a plurality of medical images stored in the teaching file DB 155.

  The CAD processing device 20 performs image analysis of a medical image, calculates an image feature amount, and detects an abnormal shadow candidate. Then, the CAD processing device 20 generates position information of abnormal shadow candidates, image feature amount information, and the like. The position information of the abnormal shadow candidate may be information indicating an area in the image (in the example of the breast image, the name of the U / M / L area of the MLO, the O / I area of the CC, etc.), May be the coordinate position. The CAD processing device 20 transmits position information, image feature amount information, and the like of abnormal shadow candidates to an external device such as the medical image processing device 10.

  The CAD detection sensitivity in the CAD processing apparatus 20 can be changed. Specifically, the detection sensitivity is changed by changing the threshold value of each parameter that is a criterion for determining whether or not to detect as an abnormal shadow candidate. When CAD processing is performed on the same medical image, normally, the higher the sensitivity of CAD, the more abnormal shadow candidates are detected.

  When the processing target is a breast image, the CAD processing apparatus 20 detects abnormal shadow candidates such as a tumor and microcalcification from the breast image. As an abnormal shadow candidate detection algorithm, a known algorithm can be applied. As an algorithm for detecting a tumor in a breast image, for example, a method using an iris filter, a Laplacian filter, or the like is applicable. As a detection algorithm for microcalcification, for example, a method using a morphological filter, a Laplacian filter, a triple ring filter, or the like can be applied.

  The medical image server 30 is configured by PACS (Picture Archiving and Communication System), stores image files of medical images generated by various modalities, and provides image files in response to requests from external devices.

  The electronic medical record system 40 stores and manages patient basic information, information indicating a medical condition and a diagnosis result, information on prescription, and the like for each patient.

Next, the operation will be described.
FIG. 3 is a flowchart showing teaching file creation processing executed in the medical image processing apparatus 10. This process is a process for adding a medical image stored in the medical image server 30 to the teaching file DB 155, and the teaching file creation processing program 151 stored in the CPU of the control unit 11 and the storage unit 15. It is realized by software processing in cooperation with.

  First, a medical image to be a target of a teaching file is selected from medical images stored in the medical image server 30 by an operation from the operation unit 12 by the user. Next, the selected medical image is acquired from the medical image server 30 by the control unit 11 via the communication unit 14, and basic patient information of the patient corresponding to the medical image is acquired from the electronic medical record system 40. And the acquired medical image and patient basic information are matched by the control part 11, and are registered into teaching file DB155 (step S1).

  Next, CAD sensitivity is set by the user's operation from the operation unit 12 (step S2). Specifically, as the sensitivity of CAD, one of high sensitivity, medium sensitivity, and low sensitivity is selected.

  Next, the CAD processing device 20 performs CAD processing at the sensitivity set in step S2 on the medical image that is the target of the teaching file. Then, the control unit 11 acquires the position information and the image feature amount information of the abnormal shadow candidate for the medical image that is the target of the teaching file from the CAD processing device 20 via the communication unit 14, and corresponds to the medical image. It is attached and registered in the teaching file DB 155 (step S3).

  Next, the control unit 11 displays a medical image as a teaching file target on the display unit 13 and adds a CAD mark to the position of the abnormal shadow candidate on the medical image based on the position information of the abnormal shadow candidate. Is displayed (step S4).

  Next, finding information is input by an operation from the operation unit 12 by the user (step S5). The input finding information is registered in the teaching file DB 155 by the control unit 11 in association with the medical image.

  Next, the position of the lesion is recorded as a schema image by the user's operation from the operation unit 12 (step S6). Specifically, a position that is determined to be a lesion is recorded in a schematic diagram showing an imaging region. The recorded schema image is registered in the teaching file DB 155 by the control unit 11 in association with the medical image.

4A and 4B show examples of input screens for finding information and schema images displayed on the display unit 13.
The input screen 50 shown in FIG. 4A includes a finding input field 51, a shape input field 52, a margin input field 53, a density input field 54, a category input field 55, and a schema image input field 56.
In the finding input field 51, a doctor's findings regarding the medical image are input.
In the shape input field 52, the shape of an abnormal shadow (tumor) is input. Examples of the shape include a round shape, an oval shape, a polygonal shape, a lobular shape, and an irregular shape.
In the margin input field 53, margin information of an abnormal shadow (tumor) is input. The edge information includes, for example, well-defined smooth (circumscribed), microlobulated, spiculated, indistinct, and difficult to evaluate (obscured).
In the density input field 54, density information of an abnormal shadow (tumor) is input. Concentration information includes fat-containing, equal density, low density, high density, and the like.
In the category input field 55, the grade of malignancy of the lesion classified into five levels is input. Category 1 is “no abnormality”, category 2 is “benign findings”, category 3 is “benign but cannot be denied malignancy”, category 4 is “suspected malignant”, and category 5 is “malignant”.
A schema image is displayed in the schema image input field 56, and the position of a lesion is recorded on the schema image.

The input screen 60 shown in FIG. 4B includes a finding input field 61, a form input field 62, a distribution input field 63, a category input field 64, and a schema image input field 65.
In the finding input field 61, a doctor's findings regarding the medical image are input.
In the form input column 62, form information of abnormal shadow (calcification) is input. Morphological information includes small round, punctate, amorphous or indistinct, polymorphism or heterogeneous, fine linear or fine branch (linear or fine) linear).
In the distribution input field 63, distribution information of abnormal shadow (calcification) is input. Distribution information includes diffuse / scattered, regional, grouped or clustered, linear, segmental, and the like.
In the category input field 64, the grade of malignancy of lesions classified into five levels is input.
The schema image is displayed in the schema image input field 65, and the position of the lesion is recorded on the schema image.

  Next, the user's operation from the operation unit 12 selects whether the CAD mark displayed on the display unit 13 is TP or FP (step S7). Information indicating whether it is TP or FP (true / false information of the CAD mark) is registered in the teaching file DB 155 by the control unit 11 in association with the medical image.

  For example, as shown in FIG. 5, with the CAD marks 71 and 72 displayed on the medical image 70, a TP / FP selection field 73 is displayed at a position associated with the CAD mark 71, and the CAD mark 72 and A TP / FP selection column 74 is displayed at the associated position. In the TP / FP selection field 73, “TP” is selected, and in the TP / FP selection field 74, “FP” is selected.

Next, the user determines whether or not the CAD result includes FN (missing) (step S8). When the CAD result includes FN (step S8; YES), a mark is added to the position of FN on the medical image displayed on the display unit 13 by the user's operation from the operation unit 12 (step S9). ). The position information of the FN is registered in the teaching file DB 155 by the control unit 11 in association with the medical image.
For example, as shown in FIG. 6, an FN mark 81 is added on the medical image 80.

  In step S8, when there is no FN in the CAD result (step S8; NO), or after step S9, a comment regarding each CAD mark is input by an operation from the operation unit 12 of the user (step S10). The input comment regarding the CAD mark is registered in the teaching file DB 155 by the control unit 11 in association with the medical image.

FIGS. 7A and 7B show examples of comments regarding the CAD mark (FP).
In the example shown in FIG. 7A, the CAD column 91 on the medical image 90 is misrecognized in the comment field 92 as “a part of the MLO U region where the mammary glands are scattered” as “construction disorder”. Is detected ".

  In the example shown in FIG. 7B, the CAD column 94 on the medical image 93 is erroneously detected as a “tumor” because “a density increase is observed locally and a left-right difference is also observed in the comment field 95. Is entered ".

  Next, the control unit 11 determines whether or not the operation unit 12 has performed an operation for changing the sensitivity (step S11). When an operation for changing the sensitivity is performed (step S11; YES), the process returns to step S2, and the same processing is repeated for different sensitivities.

  If the operation for changing the sensitivity is not performed in step S11 (step S11; NO), the registration completion button is pressed by the operation from the operation unit 12 by the user (step S12), and the teaching file is created. Processing ends.

  FIG. 8 is a flowchart showing CAD training processing executed in the medical image processing apparatus 10. This process is a process performed when the user learns the CAD function, and is realized by a software process in cooperation with the CPU of the control unit 11 and the CAD training process program 152 stored in the storage unit 15. .

  First, a sample image to be subjected to CAD processing is selected from sample images stored in the sample image DB 154 by an operation from the operation unit 12 by the user (step S21).

  Next, the selected sample image is read from the sample image DB 154 by the control unit 11, and the read sample image is displayed on the display unit 13 (step S22).

  Next, the CAD sensitivity is set by the user's operation from the operation unit 12 (step S23). Specifically, as the sensitivity of CAD, one of high sensitivity, medium sensitivity, and low sensitivity is selected.

  FIG. 9A shows an example of the sensitivity setting screen 110. On the sensitivity setting screen 110, one of high sensitivity, medium sensitivity, and low sensitivity is selected. FIG. 9B shows an example of the detection result of the abnormal shadow candidate when high sensitivity is selected. CAD marks 112, 113, and 114 are added on the sample image 111. FIG. 9C shows an example of detection results of abnormal shadow candidates when low sensitivity is selected. A CAD mark 116 is added on the sample image 115.

  Next, for the set sensitivity, the position information of the abnormal shadow candidate corresponding to the selected sample image, the authenticity information of the CAD mark, and the comment regarding the CAD mark are read from the sample image DB 154 for the set sensitivity. Then, based on the read position information of the abnormal shadow candidate, the control unit 11 adds and displays a CAD mark at the position of the abnormal shadow candidate on the sample image (step S24).

  FIG. 10 shows an example in which a CAD mark 121 is displayed on the sample image 120. FIG. 10 is an example (FP) in which the mammary gland in the U region (upper part of the breast) of the breast MLO image is erroneously detected as “construction disorder”, which is a finding of suspicion of breast cancer.

  Next, the control unit 11 determines whether or not the operation unit 12 has performed an operation for displaying the authenticity of the CAD mark (step S25). When an operation for displaying the authenticity of the CAD mark is performed (step S25; YES), the control unit 11 performs display based on the authenticity information of the CAD mark read from the sample image DB 154. Whether each of the CAD marks displayed at 13 is TP or FP is displayed (step S26). Further, a comment regarding the CAD mark may be displayed in correspondence with each CAD mark.

  In step S25, when an operation for displaying the authenticity of the CAD mark is not performed (step S25; NO), or after step S26, the control unit 11 searches the operation unit 12 for a similar case. It is determined whether or not an operation to that effect has been performed (step S27). When an operation for searching for a similar case is performed (step S27; YES), the control unit 11 performs a similar case search process (step S28).

Here, the similar case search process will be described with reference to FIG.
First, the control unit 11 calculates each image feature amount from the selected sample image (step S41). Specifically, the control unit 11 analyzes the sample image, and calculates an average or variance of pixel values obtained from the density histogram, a light / dark gradient, a degree of concentration of the line structure, and the like.

  Next, the control unit 11 extracts from the teaching file DB 155 the position information of the abnormal shadow candidate corresponding to the selected sample image, the authenticity information of the CAD mark, and the medical image that matches the comment related to the CAD mark (step S42). ). For example, regarding the position information of the abnormal shadow candidate, the region where the abnormal shadow candidate of both images exists between the sample image and the medical image in the teaching file DB 155 (in the case of a breast image, the U / M / L region of the MLO) , CC O / I regions, etc.) are determined to match. Further, regarding the authenticity information of the CAD mark, it is determined whether or not the abnormal shadow candidate detected from the sample image and the medical image in the teaching file DB 155 is TP or FP. The In addition, regarding the comment regarding the CAD mark, it is determined whether or not the keywords included in the comment match.

  Next, only the medical image extracted in step S42 is targeted by the control unit 11, and based on the image feature amount of the sample image calculated in step S41 and the image feature amount of the extracted medical image, A predetermined number (one or more) of medical images are extracted in order from the highest similarity (step S43). That is, medical images having high similarity to the sample image are narrowed down from the medical images extracted in step S42.

Here, a method for calculating the similarity between two images using the cos scale will be described.
Assign image feature values a1 to an of the sample image as a1 = average of density histogram, a2 = distribution of density histogram, a3 = gradient gradient, a4 = concentration of line structure,. A feature vector A is assumed.
A = (a1, a2, a3, a4, ..., an)

Further, the image feature amounts b1 to bn of the medical image are expressed as b1 = average of density histogram, b2 = distribution of density histogram, b3 = density gradient, b4 = concentration degree of line structure,. Assign and feature vector B.
B = (b1, b2, b3, b4, ..., bn)

  Cos θ (maximum value 1) is calculated based on the feature vectors A and B according to the equation (1), and the similarity between the two images is assumed to be higher in descending order of the calculated cos θ value.

  Returning to FIG. 8, the control unit 11 reads the medical image extracted as a similar case and the position information of the abnormal shadow candidate corresponding to the medical image from the teaching file DB 155, and displays the medical image extracted as the similar case. The CAD mark is added to the position of the abnormal shadow candidate on the medical image and displayed (step S29). Further, other information related to similar cases (comments regarding CAD marks, schema images, findings information, and FN position information) may be displayed.

  12A and 12B show examples of medical images extracted as similar cases. Specifically, this is an example in which the abnormal shadow candidate detected in the U region of the breast MLO image is FP. CAD marks 131 and 132 are displayed on the medical image 130 shown in FIG. 12A, and CAD marks 134 and 135 are displayed on the medical image 133 shown in FIG.

  In step S27, when an operation for searching for similar cases is not performed (step S27; NO), or after step S29, the control unit 11 performs an operation for changing the sensitivity in the operation unit 12. It is determined whether or not it has been performed (step S30). When an operation for changing the sensitivity is performed (step S30; YES), the process returns to step S23, and the same processing is repeated for different sensitivities.

  In step S30, when an operation for changing the sensitivity is not performed (step S30; NO), whether or not an operation for ending the CAD training process has been performed by the control unit 11 in the operation unit 12. Is determined (step S31). When the operation to end the CAD training process is not performed (step S31; NO), the process returns to step S21, and the same process is repeated for the other sample images to be CAD processed.

  In step S31, when an operation to end the CAD training process is performed (step S31; YES), the CAD training process ends.

  FIG. 13 is a flowchart showing an understanding level confirmation test execution process executed in the medical image processing apparatus 10. This process is a process performed when confirming the result of the CAD training process, and is a software process in cooperation with the CPU of the control unit 11 and the understanding level confirmation test execution processing program 153 stored in the storage unit 15. It is realized by.

  When the test is started (step S51), the control unit 11 causes the test image DB 156 to store any one of the test images and the position information of the abnormal shadow candidate corresponding to the test image, the authenticity information of the CAD mark, A schema image and finding information are read out. Then, the read test image set as the test target is displayed on the display unit 13 (step S52). Here, the user (test taker) performs normal interpretation of the test image.

  Next, based on the position information of the abnormal shadow candidate corresponding to the test image, the control unit 11 adds and displays a CAD mark at the position of the abnormal shadow candidate on the test image (step S53). Next, the user's operation from the operation unit 12 selects whether the CAD mark displayed on the display unit 13 is TP or FP (step S54). Specifically, the control unit 11 displays a TP / FP selection field for selecting a TP or FP at a position associated with each CAD mark, and the TP or FP is selected by an operation from the operation unit 12. Is done.

  Next, a finding report is input by an operation from the operation unit 12 by the user (step S55). Specifically, a lesion position (such as a schema image) and finding information are input as a finding report.

  Next, based on the authenticity information of the CAD mark corresponding to the test image, the control unit 11 determines whether the selection of TP or FP in step S54 is correct (step S54). S56). Specifically, the information indicating whether the CAD mark is TP or FP for each CAD mark and the authenticity information of each CAD mark corresponding to the test image selected by the control unit 11 in step S54. A comparison is made to determine whether or not they match.

  When the selection of TP or FP is correct (step S56; YES), the finding report is correct based on the schema image and the finding information corresponding to the test image by the control unit 11. Is determined (step S57). Specifically, the control unit 11 compares the lesion position input as the finding report in step S55 with the schema image corresponding to the test image to determine whether the lesion position matches. . In addition, the control unit 11 compares the finding information input as the finding report with the finding information corresponding to the test image, and determines whether or not the keywords match.

  When the finding report is correct (step S57; YES), the controller 11 counts up the number of correct answers (step S58).

  Here, the control unit 11 determines whether or not a predetermined number of problems have been completed (step S59). If the prescribed number of problems has not ended (step S59; NO), if the selection of TP or FP is not correct in step S56 (step S56; NO), or in step S57 If the finding report is not correct (step S57; NO), the process returns to step S52, and the processes of steps S52 to S59 are repeated.

  In step S59, when the prescribed number of problems is finished (step S59; YES), the control unit 11 determines whether or not the correct answer rate has reached the acceptance criterion (step S60). When the correct answer rate does not reach the acceptance standard (step S60; NO), the process returns to step S51, and the processes of steps S51 to S60 are repeated.

  In step S60, when the accuracy rate has reached the acceptance standard (step S60; YES), it is possible to log in to the interpretation system (step S61), and the understanding level confirmation test execution process ends.

  As described above, according to the medical image processing apparatus 10 in the present embodiment, in the CAD training process (see FIG. 8), one or a plurality of medical images (similar to the sample image and the abnormal shadow candidate detection result) are similar. Case) is extracted, it is possible to improve the understanding of the abnormal shadow candidate detection function. The user can easily learn the CAD function by interpreting sample images and similar cases.

  Further, based on the positional information of the abnormal shadow candidate and the image feature amount of the sample image, one or a plurality of medical images having similar detection results of the sample image and the abnormal shadow candidate can be extracted.

  Further, by distinguishing between true positive and false positive, it is possible to extract one or a plurality of medical images having similar detection results of the sample image and the abnormal shadow candidate. Thereby, the user can grasp the tendency of true positive / false positive detected by CAD.

  In addition, since a medical image stored in the medical image server 30 can be added to the teaching file DB 155 by teaching file creation processing (see FIG. 3), information suitable for each facility can be added.

  Further, by performing training of the CAD function at each sensitivity, the user can experience the difference in the CAD sensitivity setting and can cope with the CAD function at all sensitivities.

  In addition, the CAD training process (see FIG. 8) and the comprehension level confirmation test execution process (see FIG. 13) are performed periodically other than when the CAD processing apparatus 20 is first introduced or when the CAD processing software is upgraded. Thus, the interpretation accuracy of the user can be maintained.

  The description in the above embodiment is an example of the medical image processing apparatus according to the present invention, and the present invention is not limited to this. The detailed configuration and detailed operation of each part constituting the apparatus can be changed as appropriate without departing from the spirit of the present invention.

  For example, when a plurality of abnormal shadow candidates are detected for a sample image and a plurality of CAD marks are added to the sample image, cases similar to each of all existing CAD marks are searched. However, the user may be able to search for similar cases for each CAD mark by designating any one CAD mark by, for example, clicking a part of the CAD mark with a mouse.

  In the above description, an example in which a hard disk or a non-volatile memory is used as a computer-readable medium storing a program for executing each process is disclosed, but the present invention is not limited to this example. As another computer-readable medium, a portable recording medium such as a CD-ROM can be applied. A carrier wave may be applied as a medium for providing program data via a communication line.

DESCRIPTION OF SYMBOLS 10 Medical image processing apparatus 11 Control part 12 Operation part 13 Display part 14 Communication part 15 Storage part 16 Bus | bath 20 CAD processing apparatus 30 Medical image server 40 Electronic medical record system 100 Medical information system 151 Teaching file creation process program 152 CAD training process program 153 Comprehension confirmation test execution processing program 154 Sample image DB
155 Teaching file DB
156 Test DB
N network

Claims (7)

First storage means for storing a sample image prepared for each target of the abnormal shadow candidate detection function and positional information of the abnormal shadow candidate detected from the sample image in association with each other;
Second storage means for storing each of a plurality of medical images and positional information of abnormal shadow candidates detected from the respective medical images in association with each other;
Operating means for selecting any one of the sample images stored in the first storage means;
The selected sample image and the position information of the abnormal shadow candidate corresponding to the sample image are read from the first storage means, and the selected sample image and the mark indicating the abnormal shadow candidate corresponding to the sample image are displayed. One or a plurality of medical images that are similar to the selected sample image and the detection result of the abnormal shadow candidate are extracted from the plurality of medical images that are displayed on the second storage unit and extracted. Control means for causing the display means to display a mark indicating one or a plurality of medical images and an abnormal shadow candidate corresponding to the medical image;
A medical image processing apparatus comprising: The control means, based on the position information of the abnormal shadow candidate corresponding to the selected sample image and the image feature amount of the sample image, from among a plurality of medical images stored in the second storage means, Extracting one or a plurality of medical images having similar detection results of the selected sample image and abnormal shadow candidates;
The medical image processing apparatus according to claim 1. The first storage means further stores information indicating whether each of the abnormal shadow candidates detected from the sample image is true positive or false positive,
The second storage means further stores information indicating whether each of the abnormal shadow candidates detected from each of the medical images is true positive or false positive,
When the abnormal shadow candidate detected from the selected sample image is true positive, the control means is true positive among the plurality of medical images stored in the second storage means. When only a certain object is targeted and the abnormal shadow candidate detected from the selected sample image is false positive, the abnormal shadow candidate is false among the plurality of medical images stored in the second storage means. Extracting one or a plurality of medical images that are similar to the selected sample image and the detection result of the abnormal shadow candidate, targeting only those that are positive,
The medical image processing apparatus according to claim 1 or 2. The control unit adds information in a facility where the medical image processing apparatus is installed as a plurality of medical images stored in the second storage unit.
The medical image processing apparatus according to any one of claims 1 to 3. Computer
First storage means for storing a sample image prepared for each target of the abnormal shadow candidate detection function and positional information of the abnormal shadow candidate detected from the sample image in association with each other;
Second storage means for storing each of a plurality of medical images and positional information of abnormal shadow candidates detected from the respective medical images in association with each other;
Operating means for selecting any one of the sample images stored in the first storage means;
The selected sample image and the position information of the abnormal shadow candidate corresponding to the sample image are read from the first storage means, and the selected sample image and the mark indicating the abnormal shadow candidate corresponding to the sample image are displayed. One or a plurality of medical images that are similar to the selected sample image and the detection result of the abnormal shadow candidate are extracted from the plurality of medical images that are displayed on the second storage unit and extracted. Control means for causing the display means to display a mark indicating one or a plurality of medical images and abnormal shadow candidates corresponding to the medical images,
Program to function as. The control means, based on the position information of the abnormal shadow candidate corresponding to the selected sample image and the image feature amount of the sample image, from among a plurality of medical images stored in the second storage means, One or a plurality of medical images having similar detection results of the selected sample image and abnormal shadow candidates are extracted;
The program according to claim 5. The first storage means further stores information indicating whether each of the abnormal shadow candidates detected from the sample image is true positive or false positive,
The second storage means further stores information indicating whether each of the abnormal shadow candidates detected from each of the medical images is true positive or false positive,
When the abnormal shadow candidate detected from the selected sample image is true positive, the control means is true positive among the plurality of medical images stored in the second storage means. When only a certain object is targeted and the abnormal shadow candidate detected from the selected sample image is false positive, the abnormal shadow candidate is false among the plurality of medical images stored in the second storage means. For only positive ones, one or more medical images with similar detection results of the selected sample image and abnormal shadow candidates are extracted.
The program according to claim 5 or 6.