JP2016105796A - Medical diagnosis support device and medical diagnosis support method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To cause a doctor to present, as proper determinable information, how much a CAD remark becomes helpful in a diagnosis.SOLUTION: A medical diagnosis support device includes: a CAD remark creation unit 102 for analyzing examination data of a subject to create a CAD remark; an interpretation remark acquisition unit 105 for acquiring an interpretation remark of a doctor relative to the examination data; a remark management unit 103 for acquiring information related to a difference between the CAD remark and the interpretation remark; and a reliability evaluation unit 106 for evaluating reliability of the CAD remark on the basis of the information related to the difference between the CAD remark and the interpretation remark.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a technique for providing information that supports medical diagnosis.

  In the medical field, a doctor displays a medical image obtained by imaging a patient on a monitor, interprets the medical image displayed on the monitor, and observes the state of a lesioned part and changes over time. In an image diagnosis using a medical image, a doctor discovers an abnormal shadow or the like from a medical image that is a diagnosis target and diagnoses what the abnormal shadow or the like is. The interpretation report created by the doctor at the time of image diagnosis describes the result (estimated disease name) of determining what the detected abnormal shadow or the like is, and the interpretation findings that are the basis for the determination. At this time, interpretation findings and estimated disease names written by doctors were not always correct in the following cases. For example, if a doctor who is very busy and cannot spend a lot of time on a case or a doctor who has little experience in interpretation observes an abnormal shadow that is difficult to find on the image or an abnormal shadow that is difficult to detect, then the abnormal shadow There was a possibility of oversight, misinterpretation of interpretation findings and presumed disease names. In the following description, the estimated disease name is also considered as one of the “findings (interpretation findings)” in a broad sense.

  Therefore, for the purpose of providing information that improves the accuracy of image diagnosis by a doctor, medical image processing that automatically detects a diseased part or the like by digitizing a medical image and performing image analysis to perform computer-aided diagnosis Equipment has been developed. Hereinafter, the computer-aided diagnosis is referred to as CAD (Computer-Aided Diagnosis). In such CAD, abnormal shadow candidates are automatically detected as disease sites. In this abnormal shadow detection process, abnormal image shadows due to cancer or the like, high-density microcalcification shadows, and the like are detected by computer processing of medical image data. By presenting the detection result, it is possible to reduce the burden on the doctor's interpretation and improve the accuracy of the interpretation result.

  For example, in the apparatus described in Patent Document 1 below, after detecting an abnormal shadow by CAD, a finding (a part thereof) corresponding to a finding used for diagnosis by a doctor is automatically generated from the detected abnormal shadow. Then, the automatically generated findings and the interpretation findings written by the doctor with the first impression are automatically compared, and warning information is presented when the comparison results are different. Here, the findings automatically generated using CAD are referred to as “CAD findings”. With this technique, when the interpretation findings described by the doctor in the first impression are different from the CAD findings, the doctor can be encouraged to review the interpretation findings, and a highly accurate diagnosis can be realized.

Japanese Patent No. 3085724

  However, in Patent Document 1, if the CAD findings are always correct, a warning is always displayed if there is an error in the interpretation findings written by the doctor, but in reality, the CAD findings are not necessarily correct. Therefore, in actuality, even if the CAD findings are wrong and the doctor's interpretation findings are correct, a warning is presented, and even if the warning is presented, the doctor determines that the CAD findings at that time are correct. It wasn't possible. In other words, the CAD findings presented were not always useful for doctors' diagnosis.

  The present invention has been made in view of such problems, and an object of the present invention is to allow doctors to present to what extent CAD findings are useful as a reference for diagnosis.

The medical diagnosis support apparatus according to the present invention includes: a first generation unit configured to generate a plurality of items for diagnosis of the subject as a first generation finding by analyzing first test data of the subject; First acquisition means for acquiring an item for diagnosis of the subject based on the first examination data as a first interpretation finding based on an input from an operation unit, the first generation finding, and the A difference information acquisition unit configured to acquire information regarding a difference in each item included in the first interpretation finding; and analyzing the second examination data of the subject that is newer than the first examination data. A second generation unit that generates a plurality of items for specimen diagnosis as a second generation finding; and a finding presentation unit that presents information regarding the difference together with the second generation finding.
The present invention also includes a medical diagnosis support method and a medical diagnosis support system using the above-described medical diagnosis support apparatus.

  According to the present invention, it is possible to present to what extent an automatically generated finding (CAD finding) is a reference for diagnosis as information that can be appropriately determined by a doctor. Thereby, it becomes possible to improve the accuracy of a doctor's diagnosis.

It is a block diagram which shows an example of a function structure of the medical diagnosis assistance apparatus which concerns on 1st Embodiment. It is a block diagram which shows an example of the hardware constitutions of the medical diagnosis assistance apparatus which concerns on 1st Embodiment. It is a flowchart which shows an example of the process sequence of the medical diagnosis support method in the medical diagnosis support apparatus which concerns on 1st Embodiment. It is a schematic diagram which shows 1st Embodiment and shows the relationship between CAD findings and interpretation findings. It is a schematic diagram which shows 1st Embodiment and shows an example of the finding input screen in step S305 of FIG. It is a schematic diagram which shows 1st Embodiment and shows an example of the finding input screen after a doctor inputs an interpretation finding in step S309 of FIG. It is a flowchart which shows an example of the detailed process sequence in FIG.3 S306. It is a figure which shows 1st Embodiment and put together the reliability information of CAD finding for every finding item. FIG. 10 is a characteristic diagram illustrating the second embodiment and plotting feature amounts S in time series. It is a schematic diagram which shows 1st Embodiment and shows an example of the finding input screen in step S308 of FIG.

  Hereinafter, embodiments (embodiments) for carrying out the present invention will be described with reference to the drawings.

[First Embodiment]
In the medical diagnosis support apparatus according to the first embodiment, first, a CAD image (first automatic generation finding) is automatically generated by image analysis of a medical image (first examination data) including an abnormal shadow of a subject. . The means of the medical diagnosis support apparatus that generates the first automatic generation finding constitutes “first generation means”. In the medical diagnosis support apparatus according to the first embodiment, an interpretation finding (first interpretation finding) by a doctor with respect to the first examination data is acquired. The means of the medical diagnosis support apparatus that acquires the first interpretation finding constitutes “first acquisition means”.

  Subsequently, in the medical diagnosis support apparatus according to the first embodiment, a CAD finding (first automatic generation finding) is presented as an initial value of an interpretation finding (first interpretation finding). The means of the medical diagnosis support apparatus that presents the CAD findings (first automatically generated findings) constitutes “first presentation means”. In the medical diagnosis support apparatus according to the first embodiment, the doctor who refers to the CAD finding (first automatic generation finding) reads the CAD finding (first automatic finding) as the CAD finding (first automatic finding). Is acquired and recorded as information relating to the difference between the first automatic generation finding and the first interpretation finding. The means of the medical diagnosis support apparatus for acquiring information regarding the difference constitutes “difference information acquiring means”.

  Further, in the medical diagnosis support apparatus according to the first embodiment, when a new medical image (second examination data) of the same abnormal shadow of the same subject becomes a diagnosis target, the second examination data The CAD analysis (second automatic generation finding) is automatically generated by image analysis. The means of the medical diagnosis support apparatus that generates the second automatic generation finding constitutes “second generation means”. Further, in the medical diagnosis support apparatus according to the first embodiment, an interpretation finding (second interpretation finding) by a doctor with respect to the second examination data is acquired. The means of the medical diagnosis support apparatus for acquiring the second interpretation finding constitutes “second acquisition means”.

  Then, in the medical diagnosis support apparatus according to the first embodiment, when a new medical image (second examination data) of the same abnormal shadow of the same subject becomes a diagnosis target, the past first examination The reliability of the CAD finding (first automatic generation finding) is evaluated based on information on the difference between the CAD finding (first automatic generation finding) and the interpretation finding (first interpretation finding) in the data. Then, in the medical diagnosis support apparatus according to the first embodiment, the reliability information of the CAD findings (first automatically generated findings) in the evaluated first examination data is converted into the CAD findings (first in the second examination data). (2) Auto-generated findings) The means of the medical diagnosis support apparatus that makes this presentation constitutes “second presentation means”. Thereby, in the medical diagnosis support apparatus according to the first embodiment, the doctor can grasp the reliability of the CAD finding (second automatic generation finding) that is currently presented, thereby serving as a reference for diagnosis. Provide information.

FIG. 1 is a block diagram illustrating an example of a functional configuration of a medical diagnosis support apparatus 10 according to the first embodiment.
A medical diagnosis support apparatus 10 illustrated in FIG. 1 includes a medical information acquisition unit 100, an abnormal shadow detection unit 101, a CAD finding generation unit 102, a finding management unit 103, a finding presentation unit 104, an interpretation finding acquisition unit 105, and a reliability evaluation. Each functional configuration of the unit 106 is included.

  The medical information acquisition unit 100 acquires medical information (such as medical images and electronic medical record information) related to a case to be diagnosed from a server (not shown). Alternatively, the medical information acquisition unit 100 may connect an external storage device such as an FDD, HDD, CD drive, DVD drive, MO drive, ZIP drive, etc., and acquire medical information from these drives. Then, the medical information acquisition unit 100 outputs the acquired medical information to the abnormal shadow detection unit 101.

  The abnormal shadow detection unit 101 performs image analysis on the medical image that is a diagnosis target acquired from the medical information acquisition unit 100 to detect an abnormal shadow, and generates a CAD finding of a detection result (for example, information on a region of interest surrounding the abnormal shadow). To the unit 102.

  The CAD finding generation unit 102 generates a CAD finding from the detection result of the abnormal shadow acquired from the abnormal shadow detection unit 101 and outputs the CAD finding to the finding management unit 103.

  The finding management unit 103 includes information on CAD findings acquired from the CAD finding generating unit 102, information on interpretation findings acquired from the interpretation finding acquisition unit 105 described later (specifically, change information on CAD findings (hereinafter referred to as “finding change”). The finding management unit 103 stores the acquired information on the magnetic disk as finding information that forms part of the interpretation report regarding the abnormal shadow. Outputs the information to the observation presentation unit 104 and the reliability evaluation unit 106.

  The finding presenting unit 104 displays and presents the CAD findings acquired from the finding managing unit 103. In addition, when the finding presentation unit 104 acquires the reliability information of the CAD findings regarding each finding item from the reliability evaluation unit 106, the finding presentation unit 104 displays the information in association with the CAD findings.

  The interpretation finding acquisition unit 105 acquires interpretation finding information (specifically, in this embodiment, CAD finding change information (finding change information)) written by a doctor who refers to the information displayed by the finding presentation unit 104. And output to the finding management unit 103.

  When there is an image interpretation report obtained by previously interpreting an image having the same abnormal shadow as the abnormal shadow to be diagnosed, the reliability evaluation unit 106 stores the abnormal shadow stored in the past image interpretation report. Finding information is obtained from the finding management unit 103. And the reliability evaluation part 106 evaluates the reliability of CAD findings regarding each finding item based on the acquired finding information of the abnormal shadow, and obtains reliability information (details will be described later). Then, the reliability evaluation unit 106 outputs the evaluated reliability information to the finding presentation unit 104.

  Note that at least a part of each functional component of the medical diagnosis support apparatus 10 shown in FIG. 1 may be realized as an independent apparatus. Moreover, each function structure part of the medical diagnosis assistance apparatus 10 may be implement | achieved as software which each implement | achieves a function. In the present embodiment, each functional component of the medical diagnosis support apparatus 10 illustrated in FIG. 1 is realized by software.

  FIG. 2 is a block diagram illustrating an example of a hardware configuration of the medical diagnosis support apparatus 10 according to the first embodiment. That is, FIG. 2 is a diagram showing a basic configuration (hardware configuration) of a computer for realizing each function of each functional configuration unit shown in FIG. 1 by software.

  The medical diagnosis support apparatus 10 shown in FIG. 2 has hardware configurations of a CPU 200, a main memory 201, a magnetic disk 202, a display memory 203, a monitor 204, a mouse 205, a keyboard 206, and a common bus 207.

  The CPU 200 mainly controls the operation of each component, and comprehensively controls the operation in the medical diagnosis support apparatus 10.

  The main memory 201 stores a control program executed by the CPU 200 and provides a work area when the CPU 200 executes the program.

  The magnetic disk 202 stores an operating system (OS), device drives of peripheral devices, various application software including programs for performing processing described later, and the like.

  The display memory 203 temporarily stores display data generated by the finding presenting unit 104 shown in FIG.

  The monitor 204 is, for example, a CRT monitor or a liquid crystal monitor, and displays an image, text, or the like based on display data from the display memory 203.

  The mouse 205 and the keyboard 206 are used for pointing input and character input by the user, respectively.

  The common bus 207 connects the above-described components (200 to 206) so that they can communicate with each other.

  Here, for example, the medical information acquisition unit 100, the CAD finding generation unit 102, and the reliability evaluation unit 106 illustrated in FIG. 1 are stored in the CPU 200, the main memory 201, and / or the magnetic disk 202 illustrated in FIG. Is made up of programs. Further, for example, the abnormal shadow detection unit 101 and the finding management unit 103 illustrated in FIG. 1 include the CPU 200, the main memory 201, and / or the program stored in the magnetic disk 202 and the magnetic disk 202 illustrated in FIG. Has been. Further, for example, the finding presenting unit 104 shown in FIG. 1 includes the CPU 200, the main memory 201 and / or the program stored in the magnetic disk 202, the display memory, and the monitor shown in FIG. Further, for example, the interpretation finding acquisition unit 105 shown in FIG. 1 includes the CPU 200, the main memory 201 and / or the program stored in the magnetic disk 202, the mouse 205, and the keyboard 206 shown in FIG.

  Next, the overall processing performed by the medical diagnosis support apparatus 10 will be described using the flowchart of FIG.

  FIG. 3 is a flowchart illustrating an example of a processing procedure of a medical diagnosis support method in the medical diagnosis support apparatus 10 according to the first embodiment. In the present embodiment, the flowchart of FIG. 3 is realized by the CPU 200 executing a program that realizes the function of each unit stored in the main memory 201.

  First, in step S301 in FIG. 3, the medical information acquisition unit 100 acquires medical information (such as medical images and electronic medical record information) input to the medical diagnosis support apparatus 10 from a server (not shown) or the like. Then, the medical information acquisition unit 100 outputs the acquired medical information to the abnormal shadow detection unit 101.

  Subsequently, in step S <b> 302, the abnormal shadow detection unit 101 performs image analysis on the medical image to be diagnosed included in the medical information acquired from the medical information acquisition unit 100 and detects an abnormal shadow. Here, in the present embodiment, the medical image is a chest CT image, from which an abnormal shadow (tumor shadow) of the lung is detected. Then, for example, the abnormal shadow detection unit 101 automatically detects the presence position of the abnormal shadow by performing image analysis of the medical image by the CAD technique described in Patent Document 1, and further surrounds the abnormal shadow. Set the region of interest automatically. Then, the abnormal shadow detection unit 101 outputs the automatically set image of the region of interest to the CAD finding generation unit 102. In addition, the abnormal shadow detection unit 101 stores information indicating the range of the region of interest set automatically in the magnetic disk 202. A doctor may perform detection of abnormal shadows and setting of a region of interest. Further, when an abnormal shadow is detected in advance and stored together with medical information in a server (not shown) or the like, the configuration may be such that it is acquired together with the medical information.

  In step S303, the CAD finding generating unit 102 extracts an image feature amount from the region of interest of the abnormal shadow acquired from the abnormal shadow detecting unit 101, and generates a finding that can be used for diagnosis by the doctor based on this. Is a CAD finding. The CAD finding generation process may be performed based on any method that can be implemented in this field. For example, the CAD finding generation unit 102 generates a finding of finding item name: {size} from the size (maximum diameter) of a tumor that is one of the image feature amounts. Further, the CAD finding generation unit 102 generates a finding of the finding item name: {whole shape} using the unevenness degree of the edge that is one of the image feature amounts. Further, the CAD finding generation unit 102 generates a finding of the finding item name: {notch} based on the edge distance, which is one of the image feature amounts. Further, the CAD finding generation unit 102 can perform processing such as selecting the finding of finding item name: {radial} based on the dispersion of the edge distance, which is one of the image feature amounts. Since these processes do not represent the features of the present embodiment, description thereof will be omitted.

  Further, the CAD finding generation unit 102 infers the finding of the finding item name: {presumed disease name} using the image feature amount set extracted above. In this embodiment, the disease is classified into three disease names: {estimated disease name-primary lung cancer}, {estimated disease name-metastatic lung tumor}, and {presumed disease name-others}. For example, a support vector machine, a neural network, or a Bayesian network can be used as an inference unit for estimating a disease name. However, the reasoning device is not limited to this method.

  Then, the CAD finding generating unit 102 outputs the CAD finding generated above to the finding managing unit 103.

  In step S <b> 304, the finding management unit 103 determines whether or not a past interpretation report related to the abnormal shadow that is the same as the abnormal shadow that is the diagnosis target is stored in the magnetic disk 202. Here, the association between the abnormal shadow to be diagnosed and the abnormal shadow in the past interpretation report is performed by a known method, for example, based on the position information of the abnormal shadow in the lung field.

  If the result of determination in step S304 is that there is a past interpretation report regarding the same abnormal shadow as the abnormal shadow to be diagnosed, the finding information of the abnormal shadow is obtained from the magnetic disk 202 and the reliability evaluation unit 106, and the process proceeds to step S306. On the other hand, if there is no past interpretation report regarding the same abnormal shadow as the abnormal shadow to be diagnosed, the process proceeds to step S305.

  When the process proceeds to step S305 (when there is no past interpretation report), the finding presenting unit 104 displays the CAD finding acquired from the finding managing unit 103 together with the abnormal shadow detection result stored in the magnetic disk 202 in S302. The image is displayed on the monitor 204 in a form that allows interpretation. Then, the process proceeds to step S309.

FIG. 5 is a schematic diagram illustrating the first embodiment and an example of the finding input screen 500 in step S305 of FIG.
In the observation input screen 500 of FIG. 5, a medical image 501 to be diagnosed, a region of interest 502 surrounding an abnormal shadow detected by CAD, a region for inputting observation information (finding input region) 503, and an instruction to complete the input of interpretation findings An input completion button 504 for inputting is provided. The doctor interprets the image while referring to the medical image 501 displayed on the finding input screen 500, and refers to the region of interest 502 set by the CAD. At this time, when the doctor designates the region of interest 502, a CAD finding is displayed in the finding input region 503 as an initial value for the doctor to input an interpretation finding using a template-type input support method. Thus, by displaying the CAD findings as initial values for the doctor to input the interpretation findings, it is possible to save the trouble of the doctor inputting the interpretation findings.

Here, it returns to description of FIG. 3 again.
When the process of step S305 ends, the process proceeds to step S309.
In step S309, the interpretation finding acquisition unit 105 acquires the interpretation finding after the doctor changes the CAD finding with reference to the medical image and the CAD finding displayed on the monitor 204. In this embodiment, a doctor inputs interpretation findings of a medical image displayed on the monitor 204 using an operation unit such as a mouse 205 or a keyboard 206. Note that this process is realized by providing a function that can be selected by GUI using a template-type interpretation finding input support method, for example.

FIG. 6 is a schematic diagram illustrating an example of the finding input screen 600 after the doctor inputs an interpretation finding in step S309 in FIG. 3 according to the first embodiment. In FIG. 6, 600 to 604 represent the same as 500 to 504 in FIG.
In FIG. 6, interpretation findings 605, 606, and 607 respectively represent the interpretation findings after the CAD findings are changed by the doctor. In the example of FIG. 6, regarding the interpretation findings 605, the doctor changes the CAD findings {size-small} to {size-medium}. Further, regarding the interpretation findings 606, the doctor changes the CAD findings {radial-very strong} to {radial-moderate}. In addition, regarding the interpretation findings 607, the doctor has changed the CAD findings {estimated disease name-primary lung cancer} to {estimated disease name-metastatic lung tumor}. The doctor presses the input completion button 604 to complete the input of the interpretation findings.

  Then, the interpretation finding acquisition unit 105 acquires the interpretation finding after changing the CAD finding in the finding input area 603 when the doctor presses the input completion button 604, and outputs it to the finding management unit 103.

Here, it returns to description of FIG. 3 again.
When the process of step S309 ends, the process proceeds to step S310.
In step S310, the finding management unit 103 compares the interpretation finding input through the operation unit acquired in step S309 with the CAD finding, thereby changing the finding change information (what item finding is changed from what to what). Information indicating whether it has been done). Then, the CAD findings, interpretation findings, and finding change information acquired in the processes so far are stored and recorded in the magnetic disk 202 as findings information forming a part of the interpretation report regarding the abnormal shadow. In addition, in the finding change information, information “not considered changed” is recorded even when there is no actually changed finding. In this case, since the finding change information itself exists, when the determination process in step S304 is performed in a state where this information is recorded with the past abnormal shadow, the past interpretation report is present. A determination is made and then the process proceeds to step S306.

The above is the description of the process (the process after step S305) when there is no past interpretation report regarding the same abnormal shadow as the abnormal shadow to be diagnosed.
Next, processing (processing after step S306) when there is a past interpretation report regarding the same abnormal shadow as the abnormal shadow to be diagnosed will be described.

  In step S306 (when there is a past interpretation report relating to the same abnormal shadow as the abnormal shadow to be diagnosed), the reliability evaluation unit 106 relates to each finding item based on the past finding information acquired in step S304. Evaluate the reliability of CAD findings.

  In this step, if a CAD finding is changed for a certain finding item, it is considered that the doctor has determined that the CAD finding for that finding item is incorrect, and it is determined that the reliability of the CAD finding is not certain. To do. On the other hand, when the CAD finding is not changed with respect to a certain finding item, it is considered that the doctor has judged that the CAD finding of the finding item is correct, and it is determined that “the reliability of the CAD finding is reliable”.

Next, the determination process (the reliability evaluation process for CAD findings in step S306 in FIG. 3) performed by the reliability evaluation unit 106 described above will be described using the flowchart in FIG.
FIG. 7 is a flowchart showing an example of a detailed processing procedure in step S306 of FIG.

  When the process of step S306 in FIG. 3 starts, first, in step S701 of FIG. 7, the reliability evaluation unit 106 sets the finding number i = 1 of the CAD finding to be processed and obtains the number n of findings. .

  Subsequently, in step S702, the reliability evaluation unit 106 acquires the i-th CAD finding (CAD finding i) from the finding change information, and determines whether there is a record changed by the doctor. If the CAD finding i is changed as a result of the determination, the process proceeds to step S703. On the other hand, if the CAD finding i is not changed, the process proceeds to step S704.

  Considering the case where the example shown in FIG. 6 is a past abnormal shadow corresponding to the abnormal shadow to be diagnosed, the findings 1 {size}, findings 5 {radial}, and findings 20 {estimated disease name} , The doctor has changed the CAD findings. Therefore, if i = 1, 5, 20, the process proceeds to step S703. On the other hand, regarding findings 2 {overall shape}, findings 3 {clear bronchial image}, and findings 4 {cuts}, the doctor has not changed the CAD findings. Accordingly, if i = 2, 3, and 4, the process proceeds to step S704.

  In step S703, the reliability evaluation unit 106 receives the determination result that the CAD finding i has been changed in the past by the doctor, and determines that “the CAD finding i is not reliable”. Then, after the reliability evaluation unit 106 calculates the reliability indicating how reliable the CAD finding i is, the process proceeds to step S705.

In this embodiment, it is considered that the reliability of CAD findings is higher as the degree of change of CAD findings by the doctor is smaller, and conversely, the reliability of CAD findings is lower as the degree of change of CAD findings is larger. I reckon. Here, when the value of CAD finding i, which is a finding item, is expressed in terms of degree or probability, the maximum possible value of the value is Vi _MAX , the value of CAD finding i is Vi _PRE , and the interpretation finding after the change. Let the value of i be Vi _POST . At this time, the reliability Ri _CAD of the CAD finding i is expressed by the following equation.
Ri _CAD = (Vi _MAX − | Vi _POST −Vi _PRE |) / Vi _MAX (1)
However, if the CAD finding i itself cannot be expressed in terms of degree or probability, the degree of interpretation findings changed by the doctor cannot be evaluated, and the reliability is not calculated.

FIG. 4 is a schematic diagram illustrating the relationship between CAD findings and interpretation findings according to the first embodiment. In the example of FIG. 4A, for the finding 1 {size}, the doctor changes the CAD finding {size-small} to the interpretation finding {size-medium}. Therefore, the reliability is obtained as R1 _CAD = (4- | 2-1 |) /4=0.75. In the example of FIG. 4B, for the finding 5 {radial}, the doctor changes the CAD finding {radial-very strong} to the interpretation finding {radial-moderate}. Therefore, the reliability, R5 _CAD = (4- | 4-2 |) /4=0.5 is obtained. In other words, in the present embodiment, the reliability evaluation unit 106, as shown in FIG. 4, determines the reliability of the CAD findings based on the degree to which the doctor changed the CAD findings as interpretation findings after the CAD findings were presented. I try to evaluate the degree. In addition, regarding the finding 20 {presumed disease name}, the reliability cannot be calculated because the state of the finding cannot be expressed in terms of degree or probability.

Here, it returns to description of FIG. 7 again.
In step S704 in FIG. 7, the reliability evaluation unit 106 receives the determination result that the CAD finding has not been changed in the past by the doctor, and determines that “the CAD finding i is reliable”. Then, after calculating the CAD finding reliability as Ri _CAD = 1.0, the process proceeds to step S705. Here, in the example shown in FIG. 6, regarding the finding 2 {overall shape}, the finding 3 {transparent bronchial image}, and the finding 4 {incision}, the reliability of the CAD finding i is 1.0 (100%). Become.

  In step S705, the reliability evaluation unit 106 adds 1 to the finding number i.

  Subsequently, in step S706, the reliability evaluation unit 106 compares the finding number i with the number n of findings, and when the finding number i is larger than the number n of findings, the process of the flowchart in FIG. On the other hand, if the finding number i is equal to or less than the finding number n, the process proceeds to step S702. Thereafter, the processing after step S702 is performed again.

This is the end of the description of the flowchart of FIG.
As described above, the reliability of CAD findings in past images (past medical images) can be evaluated for each finding item.

  FIG. 8 shows the first embodiment, and is a table summarizing information on the reliability of CAD findings for each finding item. The reliability evaluation unit 106 outputs the CAD finding reliability information for each finding item described in FIG. 8 to the finding presenting unit 104.

Here, it returns to description of FIG. 3 again.
When the process of step S306 in FIG. 3 ends, the process proceeds to step S307.
In step S 307, the finding presenting unit 104 associates the reliability information of the CAD finding in the past image acquired from the reliability evaluating unit 106 with the CAD finding in the diagnosis target image acquired from the finding managing unit 103.

  In this step, if the reliability of the CAD finding in the past image is high, it is determined that the CAD finding is reliable in the diagnosis target image with respect to the same finding item.

  Therefore, in this step, reliability of CAD findings obtained from past images is associated with CAD findings for the diagnosis target image for each finding item.

  For example, regarding the finding item 1 {size}, information on “reliability of CAD findings: not certain (reliability 75%)” is associated with CAD findings in the diagnosis target image. Similarly, regarding the finding item 4 {notch}, “CAD finding reliability: certainty (reliability 100%)” is associated with the CAD finding of the diagnosis target image.

  Subsequently, in step S308, the finding presenting unit 104 displays on the monitor 204 the CAD finding of the diagnosis target image that has been subjected to the association processing of the reliability information of the CAD finding in the past image in step S307.

  FIG. 10 is a schematic diagram illustrating an example of the finding input screen 1000 in step S308 in FIG. 3 according to the first embodiment. In FIG. 10, reference numerals 1001 to 1004 denote the same as 601 to 604 in FIG. Specifically, the finding input area 1003 shows CAD findings in the diagnosis target image. Also, the finding input screen 1000 shown in FIG. 10 shows a warning message 1005 that warns that there is an error in the CAD finding, and CAD finding finding items 1006, 1007, and 1008 that are warned by the warning message 1005. . Further, the finding input screen 1000 shown in FIG. 10 shows messages 1009 and 1010 representing the reliability of CAD findings corresponding to the finding items 1006 and 1007 warned by the warning message 1005, respectively. At this time, regarding the finding item 1008 warned by the warning message 1005, the finding item is an estimated disease name, and the reliability is not calculated. Therefore, a message indicating the reliability of the CAD finding is not displayed.

  In FIG. 10, since the finding items 1006, 1007, and 1008 have been corrected (changed) in the past images and the reliability of the CAD findings in the diagnosis target image is determined to be uncertain, the warning message 1005 You have been warned. On the other hand, regarding the finding items 2, 3, and 4, the CAD findings are not corrected (changed) in the past images, and it is determined that the CAD findings are reliable in the diagnosis target images, and therefore, a warning is presented. Absent. In this way, by presenting a warning only for finding items for which reliability is determined to be uncertain, the doctor can determine how much the CAD findings are helpful.

Here, it returns to description of FIG. 3 again.
When the process of step S308 in FIG. 3 ends, the process proceeds to step S309.
In step S309, the interpretation finding acquisition unit 105 refers to the medical image displayed on the monitor 204 and the CAD finding, and acquires the interpretation finding after changing the CAD finding. Then, the interpretation finding acquisition unit 105 acquires the interpretation finding after changing the CAD finding in the finding input area 1003 when the doctor presses the input completion button 1004, and outputs it to the finding management unit 103.

  Subsequently, in step S310, the finding management unit 103 compares the interpretation finding acquired in step S309 with the CAD finding, thereby obtaining finding change information (information indicating which item has been changed from what to what). Extract. Then, the CAD findings, interpretation findings, and finding change information acquired in the processes so far are stored and recorded in the magnetic disk 202 as findings information forming a part of the interpretation report regarding the abnormal shadow. In addition, in the finding change information, information “not considered changed” is recorded even when there is no actually changed finding.

  This is the end of the description of the flowchart of FIG.

  In the present embodiment, whether or not to present a warning is changed depending on whether or not the reliability of CAD findings is reliable, but the display method is not limited to this. For example, reliability information associated with CAD findings may be always presented in step S306 in FIG. At this time, if the reliability is determined to be reliable, “reliability: certain” is always displayed, so that the doctor can recognize that the CAD findings of the findings are helpful.

  According to the configuration described above, the reliability information of the CAD findings is presented in association with the CAD findings for the diagnosis target image, so that the doctor can be used as a reference for determining the reliability of the CAD findings, which is useful for diagnosis. be able to.

(Modification 1 of the first embodiment)
In the first embodiment described above, the CAD findings are presented as initial values for inputting the interpretation findings in a template format, and information on the difference between the CAD findings and the interpretation findings depending on whether or not the doctor has changed the CAD findings. Was something to get. However, the method for acquiring information on the difference between CAD findings and interpretation findings is not limited to this. For example, a method may be used in which a doctor interprets an interpretation finding created without using a CAD finding as an initial value and a separately generated CAD finding and obtains information on the difference between the two. In this case, as long as an interpretation finding created by a doctor is recorded in advance, a CAD finding can be generated retrospectively from a past medical image and compared with the recorded interpretation finding. Therefore, in the case of a past medical image, since an interpretation finding is usually given, information regarding a difference between a CAD finding and an interpretation finding can be acquired for any past case.

(Modification 2 of the first embodiment)
In the first embodiment described above, finding change information of a past image is recorded in the finding management unit 103, and when a new image becomes a diagnosis target, the finding change information is read and reliability is evaluated. Thus, the evaluated reliability information is presented in association with the finding information. However, the flow up to the presentation of the reliability information is not limited to this method. For example, by recording the result of evaluating the reliability of CAD findings from the findings change information of past images, when a new image becomes a diagnosis target, the reliability information recorded in advance is read. May be presented in association with the finding information.

  Thereby, when a diagnosis object image is input, the trouble of evaluating the reliability of CAD findings can be saved, and the processing can be made efficient.

[Second Embodiment]
In the first embodiment, the reliability of CAD findings is evaluated from one of past images of the same subject (same subject). However, the method for evaluating the reliability of CAD findings is not limited to this, and other methods may be used. In the present embodiment, when there are a plurality of past images of the same subject (same subject), diagnosis is made from the plurality of past images (hereinafter referred to as “time-series past images (time-series examination data)”). A CAD finding of the past image related to the target image is selected. Then, the reliability of the CAD findings is evaluated based only on the selected CAD findings. For example, when the feature of the medical image of the subject (subject) has changed significantly due to surgery to remove the tumor in the past, the findings of the past image related to the diagnostic target image (postoperative) The information is selected so that only the finding information in the past image after the operation is selected. Thereby, the reliability of CAD findings is evaluated with high accuracy by excluding pre-operative past images whose features are significantly different from those of the diagnosis target images from the evaluation targets.

  The functional configuration of the medical diagnosis support apparatus 10 according to the present embodiment is the same as that in FIG. 1 in the first embodiment. However, the CAD finding generation unit 102 stores the image feature amount extracted from the medical image, the finding management unit 103 manages CAD findings in the time-series past image, and the reliability evaluation unit 106 diagnoses from the time-series past image. The point which evaluates the reliability of CAD findings based only on the CAD findings in the past image related to the target image is different from the first embodiment. The basic configuration (hardware configuration) of the computer for realizing the functions of the respective functional configuration units shown in FIG. 1 by software is the same as that in FIG. 2 in the first embodiment.

  The flowchart of the entire processing procedure performed by the medical diagnosis support apparatus 10 is the same as that in FIG. However, part of the processing in steps S303, S304, and S306 is different from that in the first embodiment. Only differences from the first embodiment will be described below.

  The medical information acquisition process in step S301 in FIG. 3 and the abnormal shadow detection process in step S302 are the same as those in the first embodiment.

  In the present embodiment, subsequently, in step S303, the CAD finding generation unit 102 extracts the image feature amount from the region of interest of the abnormal shadow acquired from the abnormal shadow detection unit 101, and obtains a finding that the doctor can use for diagnosis. Generate this and make this a CAD finding. Since the CAD finding generation method is the same as that in the first embodiment, a description thereof will be omitted. In the present embodiment, the image feature amount used for generating the CAD findings is stored in the magnetic disk 202 in association with the ID of the medical image. For example, the IDs are assigned to the medical images of the same subject (same subject) as 1, 2, 3,...

Subsequently, in step S304, the finding management unit 103 first stores the CAD finding for the acquired diagnosis target image in the magnetic disk 202 in association with the ID of the diagnosis target image. Then, the finding management unit 103 outputs the CAD finding of the diagnosis target image to the finding presenting unit 104.
Next, the finding management unit 103 determines whether or not a past interpretation report relating to the same abnormal shadow as the abnormal shadow to be diagnosed is stored in the magnetic disk 202 and exists. Specifically, it is determined whether or not finding change information indicating whether or not a doctor has changed CAD findings in the past for the same subject (same subject) is stored in the magnetic disk 202.
If the finding change information is stored in the magnetic disk 202 as a result of this determination, the finding managing unit 103 performs CAD findings for all time-series past images in which the finding change information is recorded. Acquired from the magnetic disk 202 together with the finding change information. Then, the finding management unit 103 outputs the CAD finding and finding change information about the acquired time-series past image and the CAD finding of the diagnosis target image to the reliability evaluating unit 106, and the process proceeds to step S306.
On the other hand, if the finding change information is not stored in the magnetic disk 202, the process proceeds to step S305 as it is.

  In this embodiment, when the process proceeds to step S306, the reliability evaluation unit 106 acquires the image feature amount stored in the magnetic disk 202 by the CAD finding generation unit 102 in step S303 for all time-series past images. Then, the reliability evaluation unit 106 selects a past image related to the diagnosis target image from the time-series past images based on the acquired image feature amount, and extracts the CAD findings. Next, the reliability evaluation unit 106 evaluates the reliability of each CAD finding based only on the extracted CAD findings.

Here, first, a method of extracting CAD findings of past images related to the diagnosis target image based on the image feature amount will be specifically described.
In this embodiment, based on at least one type of image feature amount extracted for generating CAD findings, CAD findings of past images related to the diagnosis target image are extracted. In step S303, at least one type of image feature amount stored in the magnetic disk 202 is selected, and all the corresponding image feature amounts are acquired from the time-series past image and the diagnosis target image. For example, the feature amount S of the size (maximum diameter) of the tumor extracted to generate the finding item 1: {size} is acquired for all the time-series past images and the diagnosis target image. Here, when the ID of an image included in the time-series past image is i, the feature quantity of the size of the tumor corresponding to the image ID: i is represented as Si. Further, the feature quantity of the size of the tumor corresponding to the diagnosis target image is expressed as St. At this time, a set of feature quantities related to the feature quantity St is set as Sgr, and all feature quantities Si satisfying both the following conditions 1 and 2 are selected.

Condition 1: all feature amounts Si whose difference from the feature amount St falls within a predetermined range
Condition 2: when the time series is traced back starting from the feature quantity St, all the feature quantities Si included until the magnitude of the rate of change between two feature quantities adjacent in the time series first exceeds a predetermined threshold value

  The above condition 1 is intended to select only past images having an image feature amount close to the image feature amount of the diagnosis target image. That is, the above condition 1 selects a past image related to the diagnosis target image from the plurality of time series past images based on the difference between the characteristics of the diagnosis target image and the characteristics of the plurality of time series past images. To do.

  In addition, the above condition 2 is intended to select only past images after that when there is a time when the image feature amount has suddenly changed between time-series past images. That is, the above condition 2 selects a past image related to the diagnosis target image from the plurality of time series past images based on the temporal change in the characteristics of the image series including the diagnosis target image and the plurality of time series past images. To do.

FIG. 9 is a characteristic diagram illustrating the second embodiment and plotting the feature amount S in time series.
A specific example of selecting the set Sgr that satisfies the above conditions 1 and 2 will be described with reference to FIG.

  In FIG. 9, time-series feature amounts S1 to S10 and a diagnosis target image feature amount St are plotted on a graph. The range D represents a range of a predetermined feature value with the feature value St as the center. At this time, since the feature amount Si satisfying the condition 1 is a point included in the range D centering on St, there are six points {S2, S3, S4, S8, S9, S10}.

  Subsequently, when two adjacent feature quantities in time series are Sj and Sj + 1, the change rate between the feature quantities Sj and Sj + 1 is Gj. G5 in FIG. 9 represents the rate of change between the feature quantity S5 and the feature quantity S6. Here, a predetermined threshold regarding the change rate Gj is defined as Gth (not shown). At this time, the feature amount Si satisfying the condition 2 is all feature amounts included between Sj + 1 and St when a certain change rate Gj first satisfies Gj> Gth when going back in time series starting from St. Represents. In the example of FIG. 9, the rate of change G5 first satisfies G5> Gth, and the feature quantities satisfying condition 2 are five points {S6, S7, S8, S9, S10}.

  Therefore, the feature amount set Sgr satisfying both the condition 1 and the condition 2 is three points {S8, S9, S10}. Then, the CAD findings assigned to the images with the image IDs: 8, 9, and 10 corresponding to the feature amount set Sgr are acquired (extracted) as time-series past images related to the diagnosis target image.

  As described above, when the image feature of the region of interest is significantly changed by surgery or the like by selecting the feature amount set Sgr related to the feature amount St using the condition 1 and the condition 2, the subsequent time-series past Only images can be selected. Furthermore, among them, it is possible to select only time-series past images whose image feature values are close (similar) to the diagnosis target image.

  In the present embodiment, as a method for selecting a feature amount related to a diagnosis target image in consideration of a change in the feature amount, a feature amount is selected by setting a threshold value for a change rate between adjacent feature amounts in time series. The feature quantity selection method is not limited to this. For example, a specific function model may be fitted based on the feature amount St of the diagnosis target image, and only the feature amount Si existing in a predetermined range may be selected from the fitted model. As this model, for example, a straight line model may be applied. This is applied when it is assumed that the disease progresses naturally, or that the change when treated by medication etc. can be approximated linearly. Alternatively, a non-linear function model representing disease progression or treatment change may be constructed based on medical statistical information and applied. In this way, by using a function model, whether or not the change between adjacent feature amounts in time series is in line with the change when the disease progresses naturally or is treated by medication, etc. The feature amount Si of the related image can be selected.

  In the present embodiment, the method of selecting a past image related to the diagnosis target image using the image feature amount corresponding to the finding item 1 {size}: the size of the tumor has been described. Is not limited to this method. For example, past images related to a plurality of other finding items may be selected in the same manner as described above. In addition, by extracting only the past images that are common among the past images obtained for each of the plurality of finding items, the reliability is evaluated using the past images that are common to all the finding items. Also good. Alternatively, the reliability may be evaluated independently using a past image obtained for each of the plurality of finding items.

  Furthermore, in the present embodiment, the past image related to the diagnosis target image is selected based on the image feature amount, but the determination material for selecting the past image is not limited thereto. For example, the selection may be made based on a CAD finding generated from the image feature amount or an interpretation finding written by a doctor. In this case, the vertical axis in FIG. 9 can be selected by the same method as described above by replacing the vertical axis with the magnitude of the finding item instead of the feature amount.

Next, a method for evaluating the reliability of the CAD findings based on the CAD findings of the past images acquired above will be described.
In the above example, the reliability is evaluated based on the CAD findings of the past images corresponding to the image IDs: 8, 9, and 10. First, the reliability of CAD findings is evaluated for each individual image. Since this method is the same method as step S306 in FIG. 3 in the first embodiment, the description thereof is omitted. Then, the reliability information evaluated for each individual image is integrated to evaluate the final reliability information. In the present embodiment, the final reliability information is obtained by taking the average value of the reliability information of the obtained individual images. In the example of FIG. 9, it is assumed that the evaluation results of the reliability of the image IDs: 8, 9, and 10 for the finding item 1: {size} are as follows.
-ID: 8 Reliability: Not sure Reliability R1 _CAD : 75%
-ID: 9 Reliability: Certainty Reliability R1 _CAD : 100%
-ID: 10 Reliability: Certainty Reliability R1 _CAD : 100%
At this time, if the reliability of the final CAD finding is R1 _{CAD, ALL} , R1 _{CAD, ALL} = (1.0 + 1.0 + 0.75) /3≈0.92. Since the reliability is not 100%, the reliability evaluation unit 106 determines that the reliability is “not sure”. Therefore, the reliability information of the final CAD findings is as follows.
・ Overall reliability: Uncertain reliability R1 _{CAD, ALL} : 92%

  However, in this embodiment, the final reliability is obtained from the average value of the individual reliability, but the reliability evaluation method is not limited to this method. For example, weighting may be performed for each image ID depending on how close the imaging time of each image is to the diagnosis target image (the older the diagnosis target image, the smaller the weighting), and then an average value may be taken. Thereby, since there is a high possibility that the image is older than the diagnosis target image, the influence on the reliability can be reduced. Further, the mode value of individual reliability may be used. As a result, when the reliability of most data is “certain” but one of the reliability is “unreliable”, it is determined as “unreliable” and the warning is displayed. Can be prevented.

  Then, the reliability evaluation unit 106 outputs the reliability information of the CAD finding for each finding item evaluated as described above to the finding presenting unit 104.

  According to the configuration described above, by evaluating the reliability of the CAD findings based only on the CAD findings of the past images related to the diagnosis target image, it is possible to evaluate the past images that are significantly different in characteristics from the diagnosis target images due to the influence of surgery or the like. Can be removed from the subject. As a result, the reliability of CAD findings can be evaluated with high accuracy.

[Other Embodiments]
In the above-described embodiment, the interpretation findings are input via the operation unit such as the mouse 205 and the keyboard 206. However, the present invention is not limited to this, and the touch panel function of the display unit may be used. Moreover, it is good also as detecting the gesture of a doctor, etc. and voice input, and acquiring the information regarding an interpretation finding. As in these examples, a user interface for inputting information from a person engaged in diagnosis is also included in the operation unit of the present invention.
The present invention can also be realized by executing the following processing.
That is, software (program) that realizes the functions of the above-described embodiments is supplied to a system or apparatus via a network or various storage media, and a computer (or CPU, MPU, or the like) of the system or apparatus reads the program. It is a process to be executed.
This program and a computer-readable recording medium storing the program are included in the present invention.

  Note that the above-described embodiments of the present invention are merely examples of implementation in practicing the present invention, and the technical scope of the present invention should not be construed as being limited thereto. It is. That is, the present invention can be implemented in various forms without departing from the technical idea or the main features thereof.

DESCRIPTION OF SYMBOLS 10 Medical diagnosis support apparatus, 100 Medical information acquisition part, 101 Abnormal shadow detection part, 102 CAD finding production | generation part, 103 Finding management part, 104 Finding presentation part, 105 Interpretation finding acquisition part, 106 Reliability evaluation part

Claims (11)

First generation means for generating, as a first generation finding, a plurality of items for diagnosis of the subject by analyzing first test data of the subject;
First acquisition means for acquiring an item for diagnosis of the subject based on the first examination data as a first interpretation finding based on an input from an operation unit;
Difference information acquisition means for acquiring information regarding differences in the respective items included in the first generation finding and the first interpretation finding;
Second generation means for generating a plurality of items for diagnosis of the subject as second generation findings by analyzing second test data of the subject newer than the first test data; ,
A finding presenting means for presenting information relating to the difference together with the second generation finding;
A medical diagnosis support apparatus characterized by comprising: The first examination data is a plurality of time-series examination data of the subject,
The first generation means generates a generation finding in the plurality of time-series inspection data as the first generation finding,
The medical diagnosis support apparatus according to claim 1, wherein the first acquisition unit acquires an interpretation finding in the plurality of time-series examination data as the first interpretation finding.   3. The difference information acquisition unit selects inspection data related to the second inspection data from the plurality of time-series inspection data based on the second inspection data. The medical diagnosis support apparatus described.   When the second examination data includes a medical image of a site where the subject has undergone surgery, the difference information acquisition unit is configured to start from the time when the surgery is performed among the plurality of time-series examination data. 4. The medical diagnosis support apparatus according to claim 3, wherein new test data is selected. The generation finding generated by either the first generation unit or the second generation unit is an automatic generation finding automatically generated by analyzing the image of the subject,
5. The medical diagnosis support apparatus according to claim 1, further comprising a finding presentation unit that presents the generated finding together with information indicating that the generated finding is an automatically generated finding. 6. A second acquisition means for acquiring an item for diagnosis of the subject based on the second examination data as a second interpretation finding based on an input from the operation unit;
Interpretation findings acquired by either the first acquisition means or the second acquisition means are acquired when an instruction to change the generated findings presented by the observation presentation means is input from the operation unit. The medical diagnosis support apparatus according to claim 5, wherein the medical diagnosis support apparatus is used. A second acquisition means for acquiring an item for diagnosis of the subject based on the second examination data as a second interpretation finding based on an input from the operation unit;
The interpretation finding acquired by either the first acquisition unit or the second acquisition unit is input from the operation unit as an instruction to finish changing the generation finding presented by the finding presentation unit. The medical diagnosis support apparatus according to claim 5, wherein the medical diagnosis support apparatus is acquired in response.   The finding presenting means is the automatic finding in response to an instruction to change the generated finding generated by either the first generating means or the second generating means being input from the operation unit. The medical diagnosis support apparatus according to any one of claims 5 to 7, wherein presentation of information indicating that is stopped. A first generation step of generating a plurality of items for diagnosis of the subject as a first generation finding by analyzing the first examination data of the subject;
A first acquisition step of acquiring an item for diagnosis of the subject based on the first examination data as a first interpretation finding based on an input from an operation unit;
A difference information acquisition step of acquiring information regarding differences in the respective items included in the first generation finding and the first interpretation finding;
A second generation step of generating a plurality of items for diagnosis of the subject as second generation findings by analyzing second test data of the subject newer than the first test data; ,
A finding presenting step for presenting information regarding the difference together with the second generating finding;
A medical diagnosis support method characterized by comprising: First generation means for generating, as a first generation finding, a plurality of items for diagnosis of the subject by analyzing first test data of the subject;
First acquisition means for acquiring an item for diagnosis of the subject based on the first examination data as a first interpretation finding based on an input from an operation unit;
Difference information acquisition means for acquiring information regarding differences in the respective items included in the first generation finding and the first interpretation finding;
Second generation means for generating a plurality of items for diagnosis of the subject as second generation findings by analyzing second test data of the subject newer than the first test data; ,
A finding presenting means for presenting information relating to the difference together with the second generation finding;
A medical diagnosis support system characterized by comprising: Image acquisition means for acquiring an image of the subject;
Detection means for detecting a feature region included in the image acquired by the image acquisition means;
Generating means for automatically generating a plurality of findings for the feature region detected by the detecting means;
Finding acquisition means for acquiring information of a plurality of findings about past images related to the image acquired by the image acquisition means;
Information acquisition means for acquiring statistical information on information of a plurality of findings acquired by the finding acquisition means;
The image acquired by the image acquisition means, a plurality of findings automatically generated by the generation means, a display showing the feature region detected by the detection means on the image, and the information acquisition means Presenting means for presenting the acquired statistical information;
A medical diagnosis support system characterized by comprising:

Images