JP7154492B2 - MEDICAL IMAGE PROCESSING APPARATUS, MEDICAL IMAGE PROCESSING METHOD AND MEDICAL IMAGE PROCESSING PROGRAM - Google Patents

Description

TECHNICAL FIELD The present invention relates to a medical image processing apparatus, a medical image processing method, and a medical image processing program, and more particularly to a medical image processing apparatus, a medical image processing method, and a medical image processing for processing medical images generated by photographing a patient. Regarding the program.

In recent years, AI (artificial intelligence) has made remarkable progress, and AI is being applied to the medical field (see, for example, Patent Document 1). It has a wide range of applications in the medical field, including diagnostic support through analysis of medical images captured by modalities.

On the other hand, according to the DICOM standard, tag information describing the examination date, patient name, age of the patient at the time of examination, etc. is attached to the medical image. Therefore, by analyzing medical images with reference to the tag information, it is expected to improve the performance of AI-based image diagnosis support.

JP 2018-14059 A

However, it is impossible to specify a detailed imaging site or imaging method from the tag information.

SUMMARY OF THE INVENTION Therefore, a primary object of the present invention is to provide a medical image processing apparatus capable of enhancing image diagnosis support performance.

A medical image processing apparatus (40: corresponding reference numeral in the embodiments; hereinafter the same) according to the present invention is a medical image processing apparatus (40) for processing medical images generated by photographing a patient. First collation means (S04 to S06) for collating the first feature amount representing the feature of the reference medical image belonging to each of the first classification (head CT, chest CT, abdomen CT) with the feature amount of the medical image , and a second feature quantity representing a feature of a reference medical image belonging to each of a plurality of second classifications (simple head CT, head CT with contrast, simple chest CT, contrast-enhanced chest CT, simple abdomen CT, contrast-enhanced abdomen) Among them, second matching means (S07 to S10, S17 to S18, S20 to S21, S23 to S24) for matching a second feature amount of a specific second classification with a feature amount of a medical image, and a plurality of second Each of the two classifications belongs to one of the plurality of first classifications, and the second collation means identifies any one of the plurality of first classifications based on the collation result of the first collation means. identifying means (S07); and second classification identifying means (S08 to S10) for identifying, as a specific second classification, a second classification belonging to the first classification identified by the first classification identifying means among the plurality of second classifications including.

Preferably, the number of specific second classifications is two or more, and identifying means (S19, S22, S25) for identifying any one of the specific second classifications based on the collation result of the second collating means, and An analysis means (S26 to S28) is further provided for analyzing the patient's disease by collating the feature amount of the reference medical image belonging to the second classification identified by the identification means with the feature amount of the medical image.

Preferably, the plurality of first categories are categories based on one of the imaging region and the imaging method, and the plurality of second categories are categories based on both the imaging region and the imaging method.

Preferably, the reference medical image is a medical image generated by imaging each of a plurality of patients.

Preferably, the patient's medical image is attached with tag information, and the tag information lacks information that serves as a reference for each of the first classification and the second classification.

A medical image processing method according to the present invention is a medical image processing method executed by a medical image processing apparatus (40) for processing medical images generated by photographing a patient, wherein a plurality of first classifications (head a first matching step (S04 to S06) of matching a first feature amount representing a feature of a reference medical image belonging to each of the abdominal CT, chest CT, and abdominal CT with the feature amount of the medical image; Specific second a second matching step (S07-S10, S17-S18, S20-S21, S23-S24) for matching the second feature quantity of the classification with the feature quantity of the medical image; belongs to any one of the first classifications, and the second matching step is a first classification identifying step (S07) of identifying any one of a plurality of first classifications based on the matching result of the first matching step; and a second class specifying step (S08 to S10) for specifying, as a specific second class, a second class belonging to the first class specified by the first class specifying step among the plurality of second classes.

A medical image processing program according to the present invention provides a plurality of first classifications (head CT, chest CT, , Abdominal CT), a first matching step (S04 to S06) of matching a first feature amount representing a feature of a reference medical image with a feature amount of a medical image, and a plurality of second classifications (head simple CT, head contrast-enhanced CT, chest simple CT, chest contrast-enhanced CT, abdominal plain CT, abdominal contrast-enhanced CT, among the second feature amounts representing the features of the reference medical images belonging to each, the second feature of the specific second classification A medical image processing program for executing a second matching step (S07-S10, S17-S18, S20-S21, S23-S24) of matching a quantity with a feature quantity of a medical image, the program comprising a plurality of second Each of the classifications belongs to one of the plurality of first classifications, and the second collation step identifies any one of the plurality of first classifications based on the collation result of the first collation step. a step (S07), and second classification identifying steps (S08 to S10) for identifying, as a specific second classification, a second classification belonging to the first classification identified by the first classification identifying step among the plurality of second classifications. include.

Each of the plurality of second classifications belongs to one of the plurality of first classifications, and the feature amount of the patient's medical image is matched with the first feature amount of each of the plurality of first classifications, and the plurality of Among the second classifications, matching is performed with a second feature quantity of a different second classification according to the matching result with the first feature quantity. As a result, even if the information of the first classification and the second classification is not attached to the patient's medical image, the information can be specified by the matching process, and the image diagnosis support performance can be enhanced.

The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of the embodiments with reference to the drawings.

1 is a block diagram showing the configuration of a medical diagnosis support system applied to this embodiment; FIG. FIG. 4 is an illustrative view showing one example of the configuration of a disease candidate table referenced by the machine learning system; (A) is an illustrative view showing an example of a mapping state of a certain memory area provided in a database constituting the machine learning system, and (B) is another memory area provided in the database constituting the machine learning system; FIG. 11C is an illustrative view showing one example of a mapping state of , and (C) is an illustrative view showing one example of a mapping state of other memory areas provided in a database that constitutes the machine learning system. FIG. 4 is an illustrative view showing one example of the structure of a DICOM file created by a modality; FIG. 4 is an illustrative view showing one example of a diagnostic support screen displayed on a main monitor of the diagnostic imaging support apparatus; FIG. 10 is an illustrative view showing another example of a diagnostic support screen displayed on the main monitor of the diagnostic imaging support apparatus; FIG. 11 is an illustrative view showing another example of the diagnostic support screen displayed on the main monitor of the diagnostic imaging support apparatus; FIG. 4 is an illustrative view showing one example of a diagnostic support screen displayed on a sub-monitor of the diagnostic imaging support apparatus; FIG. 11 is an illustrative view showing still another example of the diagnostic support screen displayed on the main monitor of the diagnostic imaging support apparatus; FIG. 10 is an illustrative view showing another example of a diagnostic support screen displayed on the main monitor of the diagnostic imaging support apparatus; FIG. 4 is a flowchart showing part of the operation of a CPU provided in the machine learning system; FIG. 4 is a flowchart showing another part of the operation of the CPU provided in the machine learning system;

Referring to FIG. 1, the medical diagnosis support system of this embodiment comprises an image diagnosis support apparatus 10, a modality 20, a PACS 30 and a machine learning system (medical image processing apparatus) 40, which are interconnected via an in-hospital LAN 50. be.

In image diagnosis support apparatus 10, bus BS1 is connected to communication I/F 12cm, CPU 12pr, keyboard/mouse 12km, DRAM 12mm, HDD 12hd, main monitor 12m1 and sub-monitor 12m2. The sub monitor 12m2 accompanies the main monitor 12m1. The radiologist faces the main monitor 12m1 and the sub-monitor 12m2, operates the keyboard/mouse 12km, and performs image diagnosis.

In machine learning system 40, communication I/F 42cm, CPU 42pr, DRAM 42mm and database 42db are connected to bus BS2. The database 42db stores a disease candidate table TBL1 shown in FIG. 2, as well as a memory area MM1 shown in FIG. 3A, a memory area MM2 shown in FIG. A memory area MM3 is provided.

According to FIG. 2, head diseases include "subarachnoid hemorrhage", "cerebral hemorrhage", "chronic otitis media", "cranium fracture", "sinusitis", "cholesteatoma otitis media", "cerebral infarction", . "No findings" were registered, and the chest diseases were "pneumothorax", "pneumoconiosis", "pulmonary embolism", "thymoma", "malignant lymphoma", "pneumomediastinum", "pulmonary edema", ..." "No findings" were registered, and "cirrhosis", "hepatocellular carcinoma", "acute cholecystitis", "gallstones", "pancreatic cancer", "acute appendicitis", "intestinal obstruction", ... "no findings" were registered as abdominal diseases. is registered.

According to FIG. 3A, the memory area MM1 stores the dimension-reduced head CT feature amount, the dimension-reduced chest CT feature amount, and the dimension-reduced abdominal CT feature amount. According to FIG. 3B, the memory area MM2 stores the dimensionality-reduced head simple CT feature amount and head enhanced CT feature amount, and the dimensionality-reduced chest simple CT feature amount and chest enhanced CT feature amount. , dimensionality-reduced abdominal simple CT features and abdominal enhanced CT features are stored. According to FIG. 3C, the memory area MM3 stores a plurality of simple head CT feature amounts, a plurality of head enhanced CT feature amounts, a plurality of chest simple CT feature amounts, and a plurality of chest enhanced CT feature amounts. A quantity, a plurality of plain abdominal CT feature amounts, and a plurality of abdominal enhanced CT feature amounts are stored.

The feature amount stored in the memory area MM3 is the feature amount of each of a plurality of reference medical images (teacher images). Specifically, the head simple CT feature amount is a feature amount of a medical image generated by head plain CT imaging for each of a plurality of patients with head diseases, and the head contrast-enhanced CT feature amount is: It is a feature quantity of a medical image generated by contrast-enhanced CT imaging of the head for each of a plurality of patients having diseases in the head.

The chest plain CT feature quantity is a feature quantity of a medical image generated by chest plain CT imaging for each of a plurality of patients with chest diseases, and the chest enhanced CT feature quantity is a feature quantity of a plurality of patients with chest diseases. It is a feature amount of a medical image generated by chest contrast-enhanced CT imaging for each.

Abdominal plain CT feature quantity is a feature quantity of a medical image generated by plain abdominal CT imaging for each of a plurality of patients with abdominal disease, and abdominal enhanced CT feature quantity is an abdominal contrast-enhanced feature quantity for a patient with abdominal disease. It is a feature amount of a medical image generated by CT imaging.

The feature quantity stored in the memory area MM2 is a feature quantity generated by reducing the dimension of the feature quantity stored in the memory area MM3. The feature amount stored in the memory area MM1 is a feature amount generated by further reducing the dimension of the feature amount stored in the memory area MM2.

Specifically, the head simple CT feature amount in the memory area MM2 is generated by performing dimensionality reduction on any head simple CT feature amount in the memory area MM3. The head-enhanced CT feature quantity in the memory area MM2 is generated by performing dimensionality reduction on any head-enhanced CT feature quantity in the memory area MM3.

The chest simple CT feature amount in the memory area MM2 is generated by performing dimensionality reduction on any chest simple CT feature amount in the memory area MM3. The chest contrast-enhanced CT feature quantity in the memory area MM2 is generated by performing dimensionality reduction on any chest contrast-enhanced CT feature quantity in the memory area MM3.

Abdominal simple CT feature amounts in the memory area MM2 are generated by performing dimensionality reduction on any abdominal simple CT feature amount in the memory area MM3. Abdominal contrast-enhanced CT feature quantities in the memory area MM2 are generated by performing dimensionality reduction on any abdominal contrast-enhanced CT feature quantity in the memory area MM3.

The head CT feature amount in the memory area MM1 is generated by performing dimensionality reduction on the head simple CT feature amount or the head enhanced CT feature amount in the memory area MM2. The chest CT feature amount in the memory area MM1 is generated by performing dimensionality reduction on the chest simple CT feature amount or the chest enhanced CT feature amount in the memory area MM2. Abdominal CT feature amounts in the memory area MM1 are generated by performing dimensionality reduction on the abdominal simple CT feature amounts or the abdominal enhanced CT feature amounts in the memory area MM2.

Therefore, the three classifications of head CT, chest CT and abdominal CT are defined as "first classification", and simple head CT, head CT with contrast, chest simple CT, chest contrast CT, abdominal simple CT, abdominal contrast CT are defined as "second classification", the first classification corresponds to the classification based on the imaging region, and the second classification corresponds to the classification based on both the imaging region and the imaging method.

Also, the second classification is positioned as a subordinate concept of the first classification, and each of the six second classifications belongs to one of the three first classifications. Specifically, two of head plain CT and head contrast-enhanced CT belong to head CT, two of chest plain CT and chest contrast-enhanced CT belong to chest CT, and two of abdomen plain CT and abdomen contrast-enhanced CT belong to chest CT. belongs to abdominal CT.

Then, the feature amount stored in the memory area MM1 can be said to be the feature amount representing the feature of the reference medical image belonging to each of the three first classifications. Also, the feature amount stored in the memory area MM2 can be said to be the feature amount representing the feature of the reference medical image belonging to each of the six second classifications.

The laboratory technician operates the modality 20 based on the requested comment from the attending physician. The modality 20 performs simple CT imaging and contrast-enhanced CT imaging multiple times on the patient's head, chest, or abdomen, and a first series of medical images generated by the simple CT imaging. and a second series of images composed of a plurality of medical images generated by contrast-enhanced CT imaging.

Of the simple CT imaging and the contrast-enhanced CT imaging, only the simple CT imaging is performed depending on the examination. In this case, only the series images of the first series out of the first series and the second series are contained in the DICOM file.

A DICOM file has the data structure shown in FIG. 4 and is stored in the PACS 30 via the hospital LAN 50 . Referring to FIG. 4, the DICOM file stores tag information in addition to the series images of the first series and the series images of the second series. Here, the tag information is composed of a plurality of items (examination date, patient name, patient's age at examination, examination instance UID, etc.) each indicating an examination attribute. However, the tag information does not include the items of imaging region and imaging method. That is, the imaging site and imaging method are missing from the tag information.

When a radiologist operates the image diagnosis support apparatus 10 to diagnose a patient, the top screen shown in FIG. 5 is developed in the DRAM 12mm by the CPU 12pr. The expanded top screen is read and displayed by the main monitor 12m1. According to FIG. 5, the top screen has an index column IDX in which a plurality of indexes representing a plurality of examinations are listed. Also, on the top screen shown in FIG. 5, the interpretation report corresponding to any heading is set as "pending".

A headline selection operation (=an operation of selecting any one of a plurality of headlines listed in the headline column IDX, for example, an operation of clicking on the headline enclosed in a thick frame in FIG. 6) is performed with the keyboard/mouse 12 km. Then, the interpretation report screen (see FIG. 7) corresponding to the headline is displayed on the main monitor 12m1 instead of the top screen.

According to FIG. 7, the interpretation report screen is provided with a requested comment column REQ, a finding column RMK, a diagnosis column DGN, and an attached image column IMG. Among them, the character string "chest pain" in the request comment field REQ was written by the attending physician.

The CPU 12pr then accesses the PACS 30 via the communication I/F 12cm, and acquires from the PACS 30 the DICOM file corresponding to the study instance UID assigned to the heading being selected. The CPU 12pr also expands any medical image in the acquired DICOM file to the DRAM 12mm. The developed medical image is read by the sub-monitor 12m2 and displayed as shown in FIG.

When a remark input operation (an operation of designating a remark column RMK and inputting a desired character string) is performed using the keyboard/mouse 12km, the CPU 12pr displays the input character string in the remark column RMK as shown in FIG. do. According to FIG. 9, the finding field RMK displays a character string of "Calcification is found at the coronary artery origin."

Also, when a diagnostic comment input operation (an operation of designating the diagnosis field DGN and inputting a desired character string) is performed using the keyboard/mouse 12km, the CPU 12pr inputs the input character string to the diagnosis field as shown in FIG. Display on DGN. According to FIG. 9, the diagnosis column DGN displays a character string of "1. suspected angina pectoris".

Furthermore, when an image attachment operation (an operation to move the medical image displayed on the sub-monitor 12m2 to the attached image field IMG by a drag-and-drop operation) is performed using the keyboard/mouse 12km, the CPU 12pr attaches the medical image displayed on the sub-monitor 12m2. and displays the created thumbnail image in the attached image field IMG in the manner shown in FIG.

Also, when an image update operation (an operation to update the medical image on the sub-monitor 12m2) is performed using the keyboard/mouse 12km, the CPU 12pr expands another medical image in the DICOM file acquired from the PACS 30 to the DRAM 12mm. As a result, the medical image displayed on the sub-monitor 12m2 is updated to another medical image.

When the "Confirm" button displayed near the attached image field IMG is clicked with the keyboard/mouse 12km, the CPU 12pr transmits an analysis request to the machine learning system 40. FIG. All medical images (the first series of images and the second series of images) in the DICOM file acquired from the PACS 30 are attached to the analysis request.

The machine learning system 40 has artificial intelligence similar to that of a physician experienced in diagnosing patient disease based on medical images, and analyzes the patient disease based on the medical image accompanying the analysis request. Specifically, the machine learning system 40 identifies disease candidates from the disease candidate table TBL1 shown in FIG. An analysis report containing the information is sent back to the diagnostic imaging support apparatus 10 .

When the analysis report is returned, the CPU 12pr develops the analysis report screen in the DRAM 12mm. The expanded analysis report screen is read out by the main monitor 12m1, and multiplexed (additionally displayed) on the interpretation report screen in the manner shown in FIG.

Referring to FIG. 10, disease candidates and corresponding probabilities based on medical images generated by plain CT imaging of the chest are displayed on the left side of the analysis report screen. Further, disease candidates and corresponding probabilities based on medical images generated by contrast-enhanced chest CT imaging are displayed on the right side of the analysis report screen.

Here, disease candidates based on medical images generated by plain chest CT imaging and their corresponding probabilities are "pleuritis 80%", "angina pectoris 50%", "acute myocarditis 15%", and "lung cancer 10%". Line up in order. In addition, disease candidates based on medical images generated by contrast-enhanced CT imaging of the chest and corresponding probabilities are "angina pectoris 80%", "myocardial infarction 40%", "aortic valve disease 30%", and "acute myocarditis 10%". ” in order.

The character strings displayed in each of the findings column RMK and the diagnosis column DGN and the thumbnail images attached to the attached image column IMG can be changed by operating the keyboard/mouse 12 km after the analysis report screen is displayed. It is possible.

When the "Confirm" button displayed at the upper right of the interpretation report screen is clicked with the keyboard/mouse 12km, the confirmation process is performed. As a result, the display on the main monitor 12m1 returns from the interpretation report screen to the top screen. However, instead of "pending", the character string "confirmed" is attached to the headline corresponding to the interpretation report that has undergone the confirmation process.

A CPU 42pr provided in the machine learning system 40 executes processing for analyzing a patient's disease according to the flowcharts shown in FIGS. A control program corresponding to this flow chart is stored in the DB 42db.

Referring to FIG. 11, in step S01, whether or not an analysis request has been received from image diagnosis support apparatus 10 is repeatedly determined based on the processing status of communication I/F 42cm, and the determination result is updated from NO to YES. and proceed to step S02. The series images accompanying the analysis request are written in the DRAM 42mm by the communication I/F 42cm.

In step S02, the number of series images stored in the DRAM 42mm is set to a constant Lmax, and the variable L is set to "1". When the series images of the first series and the series images of the second series are stored in the DRAM 42mm, the constant Lmax is set to "2". In step S03, the number of medical images forming the L-th series image is set to a constant Kmax, and the variable K is set to "1".

In step S04, the feature amount of the K-th medical image forming the L-th series image is collated with the head CT feature amount stored in the memory area MM1. In step S05, the feature amount of the K-th medical image forming the L-th series image is collated with the chest CT feature amount stored in the memory area MM1. In step S06, the feature amount of the K-th medical image forming the L-th series image is collated with the abdominal CT feature amount stored in the memory area MM1.

In step S07, based on the matching results of steps S04 to S06, the imaged region of the K-th medical image is specified. In step S08, it is determined based on the processing result of step S07 whether or not the imaging region of the K-th medical image is the head, and in step S09, the imaging region of the K-th medical image is the chest. Based on the result of processing in step S07, it is determined whether or not the imaged region of the K-th medical image is the abdomen based on the result of processing in step S07 in step S10.

If the determination result in step S08 is YES, the process proceeds to step S17 to execute matching processing for the simple head CT feature amount and the head enhanced CT feature amount. If the determination result in step S09 is YES, the process proceeds to step S20 in order to execute matching processing for the chest simple CT feature amount and the chest enhanced CT feature amount. If the determination result in step S10 is YES, the process proceeds to step S23 in order to execute matching processing for the abdominal simple CT feature amount and the abdominal contrast-enhanced CT feature amount.

As described above, the first classification is based on the imaging site (head CT, chest CT, and abdomen CT), and the second classification is based on both the imaging site and the imaging method (head simple CT, head CT, head CT). contrast-enhanced CT, chest simple CT, chest contrast-enhanced CT, abdominal simple CT, and abdominal contrast-enhanced CT), the process of step S07 corresponds to the process of specifying any one of the first classifications, The process of steps S08, S09 or S10 corresponds to the process of identifying two second classes belonging to the first class identified in step S07.

By determining YES in step S08, the two second classifications of simple head CT and enhanced head CT are identified, and by determining YES in step S09, the two second classifications of simple chest CT and enhanced chest CT are identified. Two second classifications are identified, and two second classifications of abdominal simple CT and abdominal contrast-enhanced CT are identified by determining YES in step S10.

In step S17, the feature amount of the K-th medical image is collated with the head simple CT feature amount stored in the memory area MM2. It is collated with the cephalometric CT feature amount stored in MM2. In step S19, the imaging method of the K-th medical image is specified based on the matching results of steps S17 and S18.

In step S20, the feature amount of the K-th medical image is collated with the chest plain CT feature amount stored in the memory area MM2. is compared with the chest contrast-enhanced CT feature value stored in . In step S22, the imaging method for the K-th medical image is specified based on the matching results in steps S20 and S21.

In step S23, the feature amount of the K-th medical image is collated with the abdominal simple CT feature amount stored in the memory area MM2. is compared with the feature amount of abdominal contrast-enhanced CT stored in . In step S25, the imaging method for the K-th medical image is specified based on the matching results in steps S23 and S24.

In step S26, it is determined whether or not the imaging method of the K-th medical image is simple CT imaging based on the processing result of steps S19, S22, or S25. Also, in step S27, it is determined whether or not the imaging method of the K-th medical image is contrast-enhanced CT imaging based on the processing results of steps S19, S22, or S25. If either the determination result of step S26 or the determination result of step S27 is YES, the process proceeds to step S28.

In step S28, a feature amount corresponding to the imaging region specified in step S07 and the imaging method specified in step S19, S22 or S25 is specified from among the feature amounts stored in the memory area MM3, and the K sheets are selected. The patient's disease is analyzed by collating the feature amount of the medical image of the eye with the feature amount specified from the memory area MM3.

Specifically, if the specified imaging region and imaging method are the head and simple CT, respectively, the feature amount of the K-th medical image is obtained from the plurality of head simple CT feature amounts stored in the memory area MM3. The patient's disease is analyzed by matching with each of If the specified imaging region and imaging method are the head and contrast-enhanced CT, respectively, the feature amount of the K-th medical image is collated with each of the plurality of head-enhanced CT feature amounts stored in the memory area MM3. By doing so, the patient's disease is analyzed.

If the specified imaging region and imaging method are chest and simple CT, respectively, the feature amount of the K-th medical image is compared with each of the plurality of simple chest CT feature amounts stored in memory area MM3. , to analyze the patient's disease. If the specified imaging region and imaging method are chest and contrast-enhanced CT, respectively, the feature quantity of the K-th medical image is compared with each of the plurality of chest contrast-enhanced CT feature quantities stored in memory area MM3. , to analyze the patient's disease.

If the specified imaging region and imaging method are the abdomen and the imaging method, respectively, by comparing the feature amount of the K-th medical image with each of the plurality of abdominal simple CT feature amounts stored in the memory area MM3, , to analyze the patient's disease. If the specified imaging region and imaging method are the abdomen and contrast-enhanced CT, respectively, the feature amount of the K-th medical image is compared with each of the plurality of abdominal contrast-enhanced CT feature amounts stored in the memory area MM3. , to analyze the patient's disease.

When the process of step S28 is completed, the process proceeds to step S12. If the determination result of any of steps S08 to S10 is NO, error processing is performed in step S11, and then the process proceeds to step S12. If both the determination result of step S26 and the determination result of step S27 are NO, error processing is performed in step S29, and then the process proceeds to step S12.

In step S12, the variable K is updated to "K+1", and in step S13, it is determined whether or not the updated variable K exceeds the constant Kmax. If the determination result is NO, the process returns to step S04, and if the determination result is YES, the process proceeds to step S14. In step S14, the variable L is updated to "L+1", and in step S15, it is determined whether or not the updated variable L exceeds the constant Lmax. If the determination result is NO, the process returns to step S03, and if the determination result is YES, the process proceeds to step S16.

In step S16, a disease candidate is specified from the disease candidate table TBL1 shown in FIG. 2 based on the processing result of step S28, and statistics are obtained for the disease candidate, and the disease candidate and the patient's disease correspond to the disease candidate. and an analysis report including the probability that The analysis process ends after returning the analysis report.

As can be seen from the above description, when processing a medical image generated by photographing a patient and analyzing the patient's disease, the CPU 42pr of the machine learning system 40 uses reference medical images belonging to each of a plurality of first classifications. A first matching process is executed to match the first feature amount (head CT feature amount, chest CT feature amount, abdominal CT feature amount) representing the feature of the image with the feature amount of the medical image (S04 to S06). The CPU 42pr also calculates a second feature quantity (head simple CT feature quantity, head enhanced CT feature quantity, chest simple CT feature quantity, chest enhanced CT feature quantity , abdominal simple CT feature amount, abdominal enhanced CT feature amount), the second matching process is performed to match the second feature amount of a specific second classification with the feature amount of the medical image (S07 to S10, S17 ~S18, S20~S21, S23~S24).

Here, each of the plurality of second classifications belongs to any one of the plurality of first classifications. Further, in the second collation process, one of the plurality of first classifications is identified based on the result of the first collation process (S07), and the first classification among the plurality of second classifications is identified. The second class to which it belongs is specified as a specific second class (S08-S10).

Each of the plurality of second classifications belongs to one of the plurality of first classifications, and the feature amount of the patient's medical image is matched with the first feature amount of each of the plurality of first classifications, and the plurality of Among the second classifications, matching is performed with a second feature quantity of a different second classification according to the matching result with the first feature quantity. As a result, even if the information of the first classification and the second classification is not attached to the patient's medical image, the information can be specified by the matching process, and the image diagnosis support performance can be enhanced.

In this embodiment, a CT device is used as the modality 20, but MRI may be used instead. Further, in the present embodiment, the classification based on the imaging region is set as the first classification, and the classification based on both the imaging region and the imaging method is set as the second classification. However, the classification based on the imaging direction may be the first classification, and the classification based on both the imaging region and the imaging method may be the second classification.

DESCRIPTION OF SYMBOLS 10... Image diagnosis assistance apparatus 40... Machine learning system 12pr, 42pr... CPU
12hd …HDD
12m1 ... Main monitor 12m2 ... Sub monitor 42db ... Database

Claims (6)

A medical image processing apparatus for processing medical images generated by imaging a patient with a single modality ,
A first collation means for collating a first feature quantity representing a feature of a reference medical image belonging to each of a plurality of first classifications based on one of an imaging region and an imaging method with the feature quantity of the medical image. ,and
a second feature quantity of a specific second classification among second feature quantities representing features of a reference medical image belonging to each of a plurality of second classifications based on both the imaging region and the imaging method ; , a second matching means for matching the feature amount of the medical image,
Each of the plurality of second classifications belongs to any one of the plurality of first classifications,
The second verification means is
a first classification identifying means for identifying any one of the plurality of first classifications based on the collation result of the first collating means; and one of the plurality of second classifications identified by the first classification identifying means. including a second classification identifying means for identifying a second classification belonging to the first classification as the specific second classification,
The number of the specific second classifications is 2 or more,
A medical image processing apparatus, comprising specifying means for specifying any one of the specific second classifications based on a collation result of the second collating means . 2. The medical image according to claim 1, further comprising analysis means for analyzing the patient's disease by collating the feature amount of the reference medical image belonging to the second classification identified by the identification means with the feature amount of the medical image. processing equipment. 3. The medical image processing apparatus according to claim 1, wherein said reference medical image is a medical image generated by photographing each of a plurality of patients. tag information is attached to the medical image of the patient;
4. The medical image processing apparatus according to any one of claims 1 to 3 , wherein reference information for each of said first classification and said second classification is missing in said tag information. A medical image processing method performed by a medical image processing apparatus that processes medical images generated by imaging a patient with a single modality , comprising:
A first collation step of collating a first feature quantity representing a feature of a reference medical image belonging to each of a plurality of first classifications based on one of an imaging region and an imaging method with the feature quantity of the medical image. ,and
a second feature quantity of a specific second classification among second feature quantities representing features of a reference medical image belonging to each of a plurality of second classifications based on both the imaging region and the imaging method ; , a second matching step of matching the feature amount of the medical image,
Each of the plurality of second classifications belongs to any one of the plurality of first classifications,
The second verification step includes:
a first classification identifying step of identifying any one of the plurality of first classifications based on the collation result of the first collating step; and one of the plurality of second classifications identified by the first classification identifying step including a second classification identifying step of identifying a second classification belonging to the first classification as the specific second classification;
The number of the specific second classification is 2 or more,
A medical image processing method, comprising a specifying step of specifying any one of the specific second classifications based on the matching result of the second matching step . to a processor of a medical image processing device that processes medical images generated by imaging a patient with a single modality ;
A first collation step of collating a first feature quantity representing a feature of a reference medical image belonging to each of a plurality of first classifications based on one of an imaging region and an imaging method with the feature quantity of the medical image. ,and
a second feature quantity of a specific second classification among second feature quantities representing features of a reference medical image belonging to each of a plurality of second classifications based on both the imaging region and the imaging method ; , a medical image processing program for executing a second matching step of matching the feature amount of the medical image,
Each of the plurality of second classifications belongs to any one of the plurality of first classifications,
The second verification step includes:
a first classification identifying step of identifying any one of the plurality of first classifications based on the collation result of the first collating step; and one of the plurality of second classifications identified by the first classification identifying step a second classification identifying step of identifying a second classification belonging to the first classification as the specific second classification ;
The number of the specific second classifications is 2 or more,
A medical image processing program, comprising a specifying step of specifying any one of the specific second classifications based on a matching result of the second matching step .

Images