Diagnostic information processing device and diagnostic information processing program

Abstract

The database represents a complex, intertwined hierarchical structure, and, given a specific imaging finding, lists similar differential diagnoses. [Solution] The system includes a disease database 120 that stores disease lesion information 210, an input unit 140 that inputs case lesion information 220, and a case information storage unit 150 that stores case lesion information 130. Lesion information 111, 112 and lesion information 131, 132 have a common structure that describes the characteristics of the lesions. A comparison unit 160 compares the lesion characteristics of the disease lesion information 110 and the case lesion information 130 and determines the similarity on an element-by-element basis. A determination unit 170 determines the similarity between the disease lesion information 110 and the case lesion information 130 for each case based on the similarity on an element-by-element basis determined by the comparison unit 160.

Description

[Technical Field]

[0001] The present invention relates to a diagnostic information processing device and a diagnostic information processing program. [Background technology]

[0002] In radiological diagnosis, the primary task of a diagnostic radiologist is to provide a differential diagnosis based on imaging findings. The differential diagnosis usually involves a wide range of diseases, and the diagnostic radiologist must rely on their memory of typical imaging findings, clinical findings, symptoms, and patient backgrounds to differentiate between these diseases. In the case of rare diseases or rare findings, they must conduct literature research each time.

[0003] However, there are countless differential diagnoses, including rare diseases, and it is difficult to remember all of the typical imaging findings, epidemiology, and patient backgrounds associated with these diseases. Therefore, even experienced radiologists often need to conduct comprehensive research to identify appropriate differential diagnoses. Furthermore, these imaging findings and disease concepts are frequently updated, making it difficult for anyone other than an expert in a specific field to keep up.

[0004] For example, Patent Documents 1 and 2 present programs that assist in recalling possible disease candidates for symptoms and findings during medical differential diagnosis. These programs propose a method for listing possible diseases, including rare diseases, based on the patient's clinical information and background. However, these programs are intended for medical differential diagnosis and do not support radiological differential diagnosis.

[0005] To assist in radiological differential diagnosis, books have been published that provide a table of correspondence between imaging findings and differential diagnoses, such as Non-Patent Document 1. While these books are comprehensive, they lack searchability, as users must refer to every page each time to find the differential diagnosis corresponding to a finding. Therefore, a search function specialized for radiological image diagnosis is needed. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent No. 7165371 [Patent Document 2] Japanese Patent Publication No. 2020-17137 [Non-patent literature]

[0007] [Non-Patent Document 1] "New Edition: Brain MRI Findings" (edited by Kazuhiro Tsuchiya et al.), Gakken Shujunsha, 2008 Summary of the Invention [Problem to be solved by the invention]

[0008] The difficulty in creating a database of radiological diagnostic images lies primarily in the complexity of anatomical structures. For example, the cerebrum alone is divided into the frontal lobe, parietal lobe, and other parts, each of which forms a layered structure consisting of the cortex, subcortical white matter, and deep white matter. This complex intertwining of layers must be represented in the database. Furthermore, appropriate structural design is required when representing tumorous lesions in the database. Tumorous lesions are space-occupying lesions that differ from existing anatomical structures, and these must be represented appropriately within the anatomical structure of the database.

[0009] Furthermore, when describing imaging findings, it is not enough to simply indicate that there is an abnormal signal on MRI in a specific anatomical region; various characteristics must be expressed, such as symmetry, bilaterality, border morphology, contrast enhancement, etc. Furthermore, given a specific imaging finding, in order to list similar differential diagnoses, it is necessary to score the degree to which the imaging finding resembles each disease. [Means for solving the problem]

[0010] A diagnostic information processing device according to one embodiment of the present invention comprises a disease database that stores disease lesion information, which is lesion information for each disease, and an input unit that inputs case lesion information, which is lesion information for each area diagnosed by a doctor as pathological, wherein the disease lesion information and the case lesion information have a common structure that describes the characteristics of the lesion. The device further comprises a comparison unit that compares the lesion characteristics of the disease lesion information and the case lesion information and determines the similarity on an element-by-element basis, and a determination unit that determines the similarity between the disease lesion information and the case lesion information for each case based on the similarity on an element-by-element basis determined by the comparison unit. [Brief explanation of the drawings]

[0011] [Figure 1] FIG. 2 is a block diagram showing a functional configuration of the diagnostic information processing device. [Figure 2] FIG. 2 is a block diagram showing the data configuration of disease information and case information. [Figure 3] FIG. 2 is a block diagram showing the data structure of lesion information. [Figure 4] FIG. 1 illustrates a tree structure of anatomical regions. [Figure 5] This is an input screen for anatomical parts. [Figure 6] FIG. 2 is a block diagram showing the configuration of image finding data. [Figure 7] 1 shows an input screen and a display screen for image findings. [Figure 8] 10 is a flowchart showing a process of evaluating the similarity of lesion information. [Figure 9] FIG. 10 is a diagram illustrating a process for calculating the similarity between a case and a disease. [Figure 10] FIG. 10 is a diagram illustrating a similarity score calculation process. [Figure 11] FIG. 10 is a diagram showing an outline of calculation of similarity between a disease and a case. [Figure 12] FIG. 10 is a diagram showing an outline of output processing of a list of differential diseases. DETAILED DESCRIPTION OF THE INVENTION

[0012] 1, the functional configuration of the diagnostic information processing device 100 according to this embodiment includes a disease database 120 that stores disease lesion information 210, which is lesion information for each disease 110, an input unit 140 that inputs case lesion information 220, which is lesion information for a site diagnosed as pathological by a doctor for each case 130, and a case information storage unit 150 that stores the case lesion information 220. Here, as shown in FIG. 2, the disease lesion information 210 and the case lesion information 220 have a common structure that describes the characteristics of the lesion.

[0013] The diagnostic information processing device 100 further includes a comparison unit 160 that compares the lesion characteristics of the disease lesion information 110 and the case lesion information 130 and determines the similarity on an element-by-element basis, and a determination unit 170 that determines the similarity between the disease lesion information 110 and the case lesion information 130 for each case based on the element-by-element similarity determined by the comparison unit 160.

[0014] The control unit of the diagnostic information processing device 100 according to this embodiment includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), an image processing unit, and a memory. The CPU, ROM, RAM, image processing unit, and memory are interconnected via a bus.

[0015] The CPU executes various processes according to programs stored in the ROM or programs loaded from memory into the RAM. The RAM also stores data necessary for the CPU to execute various processes.

[0016] The image processing unit is composed of a DSP (Digital Signal Processor), a VRAM (Video Random Access Memory), etc., and works in cooperation with the CPU 31 to perform various image processing operations on image data.

[0017] The memory may be any storage medium such as a DRAM, cache memory, magnetic disk, optical disk, magneto-optical disk, or semiconductor memory. Memory includes not only those connected via a bus, but also those that are read and written via a drive or the Internet. In this embodiment, the data stored will be described as being temporarily stored in this memory, whether it is temporary storage or long-term storage using non-volatile memory.

[0018] Fig. 1 is a block diagram showing the functional configuration of a diagnostic information processing device. The memory of the diagnostic information processing device 100 shown in Fig. 1 includes a disease database 120 and a case information storage unit 150. The control unit of the diagnostic information processing device 100 includes an input unit 140, a comparison unit 160, and a determination unit 170.

[0019] Fig. 2 is a block diagram showing the data configuration of disease information and case information. Disease information 110 stored in the disease database 120 is a database of general information about diseases described in specialized books, papers, etc., and includes at least one, or as shown in Fig. 2, multiple pieces of disease lesion information 210.

[0020] On the other hand, for the case information 130, when a doctor diagnoses a site as pathological, the diagnosed site is input as case lesion information 220 via the input unit 140. The input data is stored in the case information storage unit 150. For the case information 130, findings observed in cases at clinical sites are compiled into a database. Here, the basic configuration of this embodiment will be described with reference to Figs. 1 and 2.

[0021] Here, the disease information 110 has a database of lesion information 210 of pre-registered diseases. In this embodiment, the disease information 110 shown in Fig. 1 will be described focusing on the lesion information 210, but may also include other information about the corresponding disease.

[0022] A doctor inputs diagnostic information such as image findings from patient images into the input unit 140. The input information is stored in the case information storage unit 150 as case lesion information 220. Similarly, the case information 130 will be described focusing on the case lesion information 220, but may also include other information about the relevant case.

[0023] Disease lesion information 210 is a database of commonly known disease concepts, and case lesion information 220 is a doctor's diagnosis, but both have a common structure that describes the characteristics of the lesion for comparative assessment. As shown in Figure 3, disease lesion information 210 and case lesion information commonly include information on the lesion type, anatomical location, and imaging findings. Disease lesion information 210 is a list of already established disease concepts, while case lesion information 220 is only at the stage of a pathological diagnosis, but because both share the same lesion information structure, they will be described together.

[0024] The comparison unit 160 compares the lesion features of the disease lesion information 210 and the case lesion information 220 to determine the similarity on an element-by-element basis. The comparison unit 160 particularly determines the similarity by comparing the lesion type, anatomical location, and image findings of the disease lesion information and the case lesion information. The determination unit 170 determines the similarity between the disease lesion information 210 and the case lesion information 220 for each case based on the element-by-element similarity determined by the comparison unit 160.

[0025] FIG. 3 is a block diagram showing the data structure of lesion information. Because disease lesion information 210 and case lesion information 220 share a common data format, they will be hereinafter referred to as lesion information 300. Lesion information 300 includes information on lesion type 310, anatomical location 320, and imaging findings 330. Lesion type 310 indicates whether a lesion is neoplastic or not. Anatomical location 320 is information that identifies the part of the body where the lesion occurred from an anatomical perspective. Imaging findings 330 is information that describes the doctor's diagnosis and can list multiple pieces of information from different perspectives.

[0026] Figure 4 shows a tree structure of anatomical regions. Anatomical structures are embedded in a graph structure called a directed acyclic graph. A directed acyclic graph is a directed graph without closed loops. Because there are no cycles, a vertex cannot return to the same vertex, making it a suitable structure for representing anatomical structures. When a parent structure is selected, its subordinate anatomical structures are enumerated. Anatomical regions are first identified from the parent structure, and then the child structures corresponding to the parent structure are sequentially identified. Regarding the parent-child relationship of anatomical regions, for example, the cerebrum has child structures such as gray matter, white matter, frontal lobe, parietal lobe, temporal lobe, and occipital lobe, and the cerebellum and basal ganglia also have child structures.

[0027] On the other hand, cerebral gray matter can be a child structure of multiple parents. For example, it is a child structure of the cerebrum, and can be a child structure of the frontal lobe, parietal lobe, temporal lobe, and occipital lobe, which are child structures of the cerebrum, and can therefore be a grandchild structure of the cerebrum. Because of this parent-child relationship through multiple pathways, anatomical regions are represented by a directed acyclic graph, as shown in Figure 4. In this way, by selecting a structure corresponding to an imaging finding, it is possible to store the pathway leading to that lower-level structure as an anatomical structure.

[0028] Figure 5 shows the input screen for anatomical regions. Anatomical regions are ultimately listed in the order of parent structure > child structure > grandchild structure. For example, following the structure shown in Figure 4, structures are expressed in the form of cerebrum > cerebral white matter, or cerebrum > frontal lobe > cerebral gray matter, etc. Therefore, anatomical regions must be input in this order. Candidates for parent structures are listed on the parent structure selection screen 410, and a parent structure 420 is selected from among them. When parent structure 420 is selected, a confirmed candidate 430 is displayed. Then, a child structure selection screen 440 for parent structure 420 is displayed. A child structure displayed on the child structure selection screen 440 is then selected.

[0029] When a child structure is selected, the child structure selection screen 440 moves to the left and takes the position of the parent structure selection screen 410, and the selection proceeds in the form of a new parent structure 420, a confirmed candidate 430, and so on, until the final structure is identified and input is complete.

[0030] Fig. 6 is a block diagram showing the configuration of image finding data. The lesion information 300 shown in Fig. 3 includes data on image findings 330, but may include one or more image findings 330. As shown in Fig. 6, the image findings 330 are made up of modality information 610, finding information 620, and modifiers 630. The modifiers 630 may be zero or multiple.

[0031] FIG. 7 shows the input screen and display screen for imaging findings. The input results of imaging findings 330 entered on input screen 700 are reflected on display screen 750. There are a wide variety of modalities 710 to choose from, even just for CT and MRT, and the modality that captured the image is identified from these. Next, findings 720 and modifiers 730 are input in this order. The results are reflected on display screen 750 as imaging findings 760. If imaging findings 770 are to be added thereafter, input is repeated again from input screen 700 in the order of modality 710, findings 720, and modifiers 730. Once all imaging findings 760, 770, etc. have been input, the input screen is closed.

[0032] As described above, by inputting data from the input unit 140, the image findings 330 shown in Fig. 6 are created, and by inputting and creating all of the image findings 330, the lesion information 330 shown in Fig. 3 is constructed, and the case lesion information 220 shown in Fig. 2 is input. On the other hand, since the disease lesion information 210 is already prepared as a database, it is possible to evaluate cases diagnosed by a doctor by comparing it with the input case lesion information 220. Therefore, the process of calculating the similarity score of lesion information will now be described.

[0033] 8 is a flowchart showing the process of calculating the similarity score of lesion information. When the evaluation process starts, the comparison unit 160 first compares the lesion types (step S1010). As shown in FIG. 3, both the disease lesion information 210 and the case lesion information 220 have the lesion type 310. When the lesion types of both are compared and do not match, a score of 0 is assigned and the series of processes is terminated.

[0034] If there is a match, the comparison unit 160 then proceeds to anatomical score evaluation. At this time, the comparison unit 160 determines whether the lesion is non-neoplastic or neoplastic. Scoring of the degree of match between anatomical locations differs between non-neoplastic and neoplastic lesions. Because the anatomical location in which a non-neoplastic lesion is located is important, the degree of match is evaluated using a demerit system. On the other hand, with some exceptions, neoplastic lesions may be located in any anatomical location, and if they are located in a typical location, they are evaluated using a bonus system. This is explained in detail below. If the lesion type determined in step S1010 is a non-neoplastic lesion, the anatomical locations 320 of both lesions are compared (step S1020). If there is a mismatch, a score of 0 is assigned and the process ends. However, this determination may result in a match or a partial match. Since the anatomical location 320 is represented as a directed acyclic graph as shown in FIG. 4, it is represented by multiple structure names, such as parent structure, child structure, and grandchild structure.

[0035] Next, the comparison unit 160 obtains a score for the anatomical location 320 (step S1030). That is, a value is assigned by multiplying the match rate, which takes into account the parent-child structural relationship of the anatomical location 320, by a weight. For example, if the anatomical location 320 is cerebrum > cerebral white matter in the disease lesion information 210 and cerebrum in the case lesion information 220, one item matches in two, and 1 / 2 points are assigned. In the case of a perfect match, 1 point is calculated. On the other hand, if the lesion type determined in step S1010 is a neoplastic lesion, even if the anatomical locations do not match, the process is not terminated and a basic score of 1 point is assigned. Then, if the anatomical locations match completely or partially, the degree of match is evaluated in the same manner as for the non-neoplastic lesion, and the degree of match is added to the basic score of 1 point (step S1035).

[0036] After completing the anatomy score evaluation in step S1030 or step S1035, the comparison unit 160 proceeds to comparing the image findings 330. As the image findings 330 have multiple elements as described in FIG. 6, each of these elements is compared. First, the imaging methods are compared (step S1040). The imaging method refers to the modality information 610. If there is a mismatch, the score is set to 0 and the series of processes ends. If there is a match, the finding information 620 is judged (step S1050). If there is a mismatch, the score is set to 0 and the series of processes ends.

[0037] Here, since the modality information 610 and the finding information 620 match, the finding score is set to 1 point (step S1060). Furthermore, the comparison unit 160 compares the modifiers 630 and adds the number of matches to the finding score (step S1070). If there is at least one modifier 630 and it matches, 1 point is added; if both modifiers 630 have multiple modifiers and some of them match, 1 point is added according to the number of matches. In this way, the finding score is calculated.

[0038] Finally, the comparison unit 160 multiplies the anatomical score calculated in step S1030 by the findings score calculated in steps S1040 to S1070 (step S1080). The value obtained by the multiplication is set as the similarity score.

[0039] As described above, the disease lesion information 210 and the case lesion information 220 are compared one by one, but it is assumed that there are multiple disease lesion information 210 and multiple case lesion information 220, and the similarity is evaluated and calculated for all combinations. This process of comparing multiple disease lesion information 210 and multiple case lesion information 220 one by one will be described with reference to FIG. 9.

[0040] 9 is a diagram showing the process of calculating the similarity between a case and a disease. Since the case will be described where there are multiple pieces of disease lesion information 210 and case lesion information 220, for convenience, lesion information IDs 1, 2, ... are assigned to each piece. The disease lesion information 210 and case lesion information 220 are compared in the order of combinations 810 to 840.

[0041] The comparison unit 160 proceeds with the comparison process as shown in combinations 810 to 840. That is, combination 810 compares the disease lesion information 210 having a lesion information ID of 1 with the case lesion information 220 having a lesion information ID of 1 to obtain a similarity score of 10. Combination 820 compares the disease lesion information 210 having a lesion information ID of 1 with the case lesion information 220 having a lesion information ID of 2 to obtain a similarity score of 5. Combination 830 compares the disease lesion information 210 having a lesion information ID of 2 with the case lesion information 220 having a lesion information ID of 1 to obtain a similarity score of 20. Combination 840 compares the disease lesion information 210 having a lesion information ID of 2 with the case lesion information 220 having a lesion information ID of 2 to obtain a similarity score of 15.

[0042] The comparison unit 160 obtains a list of similarity scores, as shown in list 850, as a result of comparing the combinations 810 to 840 for the lesion IDs of the cases and diseases.

[0043] 10 is a diagram showing the similarity score calculation process. The comparison unit 160 has obtained similarity scores by comparing the disease lesion information 210 with the case lesion information 220 in the order of combinations 810 to 840, so now we will explain the overall similarity evaluation based on the individual evaluations.

[0044] The determination unit 170 sorts the list 850 shown in FIG. 9 that has already been obtained in order of similarity score. As a result of sorting, the determination unit 170 obtains a sorted list 860. As a result of sorting, it is found that combination 830 has the highest similarity score, and therefore determines that the case indicated by lesion information ID2 is most similar to the disease indicated by lesion information ID1. As a result of the determination, the determination unit 170 adopts combination 910 (combination 830) and assigns the disease indicated by lesion information ID1 to the case indicated by lesion information ID2.

[0045] Since assignment has been completed for the case indicated by lesion information ID2, combination 830 and combination 840, which are the cases indicated by lesion information ID2, are deleted from sort list 860. Meanwhile, since a corresponding case has also been identified for the disease indicated by lesion information ID1, this is deleted from sort list 860. As a result, determination unit 170 obtains sort list 870. As shown in remaining combination 920 (combination 820), the case indicated by lesion information ID1 is assigned to the disease indicated by lesion information ID2.

[0046] FIG. 11 is a diagram showing an overview of how similarity between a disease and a case is calculated. Combination 910 and combination 920 are combinations of confirmed lesion information, with combination 910 having a similarity of 20 and combination 920 having a similarity of 5. All of these combinations are added together, and 20 + 5 = 25 is the similarity value between the disease and the case. In the example of FIG. 11, a simple addition is described, but the similarity value may also be calculated by squaring each value, for example, 202 + 52 = 425, so that the weight increases as the similarity increases. The following explanation will continue assuming that the similarity value is 20 + 5 = 25.

[0047] Figure 12 is a diagram showing an overview of the output process for a list of differential diseases. Based on the above explanation, similarities can be calculated for each disease for cases diagnosed by a doctor. As shown in Figure 12, for case 1, similarity scores of 25 for disease 1, 40 for disease 2, and 3 for disease 3 can be calculated, and the similarity scores for each disease can be sorted and presented in descending order. While the final diagnosis is made by a doctor, providing the doctor with scoring data to aid in this process in advance can help the doctor make a more accurate diagnosis.

[0048] This allows for flexible expression of anatomical structures, making it possible to express them clinically in a natural way. Furthermore, by combining modifiers, it becomes possible to express a wide variety of imaging findings. This allows for consistent expression of many anatomical structures and imaging findings, and by searching based on this, it becomes possible to quickly and thoroughly list differential diagnoses.

[0049] Although the present invention has been described above using examples, the technical scope of the present invention is not limited to the scope described in the above examples. It will be apparent to those skilled in the art that various modifications and improvements can be made to the above examples. It is clear from the claims that such modifications and improvements can also be included within the technical scope of the present invention. [Explanation of symbols]

[0050] A diagnostic information processing device 100, a disease database 120, an input unit 140, a case information storage unit 150, a comparison unit 160, and a determination unit 170.

Claims

1. a disease database that stores disease lesion information, which is lesion information for each disease; an input unit for inputting case lesion information, which is lesion information of a site diagnosed as pathological by a doctor for each case; The disease lesion information and the case lesion information have a common structure that describes the characteristics of the lesion, and further a comparison unit that compares the characteristics of the lesions in the disease lesion information and the case lesion information to determine a similarity in element units; a determination unit that determines the similarity between the disease lesion information and the case lesion information for each case based on the element-by-element similarity determined by the comparison unit; A diagnostic information processing device comprising:

2. The disease lesion information and the case lesion information commonly include information on a lesion type, an anatomical location, and an imaging finding, The diagnostic information processing apparatus according to claim 1 , wherein the comparison unit obtains the degree of similarity by comparing the disease lesion information and the case lesion information in terms of lesion type, anatomical location, and imaging findings.

3. The diagnostic information processing device according to claim 2 , wherein the anatomical region is described with respect to other anatomical regions by a directed acyclic graph.

4. The imaging findings commonly include information on modality, a summary of findings, and modifier expressions; The diagnostic information processing device according to claim 2 or 3, wherein the comparison unit obtains the degree of similarity by comparing information on modality, summary of findings, and modifier expressions possessed by the disease lesion information and the case lesion information, respectively.

5. Both the disease lesion information and the case lesion information include at least one or more lesion information, the comparison unit compares the disease lesion information and the case lesion information one by one, calculates the similarity for each combination of lesion information, calculates the combination with the highest similarity as the similarity of the lesion information in the disease lesion information, and sums up the similarities of the lesion information in the disease lesion information to determine the similarity between the disease lesion information and the case lesion information; The diagnostic information processing device according to claim 1 .

Images