System for improving the efficiency of opinion creation work

Abstract

It is possible to provide a system, method, and program that utilizes artificial intelligence to improve the efficiency of findings processing in the workflow when requesting clinical tests or sharing pathological findings among multiple parties. [Solution] According to the present invention, by comparing a first finding statement created by a medical professional with a second finding statement created based on image data corresponding to the first finding statement, it is possible to automatically confirm whether the first finding statement is correctly related to the requested test and whether there are any typos, omissions, or errors in the diagnostic content of the finding statement, thereby providing an observation work efficiency improvement system that can streamline work related to finding statements by medical professionals.

Description

[Technical Field]

[0001] The present invention relates to a system for improving the efficiency of a finding work, a method for improving the efficiency of a finding work, and a program. [Background technology]

[0002] Clinical testing involves assessing a patient's condition by collecting samples such as urine, blood, and tissue, or by directly examining bodily functions through electroencephalograms and electrocardiograms. Among these, pathological testing is an important test used to make a definitive diagnosis. Pathological testing is carried out in the following steps: a doctor collects the sample, a clinical laboratory technician prepares the specimen, a pathologist examines the specimen under a microscope, and a diagnosis is made.

[0003] Furthermore, since only the collection of specimens during the pathology testing process, which is conducted at external testing facilities such as clinical testing centers, constitutes a medical procedure, some medical institutions collect specimens and then send them to clinical testing centers to perform clinical testing.

[0004] As such, the clinical testing process involves the exchange of specimens, samples, and accompanying documents between multiple departments, facilities, and doctors. However, because clinical testing is conducted in a medical setting, any unnecessary time and effort can have a significant impact, so it is important to streamline the work time of each individual involved. Note that these procedures are not limited to clinical testing, and similar issues occur in all situations involving medical professionals.

[0005] As a technology related to the efficient generation of findings in such clinical tests, Patent Document 1 discloses a method that makes it easy to grasp the output layout while editing a medical report or the like when creating it. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2010-160590 Summary of the Invention [Problem to be solved by the invention]

[0007] Human errors are likely to occur when work spans departments and facilities, and discrepancies between the text information submitted by doctors and the diagnosis results, spelling mistakes, typos, omissions in findings, and diagnostic errors can hinder efficiency. However, Patent Document 1 does not disclose how findings are exchanged between multiple locations and multiple doctors.

[0008] Therefore, an object of the present invention is to provide a system, method, and program that utilizes artificial intelligence to improve the efficiency of finding work in the workflow when requesting clinical tests or sharing pathological diagnosis findings among multiple parties. [Means for solving the problem]

[0009] The present invention is a system for improving the efficiency of a diagnosis work used between a first terminal and a second terminal, Medical Reports an acquisition unit that acquires a first finding sentence, which is a document, and image data corresponding to the first finding sentence; a finding generation unit that generates a second finding sentence based on source data including at least the acquired image data; a similarity calculation unit that confirms the certainty of the first finding sentence by calculating a similarity between the first finding sentence and the second finding sentence; and an output unit that outputs the first finding sentence as an output finding sentence when the similarity between the first finding sentence and the second finding sentence calculated by the similarity calculation unit exceeds a predetermined threshold. an integration unit that integrates and stores the output finding statement output by the output unit and at least the image data; a determination unit that determines whether the output finding statement has been accepted by a user; and a library that saves the output finding statement and the image data stored in the integration unit when it is determined that the output finding statement has obtained a reward; The present invention is a system for improving the efficiency of findings work, comprising:

[0010] Furthermore, the present invention preferably further includes an editing unit that accepts edits to the first finding sentence or the second finding sentence when the similarity between the first finding sentence and the second finding sentence calculated by the similarity calculation unit does not exceed a predetermined threshold, and the output unit outputs the finding sentence edited by the editing unit as an output finding sentence when the editing unit functions.

[0011] In the present invention, it is preferable that the apparatus further comprises an integration unit that integrates the output finding sentence output by the output unit with at least the image data and stores the integrated result.

[0012] Furthermore, it is preferable that the present invention further comprises a reward determination unit that determines whether the output finding statement has been accepted by the user, and a library that saves the output finding statement and the image data stored in the integration unit when it is determined that the output finding statement has received a reward.

[0013] Although the present invention is in the category of computer systems, it also exerts similar effects and advantages in other categories such as electronic signature methods and programs. [Effects of the Invention]

[0014] According to the present invention, by calculating the similarity between a first finding statement obtained by a certain medical professional and a second finding statement generated by the finding generation unit, it is possible to confirm whether the first finding statement is correctly related to the requested test and to point out typos, omissions, or errors in the diagnostic content of the finding statement, thereby making it possible to improve the efficiency of medical professionals' work related to finding statements. [Brief explanation of the drawings]

[0015] [Figure 1] FIG. 1 is a schematic diagram of a system 0 according to the present invention. [Figure 2] 1 is an overall configuration diagram of a system 0 according to a first embodiment of the present invention. [Figure 3] FIG. 2 is a flowchart showing a process executed by the system 0 according to the first embodiment of the present invention. [Figure 4] FIG. 2 is a diagram showing an example of an LLM4 model provided in the finding generation module 21 of the present invention. [Figure 5] FIG. 10 is a schematic diagram illustrating the selection of LLM4 provided in the finding generation module 21 of the present invention. [Figure 6]10 is an example of a second finding sentence generated by the finding generation module 21 of the present invention. [Figure 7] FIG. 2 is a schematic diagram showing a comparison of sentence vectors performed in the similarity calculation module 22 of the present invention. [Figure 8] FIG. 1 is an overall configuration diagram of a system 0 according to a second embodiment of the present invention. [Figure 9] FIG. 10 is a flowchart showing a process executed by the system 0 according to the second embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] The present invention will be specifically described using preferred embodiments. However, the following embodiments are merely examples of the present invention, and the effects of the present invention are not limited to those described in the embodiments of the present invention.

[0017] First, an overview of the system 0 for improving the efficiency of findings work according to the present invention will be described with reference to Fig. 1. Fig. 1 is a schematic diagram of the system 0 according to the present invention. The system 0 is used when exchanging observation statements between multiple users, and is composed of a first terminal 1 used by a first user and a second terminal 2 used by a second user. In the present invention, the term "findings" refers to documents such as diagnostic results and medical reports in which a medical professional such as a doctor gives his or her opinion on a disease or physical condition based on examination and test results. The first user and the second user are assumed to be medical personnel such as clinical laboratory technicians, pathologists, clinicians, nurses, etc. who belong to institutions where pathological specimens are transmitted and received, such as medical institutions and testing centers. As a typical example, a clinician who examines a patient is assumed to be the first user, and a pathologist who makes a pathological diagnosis at the request of the clinician is assumed to be the second user. In addition, the system 0 may include other terminals and devices in addition to the first terminal 1 and the second terminal 2, and the number, type and functions thereof are not particularly limited and can be designed as appropriate.

[0018] The first terminal 1 and the second terminal 2 may be realized by, for example, a mobile terminal such as a smartphone or a tablet terminal, or a computer such as a desktop computer or a notebook computer. Furthermore, the first terminal 1 and the second terminal 2 may be realized by, for example, a single computer, or may be realized by multiple computers such as a cloud computer. In this specification, a cloud computer refers to a computer that uses any computer in a scalable manner to perform a specific function, or a computer that includes multiple functional modules to realize a system, has those functions, and is equipped with a device, etc. as a communication unit, that enables communication with other terminals, devices, etc. In this case, the system 0 executes each process described below using the first terminal 1, the second terminal 2, and / or a combination of the other included terminals, devices, etc. Furthermore, the first terminal 1 and the second terminal 2 may be realized by the same device.

[0019] Necessary data and information are transmitted and received between the first terminal 1 and the second terminal 2, or between the second terminal 2 and other terminals, devices, etc., via information communication lines such as 4G / LTE and WiFi, or public line networks, etc. (referred to as network 4 in the figure). This communication may be wired or wireless.

[0020] Next, the system configuration of the system 0 in the first embodiment, which is a preferred embodiment of the present invention, will be described with reference to Fig. 2. Fig. 2 is a diagram showing the overall configuration of the system 0 in the first embodiment of the present invention.

[0021] The first terminal 1 and the second terminal 2 each include a central processing unit (CPU), a graphics processing unit (GPU), a random access memory (RAM), a read only memory (ROM), and the like as a control unit. Furthermore, the first terminal 1 and the second terminal 2 each include a data storage unit such as a hard disk, semiconductor memory, recording medium, memory card, etc. The data may be stored in a cloud service, database, etc. Furthermore, the first terminal 1 and the second terminal 2 each include a device as a communication unit that enables communication with other terminals, devices, etc. The communication method may be wireless or wired. Furthermore, the first terminal 1 and the second terminal 2 are assumed to have, as input units, functions necessary to enable operation of the first terminal 1 and the second terminal 2. Examples include an LCD display that realizes a touch panel function, a keyboard, a mouse, hardware buttons on the device, a microphone for voice recognition, etc. Furthermore, the first terminal 1 and the second terminal 2 are to have, as output units, functions necessary to enable operation of the first terminal 1 and the second terminal 2. Examples include display and audio output such as on an LCD display, a PC display, or projection on a projector.

[0022] The control unit of the second terminal 2 reads a predetermined program, and thereby cooperates with the communication unit to realize an acquisition module 20, a finding generation module 21, a similarity calculation module 22, an output module 23, and an editing module 24. Furthermore, these modules may be realized by a terminal on the cloud, and if the second terminal 2 is configured by a plurality of computers including a cloud computer, some of the modules may be realized on the cloud computer. It should be noted that it is not necessary for the second terminal 2 to implement all modules at once, and the system 0 may be used with only specific modules implemented.

[0023] An overview of the processing executed by the system 0 will be described with reference to Fig. 3. Fig. 3 is a diagram showing a flowchart of the processing executed by the system 0 according to the first embodiment of the present invention. As shown in FIG. 3, the system 0 of the first embodiment is composed of steps S1 to S9, and includes an acquisition module 20 (step S1) that acquires a first finding sentence and image data corresponding to the first finding sentence from a first terminal; a finding generation module 21 (steps S2 to S4) that generates a second finding sentence based on source data including at least the acquired image data; a similarity calculation module 22 (steps S5 to S8) that calculates the similarity between the first finding sentence and the second finding sentence; an output module 23 (steps S9-1, S10) that outputs the first finding sentence when the similarity between the first finding sentence and the second finding sentence calculated by the similarity calculation module 22 exceeds a predetermined threshold; and an editing module 24 (step S9-2) that accepts edits to the first finding sentence or the second finding sentence when the similarity between the first finding sentence and the second finding sentence calculated by the similarity calculation module 22 does not exceed the predetermined threshold.

[0024] In this specification, each process may be executed as a function that the process content has, or may be executed via a predetermined application, or may be executed by loading a predetermined program that includes the process.

[0025] The processing performed by System 0 will now be described in detail. The acquisition module 20 acquires the first finding sentence and source data 3 from which the finding sentence can be generated from the first terminal 1 (step S1). At this time, the communication method between the first terminal 1 and the second terminal 2 may be wired or wireless, and the first finding sentence and image data corresponding to the first finding sentence may be acquired using an external service or an external application such as email or cloud drive. The first finding statement acquired by the acquisition module 20 is, for example, the condition, symptoms, possible disease names, etc. of the patient from whom the specimen was taken, and is prepared by the medical institution or clinician that requests the pathological examination, etc. The source data 3 includes at least image data corresponding to the first finding statement, and is data that will be the source of the second finding statement in the subsequent flow. The image data corresponding to the first finding statement is image data captured for medical and research purposes, and includes at least one of the following as the subject of the image capture: an organ location, a description indicating pathological features, an examination result, and a diagnosis result. The file format of the first finding sentence and the source data 3 acquired by the acquisition module 20 is preferably some kind of image data, but the first finding sentence may be text data, and the file format of the image data and text data is not important.

[0026] The finding generation module 21 generates a second finding sentence based on the original data 3 acquired by the acquisition module 20 (steps S2 to S4). Note that this generation may be performed using only a part of the original data 3. Specifically, the finding generation module 21 includes LLMs (Large Language Models) 4. The LLMs included in the finding generation module 21 may be, for example, GPT (Generative Pre-trained Transformer), BERT (Bidirectional Encoder Representations from Transformers), or the like, and the type of method is not limited. The configuration and function of LLM4 will be explained based on Figure 4. Figure 4 shows an example of an LLM4 model provided in the finding generation module 21. LLM4 accepts words, sentences, or image data containing them as input data. First, the vector conversion layer converts the input data into vectors that can be calculated. The vector conversion layer also includes a process for vector conversion of the position information of words in each input data. Next, attention layers 1 and 2 determine which parts of this vector quantity are important and perform normalization. The feedforward layer then weights the calculation results. This process from attention layer 1 to the feedforward layer is a series of steps, and this process is repeated multiple times depending on the scale of the model. Finally, the output layer generates output data by adjusting the calculation results using linear functions, softmax functions, etc. to make them usable numerical values. The LLM4 in the present invention is in a state where it has previously been fully loaded with image data such as an organ location, a description indicating pathological characteristics, test results, and diagnosis results, as well as findings created based on that image data, as training data. By loading source data 3 into the trained LLM4, a second finding sentence corresponding to the source data 3 is generated. Note that the LLM4 can be retrained at any time.

[0027] Furthermore, the finding generation module 21 may include multiple LLMs 4. In this case, the second user of the second terminal 2 selects an LLM 4 specialized for one of the cases based on the first finding statement and the information described in the generation source data 3 (step S2). Specifically, an example is a case in which the second user selects an LLM 4B to which the clinical test to be performed (say, a blood test) belongs from among LLMs 4A specialized for cancer tumors, LLMs 4B specialized for blood diseases, LLMs 4C specialized for EEG abnormalities, etc. 5, a convolutional neural network (CNN) may be used when the second user selects an LLM4. In this case, the source data 3 is input into the CNN to extract features contained in the source data 3, and an LLM4 using training data that is closest to the extracted features is selected. When different source data 3' is input, an LLM4 different from the one used when the source data 3 was input is selected.

[0028] To summarize the processing of the finding generation module 21, first, the selection of an LLM4 is accepted from the second user (step S2), the source data 3 is imported into the selected LLM4 (step S3), and a second finding statement is generated in accordance with the information described in the imported source data 3 (step S4). An example of a second finding sentence generated by the processing of the finding generation module 21 is shown in Figure 6. The green shaded area in Figure 6(a) shows the source data 3 input into LLM4. By importing this data, the second finding sentence shown in Figure 6(b) is generated, such as "We see granulation tissue with glandular ductal epithelia and inflammatory cells lamina propria infiltration inflammatory cells are prominently present active inflammatory."

[0029] The similarity calculation module 22 calculates the similarity between the first finding sentence acquired by the acquisition module 20 and the second finding sentence generated by the finding generation module 21 (steps S5 to S8). An example of similarity calculation in the similarity calculation module 22 will be specifically described with reference to Fig. 7. Fig. 7 is a schematic diagram showing the comparison of sentence vectors performed in the similarity calculation module 22. The similarity calculation module 22 extracts keywords related to the positive and negative results of tests and the type of pathology from the first and second finding statements (step S5). The keywords extracted at this time include the type of histopathological type, the positive / negative determination result, and a scale indicating the progression of the disease, such as the stage category of Genomics England. A sentence vector is calculated for each extracted keyword, and a sentence vector group 5A indicating the feature quantities of the description content of the first finding sentence and a sentence vector group 5B indicating the description content of the second finding sentence are generated (step S6). The calculated sentence vector group may consist of a single sentence vector, or may be expressed in two or more dimensions. The distance between the region to which sentence vector group 5A belongs and the region to which sentence vector group 5B belongs is measured (step S7). If the distance between the regions exceeds a certain threshold, the first and second finding sentences are determined to be similar, and if it is less than the threshold, they are determined to be dissimilar (step S8). In addition, the sentence vector group 5B may be compared not only with the sentence vector group 5A, but also with a sentence vector group similarly calculated from the learning data used for learning LLM4, which has been stored in advance in the similarity calculation module 22.

[0030] Furthermore, in the finding generation module 21, a second finding statement' may be deliberately created that has a different opinion from the second finding statement created by selecting an LLM4 different from the LLM4 initially created, and the distance between the sentence vector group 5B' and the sentence vector group 5A may be measured by a similar process to confirm that the distance between the sentence vector group 5B generated from the second finding statement initially created and the sentence vector group 5A is shorter than the distance between the sentence vector group 5B' and the sentence vector group 5A.

[0031] When the similarity calculation module 22 determines that the first finding sentence and the second finding sentence are similar, the output module 23 outputs the first finding sentence (step S9-1). This output may be displayed on the screen of the second terminal 2, transmitted to another terminal by some means via a network, or printed by an external device connected to the second terminal 2. It should be noted that the output module 23 operates after the edit module 24 has functioned if the similarity calculation module 22 has determined that the first finding sentence and the second finding sentence are dissimilar.

[0032] The editing module 24 functions only when the similarity calculation module 22 determines that the first finding sentence and the second finding sentence are dissimilar, and accepts edits to the first finding sentence or the second finding sentence from the second user (step S9-2). This editing may be performed by someone other than the second user, using an input unit included in the second terminal 2, or using an external device connected to the second terminal 2. An example of editing content is adding "Inflammatory cells almost normal colonic mucosal epithelia." to the end of the sentence "We see granulation tissue with glandular ductal epithelia and inflammatory cells lamina propria infiltration inflammatory cells are prominently present active inflammatory." In other words, corrections such as correcting typos, changing the case, or changing the category to which the case belongs are possible. The first or second finding sentence that has been edited will be called the third finding sentence.

[0033] If the editing module 24 has functioned, the output module 23 outputs the third observation sentence after the editing is completed (step S10). This output is performed in the same manner as in step S9-1 described above.

[0034] Through the above process, by comparing the first finding statement created by the medical professional with the second finding statement created based on the image data corresponding to the first finding statement, it is possible to automatically confirm whether the first finding statement is correctly related to the requested test and whether there are any typos, omissions, or errors in the diagnostic content of the finding statement, thereby making it possible to improve the efficiency of work related to finding statements by medical professionals. Furthermore, if an error is found, it is possible to edit the text directly, so corrections to the findings can be easily made.

[0035] Hereinafter, a system 0 according to a second embodiment of the present invention will be described. The same functions and configurations as those in the first embodiment will be denoted by the same reference numerals, and descriptions thereof will be omitted.

[0036] The system configuration of a system 0 according to a second embodiment, which is another embodiment of the present invention, will be described with reference to Fig. 8. Fig. 8 is a diagram showing the overall configuration of the system 0 according to the second embodiment of the present invention. In the second embodiment, as in the first embodiment, the control unit of the second terminal 2 reads a predetermined program, and in cooperation with the communication unit, realizes an acquisition module 20, a finding generation module 21, a similarity calculation module 22, an output module 23, and an editing module 24. Furthermore, the second terminal 2 differs from the first embodiment in that it implements an integration module 25, a determination module 26, and a library module 27. It is not necessary to implement all modules at once in the second terminal 2, and the system 0 may be used with only specific modules implemented.

[0037] An outline of the processing executed by the system 0 according to the second embodiment of the present invention will be described with reference to Fig. 9. Fig. 9 is a diagram showing a flowchart of the processing executed by the system 0 according to the second embodiment of the present invention. As shown in FIG. 9, the system 0 of the second embodiment is composed of steps S1 to S13, and further includes an integration module 25 (step S11) that integrates and stores the output finding statement output by the output module with at least image data, a determination module 26 (step S12) that recognizes receipt of the output finding statement by the user as a reward and determines whether the output finding statement has acquired the reward, and a library module 27 (step S13) that saves the output finding statement and the image data stored in the integration unit when it is determined that the output finding statement has acquired the reward.

[0038] In the following explanation, it is assumed that the first finding sentence or the third finding sentence edited by the editing module 24 has been output from the output module 23 by the processing up to step S10.

[0039] The integration module 25 integrates the first finding sentence or the third finding sentence (hereinafter referred to as the output finding sentence) output by the output module 23 with the image data corresponding to the first finding sentence, and stores the integrated data in a temporary storage area (step S11). This integrated data may include data other than image data corresponding to the first finding statement.

[0040] The determination module 26 determines whether the output observation sentence has been accepted by a user, including the second user (step S12). An example of a case where a review is deemed to have been accepted is when a reward is exchanged between the institutions to which the first and second users belong or the recipients of the review.

[0041] When the determination module 26 determines that the output finding sentence has acquired a reward, the library module 27 copies or moves the integrated data stored in the primary storage area of ​​the integration unit and stores it (step S13). It is preferable that the integrated data stored in the library module 27 be used for re-learning the LLM4 included in the finding generation module 21.

[0042] Through the above processing, System 0 stores the output observation sentences as correct data and uses them for learning, which may make it possible to further optimize the LLM without preparing a large amount of additional learning data.

[0043] The above-described means and functions are realized by a computer (including a CPU, an information processing device, various terminals, etc.) incorporating and executing a predetermined program. This program may be provided, for example, from the computer via a network or by a cloud service. The program may also be provided in a form recorded on a computer-readable recording medium. In this case, the computer reads the program from the recording medium, transfers it to an internal or external recording device, records it, and executes it. The program may also be pre-recorded on a recording device and provided to the computer from the recording device via a communication line.

[0044] Although the embodiments of the present invention have been described above, the present invention is not limited to these embodiments. Furthermore, the effects described in the embodiments of the present invention are merely a list of the most preferable effects resulting from the present invention, and the effects of the present invention are not limited to those described in the embodiments of the present invention. [Explanation of symbols]

[0045] 0: System 1: First terminal 2: Second terminal 3:Original data 4:LLM 5: Text vector group 20: Acquisition module 21: Findings Generation Module 22: Similarity calculation module 23: Output module 24: Editing module 25: Integration module 26: Judgment module 27: Library module

Claims

1. A system for improving the efficiency of findings used between a first terminal and a second terminal, an acquisition unit that acquires, from the first terminal, a first finding statement that is a document of a diagnosis result or a medical report and image data corresponding to the first finding statement; a finding generator that generates a second finding sentence based on source data including at least the acquired image data; a similarity calculation unit that calculates a similarity between the first finding sentence and the second finding sentence to confirm the certainty of the first finding sentence; an output unit that outputs the first finding sentence as an output finding sentence when the similarity between the first finding sentence and the second finding sentence calculated by the similarity calculation unit exceeds a predetermined threshold; an integration unit that integrates the output finding statement output by the output unit with at least the image data and stores the integrated result; a determination unit that determines whether the output remark sentence has been accepted by a user; a library that stores the output finding sentence and the image data stored in the integration unit when it is determined that the output finding sentence has obtained a reward; A system for improving the efficiency of findings work, comprising:

2. an editing unit that accepts editing of the first finding sentence or the second finding sentence when the similarity between the first finding sentence and the second finding sentence calculated by the similarity calculation unit does not exceed a predetermined threshold; Further provided with 2. The system for improving efficiency of an observation work according to claim 1, wherein the output unit outputs an observation statement edited by the editing unit as an output observation statement when the editing unit is functioning.

3. A method executed by a system for improving the efficiency of findings used between a first terminal and a second terminal, comprising: acquiring, from the first terminal, a first finding statement, which is a document of a diagnosis result or a medical report, and image data corresponding to the first finding statement; generating a second finding statement based on source data including at least the acquired image data; confirming the likelihood of the first finding sentence by calculating a similarity between the first finding sentence and the second finding sentence; a step of outputting the first finding sentence as an output finding sentence when the similarity between the first finding sentence and the second finding sentence calculated in the step of calculating the similarity exceeds a predetermined threshold; a step of integrating and storing the output finding statement and at least the image data; determining whether the output finding statement is accepted by a user; When it is determined that the output finding sentence has acquired a reward, storing the stored output finding sentence and the image data; A method comprising:

4. A computer, A step of acquiring a first finding statement, which is a document of a diagnosis result or a medical report, and image data corresponding to the first finding statement; generating a second finding sentence based on source data including at least the acquired image data; a step of confirming the likelihood of the first finding sentence by calculating a similarity between the first finding sentence and the second finding sentence; a step of outputting the first finding sentence as an output finding sentence when the similarity between the first finding sentence and the second finding sentence calculated in the step of calculating the similarity exceeds a predetermined threshold; a step of integrating and storing the output finding statement and at least the image data; determining whether the output finding statement is accepted by a user; a step of saving the stored output finding sentence and the image data when it is determined that the output finding sentence has obtained a reward; A computer-readable program for executing the program.

5. The observation work efficiency improvement system described in claim 1, characterized in that it is used when exchanging observation statements between multiple users.

Images