Diagnostic and treatment aids
By using the detection, classification and calculation unit in the diagnosis and treatment auxiliary device to analyze the time series data of the pre-detected model and calculate the performance indicators, the problem of difficult model performance in the prior art is solved, and effective evaluation and update of model performance is achieved .
Patent Information
- Application Number
- CN202010902185.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-09-02
- Filing Date
- 2020-09-01
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2040-09-01
Smart Images

Figure CN112530575B_ABST
Abstract
Description
[0001] Cross-reference to related applications: This application is based upon and claims the benefit of priority from Japanese Priority Patent Application No. 2019-159598 filed on September 2, 2019, the latter application being incorporated herein by reference in its entirety. Technical Field
[0002] The invention relates to a diagnosis and treatment auxiliary device. Background Art
[0003] Unfavorable medical events that occur to patients are collectively referred to as adverse events. Adverse events include, for example, worsening of the patient's condition caused by diseases such as heart failure and stroke that require hospitalization, and the occurrence of side effects caused by treatment such as liver and kidney dysfunction caused by drugs. In this case, it is known that there is a technology that uses actual diagnosis and treatment data to build a pre-detection model and pre-detect harmful phenomena. When a harmful phenomenon is detected in advance, a doctor or the like implements an intervention (therapeutic intervention) to prevent the harmful phenomenon.
[0004] However, depending on the accuracy of the pre-detection, harmful phenomena may not be prevented, or unnecessary intervention may be implemented. For example, there is a problem such as the following: for the same patient, it is impossible to observe at the same time when there is a pre-detection and when there is no pre-detection. Similarly, there is a problem such as the following: for the same patient, it is impossible to observe at the same time when a harmful phenomenon occurs and when it does not occur. Therefore, in the case where a harmful phenomenon does not occur, it is impossible to distinguish between a case where a harmful phenomenon does not occur by implementing an intervention according to the pre-detection, and a case where a harmful phenomenon does not occur because the pre-detection is inappropriate and has nothing to do with the implementation of the intervention. That is, there is a problem that it is impossible to verify whether the pre-detection based on the pre-detection model is appropriate for each patient. In other words, there is a problem that the performance of the pre-detection model used to prevent harmful phenomena cannot be evaluated.
[0005] Patent Document 1: Japanese Patent Application No. 2017-502983
[0006] Patent Document 2: Japanese Patent Application No. 2016-504924
[0007] Patent Document 3: Japanese Patent Application No. 2015-533084
[0008] Patent Document 4: Japanese Patent Application Publication No. 2018-156415 Summary of the invention
[0009] The problem to be solved by the present invention is to evaluate the performance of a pre-detection model for preventing harmful phenomena.
[0010] The diagnosis and treatment support device of the embodiment comprises a detection unit, a classification unit and a calculation unit. The detection unit outputs whether there is a pre-detection at each of the multiple application moments of the pre-detection model of the harmful phenomenon applied to the time series diagnosis and treatment data. The classification unit classifies the detection events of the pre-detection model related to the multiple application moments into a plurality of patterns specified by the combination of the presence or absence of the pre-detection and the presence or absence of the medical event related to the harmful phenomenon. The calculation unit calculates the performance index for evaluating the pre-detection model based on the number of each of the plurality of patterns.
[0011] Effects of the Invention
[0012] The object of the present invention is to evaluate the performance of a pre-detection model for preventing harmful phenomena. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 It is a diagram showing an example of the configuration of a diagnosis and treatment support system including the diagnosis and treatment support device according to the embodiment.
[0014] Figure 2 Yes means Figure 1 A diagram showing an example of the configuration of a medical terminal.
[0015] Figure 3 Yes means Figure 1 A diagram showing an example of the structure of a diagnosis and treatment supporting device.
[0016] Figure 4 It means in Figure 1 A flowchart of an example of processing involving performance evaluation of a pre-detection model for preventing harmful phenomena performed in a diagnosis and treatment support system.
[0017] Figure 5 Yes means Figure 4 A flowchart of an example of preliminary detection processing related to model verification data in the processing of .
[0018] Figure 6 It is used to illustrate Figure 5 Figure 1 shows a diagram of the pre-test processing related to the model validation data.
[0019] Figure 7 Yes means Figure 4 A flowchart of an example of classification processing related to a series of event groups in the processing of .
[0020] Figure 8 It is used to illustrate Figure 7 A diagram of a classification process associated with a series of event groups.
[0021] Fig. 9 It is used to illustrate Figure 7A diagram showing an example of classification results in a classification process related to a series of event groups.
[0022] Fig.10 It means in Figure 4 The display process of the performance indicators is displayed in Figure 2 A schematic diagram of an example of a display screen of a display of a medical terminal.
[0023] Fig.11 is a display of the performance index in the first variant. Figure 2 A schematic diagram of an example of a display screen of a display of a medical terminal.
[0024] Fig.12 is a display of the performance index in the second variant. Figure 2 A schematic diagram of an example of a display screen of a display of a medical terminal.
[0025] Fig.13 is a display of the performance index in the third variant. Figure 2 A schematic diagram of an example of a display screen of a display of a medical terminal.
[0026] Fig.14 is a display of the performance index in the fourth variant. Figure 2 A schematic diagram of an example of a display screen of a display of a medical terminal.
[0027] Fig.15 is displayed in the display processing of the performance index of the fifth variant. Figure 2 A schematic diagram of an example of a display screen of a display of a medical terminal.
[0028] Fig.16A This is a diagram for explaining classification processing related to a series of event groups in the sixth modification example.
[0029] Fig. 16B This is a diagram for explaining classification processing related to a series of event groups in the sixth modification example.
[0030] Explanation of symbols
[0031] 1: medical terminal; 2: diagnosis and treatment auxiliary device; 3: diagnosis and treatment information storage device; 9: diagnosis and treatment auxiliary system; 11: processing circuit; 12: input interface; 13: display; 14: memory; 15: communication interface; 21: communication interface; 22: memory; 23: processing circuit; 111: indication function; 112: display control function; 231: collection function; 232: detection function (detection unit); 233: classification function (classification unit); 234: calculation function (calculation unit); 235: generation function (generation unit); A1: icon; A2: arrow; C1: cursor, I10, I20, I30, I40, I50, I60: display screen; I11, I21, I31, I41, I51, I61: display of performance indicators; I12: display of classification results; P1~P8: multiple modes. DETAILED DESCRIPTION
[0032] Typically, according to one embodiment, the diagnosis and treatment support device of the embodiment includes a detection unit, a classification unit, and a calculation unit. The detection unit outputs whether there is a pre-detection at each of the multiple application moments of the pre-detection model of the harmful phenomenon applied to the time series diagnosis and treatment data. The classification unit classifies the detection events of the pre-detection model related to the multiple application moments into a plurality of patterns specified by a combination of the presence or absence of the pre-detection and the presence or absence of medical events related to the harmful phenomenon. The calculation unit calculates a performance index for evaluating the pre-detection model based on the number of each of the plurality of patterns.
[0033] Hereinafter, the diagnosis and treatment assisting device of the present embodiment will be described with reference to the accompanying drawings. In addition, in the following description, the same symbols are marked for the components having the same or substantially the same functions as the components already described in the accompanying drawings, and the description is repeated only when necessary. In addition, even when representing the same parts, the sizes and proportions of each other may be represented differently according to the accompanying drawings.
[0034] Figure 1 FIG. 1 is a diagram showing an example of the configuration of the diagnosis and treatment support system 9 according to the present embodiment. Figure 1 As shown, the diagnosis and treatment support system 9 includes a medical terminal 1 , a diagnosis and treatment support device 2 , and a diagnosis and treatment information storage device 3 . Figure 2 Yes means Figure 1 A diagram showing an example of the structure of the medical terminal 1. Figure 3 Yes means Figure 1 A diagram of an example of a diagnosis and treatment supporting device 2.
[0035] The following, such as Figure 1 to Figure 3As shown, the following case is used as an example for explanation: a client-server system is formed by a medical terminal 1 and a diagnosis and treatment support device 2, in which the medical terminal 1 functions as a client and the diagnosis and treatment support device 2 functions as a server. Figure 1 As shown, the medical terminal 1, the diagnosis and treatment support device 2, and the diagnosis and treatment information storage device 3 are connected to each other so as to be able to communicate with each other via, for example, a hospital network such as a Local Area Network (LAN). In addition, the medical terminal 1 and the diagnosis and treatment support device 2 may also be integrally configured. In this case, the diagnosis and treatment support system 9 may be expressed as a system including the diagnosis and treatment support device 2 further having the same functions or elements as the medical terminal 1 described below, and the diagnosis and treatment information storage device 3. That is, the diagnosis and treatment support system 9 may not include the medical terminal 1.
[0036] The diagnosis and treatment support device 2 is a device that receives a processing request from the medical terminal 1 and executes a process corresponding to the received processing request. A predetermined server application corresponding to the integrated viewer installed in the medical terminal 1 is installed in the diagnosis and treatment support device 2. The diagnosis and treatment support device 2 determines, for example, the display method of the diagnosis and treatment data and performance index data displayed on the display of the medical terminal 1. Here, the performance index data is, for example, data indicating the results of the performance evaluation related to the pre-detection model calculated in the diagnosis and treatment support device 2. Figure 3 As shown, the diagnosis and treatment supporting device 2 has a communication interface 21 , a memory 22 and a processing circuit 23 .
[0037] The communication interface 21 performs data communication with external devices such as the medical terminal 1 and the diagnosis and treatment information storage device 3. The communication interface 21 has a communication circuit for performing the data communication. The communication circuit may be a communication circuit corresponding to wired communication or a communication circuit corresponding to wireless communication such as Wi-Fi (registered trademark) communication.
[0038] The memory 22 is a storage device such as a hard disk drive (HDD), a solid state drive (SSD), an integrated circuit storage device, etc. that stores various information. In addition to HDD, SSD, etc., the memory 22 can also be a drive device that reads and writes various information between removable storage media such as an optical disk (CD), a digital versatile disk (DVD), a Blu-ray Disc (BD) (registered trademark), and a flash memory. In addition, the storage area of the memory 22 can be in the diagnosis and treatment auxiliary device 2, or in an external storage device connected via a network. The memory 22 stores, for example, performance indicator data, medical image files, and diagnosis and treatment data. The memory 22 temporarily stores various data in process. In addition, the memory 22 stores a pre-detection model.
[0039] The pre-detection model is, for example, a machine learning model. As an algorithm of the machine learning model, decision trees, decision forests, neural networks, support vector machines, clustering, self-organizing maps, Bayesian networks, etc. can be used. The pre-detection model is set to input diagnosis and treatment data such as vital sign information and output the results of pattern classification, the results of aggregation, or various performance indicators.
[0040] The processing circuit 23 controls the overall operation of the diagnosis and treatment support device 2. The processing circuit 23 executes a program related to the performance evaluation of the pre-detection model for preventing harmful phenomena (hereinafter referred to as the performance evaluation program), and generates performance index data for evaluating the performance of the pre-detection model. As hardware resources, the processing circuit 23 has processors such as a CPU (Central Processing Unit), an MPU (Micro Processing Unit), and a GPU (Graphics Processing Unit), and memories such as a ROM (Read Only Memory) and a RAM (Random Access Memory).
[0041] The processing circuit 23 of the present embodiment performs a collection function 231 , a detection function 232 , a classification function 233 , a calculation function 234 , and a generation function 235 by a processor executing a program developed in a memory.
[0042] The collection function 231 collects model verification data from the medical information storage device 3, for example. The model verification data includes various data used in the model performance evaluation process. The collected model verification data is stored in the database of the memory 22. Here, the model verification data includes medical data and event data described later.
[0043] The detection function 232 outputs whether there is a pre-detection at each of the multiple application times of the pre-detection model of the harmful phenomenon applied to the time series medical treatment data. Specifically, the detection function 232 applies the pre-detection model of the performance evaluation object to the model verification data, thereby outputting time series data indicating whether there is a pre-detection related to each model detection time. Here, the detection function 232 is an example of a detection unit.
[0044] The classification function 233 classifies the detection events of the pre-detection model related to multiple application moments into multiple patterns specified by the combination of the presence or absence of pre-detection and the presence or absence of medical events related to harmful phenomena. Specifically, the classification function 233 determines a series of event groups corresponding to the detection moments of each model, and classifies the determined series of event groups into patterns. A series of event groups is related to multiple events that occur in a specific order and period. In addition, the classification function 233 aggregates the series of event groups after pattern classification and outputs the aggregated result. Here, the classification function 233 is an example of a classification unit.
[0045] The calculation function 234 calculates a performance index for evaluating the pre-detection model based on the number of each of the multiple patterns. In other words, the calculation function 234 calculates the performance index based on the result of the total pattern classification. The performance index is a value calculated based on the number of each of the multiple patterns of whether there is pre-detection, whether there is intervention, and whether there is harmful phenomenon. The performance index includes at least one of the incidence rate of intervention, the incidence rate of harmful phenomenon, the return index, the conditional sensitivity, the conditional specificity, the model compliance rate, and the model compliance validity index. Here, the calculation function 234 is an example of a calculation unit.
[0046] The generation function 235 generates image data for display. The image data for display includes, for example, image data for displaying performance indicators. The generated image data for display is output to the medical terminal 1, for example. Here, the generation function 235 is an example of a generation unit.
[0047] The medical terminal 1 is, for example, a device capable of comprehensively viewing medical information. A comprehensive viewer is installed in the medical terminal 1, and the comprehensive viewer is an application that comprehensively presents medical information to the user. The comprehensive viewer can also be installed in any manner such as a Web application, a fat client application, or a thin client application.
[0048] like Figure 2 As shown, the medical terminal 1 includes a processing circuit 11, an input interface 12, a display 13, a memory 14, and a communication interface 15. The processing circuit 11, the input interface 12, the display 13, the memory 14, and the communication interface 15 are connected to each other for communication via a bus, for example.
[0049] The processing circuit 11 controls the overall operation of the medical terminal 1. The processing circuit 11 includes a processor such as a CPU, an MPU, or a GPU, and a memory such as a ROM or a RAM as hardware resources.
[0050] The processing circuit 11 of the present embodiment performs an instruction function 111 and a display control function 112 by a processor that executes a program expanded into a memory.
[0051] The instruction function 111 transmits, for example, a display instruction received via the input interface 12 to the medical treatment support device 2 via the communication interface 15. The display instruction includes, for example, an instruction to display medical treatment data or performance index data related to a specific patient.
[0052] The display control function 112 displays the image data for display received from the diagnosis and treatment support device 2 on the display 13 in a predetermined display format. The predetermined display format may be set in advance and stored in the memory 14 or the like. The display format may be changed according to the output of the input interface 12.
[0053] The data received from the diagnosis and treatment support device 2 is not limited to image data for display, but may be various values. In this case, the display control function 112 may generate image data for display based on the received various values, similar to the processing circuit 23 of the diagnosis and treatment support device 2 described later.
[0054] The input interface 12 is realized by, for example, a mouse, a keyboard, or a touch panel for inputting instructions by touching the operation surface. The input interface 12 receives, for example, display instructions from an operator. The input interface 12 converts the display instructions from the operator into electrical signals and outputs the electrical signals to the processing circuit 11.
[0055] The display 13 displays various information. As the display 13, various arbitrary displays can be used appropriately. For example, as the display 13, a liquid crystal display (LCD: Liquid Crystal Display), a cathode ray tube (CRT: Cathode Ray Tube) display, an organic EL display (OELD: Organic Electro Luminescence Display) or a plasma display can be used. In addition, the display 13 can be a desktop, or it can be composed of a tablet terminal that can wirelessly communicate with the main body of the medical terminal 1. In addition, as the display 13, one or more projectors can also be used.
[0056] The memory 14 is a storage device such as a HDD, SSD, or integrated circuit storage device that stores various information. In addition to the HDD, SSD, etc., the memory 14 may also be a drive device that reads and writes various information between removable storage media such as CD, DVD, BD, and flash memory. In addition, the storage area of the memory 14 may be in the medical terminal 1 or in an external storage device connected via a network. For example, the memory 14 stores information received from the diagnosis and treatment auxiliary device 2.
[0057] The communication interface 15 performs data communication with the diagnosis and treatment support device 2 and the diagnosis and treatment information storage device 3 connected via an electrical communication line such as an intra-hospital network. The communication interface 15 has a communication circuit for performing the data communication. The communication circuit can be a communication circuit corresponding to wired communication, or a communication circuit corresponding to wireless communication such as Wi-Fi (registered trademark) communication. The communication standard between the diagnosis and treatment support device 2 and the diagnosis and treatment information storage device 3 can be any standard, for example, HL7 (Health Level 7), DICOM (Digital Imaging and Communications in Medicine) or both thereof can be listed.
[0058] As the medical information storage device 3, for example, a VNA (Vendor Neutral Archive) system can be used. The VNA system as the medical information storage device 3 can be a system that stores various information, or a system in which multiple VNA systems are connected separately. In addition, for example, as the medical information storage device 3, multiple data servers connected to the network can also be used in combination. Below, as an example of the medical information storage device 3, the VNA system is described.
[0059] The medical information storage device 3 is a comprehensive archiving system that uniformly manages medical image files stored in medical image management systems (Picture Archiving and Communication Systems: PACS) manufactured by different manufacturers, as well as various medical data managed by various clinical department systems. The medical information storage device 3 is connected to a PACS (not shown) and an electronic medical record system (not shown) via a hospital network such as a LAN so as to be able to communicate with each other. In addition, the various information managed and stored by the medical information storage device 3 is not necessarily limited to information obtained from systems manufactured by different manufacturers, but may also be information obtained from a system manufactured by a single manufacturer.
[0060] The diagnosis and treatment information storage device 3, for example, periodically obtains medical image files stored in the PACS and stores them in a memory provided in the diagnosis and treatment information storage device 3. The medical image file is, for example, a file in a format that complies with the DICOM standard. The medical image file may also be referred to as DICOM data. The medical image file is generated by a medical image diagnostic device. The medical image diagnostic device performs an examination by photographing the patient. In addition, the medical image diagnostic device includes, for example, an X-ray computed tomography device, an X-ray diagnostic device, a magnetic resonance imaging device, a nuclear medicine diagnostic device, and an ultrasonic diagnostic device. The medical image file includes, for example, medical image data and accompanying information.
[0061] The medical image diagnosis apparatus collects raw data related to the patient by imaging, generates medical image data based on the collected raw data, and displays medical images based on the medical image data.
[0062] The supplementary information is to classify the corresponding medical image data and indicate the attributes, types, sources, etc. of the corresponding medical image data. The supplementary information of the medical image file includes, for example, information for identifying the medical image such as the examination UID (Unique IDentifier), series UID, patient ID, patient name, date of birth, treatment equipment code, and series description.
[0063] The examination UID is an identifier that can uniquely identify the examination. The series UID is an identifier that can uniquely identify a series of image groups, for example, obtained according to each imaging part or each imaging condition. The patient ID is an identifier assigned to each patient, for example, used to uniquely identify the patient in one hospital. The patient name indicates the name of the patient corresponding to the patient ID. The date of birth indicates the date of birth of the patient corresponding to the patient ID. The treatment equipment code is an identifier used to identify the type of treatment equipment, for example, "CT", "MR" and "US" are specified. "CT", "MR" and "US" mean medical images photographed by X-ray computed tomography, magnetic resonance imaging and ultrasonic diagnostic equipment, respectively. When there are special matters that the inspection technician should convey to the doctor in advance during the inspection (photography), the series description indicates the content.
[0064] Furthermore, the medical information storage device 3 periodically acquires information related to electronic medical records stored in the electronic medical record system, for example, and stores the information in a memory provided in the medical information storage device 3. The information related to the electronic medical records includes, for example, basic patient information and medical data.
[0065] The patient's basic information is information unique to the patient, including, for example, the patient's ID, name, date of birth, gender, and age.
[0066] Diagnostic data refers to information that medical personnel can learn about the patient's physical condition, disease condition, and treatment during the diagnosis and treatment process. Diagnostic data includes, for example, data obtained in various environments such as devices from different manufacturers, devices of different versions, and the same device but with different settings. Diagnostic data is not limited to objective data such as numerical values, but can also be non-numerical data, such as subjective data represented by text. Diagnostic data includes, for example, examination history information, image information, report information, electrocardiogram information, vital signs information, medication history information, medical record information, and nursing record information.
[0067] The examination history information is information indicating a history of examination results obtained as a result of performing a specimen examination or a bacterial examination on a patient, for example.
[0068] Image information is, for example, information indicating the location of a medical image acquired by imaging a patient, etc. Image information includes, for example, information indicating the location of a medical image file generated by a medical image diagnosis apparatus as a result of performing an examination.
[0069] The report information is, for example, information that is compiled by the radiology department's reading doctor reading medical images such as X-ray images, CT images, MRI images, and ultrasound images in response to an inspection request from a medical doctor in the diagnosis and treatment department, and summarizing the patient's condition and disease. The report information includes, for example, reading report information indicating a reading report prepared by the reading doctor with reference to the medical image file stored in the PACS. The report information includes, for example, the patient ID, patient name, and date of birth of the patient corresponding to the medical image file that is the reading target.
[0070] The electrocardiogram information is, for example, information related to an electrocardiogram waveform measured from a patient. The vital sign information is, for example, basic information related to the life of a patient.
[0071] Vital sign information includes, for example, pulse rate, respiratory rate, body temperature, blood pressure, and level of consciousness.
[0072] The medication history information is information indicating, for example, the history of medication dosages administered to a patient.
[0073] Medical record information is information input into an electronic medical record by, for example, a treating physician, etc. Medical record information includes, for example, medical records during hospitalization, the patient's medical history, and prescription records of medicines.
[0074] The nursing record information is information input into the electronic medical record by, for example, a nurse, etc. The nursing record information includes nursing records during hospitalization, etc.
[0075] In addition, the processing circuit 11, the processing circuit 23 and the processing circuit of the medical information storage device 3 can also be implemented by application specific integrated circuits (Application Specific Integrated Circuit: ASIC) and programmable logic devices (Programmable Logic Device: PLD). Here, PLD includes simple programmable logic devices (Simple Programmable Logic Device: SPLD), complex programmable logic devices (Complex Programmable Logic Device: CPLD), and field programmable gate arrays (Field Programmable Gate Array: FPGA).
[0076] In addition, the functions 111 to 112 and the functions 231 to 235 are not limited to being implemented by a single processing circuit, but may be implemented by combining a plurality of independent processors to form a processing circuit and by executing a program on each processor.
[0077] Next, the performance evaluation of the pre-detection model for preventing harmful phenomena (hereinafter referred to as model performance evaluation) implemented in the diagnosis and treatment support system 9 of the present embodiment will be described with reference to the drawings.
[0078] Here, the model performance evaluation of the present embodiment is a performance evaluation related to "a model that performs preliminary detection within 2 days before the occurrence of acute heart failure based on feature quantities obtained from vital signs."
[0079] In addition, adverse events refer to all medical events that are not preferred by the patient. Adverse events include deterioration of the patient's condition caused by the disease, the occurrence of side effects caused by treatment, etc. Examples of deterioration of the patient's condition caused by the disease include heart failure and stroke that require hospitalization. Examples of side effects caused by treatment include liver dysfunction and kidney dysfunction caused by drugs.
[0080] Usually, in order to make predictions (preliminary detection) and prevent harmful phenomena, it is necessary to infer counterfactuals. However, for the same patient, it is impossible to observe with and without pre-detection at the same time, so there is a problem that it is impossible to judge whether the prediction results of the pre-detection model for each patient (hereinafter referred to as model detection results) are appropriate. For example, when there is model detection and no harmful phenomenon occurs, it is impossible to distinguish whether the model detection result is incorrect or the model detection result is correct but the harmful phenomenon does not occur due to intervention. That is, there is a problem that it is impossible to judge whether the model performance has changed, especially decreased, during operation.
[0081] Therefore, as described below, in the model performance evaluation of the present embodiment, a performance index for evaluating the model performance is introduced. Figure 4 It means in Figure 1 A flowchart of an example of a process involving performance evaluation of a pre-detection model for preventing harmful phenomena (hereinafter referred to as model performance evaluation process) executed in the diagnosis and treatment support system 9.
[0082] In step S101, the detection function 232 performs a preliminary detection process (hereinafter referred to as a detection process) on the model verification data. The details of the detection process will be described later. Thereafter, the process proceeds to step S102.
[0083] In step S102, the classification function 233 performs classification processing (hereinafter referred to as classification processing) related to a series of event groups. The details of the classification processing will be described later. Thereafter, the process proceeds to step S103.
[0084] In step S103, the calculation function 234 performs a calculation process of the performance index (hereinafter referred to as the calculation process). The details of the calculation process will be described later. Thereafter, the process proceeds to step S104.
[0085] In step S104, the generation function 235 performs a display process of the performance index (hereinafter referred to as a display process). The display process will be described in detail later. After that, the model performance evaluation process ends.
[0086] (About detection processing)
[0087] Here, with reference to the accompanying drawings Figure 4 The detection processing performed in step S101 is described in detail. Figure 5 It means Figure 4 Flowchart of an example of pre-detection processing related to model verification data in the processing of . Figure 6 It is used to illustrate Figure 5 Figure 1 shows a diagram of the pre-test processing related to the model validation data.
[0088] In step S201, the collection function 231 collects time-series medical data (model verification data) covering a predetermined model performance evaluation target period from the database in the memory 22. As the medical data, for example, vital sign information is used. Figure 6 In the example shown, as the collected medical treatment data, vital signs over the model performance evaluation target period are shown. The predetermined model performance evaluation target period is, for example, one year.
[0089] In step S202, the detection function 232 applies the pre-detection model of the performance evaluation object to the vital sign information (diagnosis and treatment data) at a predetermined model execution interval. The predetermined model execution interval is, for example, one day. Figure 6 As shown, the pre-detection model is executed at a fixed time every day (model detection time), and as a result, information on whether there is pre-detection (with model detection) or no pre-detection (without model detection) can be obtained. Figure 6 As shown, when the target period of the performance evaluation related to the pre-detection model is 1 year, the pre-detection model is applied to 365 time points.
[0090] In step S203, the detection function 232 outputs the model detection result obtained by applying the pre-detection model to the diagnosis and treatment data (vital sign information) in step S202 to the memory 22. The model detection result is time series data related to the presence or absence of pre-detection, indicating whether there is pre-detection at each time of multiple model detection times. After that, the detection process ends.
[0091] (About classification processing)
[0092] Here, with reference to the accompanying drawings Figure 4 The classification processing performed in step S102 is described in detail. Figure 7 Yes means Figure 4 A flowchart of an example of classification processing related to a series of event groups in the processing of . Figure 8 It is used to illustrate Figure 7 A diagram of a classification process associated with a series of event groups. Fig. 9 It is used to illustrate Figure 7 A diagram showing an example of classification results in a classification process related to a series of event groups.
[0093] In step S301, the collection function 231 collects event data (model verification data) from the database of the memory 22. Here, the event data is time series data related to a series of event groups that occur in a specific order and during a period, such as the presence or absence of pre-detection (detection event), the presence or absence of intervention (intervention event), and the presence or absence of harmful phenomena (medical events). A series of event groups is an event group corresponding to an arbitrary model detection time among the event groups included in the event data.
[0094] In step S302, the classification function 233 determines the model detection time (model execution timing). After that, the classification function 233 determines the event group object period. The event group object period is a period in the entire period of event data that is processed as a history of interventions and harmful phenomena corresponding to any model detection results (a series of event groups). That is, whether or not each event of pre-detection, intervention, and harmful phenomena occurs is whether or not each event occurs during the event group object period. The event group object period is, for example, a period starting from the model detection time. As an event group object period, for example, there is a harmful phenomenon determination period and an intervention determination period.
[0095] The harmful phenomenon determination period is a period for determining whether a harmful phenomenon corresponding to the determined model detection result has occurred. The intervention determination period is a period for determining whether an intervention corresponding to the determined model detection result has occurred. For example, Figure 8 As shown in the first row of , the harmful phenomenon determination period and the intervention determination period are set to 2 days and 1 day respectively. In this way, the harmful phenomenon determination period (a prescribed period related to whether a harmful phenomenon occurs) and the intervention determination period (a prescribed period related to whether an intervention is implemented) are set independently.
[0096] In addition, the event group object period may also be a period starting from a specified date and time such as 0:00 a.m. every day. In this case, a pre-detection determination period may also be set as a period for determining whether there is a pre-detection (detection event). In the detection process, the pre-detection model is executed every day, so the pre-detection determination period is, for example, 1 day.
[0097] In step S303, the classification function 233 determines whether or not a pre-detection based on the pre-detection model has been performed for the determined model detection time. The result of this determination is stored in the memory 22, for example.
[0098] In step S304, the classification function 233 determines whether a harmful phenomenon has occurred during the determined harmful phenomenon determination period. Here, there is no such phenomenon that heart failure will definitely occur after a certain phenomenon occurs, and it is sometimes difficult to make a unified heart failure determination. Therefore, in this determination, when active treatment is required, that is, when a more invasive intervention for heart failure (intervention event) occurs, it is determined that a harmful phenomenon has occurred. As a more invasive intervention for heart failure, there are, for example, unscheduled emergency surgery, Extracorporeal Membrane Oxygenation (a life support method for patients with severe respiratory failure or severe heart failure: ECMO), the use of cardiotonic drugs and catecholamines, and the use of diuretics (intravenous injection). Of course, the death of the patient is determined to be the occurrence of a harmful phenomenon. The result of this determination is stored in the memory 22, for example.
[0099] In addition, in determining whether a harmful phenomenon has occurred, for example, a clear determination result of whether heart failure has occurred by a doctor or the like may be used.
[0100] In step S305, the classification function 233 determines whether it is determined in step S304 that a harmful phenomenon has occurred. If it is determined that a harmful phenomenon has occurred, the process proceeds to step S307, and if it is not determined that a harmful phenomenon has occurred, the process proceeds to step S306.
[0101] In step S306, the classification function 233 determines whether intervention is implemented during the determined intervention determination period. In step S307, the classification function 233 determines whether intervention is implemented during the determined intervention determination period, from the time of model detection to the time of occurrence of the harmful phenomenon. In the determination of whether intervention is implemented in steps S306 and S307, for example, when a less invasive intervention for heart failure (intervention event) occurs, it is determined that intervention has occurred. As less invasive intervention events for heart failure, for example, there is the use of ACE inhibitors, the use of ARBs, the use of beta-blockers, the use of diuretics (oral), etc. The results of these determinations are stored in the memory 22, for example.
[0102] In addition, in determining whether or not an intervention has been performed, for example, a clear determination result of whether or not an intervention has been performed by a doctor or the like may be used.
[0103] Furthermore, when a medication event is used as an intervention event, a threshold value related to the medication period or medication amount may be set in order to determine whether or not an intervention has been performed.
[0104] In step S308, the classification function 233 classifies a series of event groups related to the determined event group object period into multiple patterns according to the results of each judgment. A series of event groups are the model detection results corresponding to any model detection moment (whether there is model detection: detection event), the judgment results of the occurrence of intervention (whether there is intervention), and the judgment results of the occurrence of harmful phenomena (whether there is harmful phenomenon). Fig. 9 As shown in FIG, multiple modes are defined by a combination of whether or not there is a pre-detection and whether or not there is a medical event associated with a harmful phenomenon. Medical events associated with a harmful phenomenon include whether or not there is an intervention and whether or not a harmful phenomenon occurs. Fig. 9 As shown, the classification function 233 classifies a series of event groups related to each event group target period into each pattern P1 to P8 according to the presence or absence of each event. Figure 8 This classification is described in more detail.
[0105] consider Figure 8 An example of a series of event groups is shown in the second row of . In this example, a pre-detection is performed at a determined pre-detection time. In addition, an intervention occurs during the intervention determination period. In addition, a harmful phenomenon occurs during the harmful phenomenon determination period. Therefore, this series of event groups is classified as Fig. 9 Mode P8 shown.
[0106] consider Figure 8An example of a series of event groups is shown in the third row of . In this example, the pre-detection is performed at the determined pre-detection time. In addition, the implementation of the intervention occurs during the implementation of the intervention determination period. On the other hand, no harmful phenomenon occurs during the harmful phenomenon determination period. Therefore, this series of event groups is classified as Fig. 9 Mode P7 shown.
[0107] consider Figure 8 An example of a series of events is shown in the fourth row of . In this example, the pre-detection is performed at the determined pre-detection time. In addition, the implementation of the intervention occurs during the implementation of the intervention determination period. However, the harmful phenomenon occurs during the harmful phenomenon determination period and before the implementation of the intervention. Therefore, this series of events is considered to have occurred without the implementation of the intervention and the harmful phenomenon has occurred, and is classified as Fig. 9 Mode P6 shown.
[0108] consider Figure 8 An example of a series of event groups is shown in the fifth row of . In this example, no pre-detection was performed at the determined pre-detection time. In addition, no intervention occurred during the intervention determination period. On the other hand, a harmful phenomenon occurred during the harmful phenomenon determination period. Therefore, this series of event groups is classified as Fig. 9 Mode P2 shown.
[0109] In step S309, the classification function 233 determines whether the classification is completed for all event group object periods. In this embodiment, the case where the event group object period is 1 year and the pre-detection model is executed every day is used as an example for explanation, so there are 365 periods as all event group object periods. If it is not determined that the classification is completed for all event group object periods, the processing returns to step S302, and the processing of steps S302 to S309 is repeatedly performed. If it is determined that the classification is completed for all event group object periods, the processing enters step S310.
[0110] In step S310, for example, Fig. 9 As shown, the classification function 233 totals the number of a series of event groups after classification for each of the patterns P1 to P8.
[0111] In step S311, the classification function 233 outputs the total result of a series of event groups classified into the respective patterns P1 to P8 to the memory 22. Thereafter, the classification process ends.
[0112] As described above, the pre-detection model of the performance evaluation object is applied multiple times to the diagnosis and treatment data (model verification data) of the time series. The multiple applications can be the model verification data related to one patient, that is, the application of multiple moments in one time series, or the model verification data related to multiple patients, that is, the application of at least one moment related to each of the multiple time series. That is, the total number of pattern classification results is the product of the number of event group object periods and the number of patients used in the performance evaluation process of the pre-detection model.
[0113] (About calculation processing)
[0114] Here, for Figure 4 The calculation processing performed in step S103 of is described in detail. The calculation function 234 calculates the performance index IC based on the number of each of the multiple patterns obtained by the classification processing (total result). The performance index IC includes at least one of the intervention incidence rate IC1, the harmful phenomenon incidence rate IC2, the return index IC3, the conditional sensitivity IC4, the conditional specificity IC5, the model compliance rate IC6, and the model compliance effectiveness index IC7.
[0115] In the following description, the total number of events is set to N ALL , the number of events (number of cases) of each mode P1 to P8 is set to N1 to N8 respectively.
[0116] (Intervention incidence)
[0117] The intervention incidence rate IC1 is a performance indicator that shows how much intervention has occurred to prevent harmful phenomena. The intervention incidence rate IC1 is calculated, for example, by the following formula (1). Fig. 9 The intervention incidence rate IC1 calculated from the total results shown is 31.0%.
[0118]
[0119] (Incidence rate of harmful phenomena)
[0120] The harmful phenomenon occurrence rate IC2 is a performance indicator showing how many harmful phenomena have occurred. The harmful phenomenon occurrence rate IC2 is calculated, for example, by the following formula (2). For example, by formula (2) based on Fig. 9 The total result shown shows that the harmful phenomenon occurrence rate IC2 is calculated to be 11.0%.
[0121]
[0122] (returns index)
[0123] The return index IC3 is a performance index expressed as an index related to the balance between the occurrence of intervention and the occurrence of harmful phenomena. The return index IC3 is calculated, for example, by the following formula (3). Here, X is the number of harmful phenomena that occur when no intervention is implemented. For example, when X=100, the return index IC3 is calculated based on the formula (3). Fig. 9 The calculated return index IC3 of the combined results shown is 53.1%.
[0124]
[0125] (with conditional sensitivity)
[0126] The conditional sensitivity IC4 is a performance indicator that is expressed as the sensitivity of the model estimated by limiting the situation where no intervention is implemented. The conditional sensitivity IC4 is calculated, for example, by the following formula (4). For example, by formula (4) based on Fig. 9 The conditional sensitivity IC4 calculated from the total results shown is 44.4%.
[0127]
[0128] (with conditional specificity)
[0129] The conditional specificity IC5 is a performance indicator that is expressed as the specificity of the model inferred by limiting it to the case where no intervention is implemented. The conditional specificity IC5 is calculated, for example, by the following formula (5). For example, by formula (5) based on Fig. 9 The conditional specificity IC5 calculated from the total results shown was 93.3%.
[0130]
[0131] (Model compliance rate)
[0132] The model compliance rate IC6 is a performance indicator that shows how much the model detection result is followed and the intervention judgment is made. The model compliance rate IC6 is calculated, for example, by the following formula (6). For example, by formula (6) based on Fig. 9 The combined results shown show a model compliance rate IC6 of 74.0%.
[0133]
[0134] (Model adheres to effective index)
[0135] The model compliance effectiveness index IC7 is a performance indicator expressed as an index related to the difference in results when the model is followed or not followed. The model compliance effectiveness index IC7 is calculated, for example, by the following formula (7). For example, by formula (7) based on Fig. 9The calculated model compliance validity index IC7 of the combined results shown is 1.04.
[0136]
[0137] (About display processing)
[0138] Here, with reference to the accompanying drawings Figure 4 The display processing performed in step S104 is described in detail. Fig.10 It means in Figure 4 The display process of the performance indicators is displayed in Figure 2 Schematic diagram of an example of a display screen I10 of the display 13 of the medical terminal 1.
[0139] In order to display the display screen I10 including the calculated performance index, the generation function 235 generates image data for display. The generated image data for display is output to the memory 22, or to the memory 14 and / or the display 13 of the medical terminal 1. In the medical terminal 1, the display control function 112 displays the display screen I10 on the display 13 based on the received image data for display. The user can evaluate the model performance based on the displayed performance index.
[0140] Here, in Fig.10 In the example shown, the item "Performance Index" is selected in the display screen I10. The selected item is expanded to display more detailed information (display of performance index I11). Fig.10 The display example of FIG. 1 shows a display I11 of performance indicators using a graph. The vertical axis and the horizontal axis of the graph represent the value of each performance indicator and time, respectively.
[0141] exist Fig.10 In the display example, the intervention incidence rate IC1, the harmful phenomenon incidence rate IC2, and the return index IC3 are shown as examples of the performance indicator values. In addition, as time, the year, month, and day every three months are shown as an example. The year, month, and day of each plot indicates the timing of the execution of the model performance evaluation process. For example, each plot on April 1, 2018 represents the performance indicator values calculated for the detection moments between January 1, 2018 and April 1, 2018. In this way, in the display I11 of the performance indicators using a chart, for multiple detection moments, the value of the performance indicator for each of the multiple detection moments is plotted. In the display screen I10, time information such as the date and time when the performance indicator is calculated can also be displayed.
[0142] Doctors and other users can evaluate model performance based on the displayed performance indicators. For example, when the incidence rate of intervention IC1 increases but the incidence rate of harmful phenomena IC2 does not decrease, it is determined that the model performance has decreased. In addition, for example, when the incidence rate of intervention IC1 does not change but the incidence rate of harmful phenomena IC2 increases, it is determined that the model performance has decreased. That is, users can evaluate model performance based on the ratio of the incidence rate of harmful phenomena IC2 to the incidence rate of intervention IC1.
[0143] In addition, as shown in equations (1) to (3), the return index IC3 is expressed using the ratio of the harmful phenomenon occurrence rate IC2 to the implemented intervention occurrence rate IC1. Specifically, when the ratio of the harmful phenomenon occurrence rate IC2 to the implemented intervention occurrence rate IC1 increases, the value of the return index IC3 decreases. That is, the user can also judge that the model performance has decreased based on the decrease in the return index IC3.
[0144] In addition, the performance index values are not limited to the intervention incidence rate IC1, the harmful phenomenon incidence rate IC2, and the return index IC3, and other performance indexes mentioned above can also be displayed in the same manner. For example, when a drug or treatment method that was not assumed when the pre-detection model was prepared is implemented as a therapeutic intervention, the number of cases classified as no model detection and intervention implementation increases. Even in such a case, according to the technology of this embodiment, it can be detected as a change in the return index IC3 and the model compliance rate IC6, that is, a decrease in model performance.
[0145] The model performance evaluation process is performed regularly, for example, every three months. When it is determined based on the performance index that the model performance has deteriorated, the pre-detection model is updated.
[0146] The model performance evaluation process is not limited to being performed regularly, and may be performed when a drug or treatment method not assumed in advance when the model was created is implemented as a therapeutic intervention or when a patient's condition changes.
[0147] In addition, the series of processes of the model performance evaluation process may also be executed in a divided manner. For example, the classification process may be executed periodically, and the results of the summary process may be accumulated in the memory 22, etc., and the display process may be performed at a timing corresponding to the user operation. Similarly, the results of the classification process and the detection results may be accumulated, and other processes may be performed at a timing corresponding to the user operation.
[0148] In addition, a part of the processing in a series of processes involved in the model performance evaluation processing may also be implemented in the medical terminal 1, etc., that is, outside the diagnosis and treatment support device 2. For example, the calculated performance index may be sent to the medical terminal 1, and image data for display may be generated in the medical terminal 1. In addition, for example, the test results may be sent to the medical terminal 1, and the performance index may be calculated in the medical terminal 1.
[0149] In addition, each performance index can also be calculated and displayed for each patient. In this case, the judgment involved in the performance evaluation of the pre-detection model includes the judgment of whether the pre-detection model is suitable for the current state of the patient. That is, according to the technology of this embodiment, it is also possible to perform performance evaluation on the pre-detection model adjusted according to each patient.
[0150] In addition, Figure 4 In the display processing performed in step S104 of , each performance indicator value may be plotted based on the timing (period) of the year, month, and day when the event is actually observed, or may be plotted based on the model detection time. For example, each performance indicator value plotted on April 1, 2018 may be calculated based on the number of harmful phenomena that occurred between January 1, 2018 and April 1, 2018. In this case, harmful phenomena that occurred after April 2, 2018 will not be used in the calculation of the performance indicators.
[0151] In addition, Figure 4 In the display process executed in step S104, the display scale of the graph used in the display I11 of the performance index may be changed. For example, when the scale of the horizontal axis is changed, each performance index value may be recalculated according to the display interval at that time.
[0152] In addition, Figure 4 In the display process executed in step S104, the display I12 of the mode classification may be performed as a more detailed content. For example, the display I12 of the mode classification may be performed based on the user operation. For example, Fig.10 As shown, when the cursor C1 is located on a plot or when the plot is selected, the result of the classification process used in calculating the performance index corresponding to the plot is displayed as a display I12 of the mode classification. For example, these user operations are obtained through the input interface 12 of the medical terminal 1. Fig.10 In the display example, a Sankey diagram is shown as a display I12 of pattern classification. The Sankey diagram shows the number distribution of each P1 to P8 of multiple patterns. More specifically, the Sankey diagram shows the transition of the number of pre-detection, intervention, and the presence or absence of harmful phenomena, and the thickness of the arrow representing the transition is determined according to the number of cases.
[0153] According to this configuration, the user can further confirm to what extent the intervention has been implemented according to the model detection result. In addition, the user can further confirm whether the model detection is appropriate, whether the intervention has been implemented according to the model detection, whether the intervention has been implemented before the occurrence of harmful phenomena, the judgment and countermeasures made when harmful phenomena occur, and the cause of the harmful phenomena.
[0154] Thus, in the diagnosis and treatment support system 9 of the present embodiment, the performance index of the pre-detection model for preventing harmful phenomena is calculated based on a series of event groups ("whether model detection is present", "whether intervention is implemented", and "whether harmful phenomena occur") occurring in a specific order and during a period. In the calculation of the performance index, a series of event groups are classified into each mode and the total is calculated according to each mode after classification. By displaying the performance index calculated in this way, the user can evaluate the model performance.
[0155] [Modifications]
[0156] Hereinafter, the diagnosis and treatment assisting device of each modified example will be described with reference to the accompanying drawings. Here, the differences from the above-mentioned embodiment will be mainly described. In addition, in the following description, the same symbols are marked for the components having the same or substantially the same functions as those in the above-mentioned embodiment, and the description will be repeated only when necessary.
[0157] (First Modification)
[0158] exist Figure 4 In the display process executed in step S104, the graph used in the display I21 of the performance index is not limited to Fig.10 A graph showing an example of the display. Fig.11 This is the information displayed in the display process of the performance index of this modification example. Figure 2 A schematic diagram of an example of a display screen I20 of the display 13 of the medical terminal 1. Fig.11 As shown, in the graph used for the display I21 of the performance index of the display screen I20 of this modification, the vertical axis and the horizontal axis are the incidence rate IC2 of harmful phenomena and the incidence rate IC1 of intervention implementation, respectively. Fig.11 In the graph of the display example of , the dashed line indicates the return index IC3. Therefore, according to the technique of this variation, the user can judge that the model performance is degraded when the depiction of each performance index moves to the upper right or lower left, that is, when the depiction of each performance index moves away from the dashed line.
[0159] (Second Modification)
[0160] exist Figure 4 In the display processing executed in step S104, in order to determine whether the model performance has deteriorated, the user may be notified when the value of each performance indicator satisfies a predetermined condition.
[0161] Fig.12 This is the information displayed in the display process of the performance index of this modification example. Figure 2 A schematic diagram of an example of a display screen I30 of the display 13 of the medical terminal 1. Fig.12 In FIG. 1 , a dotted line and an arrow A2 are used to represent a threshold value related to each performance indicator and a difference between each performance indicator value and the threshold value.
[0162] The various threshold values and / or threshold value ranges used in this modification are, for example, preset and stored in the memory 22, etc. For example, the generation function 235 of this embodiment determines whether each performance indicator value exceeds a predetermined threshold value range. In addition, this determination is not limited to all performance indicator values, and can also be implemented for at least one performance indicator value that is preset, for example.
[0163] When it is determined that each performance indicator value exceeds the threshold range, the generation function 235 notifies the user. Fig.12 As shown, notification is made through an icon A1 displayed together with a graph of the display I31 of the performance indicator. In other words, the generating function 235 generates image data for displaying a notification image when it is determined that each performance indicator value exceeds a predetermined threshold range, and the notification image notifies the user of the performance degradation of the pre-detection model. The notification image includes the icon A1. In addition, the display of the icon A1 may also flash. In addition, as shown by the dotted line and the arrow A2, the threshold value related to each performance indicator and the difference between each performance indicator value and the threshold value may be further displayed.
[0164] Thus, according to the technology of this variation, the user can easily recognize the reduction of model performance. In addition, the processing circuit 23 can also be configured to evaluate the model performance using each performance indicator value in this way, and determine whether the model performance is reduced, and when it is determined that the model performance is reduced, it can execute an update function to update the pre-detection model.
[0165] (Third Modification)
[0166] exist Figure 4 In the display process performed in step S104, performance indicators related to multiple pre-detection models can also be displayed. Fig.13 This is the information displayed in the display process of the performance index of this modification example. Figure 2 A schematic diagram of an example of a display screen I40 of the display 13 of the medical terminal 1. Fig.13 As shown, in the graph used for the display of the performance index I41, each pre-detection model is plotted. Fig.13In the display example shown, for the model detection times from January 1, 2018 to April 1, 2018, the return index IC3 calculated using the model detection results based on model B is further displayed. In this way, according to the technology of this variation, the user can easily compare the model performance for two or more pre-detected models.
[0167] (Fourth Modification)
[0168] exist Figure 4 In the display process performed in step S104, performance indicators related to multiple pre-detection models can also be displayed. Fig.14 This is the information displayed in the display process of the performance index of this modification example. Figure 2 A schematic diagram of an example of a display screen I50 of the display 13 of the medical terminal 1. Fig.14 As shown, the display screen I50 is provided with a tab for switching the screen for each of a plurality of pre-detection models to be compared. Fig.14 The display example of shows the state where model D is selected. As the number of pre-detected models and performance indicators of the comparison object increases, the display screen may become difficult to see. Therefore, according to the technology of this variation, the user can easily compare the performance of each model for multiple pre-detected models and multiple performance indicators without compromising the easy viewing of the display screen.
[0169] (Fifth Modification)
[0170] exist Figure 4 The display process performed in step S104 is not limited to the above-mentioned performance indicators, and other performance indicators may be displayed. Fig.15 This is the information displayed in the display process of the performance index of this modification example. Figure 2 Schematic diagram of an example of a display screen I60 of the display 13 of the medical terminal 1.
[0171] As other performance indicators, economic results such as costs related to the group to which the pre-detection model is applied, and clinical results such as mortality rates are further calculated. These performance indicators (hereinafter referred to as outcome indicators) are calculated, for example, by the calculation function 234 based on data collected from the medical information storage device 3, etc. The calculated indicators are output to the memory 22. In addition, the generation function 235 generates image data for displaying a display image containing the outcome indicators. As the outcome indicator, for example, at least one of the total number of deaths, in-hospital mortality rate, re-hospitalization rate, surgical implementation rate, number of days in hospital, medical expenses, number of examinations per day, and number of medications per day is used.
[0172] In addition, the total number of deaths can also be limited to causes of death associated with a specific disease. In addition, the in-hospital mortality rate can also be limited to causes of death associated with a specific disease. In addition, the re-hospitalization rate can also be limited to unscheduled hospitalizations. In addition, the surgery implementation rate can also be limited to unscheduled surgeries. In addition, the number of days of hospitalization can also be limited to wards such as ICU and CCU. In addition, medical expenses can also be limited to specific medical expense items. In addition, the number of examinations per day can also be limited to inpatients or outpatients. In addition, the number of medications per day can also be limited to inpatients or outpatients. Fig.15 In the example display shown, the average number of medications (types) and the average length of stay per day are displayed. For example, if the average number of medications increases but the average length of stay does not change, the user may suspect that an intervention is unnecessary. That is, according to the technology of this variant, the user can evaluate the model performance of the pre-detection model from more perspectives.
[0173] In addition, the result index can also be displayed together with the performance index of the above-mentioned embodiment and the first to fourth embodiments. In addition, each result index can be calculated for each patient group or for each doctor. Fig.15 As shown, the display can be switched by tags, or it can be displayed in the same screen as the technique of the third variant. Furthermore, when calculating each result according to the patient group, the patient group set by classifying the patients to whom the pre-detection model is applied according to "with or without model detection" and "with or without intervention" can also be used.
[0174] (Sixth Modification)
[0175] In the above embodiment, the following classification process is described as an example: the presence or absence of intervention for heart failure with a higher invasiveness (intervention event) and the presence or absence of intervention for heart failure with a lower invasiveness (intervention event) are respectively used to determine the occurrence of harmful phenomena and the presence or absence of intervention. However, the classification process is not limited to the occurrence of harmful phenomena and the presence or absence of intervention, and can also be implemented based on the degree (level) of the occurrence of harmful phenomena and the implementation of intervention. Fig.16A as well as Fig. 16B Each of them is a diagram for explaining classification processing related to a series of event groups in the sixth modification example.
[0176] In the classification process of this modification, Figure 7 In step S304, the classification function 233 determines whether a harmful phenomenon occurs and its level for the determined harmful phenomenon determination period. Similarly, in steps S306 and S307, the classification function 233 determines whether an intervention is implemented and its level for each determined period. In other words, regarding whether a medical event exists, the existence of a medical event includes multiple levels. For example, as multiple levels, Fig.16A as well as Fig. 16B 3 and 2 levels are shown as examples, respectively. In addition, multiple levels can be used for either or both of the occurrence of harmful phenomena and the implementation of intervention. In addition, in determining the level of occurrence of these harmful phenomena and / or the implementation of intervention, for example, the determination results based on doctors, etc. can also be used.
[0177] Thus, in the classification process of this modification, for example, referring to Fig. 9 The above-described summary result is further divided according to the number of levels to become a more detailed result. That is, according to the technique of this modification, it is possible to calculate a performance index corresponding to the presence or absence of a series of event groups and / or levels.
[0178] According to at least one embodiment described above, it is possible to evaluate the performance of a pre-detection model for preventing harmful phenomena.
[0179] The term "processor" used in the above description refers to circuits such as CPU, GPU, Application Specific Integrated Circuit (ASIC), Programmable Logic Device (PLD), etc. PLD includes Simple Programmable Logic Device (SPLD), Complex Programmable Logic Device (CPLD), Field Programmable Gate Array (FPGA). The processor implements functions by reading and executing programs stored in the storage circuit. The storage circuit storing the program is a non-temporary recording medium that can be read by a computer. In addition, instead of storing the program in the storage circuit, it is also possible to directly load the program into the circuit of the processor. In this case, the processor implements functions by reading and executing the program loaded into the circuit. In addition, it is also possible not to execute the program, but to implement the function corresponding to the program by a combination of logic circuits. In addition, each processor of the present embodiment is not limited to the case where each processor is configured as a single circuit, and multiple independent circuits can be combined to form one processor to implement its functions. Furthermore, it is also possible to Figure 1 , Figure 2 as well as Figure 3 Multiple components in a processor are integrated into one processor to realize its functions.
[0180] In addition, the processing circuit 23 may also have a circuit configuration that realizes the same function as the machine learning model of the embodiment, which learns parameters by inputting diagnosis and treatment data such as vital sign information and outputting the result of pattern classification, the result of aggregation, or each performance index. The circuit configuration is realized by, for example, an integrated circuit such as ASIC or PLD.
[0181] Several embodiments of the present invention have been described, and these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other ways, and various omissions, substitutions, and changes can be made without departing from the scope of the subject matter of the invention. These embodiments and their variations are included in the scope and subject matter of the invention, and are included in the invention described in the claims and the scope equivalent thereto.
Claims
1. A diagnosis and treatment auxiliary device, comprising: A detection unit that applies a pre-detection model of harmful phenomena to the time series medical treatment data multiple times, thereby outputting whether a pre-detection has been performed at each of the multiple application times; a classification unit that classifies the detection events of the pre-detection model related to the multiple application moments into a plurality of patterns defined by a combination of the presence or absence of the pre-detection and the presence or absence of a medical event related to a harmful phenomenon; and a calculation unit that calculates a performance index for evaluating the pre-detection model based on the number of each of the plurality of patterns; Whether the above medical events occurred, including whether intervention was implemented and whether harmful phenomena occurred, The above-mentioned multiple modes are determined by a combination of the presence or absence of the above-mentioned pre-detection, the presence or absence of the above-mentioned intervention, and the presence or absence of the above-mentioned harmful phenomenon.
2. The diagnosis and treatment assisting device according to claim 1, wherein: Regarding the presence or absence of the above-mentioned medical events, the presence of the above-mentioned medical events includes multiple levels.
3. The diagnosis and treatment assisting device according to claim 1, wherein: The occurrence of the above-mentioned medical events, including whether intervention was carried out during the first prescribed period and whether harmful phenomena occurred during the second prescribed period, The first predetermined period and the second predetermined period are set independently of each other.
4. The diagnosis and treatment assisting device according to claim 3, wherein: When the harmful phenomenon occurs within the second prescribed period from the time when the preliminary detection is performed, the classification unit determines whether the intervention is performed within the first prescribed period from the time when the preliminary detection is performed to the time when the harmful phenomenon occurs.
5. The diagnosis and treatment assisting device according to claim 3, wherein: When the harmful phenomenon has not occurred within the second predetermined period from the time when the preliminary detection is performed, the classification unit determines whether the intervention has occurred within the first predetermined period from the time when the preliminary detection is performed.
6. The diagnosis and treatment assisting device according to claim 3, wherein: The first mode of the plurality of modes is the sequence of the presence of the pre-detection, the occurrence of the harmful phenomenon, and the implementation of the intervention. The second mode among the plurality of modes is the sequence in which the pre-detection occurs, the intervention occurs, and the harmful phenomenon occurs.
7. The diagnosis and treatment assisting device according to claim 1, further comprising: a generating unit, generating an image including the calculated performance index; and The display unit displays the above-mentioned image.
8. The diagnosis and treatment assisting device according to claim 7, wherein: The image includes a graph showing values of the performance index corresponding to each of the plurality of applications, which are plotted at the time of the plurality of applications.
9. The diagnosis and treatment assisting device according to claim 8, wherein: The image further includes a graph showing the number distribution of each of the plurality of patterns.
10. The diagnosis and treatment assisting device according to claim 7, wherein: The generating unit is configured to further generate an image for notifying a user that the performance of the pre-detection model has deteriorated when the calculated performance index exceeds a predetermined threshold range.
11. The diagnosis and treatment assisting device according to claim 7, wherein: The calculation unit further calculates the result index including economic results and clinical results. The generating unit further generates an image including the result index.
12. The diagnosis and treatment assisting device according to claim 8, wherein: The above image contains the above chart and the time information when the above performance indicators were calculated.
13. The diagnosis and treatment assisting device according to claim 1, wherein: The above-mentioned multiple applications include applications at multiple times related to one time series.
14. The diagnosis and treatment assisting device according to claim 1, wherein: The plurality of applications include applications at at least one time point associated with each of a plurality of time series.
15. The diagnosis and treatment assisting device according to claim 1, wherein: The performance index is a value calculated based on the number of occurrences of the pre-detection, the intervention, and the harmful phenomenon in the plurality of patterns.
