System and method for ECG interpretation based on longitudinal criteria

Abstract

Methods and systems for electrocardiogram (ECG) interpretation based on longitudinal medical data are presented herein. In one example, a method includes, during a development phase of an ECG analysis model, generating a set of longitudinal ECG features (306) from a plurality of electrocardiograms (ECGs) with known diagnostics; extracting a longitudinal criterion from the plurality of ECGs (410); during a deployment phase of the ECG analysis model, obtaining a plurality of ECGs (502, 506) of the patient, where the plurality of ECGs includes a current ECG and one or more historical ECGs; determining a diagnosis (510) based on the longitudinal criterion; and sending the diagnosis to the user device (514).

Description

[0001] Cross-reference to related applications

[0002] This application claims priority to Greek Patent Application No. 20240100585, filed on August 22, 2024. The entire contents of the above application are hereby incorporated by reference for all purposes. Technical Field

[0003] The embodiments of the subject matter disclosed herein relate to longitudinal ECG parameters, and more specifically to systems and methods for generating longitudinal ECG parameters and using these longitudinal ECG parameters to determine diagnoses. Background Technology

[0004] ECG is a graphical representation of the heart's electrical activity and is typically represented as waveforms. An ECG analysis model receives an ECG, determines a set of features (e.g., findings), compares this set of features to corresponding criteria in a set of rules, determines a diagnosis based on the criteria met by the ECG features, and generates a diagnostic interpretation of the ECG. Physicians can review the ECG and / or diagnostic interpretation to assess a patient's cardiac activity.

[0005] The rules or rule set of an ECG analysis model can correspond to a specific diagnosis. Furthermore, the rules can include various criteria that, if met, allow the ECG analysis model to determine a specific diagnosis. Criteria can correspond to specific features of the ECG and can include relevant thresholds. As an example, an ECG analysis model might use criteria corresponding to a QRS duration greater than or equal to 120 ms. A QRS duration greater than 120 ms can be a necessary / prerequisite for an ECG analysis model to determine a specific diagnosis (such as left bundle branch block (LBBB)). However, other diagnoses (such as left ventricular hypertrophy (LVH)) may have similar ECG patterns and meet similar criteria of the ECG model. For any ECG analysis system, distinguishing a more impactful / more severe diagnosis (e.g., LBBB) from its similar diagnoses (e.g., for LBBB, LVH) is challenging. Summary of the Invention

[0006] In one example, a method includes: during the development phase of an ECG analysis model, generating a set of longitudinal ECG features from multiple electrocardiograms (ECGs) with known diagnoses; extracting longitudinal criteria from the multiple ECGs; during the deployment phase of the ECG analysis model, obtaining multiple ECGs of a patient, wherein the multiple ECGs include a current ECG and one or more historical ECGs; determining a diagnosis based on the longitudinal criteria; and sending the diagnosis to a user device.

[0007] It should be understood that the above brief description is provided to introduce selected concepts further described in the detailed embodiments in a simplified form. This is not intended to identify key or essential features of the claimed subject matter, the scope of which is uniquely defined by the claims following the detailed embodiments. Furthermore, the claimed subject matter is not limited to specific implementations that address any shortcomings pointed out above or in any part of this disclosure. Attached Figure Description

[0008] The invention will be better understood by referring to the following description of non-limiting embodiments, in which:

[0009] Figure 1 This is a diagram of an example system for interpreting electrocardiograms (ECG) based on longitudinal parameters;

[0010] Figure 2 yes Figure 1 A diagram illustrating example components of the apparatus of this example system;

[0011] Figure 3 This is a flowchart illustrating a method for generating vertical features;

[0012] Figure 4 This is a flowchart illustrating a method for using longitudinal features to determine the diagnosis of ECG;

[0013] Figure 5 A method for determining ECG diagnosis based on longitudinal criteria is shown;

[0014] Figure 6 This is a flowchart illustrating a method for using an AI model to extract criteria for determining a diagnosis by an ECG analysis model;

[0015] Figure 7 This is an example graph of ECG parameters used for primary diagnosis;

[0016] Figure 8 It is used for secondary diagnosis. Figure 7 Another example graph of ECG parameters; and

[0017] Figure 9 This is a sample user interface for patient records. Detailed Implementation

[0018] The following description relates to various implementation schemes for interpreting electrocardiograms (ECGs) based on longitudinal features. Specifically, systems and methods for generating longitudinal features and interpreting ECGs based on these longitudinal features are provided.

[0019] ECG analysis models can be used to determine the diagnosis of an ECG based on features and criteria identified within it. For example, an ECG analysis model can be configured with a set of rules that define a set of criteria (e.g., one or more features that can be satisfied together). The ECG analysis model can ingest ECG data to determine which criteria the ECG meets and output a diagnosis based on this. Generally, ECG analysis models reduce the time spent by clinicians, specialists, and other users manually analyzing ECGs.

[0020] However, ECG interpretation based on ECG analysis models is typically performed on the characteristics and criteria of a single ECG. For example, the characteristics of a single ECG are fed into an ECG analysis model, and a diagnosis is output. Analysis of a single ECG does not consider the progression of certain criteria over time. For example, trends in certain criteria can provide clinical significance for diagnostic determination. As a non-limiting example, the criteria used to diagnose left bundle branch block (LBBB) and the criteria used to diagnose left ventricular hypertrophy (LVH) may be similar within a single ECG. However, conversely, the trends of the criteria over time may differ for LBBB versus LVH.

[0021] In practice, when longitudinal data proves useful, subject matter experts (e.g., physicians, ECG technicians, etc.) can compare the current diagnosis with one or more of the patient's previous ECGs to confirm the diagnosis. However, manual comparisons can be time-consuming and prone to human error. For example, an expert might analyze the QRS duration, an ECG parameter. In LBBB, the QRS duration typically increases abruptly, while in LVH, the QRS duration increases gradually over time. However, for a patient with LBBB, an expert may not be able to review sufficiently old records to find ECGs prior to the abrupt jump in QRS duration, leading to the conclusion of a gradual increase in QRS duration, and thus potentially resulting in an incorrect LVH diagnosis. Furthermore, experts may not have the time or easy access to longitudinal records to examine the evolution of relevant parameters. Therefore, misdiagnosis is more likely to occur when information on relevant differences between more than one plausible diagnosis is unavailable or when there is no time to properly examine differences that may arise from longitudinal ECG data.

[0022] Therefore, this paper proposes a system and method for ECG interpretation based on longitudinal ECG features. As described herein, longitudinal features can include ECG features based on comparisons of at least two ECGs. For example, the difference between the values ​​of an ECG parameter in a first ECG and a second ECG can be a first longitudinal feature, the difference between the value of an ECG parameter and the average of all values ​​of that ECG parameter for a given patient can be a second longitudinal feature, and so on. Longitudinal criteria can include a combination of one or more longitudinal ECG features that can be satisfied by patient data (e.g., the patient's current or recent ECG and one or more historical ECGs). Longitudinal criteria can correspond to different diagnoses, such that satisfying a particular longitudinal criterion produces the output of the corresponding diagnosis.

[0023] By generating one or more longitudinal ECG features, a patient's ECG can be interpreted, for example, using an ECG analysis model based on criteria extracted from both longitudinal ECG features and individual ECG features. Therefore, by considering the patient's longitudinal data, ECG interpretation and diagnostic output via a rule-based model can be more accurate.

[0024] The systems and methods disclosed herein will now be described by way of example with reference to the accompanying drawings, in which... Figure 1 A diagram of an example ECG analysis system is shown. Figure 2 It shows Figure 2 A diagram showing example components of the apparatus of the example system is provided. Figures 3 to 5 The flowchart illustrates a method for generating longitudinal features, interpreting ECG based on longitudinal features, and extracting criteria for determining a diagnosis from an ECG analysis model. Figure 6 and Figure 7 Example ECG parameters for different diagnoses are shown as graphs over time; and Figure 9 A sample user interface for patient records is shown.

[0025] from Figure 1 The diagram of an example ECG analysis system 100 is shown at the beginning. The ECG analysis system 100 can be configured to extract interpretable criteria and generate longitudinal criteria for use in determining diagnoses by an ECG analysis model. Figure 1 As shown, system 100 may include an ECG device 110, multiple electrodes 128, an ECG analysis device 112, an ECG analysis model 114, a longitudinal standard generator 116, a platform 118, an AI model 120, a user device 122, a database 124, one or more medical data repositories 130, and a network 126.

[0026] ECG device 110 can be configured to generate ECG lines for a patient. For example, ECG device 110 can be a stand-alone ECG device, a portable ECG device, or a multi-vital sign monitoring device. The ECG device can receive cardiac electrical signals via the plurality of electrodes 128 and can generate ECG lines based on these cardiac electrical signals. The ECG can be a single-lead ECG, a 3-lead ECG, a 5-lead ECG, a 6-lead ECG, a 12-lead ECG, etc. ECG device 110 can include any number of electrodes 128. For example, ECG device 110 can include electrodes for generating a 12-lead ECG. In this example, the leads can include leads I, II, III, aVF, aVR, aVL, V1, V2, V3, V4, V5, and V6. ECG device 110 can be configured to use a subset of standard 12 leads, or alternatively use ECGs with non-standard lead placements (e.g., Holter monitoring lead placements), or use synthetic ECG leads from the actual acquired lead set (e.g., GE Healthcare's 12RL algorithm for assembling 12-lead ECGs from a reduced lead set).

[0027] ECG analysis device 112 can be configured to receive ECG from ECG device 110, determine the feature set of ECG using ECG analysis model 114, compare the feature set with the corresponding criteria of a rule set using ECG analysis model 114, determine a diagnosis using ECG analysis model 114, generate a diagnostic interpretation result of ECG using ECG analysis model 114, and send the diagnostic interpretation result to another device or display, etc.

[0028] ECG analysis model 114 can be a rule-based model, a machine learning model (e.g., a deep neural network, convolutional neural network, etc.), a decision tree, or other type of model configured to ingest ECG data and output a diagnosis. For example, when ECG analysis model 114 is a rule-based model, it can be configured to receive an ECG from ECG device 110, receive one or more previous ECGs from the patient from one or more medical data repositories 130, determine a feature set (e.g., single ECG features) and longitudinal features of the ECG and one or more previous ECGs, compare the single ECG feature set and longitudinal features with corresponding criteria of a rule set (e.g., single ECG criteria and longitudinal criteria), determine a diagnosis, and generate a diagnostic interpretation of the ECG. For example, ECG analysis model 114 could be Marquette... TM 12SL ECG analysis program, etc.

[0029] The longitudinal criteria generator 116 can be configured to generate a set of longitudinal criteria and corresponding diagnoses based on population data (including diagnoses and ECG findings), individual patient data (including diagnoses and ECG findings), and subject matter expert input. The generated longitudinal criteria and corresponding diagnoses can be fed back to the ECG analysis model 114 for storage and comparison with acquired patient data (e.g., current ECGs and one or more previous ECGs).

[0030] Platform 118 can be configured to use AI model 120 to extract interpretable criteria for diagnoses determined by ECG analysis model 114. For example, platform 118 can be a server or cloud computing system, etc. The extracted interpretable criteria may include one or both of single ECG criteria and longitudinal criteria.

[0031] AI model 120 can be configured to extract interpretable criteria for the diagnosis determined by ECG analysis model 114. As noted above, interpretable criteria may include one or both of single ECG criteria and longitudinal criteria. For example, AI model 120 may be a decision tree (e.g., a classification tree or regression tree), a linear regression model, a neural network (e.g., a deep neural network (DNN), a convolutional neural network (CNN), or a recurrent neural network (RNN)), a logistic regression model, or a support vector machine, etc.

[0032] User device 122 can be configured to receive criteria for diagnosis determined by ECG analysis model 114 and to provide the criteria for diagnosis determined by ECG analysis model 114 for display. For example, user device 122 can be a smartphone, laptop computer, desktop computer, wearable device, medical device, etc.

[0033] Database 124 can be configured to store ECGs, ECG feature sets, ECG diagnoses, ECG diagnostic interpretations, modification information, patient information associated with ECGs, or ECG classification targets, etc. Furthermore, database 124 can be configured to store criteria used to determine diagnoses by ECG analysis model 114, including single ECG criteria and / or longitudinal criteria. For example, database 124 can be a hierarchical database, a network database, a relational database, etc. In some examples, the criteria stored in database 124 can be initially generated by subject matter experts. For example, database 124 could be Marquette... TM 12SL ECG analysis program or other similar program database.

[0034] The one or more medical data repositories 130 may include one or more of the following: an electronic medical record (EMR) database, an ECG database, a radiology information system (RIS), a picture archiving and communication system (PACS), or other types of databases configured to store medical records for multiple patients. For example, an ECG performed on a patient may be stored in the ECG database along with associated data, including individual ECG features, corresponding diagnostic interpretations, acquisition dates, order providers, etc.

[0035] Network 126 can be configured to allow communication between devices of system 100. For example, network 126 can be a cellular network (e.g., fifth-generation (5G) network, long-term evolution (LTE) network, third-generation (3G) network, code division multiple access (CDMA) network, etc.), public land mobile network (PLMN), local area network (LAN), wide area network (WAN), metropolitan area network (MAN), telephone network (e.g., public switched telephone network (PSTN)), private network, self-organizing network, intranet, Internet or fiber-optic based network, etc., and / or combinations of these types or other types of networks.

[0036] As an example, the longitudinal standard generator 116 can communicate with the ECG analysis device 112 to transmit the generated longitudinal features via network 126. Furthermore, the ECG analysis device 112 can access data from database 124 and / or one or more medical data repositories 130 via network 126. In this way, system 100 can be an interconnected system through which its modules communicate with each other to perform various processes and / or methods.

[0037] Figure 1 The number and arrangement of devices in the system 100 shown are provided as examples. In practice, the system 100 may include additional devices, fewer devices, different devices, or devices related to… Figure 1 The devices shown are arranged differently. Additionally or alternatively, a group of devices (e.g., one or more devices) of system 100 may perform one or more functions described as being performed by another group of devices of system 100.

[0038] Figure 2 It shows Figure 1 The diagram illustrates example components of apparatus 200 in the example ECG analysis system 100. Apparatus 200 may correspond to ECG apparatus 110, ECG analysis apparatus 112, platform 118, user device 122, longitudinal standard generator 116, and / or database 124. Figure 2 As shown, the device 200 may include a bus 210, a processor 220, a memory 230, a storage component 240, an input component 250, an output component 260, and a communication interface 270.

[0039] Bus 210 includes components that allow communication between components of device 200. Processor 220 may be implemented in hardware, firmware, or a combination of hardware and software. Processor 220 may be a central processing unit (CPU), graphics processing unit (GPU), accelerated processing unit (APU), microprocessor, microcontroller, digital signal processor (DSP), field-programmable gate array (FPGA), application-specific integrated circuit (ASIC), or another type of processing component. Processor 220 may include one or more processors capable of being programmed to perform functions. Specifically, processor 220 may include one or more processors 220 configured to perform the operations described herein. Alternatively, a plurality of processors 220 may be collectively configured to perform the operations described herein, and each of the plurality of processors 220 may be configured to perform a subset of the operations described herein. For example, a first processor 220 may perform a first subset of the operations described herein, a second processor 220 may be configured to perform a second subset of the operations described herein, and so on.

[0040] Memory 230 may include non-transitory memory, random access memory (RAM), read-only memory (ROM), and / or another type of dynamic or static storage device (e.g., flash memory, magnetic memory, and / or optical memory) that stores information and / or instructions for use by processor 220. For example, memory 230 may store executable instructions thereon that can be executed by processor 220 to perform the operations described herein.

[0041] Storage component 240 may store information and / or software related to the operation and use of device 200. For example, storage component 240 may include hard disk (e.g., magnetic disk, optical disk, magneto-optical disk and / or solid-state disk), compact disc (CD), digital versatile disc (DVD), floppy disk, cassette, magnetic tape and / or another type of non-transitory computer-readable medium and corresponding drives.

[0042] Input component 250 may include components that allow device 200 to receive information, such as via user input (e.g., a touchscreen display, keyboard, keypad, mouse, buttons, switches, camera, and / or microphone for receiving reference audio and / or visual input). Additionally or alternatively, input component 250 may include sensors for sensing information (e.g., a Global Positioning System (GPS) component, accelerometer, gyroscope, and / or actuator). Output component 260 may include components that provide output information from device 200 (e.g., a display, a speaker for outputting sound at an output sound level, and / or one or more light-emitting diodes (LEDs)).

[0043] Communication interface 270 may include transceiver-like components (e.g., a transceiver and / or separate receiver and transmitter) that enable device 200 to communicate with other devices, such as via a wired connection, a wireless connection, or a combination of wired and wireless connections. Communication interface 270 may allow device 200 to receive information from and / or transmit information to another device. For example, communication interface 270 may include an Ethernet interface, an optical interface, a coaxial interface, an infrared interface, a radio frequency (RF) interface, a universal serial bus (USB) interface, a Wi-Fi interface, a cellular network interface, etc.

[0044] Apparatus 200 can execute one or more processors and methods described herein. Apparatus 200 can perform these processes and / or methods based on processor 220 executing software instructions stored in a non-transitory computer-readable medium such as memory 230 and / or storage component 240. A computer-readable medium may be defined herein as a non-transitory memory device. A memory device may include memory space within a single physical storage device or memory space distributed across multiple physical storage devices.

[0045] Software instructions may be read into memory 230 and / or storage component 240 via communication interface 270 from another computer-readable medium or from another device. When executed, the software instructions stored in memory 230 and / or storage component 240 may cause processor 220 to perform one or more processes and / or methods described herein. Additionally or alternatively, hardwired circuitry may be used in place of or in combination with software instructions to perform one or more methods and / or processes described herein. Therefore, the specific implementations described herein are not limited to any particular combination of hardware circuitry and software.

[0046] Figure 2 The number and arrangement of components shown are provided as examples. In practice, device 200 may include additional components, fewer components, different components, or components with... Figure 2 The components shown are arranged differently. Additionally or alternatively, a group of components of device 200 (e.g., one or more components) may perform one or more functions described as being performed by another group of components of device 200.

[0047] Now for reference Figure 3 A flowchart illustrating a method 300 for generating longitudinal features is shown. Method 300 can be executed by one or more processors (e.g., processor 220) according to instructions stored in non-transitory memory (e.g., memory 230).

[0048] At 302, method 300 includes extracting criteria from a single ECG record used to interpret the endpoint. As an example, a single ECG record can be one of multiple ECG records in an ECG dataset for which the diagnosis is known. Regarding Figure 6 An exemplary method for extracting criteria is further described. The extracted criteria may include combinations of ECG features (e.g., sums and differences of parameters across different leads, or products and ratios of different parameters within the same lead) and / or sub-combinations of ECG features that collectively point to an interpretable endpoint. The interpretable endpoint may be a diagnostic interpretation or diagnosis of the output using the criteria.

[0049] At 304, method 300 includes identifying one or more relevant ECG parameters based on the extracted criteria. As an example, a single ECG may include multiple features, parameters, or findings. For example, parameters may include the amplitude and duration of the P wave, the amplitude and duration of the Q wave, the amplitude and duration of the R wave, the PR interval, the amplitude and duration of the S wave, the duration of the QRS complex, the amplitude and duration of the T wave, the QT interval, etc. Additionally or alternatively, the feature set may be defined by data from a single lead (e.g., a specific P wave amplitude or P wave duration in a single lead), or by data from multiple leads (e.g., the global QT duration from the earliest Q onset to the latest T offset across all leads, or QT dispersion, which is the difference between the shortest and longest single-lead QT interval measurements across all leads), or by data from spatial relationships across multiple leads (e.g., spatial QRS-T angles or ventricular gradients). Based on the interpretation endpoint, the most relevant parameters can be identified. For example, if the endpoint is the diagnosis of LBBB, the most relevant parameters may include the QT interval, the duration of the QRS complex, the relationship between the R wave and the S wave in lead V1, and the identified rhythm (e.g., supraventricular).

[0050] At 306, method 300 includes generating longitudinal features for the identified parameter. The individual ECG identifying the associated ECG parameter may also have one or more ECGs previously acquired in the same patient. The longitudinal features can capture a historical record of the identified parameter for a given patient. As an example, if the identified parameter is QRS duration, the generated longitudinal feature could be the QRS duration difference (QRSDD) between the current ECG (e.g., the single ECG under consideration) and the immediately preceding previous ECG or another selected previous ECG, and / or a QRSDD normalized for a given time period. Thus, the longitudinal features of the QRS duration can indicate a trend in the QRS duration over time, involving the difference of the parameter compared to a given previous value, the difference compared to the mean, and changes in the trend (e.g., if a first trend is observed in a first time period and a second, different trend is observed in a second time period, as can be seen in a diagnosis that has a sudden impact on the parameter).

[0051] In some examples, method 300 can be repeated for multiple single ECG records, for example, for various different interpretable endpoints. In this way, a set of longitudinal features corresponding to different interpretable endpoints can be created. This set of longitudinal features can be stored in memory and, in some examples, can be fed back into the ECG analysis model so that the model uses these longitudinal features to determine the diagnosis of newly acquired ECGs, as will be discussed in this paper. Figure 4 As described, by generating longitudinal features, ECGs can be analyzed not only for features of the ECG itself, but also for features related to historical ECGs, thus saving providers the time of manually comparing ECGs and improving the accuracy of diagnostic outputs.

[0052] Turn now Figure 4 This is a flowchart illustrating a method 400 for determining a diagnosis using longitudinal ECG features. Longitudinal ECG features can be as follows: Figure 3 Method 400 is constructed as described in method 300. Method 400 can be executed by one or more processors (e.g., processor 220) according to instructions stored in non-transitory memory (e.g., memory 230). In some examples, method 400 can be based on an ECG analysis model (such as...) Figure 1 The ECG analysis model 114 is used for this purpose. Method 400 can be performed as part of the development phase of the ECG analysis model.

[0053] At point 402, method 400 includes obtaining medical record data of multiple patients from a medical data repository. For example, Figure 1The ECG analysis model 114 can obtain data from one or more medical data repositories. For example, the ECG analysis model 114 can obtain data from an ECG database that stores ECGs obtained for multiple patients and corresponding information, including collection date, diagnostic interpretation, chief complaint, etc.

[0054] At 404, method 400 includes generating a subset of patients with records including more than one ECG based on the acquired medical record data. For example, the medical record data may include data from multiple patients. Some of these patients may have undergone only one ECG, while others may have undergone more than one ECG, wherein the acquired medical record data includes more than one ECG assigned to each of these patients. Therefore, the subset of patients with records including more than one ECG may include the patients and the corresponding ECG for each of these patients.

[0055] At 406, method 400 includes obtaining the constructed longitudinal feature of the ECG parameters. (See regarding...) Figure 3 As described, a set of longitudinal features can be constructed from multiple ECGs with various explanatory endpoints (e.g., known diagnoses). For example, longitudinal features corresponding to identified relevant parameters of multiple single ECGs with a given explanatory endpoint can be generated. Thus, longitudinal features can be generated for multiple explanatory endpoints, thereby constructing a set of longitudinal features.

[0056] At 408, method 400 includes determining longitudinal features of historical data from a captured subset. As an example, this set of longitudinal features may include longitudinal features of various ECG parameters and various explanatory endpoints. The subset of patients and ECGs may include subsets of multiple ECG parameters. Therefore, based on the ECG parameters within the subset to be analyzed, irrelevant longitudinal features can be filtered out, thereby reducing the processing power required for ECG analysis.

[0057] At 410, method 400 includes using longitudinal features identified as relevant to a subset to extract criteria for the selected ECG. The selected ECG may have known corresponding previous ECGs. For example, the selected ECG may be the most recently performed ECG of the selected patient. The most recently performed ECG may have one or more corresponding previous ECGs. The one or more corresponding previous ECGs may have known acquisition dates and features therein, as well as known diagnoses. Criteria may be extracted based on features identified within the selected ECG, its diagnosis, and its one or more corresponding previous ECGs. Figure 6 An exemplary method for extracting standards is given in the document.

[0058] As noted above, Method 400 can be performed during the development phase of the ECG analysis model. Therefore, criteria extraction can be performed from a retrospective database of ECGs in which the correct diagnosis for each ECG is known. Criteria are then extracted for the interpretation endpoints (e.g., targets) of the known diagnoses.

[0059] In some examples, the criteria for extracting the selected ECG may also be based on a single ECG feature rather than a longitudinal feature. For example, similar to what is described for a method used to generate this set of longitudinal features, a single ECG criterion may be extracted (e.g., similar to 302 in method 300). In such an example, at 412, method 400 includes determining the performance of the criteria extracted using the longitudinal features compared to the criteria extracted using the single ECG feature. Therefore, as will be discussed regarding... Figure 6 As described, when the performance of longitudinal feature extraction criteria is high, longitudinal criteria can be used when generating diagnostic output from the ECG analysis model during the deployment phase. This performance determination refers to determining which set of criteria is better based on analysis of a retrospective database. For example, a retrospective database of ECGs stores the ground truth for the correct interpretation of each ECG, and performance can therefore be evaluated within this database. This determination can be performed during the development phase and is not repeated during deployment, thus it is independent of newly acquired ECGs (acquired by the ECG analysis model for interpretation) and does not require a connection to the retrospective database during model deployment.

[0060] At 414, method 400 includes updating the ECG analysis model based on longitudinal criteria. For example, the ECG analysis model can be updated or otherwise configured to determine the diagnosis of a newly acquired or obtained ECG based on one or both of longitudinal criteria and single ECG criteria. Thus, a set of longitudinal features (such as those related to single ECGs) generated from a retrospective database of single ECGs is used. Figure 3 As described herein, a longitudinal set of criteria can be defined. This longitudinal set of criteria can be a fixed set, which can then be used to determine the diagnosis of newly acquired ECGs when deploying an ECG analysis model, such as regarding... Figure 5 As described.

[0061] Turn now Figure 5 A method 500 for determining an ECG diagnosis based on longitudinal criteria is illustrated. Method 500 can be executed by one or more processors of the apparatus based on instructions stored in the memory of the apparatus. For example, method 500 can be executed by the processor based on instructions of an ECG analysis model (e.g., ECG analysis model 114), which is configured to determine an ECG diagnosis based on one or both of a single ECG criterion and longitudinal criteria. (See also: Regarding...) Figure 3 and Figure 4The explained longitudinal criteria can be extracted based on a set of longitudinal features generated from a single ECG set, for which the diagnosis is known and has associated previous ECGs. The method 500 presented in this paper can be a deployment phase of the ECG analysis model, thereby based on... Figure 3 and Figure 4 The method is used to develop ECG analysis models.

[0062] At 502, method 500 includes receiving the patient's ECG for interpretation. The ECG device may be a newly acquired ECG from an ECG device or from an ECG database. For example, as per [reference to...] Figure 1 As described, the ECG analysis model can communicate with an ECG device (e.g., ECG device 110). ECGs can be acquired by the ECG device and sent directly to the ECG analysis device, which includes the ECG analysis model. In other examples, the ECG analysis model can communicate with an ECG database (e.g., ...). Figure 1 The ECG can communicate with one or more medical data storage libraries 130 (ECG databases), and the ECG can be obtained from the ECG database rather than directly from the ECG device.

[0063] At 504, the method includes determining whether one or more previous ECGs of the patient are available. For example, one or more previously acquired ECGs may be stored in an ECG database (e.g., the one or more medical data repositories 130). The one or more previously acquired ECGs may be associated with the patient based on one or more patient demographic characteristics, such as name, date, and / or medical record number. For example, the patient record associated with the patient may include one or more previous ECGs available therein. If previous ECGs are available for the patient, method 500 proceeds to 506. If no previous ECGs are available for the patient, method 500 proceeds to 510.

[0064] At 506, method 500 includes obtaining one or more previous ECGs of the patient. As noted above, the one or more previous ECGs may be stored in an ECG database or other medical data repository with which the ECG analysis device (e.g., via a network) communicates. The ECG analysis device may retrieve the one or more previous ECGs from the ECG database or other medical data repository in response to determining that the one or more previous ECGs are available. In some examples, the ECG (e.g., the most recent ECG) and the one or more previous ECGs may be retrieved substantially simultaneously, for example, when the ECG and the one or more previous ECGs are stored in the same database. In other examples, the current ECG may be obtained first, for example, directly from the ECG device, and the one or more previous ECGs may be obtained separately from the medical data repository.

[0065] At 508, method 500 determines whether the performance of the longitudinal criterion is greater than that of the single ECG criterion. As described with respect to method 400, the performance of the longitudinal criterion can be evaluated when extracting the longitudinal criterion based on the longitudinal features of the ECG generated for a given interpretation endpoint. In some examples, this performance can be compared to the performance of the single ECG criterion for the same set of ECGs and interpretation endpoints. If the performance of the longitudinal criterion is higher, method 500 proceeds to 510. If the performance of the single ECG criterion is higher, method 500 proceeds to 512.

[0066] At 512, method 500 includes determining a diagnosis for the acquired ECG (e.g., a newly acquired ECG) based on single ECG features. As noted above, if the performance of a single ECG criterion is greater than that of a longitudinal criterion, an ECG analysis model can be deployed based on a single ECG criterion. For example, a single ECG criterion can be extracted for a retrospective ECG dataset. As noted above, Figure 6 An exemplary extraction method is provided. The ECG analysis model can then determine the ECG diagnosis based on these extracted individual ECG criteria.

[0067] At point 510, when the performance of the longitudinal criteria is higher than that of a single ECG criterion, method 500 includes determining a diagnosis for the ECG based on the longitudinal criteria. As noted above, the ECG analysis model can be configured to determine a diagnosis based on ECG features and extracted criteria. In this example, the ECG analysis model can be configured to determine a diagnosis based on features of the ECG and corresponding previous ECGs, as well as extracted longitudinal criteria. It should be clarified that features can be longitudinal features of individual findings and / or trends in ECG parameters within a single ECG or other historical ECG data (e.g., QT interval, QRS complex duration, QRSDD, etc.), and criteria can be a set of features, such as a first feature and a second feature together, but the first feature alone may not satisfy the criterion. Therefore, longitudinal features identified within the ECG and one or more corresponding previous ECGs may satisfy one or more fixed longitudinal criteria extracted during model development to determine a diagnosis. Thus, the diagnosis can be a diagnosis indicated based on longitudinal features and satisfied criteria.

[0068] At 514, method 500 includes outputting a diagnostic (whether determined based on longitudinal criteria or single ECG criteria) to a user device. For example, the diagnostic may be output to be displayed on the user device (e.g., user device 122). Thus, the provider can review the output diagnostic along with the longitudinal characteristics, met criteria, and diagnostic interpretation to accept or reject the output diagnostic.

[0069] Therefore, diagnoses determined at least in part based on longitudinal features can take into account historical and temporal medical data that might otherwise be overlooked. As a non-limiting example, the QRS duration, an ECG parameter, is a primary diagnostic parameter for both LBBB and LVH. For a single ECG, a QRS duration greater than 120 can indicate either LBBB or LVH. However, it is difficult to determine which diagnosis to output based on other ECG features. Longitudinal features, however, can provide deeper insights for differentiating between these two diagnoses.

[0070] Brief Turn Figure 7 Figure 700 illustrates the QRS duration over time in patients with LVH. For example, a patient may experience multiple ECGs over a period of many years. Each of those multiple ECGs may include a defined QRS duration. The QRS duration is plotted over time. As shown, in LVH, the QRS duration gradually increases over time. For example, Figure 700 depicts an average increase of 6.2 ms / year.

[0071] In comparison, Figure 8 A graph 800 depicts the duration of the QRS over time in patients with LBBB. For example, a patient may experience multiple ECGs over a period of many years. As graph 800 shows, in LBBB, the duration of the QRS may increase sharply and abruptly, rather than gradually.

[0072] Therefore, recently acquired ECGs can include QRS durations of 140, a value that can indicate LVH or LBBB on its own. However, when longitudinal ECG data is included, QRS durations from previous ECGs can be considered. In this way, a gradual increase in QRS duration can be represented by defined longitudinal features (such as QRSDD) in patients with LVH, while a sudden increase in QRS duration can be represented by defined longitudinal features in patients with LBBB. In this way, including longitudinal data, constructing longitudinal features, and extracting longitudinal criteria allows for more accurate diagnostic determination. Furthermore, including longitudinal data in diagnostic determinations performed by ECG analysis systems reduces the time spent by providers manually comparing ECGs to differentiate diagnoses.

[0073] Turning Figure 6This is a flowchart illustrating a method 600 for extracting criteria for diagnosis determined by an ECG analysis model, such as a rule-based ECG analysis model. Method 600 can be executed by one or more processors (e.g., processor 220) according to instructions stored in non-transitory memory (e.g., memory 230). It should be understood that the method 600 described herein is exemplary in nature, and other methods for extracting criteria for diagnosis determined by an ECG analysis model, such as methods that do not use AI models, are also possible.

[0074] At 602, method 600 includes receiving training data for training an AI model to extract criteria for determining a diagnosis by the ECG analysis model. As an example, platforms (such as...) Figure 1 The ECG analysis system 100 platform 118 can receive data for training AI models (such as...). Figure 1 The training data of AI model 120.

[0075] As an example, when training an AI model to interpret a single ECG, the training data may include a single ECG as input and an ECG standard as the target. For example, the ECG may be a waveform acquired via ECG device 110. Additionally or alternatively, the input to the training data may include a feature set of the ECG. For example, the feature set may include the amplitude and duration of the P wave, the amplitude and duration of the Q wave, the amplitude and duration of the R wave, the PR interval, the amplitude and duration of the S wave, the duration of the QRS complex, the amplitude and duration of the T wave, the QT interval, etc. Additionally or alternatively, the feature set may be defined by data from a single lead (e.g., a specific P-wave amplitude or P-wave duration in a single lead), or by data from multiple leads (e.g., the global QT duration from the earliest Q onset to the latest T offset across all leads, or QT dispersion, which is the difference between the shortest and longest single-lead QT interval measurements across all leads), or by data from spatial relationships across multiple leads (e.g., spatial QRS-T angles or ventricular gradients).

[0076] As another example, when training an AI model for longitudinal ECG interpretation, the input to the training data may include multiple ECGs. For example, an ECG may be a waveform acquired via ECG device 110, stored in one or more medical data repositories, and retrieved by platform 118. Additionally or alternatively, the input to the training data may include a longitudinal feature set of ECGs. For example, the longitudinal feature set may include trend features of the aforementioned features, including differences normalized to time (e.g., the difference between the current ECG and the immediately preceding ECG, and all variations thereof), the difference between a measurement normalized to standard deviation and the mean of that measurement across all the patient's ECGs (e.g., a measure of traction), and / or a measure of the current trend, which may define the longest period of time the trend is running (e.g., the variation in normalized differences). Examples of longitudinal features include QRSDD and averaged or normalized QRSDD over a specific time period, as described above.

[0077] Other example longitudinal features used to enhance confidence in the diagnosis of LBBB include: 1) the difference in QRS area measurements in leads V1 and V2 compared to the patient's baseline or mean; and 2) the difference in T wave amplitude / area and ST segment parameters in leads V5 and V6 compared to the patient's baseline or mean. When the difference in QRS area measurements in leads V1 and V2 compared to the patient's baseline or mean becomes significantly negative (e.g., becomes statistically significant, or becomes more negative than the threshold amount), they indicate the presence of deep and wide S waves in leads V1 and V2, which is characteristic of LBBB and abnormal for the patient (compared to their baseline). When the difference in T wave amplitude or area and ST segment parameters becomes significantly negative, they indicate ST segment depression and T wave inversion in the lateral leads, which is characteristic of LBBB and abnormal for the patient (compared to their baseline).

[0078] For patients, similar longitudinal parameters on abnormal ST segments, T waves, and Q waves can identify myocardial infarction with higher confidence than a single ECG standard or parameter. More generally, “score” parameters can be synthesized from a single ECG to determine the presence of conditions such as structural heart disease. An increase in such a score parameter over time (as indicated by longitudinal features) can indicate the evolution of an underlying condition with higher confidence than a single ECG value for that score parameter alone.

[0079] The target of the training data may include a set of ECG criteria. When the AI ​​model is trained for single ECG criterion extraction, the ECG criterion can be a single ECG criterion. When the AI ​​model is trained for longitudinal criterion extraction, the ECG criterion can be a longitudinal criterion. ECG criteria may include a combination of one or more ECG features corresponding to a diagnosis. For example, the input ECG may include a first feature and a second feature, and the target criterion may be a combination of the first feature and the second feature that together correspond to a diagnosis. Therefore, for example, an AI model can be trained to ingest an ECG, identify the first feature and the second feature, and extract a criterion from the combination of the first feature and the second feature.

[0080] Therefore, training data can include diagnoses of input ECGs (e.g., the current ECG in the longitudinal example). For example, diagnoses can include atrial pacing rhythm, ventricular pacing rhythm, atrial flutter, ectopic atrial tachycardia, sinus bradycardia, junctional bradycardia, atrial fibrillation, LBBB, LVH, septal infarction, non-ST-elevation myocardial infarction, ST-elevation myocardial infarction, etc. Furthermore, the diagnostic interpretation can be the interpretation of the ECG determined by a physician. Additionally or alternatively, the diagnostic interpretation can be a diagnosis made by a physician using only the ECG, or alternatively using another source of clinical information (e.g., high-sensitivity troponin levels, echocardiographic measurements, or angiographic findings), or a combination of ECG and non-ECG clinical information.

[0081] Furthermore, in some examples, the training data may include modification information that identifies whether a diagnostic interpretation result has been modified. For example, modification information may identify whether a diagnosis determined by the ECG analysis model 114 has been accepted, removed, or replaced. As another example, modification information may identify whether a diagnosis not determined by the ECG analysis model 114 has been added. In the case of a replacement diagnosis, modification information may identify the replacement diagnosis. According to one implementation, platform 118 may determine modification information based on the diagnostic interpretation result determined by the ECG analysis model 114 and the final diagnostic interpretation result determined by the physician after the physician reviews the diagnostic interpretation result determined by the ECG analysis model 114. For example, platform 118 may compare the diagnostic interpretation result determined by the ECG analysis model 114 with the final diagnostic interpretation result determined by the physician and determine the modification based on the comparison.

[0082] In addition, training data can include patient information associated with ECG. For example, patient information can identify whether a specific diagnosis exists in the patient's previous diagnostic interpretations, whether the physician has previously modified the diagnostic interpretations, the patient's demographic information, the patient's health status (e.g., previous diagnoses, comorbidities, etc.), medications prescribed to the patient, previous surgeries performed on the patient, and the patient's previous diagnosis (the determined diagnosis), etc.

[0083] Furthermore, training data can include classification objectives for ECGs. For example, classification objectives could be ECG diagnoses, interpretations of ECG diagnoses, or information on ECG modifications. Classification objectives can be diagnostic statements such as LBBB, myocardial infarction, or atrial fibrillation. Classification objectives can also be expert modifications to diagnostic statements given by a baseline algorithm. For example, when a baseline algorithm gives a diagnostic statement such as LBBB or atrial fibrillation (e.g., a diagnosis), and there is data from a retrospective database where experts have edited / removed that diagnostic statement, in these cases, the objective could be to identify errors in the baseline algorithm's criteria to reduce the amount of expert modifications.

[0084] At 604, method 600 includes training an AI model using training data that incorporates ECG features (e.g., input) and diagnostic criteria (e.g., target). For example, Figure 1 The platform 118 of the ECG analysis system 100 can train the AI ​​model 120 based on training data. Alternatively, systems or devices other than the platform 118 can be used to generate and / or train the AI ​​model 120. For example, the system or device may include instructions for generating the AI ​​model 120 and / or instructions for training the AI ​​model 120. The system or device may provide the resulting trained AI model 120 to the platform 118 for use.

[0085] In some examples, the AI ​​model 120 may include a training phase, a deployment phase, and a monitoring phase. During the training phase, the platform 118 may receive and process training data to generate a trained AI model 120 for extracting criteria used to determine diagnoses by the ECG analysis model. The training data may include multiple training datasets, each comprising one or more of the following: ECG, ECG feature sets, ECG diagnoses, ECG diagnostic interpretations, modification information, patient information associated with ECGs, ECG classification targets, etc. Each of the multiple training datasets may be associated with a specific ECG and a specific diagnostic interpretation.

[0086] Training data may be generated, received, or otherwise obtained from internal and / or external resources. For example, training data may be generated, received, or otherwise obtained from ECG device 110, ECG analysis device 112, user device 122, and / or database 124.

[0087] Typically, AI model 120 may include a set of variables (e.g., nodes, neurons, or filters) tuned (e.g., weighted or biased) to different values ​​through the application of training data. According to one embodiment, the training process may employ supervised, unsupervised, semi-supervised, and / or reinforcement learning processes to train AI model 120. According to one embodiment, a portion of the training data may be retained during training and / or used to validate the trained AI model 120.

[0088] For a supervised learning process, training data may include labels or scores that can facilitate the training process by providing ground truth values. For example, labels or scores may indicate a classification target. Training can be performed by feeding the training dataset into AI model 120. AI model 120 may have variables set to initial values ​​(e.g., randomly, based on Gaussian noise, based on pre-trained values, etc.). AI model 120 can generate outputs. The outputs can be compared with corresponding labels or scores (e.g., ground truth values) and then backpropagated through AI model 120 to adjust the values ​​of the variables. This process can be repeated for multiple samples, at least until the determined loss or error is below a predefined threshold. According to one implementation, some of the training data may be retained and used for further validation or testing of the trained AI model 120.

[0089] For unsupervised learning processes, training data may not include pre-assigned labels or scores to aid the learning process. Instead, unsupervised learning processes can include clustering, classification, etc., to identify patterns naturally present in the training data. As an example, training data can be clustered into groups based on identified similarities and / or patterns. K-means clustering or K-nearest neighbors can also be used, and these can be supervised or unsupervised. A combination of K-nearest neighbors and unsupervised clustering techniques can also be used. For semi-supervised learning, a combination of training data with pre-assigned labels or scores and training data without pre-assigned labels or scores can be used to train an AI model120.

[0090] When reinforcement learning is employed, an agent (e.g., an algorithm) can be trained to make decisions from training data through trial and error regarding whether a diagnostic interpretation should be modified. For example, based on the decisions made, the agent can then receive feedback (e.g., a positive reward for predicting a value above a predetermined threshold), adjust its next decision to maximize the reward, and repeat until the loss function is optimized.

[0091] After being trained, the trained AI model 120 can be stored and subsequently applied by platform 118 during the deployment phase. For example, during the deployment phase, the trained AI model 120, executed by platform 118, can receive input data and generate output data. During the monitoring phase, monitoring data can be analyzed along with the output and input data to determine the accuracy of the trained AI model 120. According to one implementation, based on the analysis, platform 118 can return to the training phase, where the values ​​of one or more variables of the AI ​​model 120 can be adjusted to improve the accuracy of the AI ​​model 120.

[0092] According to one implementation, AI model 120 can be a decision tree. In this case, platform 118 can use training techniques to generate the decision tree. For example, training techniques may include random forest, boosting tree, bootstrapping, rotating forest, etc. AI model 120 may include a set of nodes. For example, the set of nodes may include a root node, one or more intermediate nodes, and leaf nodes. Platform 118 can use attribute selection metrics to generate AI model 120. For example, attribute selection metrics may be information gain, gain ratio, Gini index, etc. Platform 118 can generate the decision tree and use pruning techniques to prune the decision tree. For example, pruning techniques may be cost complexity pruning, error reduction pruning, etc.

[0093] At 606, method 600 includes extracting criteria used to determine the diagnosis by the ECG analysis model. For example, Figure 1 Platform 118 can use AI model 120 to extract criteria for diagnosis determined by ECG analysis model 114.

[0094] In some examples, platform 118 can determine the decision branch of AI model 120 corresponding to a specific target classification. For example, the target classification could be an ECG diagnosis, an ECG diagnosis interpretation result, ECG modification information, etc. The decision branch can include one or more nodes corresponding to the relevant criteria. For example, a decision branch can include a root node, one or more intermediate nodes, and leaf nodes.

[0095] Furthermore, in some examples, platform 118 can determine a decision branch of AI model 120, which includes one or more nodes associated with an attribute selection metric that satisfies a threshold. For example, the attribute selection metric could be information gain, gain ratio, Gini index, etc. As a specific example, platform 118 can determine a decision branch that includes leaf nodes corresponding to a criterion with a Gini index less than a threshold. According to one implementation, platform 118 can determine the decision branch based on a metric of the decision path. For example, the metric could be accuracy, positive prediction value, sensitivity, etc. Additionally or alternatively, platform 118 can determine the decision branch based on the performance of generalization to an external dataset.

[0096] Based on one or more examples, platform 118 can identify specific nodes among one or more nodes for criterion extraction. Platform 118 can identify specific nodes based on feature selection metrics. Feature selection metrics can correspond to the importance of features associated with the criteria corresponding to the node. According to one implementation, platform 118 can fit another AI model 120 (e.g., a decision tree) to a dataset separate from the training dataset. In this way, platform 118 can reduce the risk of overfitting and increase the statistical power of the technique.

[0097] In some examples, platform 118 can extract criteria from decision branches. For example, platform 118 can extract criteria belonging to decision branches. Each criterion may include a corresponding feature and a corresponding threshold. In some examples, platform 118 can determine the threshold for the criterion. For example, platform 118 may determine the threshold based on optimization techniques, based on input from user device 122, etc.

[0098] Furthermore, platform 118 can generate rules that include the extracted set of criteria. For example, rules may include criteria from a specific decision branch. Alternatively, rules may include criteria from different decision branches. Platform 118 can generate rules using permutations and / or combinations of criteria. According to one implementation, platform 118 can determine the metrics of the rules. For example, metrics may be accuracy, positive predictive value, sensitivity, etc. Furthermore, platform 118 can determine that rules include metrics that satisfy thresholds.

[0099] Furthermore, platform 118 can generate rules to include score thresholds. For example, if a score threshold is met, that threshold can trigger ECG analysis model 114 to determine a diagnosis. According to one or more implementations, platform 118 can determine the score threshold based on performance metrics. For example, performance metrics could be F-scores, accuracy, sensitivity, etc. Each criterion of the rule can be associated with a specific score. ECG analysis model 114 can determine the overall score based on the specific criteria being met. ECG analysis model 114 can compare the overall score to the score threshold and determine a diagnosis based on this comparison. For example, if the overall score is greater than or equal to the score threshold, ECG analysis model 114 can determine a specific diagnosis. Alternatively, if the overall score is less than the score threshold, ECG analysis model 114 may not determine a specific diagnosis. According to one implementation, platform 118 can generate rules to include one or more mandatory criteria. For example, rules can include mandatory criteria that must be met so that ECG analysis model 114 determines a diagnosis regardless of the overall score. In this way, different decision branches can be explored by requiring some criteria to be met rather than added to the score, and by varying the degree to which each criterion is required. Different decision branches that should be satisfied before giving a positive classification can be considered as accessing different subgroups and can be deployed in parallel. For example, when assessing the diagnosis of atrial fibrillation, platform 118 can consider two different subgroups (e.g., a first subgroup that detects many P waves and a second subgroup that detects very few P waves). Both subgroups can correctly identify the relevant target, but they operate in different parameter regions.

[0100] In some cases, platform 118 may receive input from user device 122 corresponding to expert knowledge used to fine-tune scores. Expert knowledge may include checks on all ECGs selected by the editing program to evaluate the effectiveness of ECG analysis model 114, which is possible based on the complete transparency of the criteria. If expert knowledge indicates that some criteria are inaccurate, or were selected due to bias caused by errors in annotations or labels, platform 118 may remove that criterion. To ensure greater control to avoid spurious decisions, platform 118 may request concomitant verification from supplementary leads and features or complementary leads and features. This control and accessibility is possible based on the interpretability of the criteria and is unparalleled in other models.

[0101] In some examples, platform 118 can validate rules based on training data. For instance, platform 118 can receive training data from database 124 and validate rules. Based on the validated rules, platform 118 can update ECG analysis model 114 to implement the rules.

[0102] At 608, method 600 includes updating the ECG analysis model to determine a diagnosis using the extracted criteria. For example, platform 118 can update ECG analysis model 114 based on these criteria. For example, ECG analysis model 114 can be updated to include the extracted longitudinal criteria, making analysis model 114 usable for determining a diagnosis based on longitudinal criteria. Platform 118 can update the ECG analysis model so that ECG analysis model 114 can use the criteria to determine a diagnosis. In this way, ECG analysis model 114 can receive ECGs and use the criteria extracted by AI model 120 to determine the diagnosis of these ECGs.

[0103] It should be understood that method 600 is provided herein as an example of how to extract the standard, and other methods may be performed without departing from the scope of this disclosure.

[0104] Turn now Figure 9 An example user interface 900 is shown. User interface 900 can be displayed on a user device (e.g., user device 122). The user device can be configured to access one or more data repositories, such as an EMR, ECG database, etc. Therefore, the user device can display various patient information, including ECG data, based on input applied to the user device. For example, user interface 900 can display ECG data, including identifying features of the ECG.

[0105] The user interface 900 includes a list of available ECGs. Figure 9 In this user interface 900, the list is sorted by patient (e.g., by patient ID (PID)). Therefore, a first patient 902 can be listed multiple times, with one entry for each ECG of the first patient 902. The user interface can also indicate the collection date 904 for each ECG in the first patient 902's ECGs.

[0106] Additionally, relevant or selected single ECG features and longitudinal features can be displayed within the user interface 900. For example, a single ECG feature column 906 can be displayed. In one example, the single ECG feature column 906 can display the QRS duration. A first longitudinal feature column 908 and a second longitudinal feature column 910 can also be displayed. The first longitudinal feature column 908 can display the average QRS duration of all available ECGs up to the corresponding ECG entry. The second longitudinal feature column 910 can display QRSDD. For each ECG entry in the list, the feature value can be displayed in the corresponding column. The features displayed within the user interface 900 can be those features ingested by the ECG analysis model to determine the diagnosis.

[0107] The technical advantage of the system and method disclosed in this paper is that longitudinal data of the patient can be considered when determining a diagnosis based on ECG. Utilizing longitudinal data for diagnosis by generating longitudinal features, extracting longitudinal criteria, and determining the diagnosis based on the met longitudinal criteria can achieve more accurate diagnoses and save physicians time in determining diagnoses and comparing ECGs to make a diagnosis.

[0108] As used herein, elements or steps listed in the singular and beginning with the word "a" or "an" should be understood to not exclude a plurality of said elements or steps unless such exclusion is explicitly stated. Furthermore, references to "an embodiment" of the invention are not intended to be construed as excluding the existence of additional embodiments that also include the referenced features. Moreover, unless explicitly stated to the contrary, embodiments that "comprise," "include," or "have" elements or multiple elements having a particular characteristic may include additional such elements that do not have that characteristic. The terms "comprise" and "in" are used as concise linguistic equivalents to the corresponding terms "comprising" and "wherein." Furthermore, the terms "first," "second," and "third," etc., are used merely as notations and are not intended to impose numerical requirements or a particular order of position on their objects.

[0109] This written description uses examples to disclose the invention, including the best mode, and also enables those skilled in the art to practice the invention, including making and using any apparatus or system and performing any included methods. The scope of patentability of the invention is defined by the claims, but may include other examples that would occur to those skilled in the art. Such other examples are intended to fall within the scope of the claims if they have structural elements that are not indistinguishable from the literal language of the claims, or if they include equivalent structural elements that have minor differences from the literal language of the claims.

Claims

1. A method, the method comprising: A set of longitudinal ECG features (306) is generated from multiple electrocardiograms (ECGs), wherein the multiple ECGs include a current ECG and one or more previous ECGs of each current ECG; Extract longitudinal criteria (410) from the multiple ECGs; The ECG analysis model (414) is updated using the aforementioned longitudinal criteria; Deploy the ECG analysis model to determine the diagnosis of the patient’s ECG based on the longitudinal criteria (510), wherein one or more corresponding previous ECGs are available for the patient; and send the diagnosis to the user device (514).

2. The method according to claim 1, wherein generating the set of longitudinal ECG features comprises: The standard for extracting multiple single ECG records (302); The most relevant ECG parameter in each of the plurality of single ECG records is identified based on the extracted criteria (304); and Generate longitudinal ECG features (306) for each of the most relevant ECG parameters among the identified most relevant ECG parameters.

3. The method of claim 2, wherein each of the plurality of single ECG records... The ECG record has one or more corresponding historical ECGs, and the longitudinal ECG features are generated based on the most relevant ECG parameters identified in each of the plurality of single ECGs and the corresponding historical ECGs (304, 306).

4. The method of claim 2, wherein an AI model is used to extract the criteria of the plurality of single ECG records, the AI ​​model being trained to extract the criteria (302, 604) based on features identified within the plurality of single ECG records.

5. The method of claim 1, further comprising comparing the performance of the standard extracted based on the longitudinal standard with the performance of the standard extracted only from the current ECG (508).

6. The method of claim 5, wherein the diagnosis is determined based on the longitudinal criteria in response to determining that the performance of the criteria extracted based on the longitudinal criteria is greater than the performance of the criteria extracted only from the current ECG (510).

7. The method of claim 1, wherein an AI model is used to extract the longitudinal criteria, the AI ​​model being trained to extract longitudinal features (410, 604) based on longitudinal features identified in the plurality of ECGs.

8. The method of claim 7, wherein the AI ​​model is a decision tree, and the longitudinal criteria are extracted by the AI ​​model based on the decision branches of the decision tree (604).

9. An apparatus comprising: A memory (230) configured to store instructions; and One or more processors (220) are configured to operate based on instructions stored in the memory: During the development phase of the ECG analysis model, multiple electrocardiograms (ECGs) with known diagnoses are received, wherein one or more of the multiple ECGs have one or more ECGs previously obtained for the same patient (402); Determine one or more longitudinal features of the plurality of ECGs (408); One or more vertical criteria (410) are extracted based on the one or more vertical features; The ECG analysis model (414) is updated based on one or more of the longitudinal criteria. During the deployment phase of the ECG analysis model, the diagnosis of newly acquired ECGs is determined using one or more longitudinal criteria extracted from the plurality of ECGs (510); as well as The diagnostic (514) is sent to a user device that is communicatively coupled to the device.

10. The apparatus of claim 9, wherein the one or more processors are further configured to: obtain medical record data of a plurality of patients from one or more medical data repositories (402), and identify a subset of patients from the plurality of patients having records including more than one ECG, the subset of patients including the patients (404).

11. The apparatus of claim 9, wherein the one or more longitudinal features determined by the plurality of ECGs are a subset of a set of longitudinal features generated by the following operation: The standard for extracting multiple single ECG records (302); The most relevant ECG parameter in each of the plurality of single ECG records is identified based on the extracted criteria (304); and Generate longitudinal ECG features (306) for each of the most relevant ECG parameters among the identified most relevant ECG parameters.

12. The device of claim 9, wherein the longitudinal feature indicates the patient's historical data.

13. The apparatus of claim 9, wherein an AI model is used to extract the one or more longitudinal criteria of the plurality of ECGs, the AI ​​model being trained to extract the longitudinal criteria based on the longitudinal features identified within the plurality of ECGs (604).

14. The apparatus of claim 9, wherein the one or more processors are further configured to determine the performance of the longitudinal criterion (412).

15. The apparatus of claim 14, wherein the one or more processors are further configured to compare the performance of the longitudinal standard with the performance of a single ECG standard (508).

Images