Ladder Gramm Generation Device and Computer Program

Abstract

We provide technology capable of generating ladder diagrams. [Solution] The ladder graph generation device comprises an acquisition unit that acquires a specific electrocardiogram of the subject, and a calculation unit. The calculation unit generates a ladder graph from the specific electrocardiogram based on machine learning using training data, which includes a training electrocardiogram and a training ladder graph corresponding to the training electrocardiogram.

Description

Technical Field

[0001] The technology disclosed in this specification relates to a technology for generating a laddergram from an electrocardiogram.

Background Art

[0002] Patent Document 1 discloses a Holter electrocardiograph used for measuring an electrocardiogram. When diagnosing heart diseases (such as arrhythmia, etc.) using an electrocardiogram, a laddergram may be created from the electrocardiogram in order to consider the complex movement of the cardiac conduction system. A laddergram is a diagram showing the transmission pattern of electrical signals to each part in the heart. By creating a laddergram, it becomes easier to grasp the abnormality of the excitation conduction in the heart.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Since a laddergram is manually created by medical staff such as doctors, it takes time from when an electrocardiogram is measured until a doctor determines a heart disease such as arrhythmia. This specification discloses a technology for automatically generating a laddergram.

Means for Solving the Problems

[0005] In a first aspect of the technology disclosed in this specification, a laddergram generation device includes an acquisition unit that acquires a specific electrocardiogram of a subject, and a calculation unit. The calculation unit generates a laddergram from the specific electrocardiogram based on machine learning using learning data including a learning electrocardiogram and a learning laddergram corresponding to the learning electrocardiogram.

[0006] In the ladder graph generation device described above, the calculation unit automatically generates a ladder graph corresponding to a specific electrocardiogram based on machine learning using a training ladder graph. Therefore, medical professionals no longer need to generate ladder graphs themselves.

[0007] Furthermore, this specification discloses a computer program for generating ladder graphs. The computer program causes the computer to function as a generator that generates a ladder graph from a specific electrocardiogram of a subject based on machine learning using training data, which includes training electrocardiograms and training ladder graphs corresponding to the training electrocardiograms. [Brief explanation of the drawing]

[0008] [Figure 1] (a) is a diagram showing the structure of the heart, and (b) is a diagram showing the relationship between an electrocardiogram and a ladder diagram. [Figure 2] A block diagram showing the functionality of the ladder graph generation system according to the embodiment. [Figure 3] A flowchart illustrating an example of machine learning processing. [Figure 4] A diagram showing the relationship between an electrocardiogram and a segmented ladder graph. [Figure 5] (a) shows the electrocardiogram of a case of second-degree atrioventricular block, and (b) shows the segmented ladder diagram corresponding to (a). [Figure 6] A flowchart illustrating an example of a generation process that generates a ladder graph from a subject's electrocardiogram using a Deep Neural Network (DNN) trained through machine learning. [Figure 7] (a) shows the electrocardiogram of a normal person (normal case), and (b) shows the segmented ladder graph corresponding to (a) and arrows indicating the direction of electrical signal (action potential) transmission. [Figure 8] (a) shows an electrocardiogram of an atrioventricular block case without escape contractions or escape rhythms from the atrioventricular junction or ventricles, and (b) shows a segmented ladder graph corresponding to (a) and arrows indicating the direction of electrical signal transmission. [Figure 9](a) shows an electrocardiogram of an atrioventricular block case with a ventricular escape beat (or escape rhythm), and (b) shows the segmented ladder graph corresponding to (a) and arrows indicating the direction of electrical signal transmission. [Figure 10] (a) shows an electrocardiogram of an atrioventricular block case with escape contractions (or escape rhythms) from the atrioventricular junction, and (b) shows the segmented ladder graph corresponding to (a) and arrows indicating the direction of electrical signal transmission. [Figure 11] This diagram illustrates how to create a linear ladder diagram. (a) shows an electrocardiogram, (b) shows a segmented ladder diagram corresponding to (a), and (c) shows a linear ladder diagram generated based on (b). [Modes for carrying out the invention]

[0009] The main features of the embodiments described below are listed below. Note that the technical elements described below are independent technical elements that exhibit technical usefulness individually or in various combinations, and are not limited to the combinations described in the claims at the time of filing.

[0010] In a second aspect of the technology disclosed herein, in the first aspect described above, the learning ladder graph may consist of time information relating to the times when excitation begins and ends at each of several locations within the heart. The calculation unit may generate a ladder graph in a specific electrocardiogram that consists of specific time information relating to specific times when excitation begins and ends at each of several locations within the heart, based on machine learning. Such a configuration makes it possible to generate a ladder graph that quantitatively shows the transmission of electrical signals in each location within the heart.

[0011] In a third aspect of the technology disclosed herein, in the second aspect described above, the training ladder graph may be drawn using segments separated by specific time intervals based on specific time information. Furthermore, in a fourth aspect of the technology disclosed herein, in the third aspect described above, the ladder graph (i.e., the generated ladder graph) may be drawn using segments separated by specific time intervals based on specific time information. By drawing the training ladder graph and the ladder graph generated using the ladder graph generation device using segments as described above, the basis for drawing the ladder graph can be made clear.

[0012] In a fifth aspect of the technology disclosed herein, in any one of the second to fourth aspects described above, the calculation unit may determine the direction of electrical signal transmission between multiple locations in a specific electrocardiogram based on specific time information. With such a configuration, a medical professional such as a physician (hereinafter also simply referred to as a medical professional) can easily grasp the direction of electrical signal transmission between multiple locations in the heart.

[0013] In a sixth aspect of the technology disclosed herein, the ladder graph generation device may further include an output unit that outputs a ladder graph generated by the calculation unit, in the fourth aspect described above. The output unit may output a ladder graph drawn using segments. With such a configuration, a ladder graph composed of specific time information is drawn using segments separated by specific time intervals. This makes it easier for medical professionals to visually grasp the timing of excitation in each part of the heart.

[0014] In a seventh aspect of the technology disclosed in this specification, in the above sixth aspect, the output unit may further output a diagram ladder diagram that shows a ladder diagram as a diagram. The diagram ladder diagram may be a graph that shows a plurality of parts on the vertical axis along the order of transmission of an electrical signal (action potential) and shows time on the horizontal axis. Each of the plurality of parts of the graph may have an upper end and a lower end. The time estimated to be the start of the part above one of the upper end and the lower end and the time estimated to be the completion of the part above the other of the upper end and the lower end may be connected by a line segment. The conventional ladder diagram manually created by medical staff is shown as a diagram. By outputting the diagram ladder diagram, it is possible to output a diagram ladder diagram, which is a form that medical staff are familiar with. Therefore, it becomes easier for medical staff to grasp the overall abnormality of electrical signal transmission between each part in the heart.

[0015] In an eighth aspect of the technology disclosed in this specification, in any one of the above first to seventh aspects, the calculation unit may specify a candidate for heart disease based on the generated ladder diagram. According to such a configuration, it is possible to provide information for assisting diagnosis to a doctor.

Example

[0016] Referring to the drawings, the laddergram generation system 1 according to the embodiment will be described. The laddergram generation system 1 generates a laddergram from the electrocardiogram of a subject. As shown in Fig. 1(a), the electrical excitation that causes a heartbeat occurs in the sinoatrial node (SA) and is transmitted in the order of the atrium (A), the atrioventricular node (AV), and the ventricle (V). A laddergram is a diagram showing the electrical propagation between the sinoatrial node (SA), the atrium (A), the atrioventricular node (AV), and the ventricle (V). Fig. 1(b) shows a part of the electrocardiogram and the laddergram corresponding to the electrocardiogram. In Fig. 1(b), the laddergram is shown as a diagram. Generally, a conventional laddergram manually created by medical staff such as doctors (hereinafter also simply referred to as medical staff) is created as a diagram. By creating a laddergram, the conduction of the stimulus between the atrium (A) and the ventricle (V) is organized, and the excitation propagation of the stimulus conduction system can be grasped at a glance. In this specification, the "electrocardiogram" may mean the waveform itself representing the electrical signal of the stimulus conduction system of the subject's heart, or may mean the data representing the waveform. Also, the laddergram is not limited to the laddergram shown as a diagram. In this embodiment, as will be described in detail below, the laddergram includes not only the laddergram shown as a diagram but also laddergrams shown in other forms (for example, the segment type laddergram described later, etc.). A laddergram can also be said to be a diagram in which the electrical propagation to each part in the heart is represented in time series. Also, in this embodiment, although candidates for arrhythmia diagnosis are specified from the laddergram generated by the laddergram generation system 1 as described later, it is not limited to such a configuration. If it is possible to specify from the generated laddergram, candidates for heart diseases other than arrhythmia may be specified.

[0017] As shown in Fig. 2, the laddergram generation system 1 includes a machine learning device 10 and a laddergram generation device 30.

[0018] The machine learning device 10 includes an arithmetic unit 12 and a communication unit 24. The arithmetic unit 12 is communicably connected to the communication unit 24 by wiring such as a bus bar.

[0019] The arithmetic unit 12 is configured using a computer that includes memory 14 and a CPU 20. Memory 14 includes hardware such as ROM and RAM. The memory 14 stores an arithmetic program, and when the CPU 20 executes this program, the CPU 20 functions as the learning processing unit 22 shown in Figure 2. The processing of the learning processing unit 22 will be described in detail later. Memory 14 includes a learning data storage unit 16 and a trained DNN storage unit 18.

[0020] The training data storage unit 16 stores data used for machine learning processing (hereinafter also referred to as training data). The training data is used in machine learning processing to generate ladder graphs, which will be described later. The training data is acquired from external devices via the communication unit 24 and stored in the training data storage unit 16.

[0021] The trained DNN memory unit 18 stores information including machine learning models learned by machine learning. In this embodiment, features for generating a ladder graph from an electrocardiogram are learned by machine learning using a deep neural network (DNN). The trained DNN memory unit 18 stores the DNN learned by machine learning (hereinafter also referred to as the trained DNN).

[0022] The ladder graph generation device 30 comprises a calculation unit 32, a communication unit 44, and an output unit 46. The calculation unit 32 is connected to the communication unit 44 and the output unit 46 via wiring such as busbars, enabling them to communicate with each other.

[0023] The arithmetic unit 32 is configured using a computer that includes memory 34 and a CPU 40. Memory 34 includes hardware such as ROM and RAM. An arithmetic program is stored in memory 34, and when the CPU 40 executes this arithmetic program, the CPU 40 functions as the generation processing unit 42 shown in Figure 2. The processing of the generation processing unit 42 will be described in detail later. Memory 34 includes a subject information storage unit 36 ​​and a trained DNN storage unit 38.

[0024] The communication unit 24 is equipped with a communication interface for the machine learning device 10 to communicate with external devices via wired or wireless means. The arithmetic unit 12 can acquire various types of data from external devices via the communication unit 24. Specifically, the arithmetic unit 12 acquires training data transmitted from external devices via the communication unit 24. The arithmetic unit 12 stores the acquired training data in the training data storage unit 16.

[0025] The subject information storage unit 36 ​​stores information about the subject. This information includes the subject's electrocardiogram (hereinafter also referred to as the specific electrocardiogram), the ladder graph of the specific electrocardiogram generated by the generation processing unit 42, and candidate arrhythmia diagnoses identified from the generated ladder graph. The subject's electrocardiogram (specific electrocardiogram) is measured using an electrocardiogram measuring device such as a Holter monitor. The calculation unit 32 acquires the specific electrocardiogram from an external device such as an electrocardiogram measuring device via the communication unit 44 and stores it in the subject information storage unit 36. The calculation unit 32 stores the ladder graph of the specific electrocardiogram generated by the generation processing unit 42 and the identified candidate arrhythmia diagnoses in the subject information storage unit 36.

[0026] The trained DNN memory unit 38 stores the trained DNN that has been trained by machine learning in the machine learning device 10. The arithmetic unit 32 acquires the trained DNN from the machine learning device 10 via the communication units 24 and 44 and stores it in the trained DNN memory unit 38.

[0027] Memory 14 also stores diagnostic information used in the process of identifying candidates for arrhythmia diagnosis, as described later, ladder graphs corresponding to the diagnostic information, and information derived from them (hereinafter also referred to as information regarding candidate arrhythmia diagnoses). The diagnostic information includes ladder graph patterns that suggest arrhythmia. Memory 14 stores these ladder graph patterns that suggest arrhythmia as information regarding candidate arrhythmia diagnoses, linking them to the type of diagnosis (i.e., the disease name).

[0028] The communication unit 44 is equipped with a communication interface for the ladder graph generation device 30 to communicate with external devices via wired or wireless means. The calculation unit 32 can acquire various data from external devices via the communication unit 44. Specifically, the calculation unit 32 acquires the subject's electrocardiogram (specific electrocardiogram) transmitted from the external device and a trained DNN transmitted from the external device (specifically, the machine learning device 10) via the communication unit 44. The calculation unit 32 stores the acquired specific electrocardiogram in the subject information storage unit 36 ​​and stores the acquired trained DNN in the trained DNN storage unit 38.

[0029] The output unit 46 outputs the ladder graph generated by the calculation unit 32 and the identified candidate arrhythmias. In this embodiment, the output unit 46 is a display device, and the generated ladder graph and identified candidate arrhythmias are displayed on the display device to present them to healthcare professionals. The output unit 46 may also be a printer, and the generated ladder graph and identified candidate arrhythmias may be printed by the printer to present them to healthcare professionals. Furthermore, the ladder graph generation device 30 may have both a monitor and a printer as the output unit 46, or it may be connectable to an external display device and printer.

[0030] Next, the process by which the laddergram generation system 1 generates a laddergram from the subject's electrocardiogram will be explained. The process by which the laddergram generation system 1 generates a laddergram from the subject's electrocardiogram is performed by a machine learning process in which the machine learning device 10 learns the features necessary for generating a laddergram from the electrocardiogram using machine learning, and a generation process in which the laddergram generation device 30 generates a laddergram from the subject's electrocardiogram using the trained DNN learned by the machine learning process.

[0031] First, the machine learning process performed by the machine learning device 10 will be described. As shown in Figure 3, the learning processing unit 22 first acquires training data (S100). The training data is stored in the training data storage unit 16. The learning processing unit 22 acquires the training data stored in the training data storage unit 16. The training data is data for learning features to generate ladder graphs from electrocardiograms, and includes training electrocardiogram data and its correct labels. The training data includes supervised data. The correct labels are ladder graphs manually generated by a specialist from the training electrocardiogram data. The learning processing unit 22 performs DNN machine learning using the training data. Note that a detailed explanation of DNN machine learning is omitted because known DNN machine learning methods can be applied.

[0032] Here, we will explain the ladder graph used as the correct label. As shown in Figure 4, in this embodiment, the ladder graph used as the correct label is a ladder graph (hereinafter also called a segmented ladder graph) that shows the transmission of electrical signals (action potentials) in each part of the heart in four layers of segments. In the segmented ladder graph of Figure 4, the vertical axis shows each part of the heart, and the horizontal axis shows time. On the vertical axis, each part is arranged from top to bottom in accordance with the order of normal excitation conduction to each part of the heart. In this embodiment, the parts arranged on the vertical axis from top to bottom are the sinoatrial node (SA), atria (A), atrioventricular junction (AVJ), and ventricle (V). Each segment shows the time from the start of excitation to the completion of excitation in the corresponding part. Specifically, the time of the start of excitation in each part is the time when it is estimated that at least one of the cells in that part was excited. Also, the time of the completion of excitation in each part is the time when it is estimated that the excitation has spread to all the cells in that part. In other words, each segment represents quantitative time information, showing the time from when excitation reaches the corresponding area to when the excitation spreads throughout the entire corresponding area.

[0033] In a segmented ladder graph, the end of a segment on the excitation completion side may overlap with the excitation start time side of the segment in the next region (the adjacent lower region on the vertical axis). For example, in Figure 4, a portion of the right side of the segment representing the atrium (A) temporally overlaps with a portion of the left side of the atrioventricular junction (AVJ). This indicates that the electrical signal was transmitted from the atrium (A) to the AVJ before the entire atrium (A) was excited. Also, the end of a segment on the excitation completion side may approximately coincide with the excitation start time side of the segment in the next region (the adjacent lower region on the vertical axis). For example, in Figure 4, the right end of the segment representing the AVJ approximately coincides with the left end of the segment representing the ventricle (V). This indicates that the excitation of the ventricle (V) began approximately simultaneously with the completion of excitation of the AVJ.

[0034] Next, the learning processing unit 22 performs machine learning processing using the training data acquired in step S100 (S110). The training data includes an electrocardiogram and a segmented ladder graph corresponding to that electrocardiogram. In the machine learning processing, the features of each segment of the segmented ladder graph and the electrocardiogram features corresponding to that segment are learned. The features learned here include not only the shape of the electrocardiogram waveform from the excitation start time to the excitation completion time of a particular segment, but also the features of the electrocardiogram waveform for a continuous time interval containing multiple segments.

[0035] As training data, we use not only electrocardiograms of people with normal rhythm (normal sinus rhythm) without arrhythmias (hereinafter simply referred to as "normal people"), but also electrocardiograms of patients with various arrhythmias. For example, Figure 5(a) shows the electrocardiogram of a patient with atrioventricular block, and Figure 5(b) shows the segmented ladder graph corresponding to Figure 5(a). As shown in Figures 5(a) and 5(b), in patients with atrioventricular block, there is a mixture of heartbeats in which excitation is transmitted along the cardiac conduction system in the order of sinoatrial node (SA), atrium (A), atrioventricular junction (AVJ), and ventricle (V), and heartbeats in which excitation is transmitted to the sinoatrial node (SA), atrium (A), and atrioventricular junction (AVJ), and then the transmission is interrupted within the atrioventricular junction (AVJ). Furthermore, since excitation is not transmitted throughout the entire atrioventricular junction (AVJ), the segment ladder graph shown in Figure 5(b) omits the segment representing the electrical signal transmitted to the AVJ. By using machine learning with electrocardiograms from patients with arrhythmias, such as those with atrioventricular block, it is possible to learn the characteristics of cases where electrical signals are not transmitted normally to various parts of the heart.

[0036] When the machine learning process in step S110 is completed, the learning processing unit 22 stores the DNN learned in step S110 in the learned DNN storage unit 18 (S120).

[0037] Next, the generation process of generating a ladder graph from a subject's electrocardiogram using a trained DNN executed by the ladder graph generation device 30 will be described. As shown in Figure 6, first, the generation processing unit 42 acquires a specific electrocardiogram stored in the subject information storage unit 36 ​​(S200). As described above, the specific electrocardiogram is, for example, the electrocardiogram of a subject measured using an electrocardiogram measuring device such as a Holter monitor. The specific electrocardiogram is stored in the subject information storage unit 36. The generation processing unit 42 acquires the specific electrocardiogram from the subject information storage unit 36. Next, the generation processing unit 42 acquires a trained DNN stored in the trained DNN storage unit 18 (S210).

[0038] Next, the generation processing unit 42 uses the trained DNN acquired in step S210 to generate a segmented ladder graph from the specific electrocardiogram acquired in step S200 (S220). The trained DNN acquired in step S210 has learned the features necessary to generate segments for each part of the heart from the electrocardiogram. Therefore, by using the trained DNN, it is possible to identify the segments corresponding to the specific electrocardiogram and generate a segmented ladder graph from the identified segments.

[0039] Next, the generation processing unit 42 identifies the direction of electrical signal transmission between each segment of the segment-type ladder gram generated in step S220 (S230). Assuming a proximal end (SA side) and distal end (V side) of the cardiac conduction system, the direction of transmission from the proximal end to the distal end is defined as forward conduction, and the direction of transmission from the distal end to the proximal end is defined as reverse conduction. Once the cardiomyocytes that make up the heart are excited, they enter a refractory period, and cannot enter the next excitation until the refractory period caused by the preceding excitation disappears. Specifically, the generation processing unit 42 determines that electrical signals have been transmitted between adjacent parts if adjacent parts on the vertical axis of the segment-type ladder gram overlap or if the endpoint and starting point approximately coincide. The generation processing unit 42 then identifies that among the adjacent parts, the electrical signal was transmitted from the part that was excited first to the part that was excited later.

[0040] Figure 7(b) shows a segmented ladder diagram for a normal person. In Figure 7(b), the direction of electrical signal transmission is indicated by arrows. In a normal person, electrical signals are transmitted sequentially from the top to the bottom of the segmented ladder diagram.

[0041] Figure 8(b) shows a segmented ladder gram of a patient with atrioventricular block without ectopic spontaneous excitation (no escape contractions or escape rhythms due to spontaneous excitation from sites other than the sinoatrial node (SA)). The heartbeat on the left side of Figure 8(b) is a block beat, where the electrical signal is transmitted in a forward direction from top to bottom to the sinoatrial node (SA) and atrium (A), and excitation conduction stops within the atrioventricular junction (AVJ). In the segmented ladder gram shown in Figure 8(b), the electrical signal transmitted partway through the AVJ is omitted. On the other hand, the heartbeat on the right side of Figure 8(b) is a normal beat, where the electrical signal transmitted in a forward direction to the sinoatrial node (SA), atrium (A), and AVJ is transmitted throughout the AVJ without stopping within the AVJ, and further transmitted in a forward direction from the AVJ to the ventricle (V).

[0042] Figure 9(b) shows a segmented ladder gram of a patient with atrioventricular block (AVJ) with ectopic spontaneous excitation (rescue contractions and escape rhythms occurring due to spontaneous excitation from a site other than the sinoatrial node (SA)). In the two heartbeats in Figure 9(b), electrical signals are transmitted in the forward direction to the sinoatrial node (SA) and atrium (A), and excitation conduction is stopped within the AVJ. Also, corresponding to the AVJ block, reverse electrical signal transmission is observed, originating from the ventricle (V) and blocked within the AVJ, moving from lower to upper. Note that in the AVJ, neither the electrical signals transmitted from the atrium (A) to the AVJ nor the electrical signals transmitted from the ventricle (V) to the AVJ are transmitted to the entire AVJ. Therefore, in the segmented ladder gram shown in Figure 9(b), the segment showing the electrical signals transmitted to the AVJ is omitted. The most rational explanation for this excitation pattern of the atria (A) and ventricles (V) is the ventricular escape rhythm associated with atrioventricular block.

[0043] Figure 10(b) is a segmented ladder gram of a different case from Figure 9(b), and is an atrioventricular block with ectopic spontaneous excitation. The source of the excitation is different from that in Figure 9(b). In the 3 seconds shown in Figure 10(b), excitation from the sinoatrial node (SA) (frequency 92-93 / min) is transmitted laterally from the sinoatrial node (SA) to the atria (A) and terminates in the atrioventricular junction (AVJ). There is a slight delay between the excitation up to partway up the AVJ and the excitation of the AVJ itself. This is presumed to be due to the generation of ectopic spontaneous excitation (rescue contraction / rescue rhythm) at a frequency of 44 / min from the AVJ, corresponding to the high-grade atrioventricular block. This ectopic spontaneous excitation generated in the AVJ is conducted laterally from the AVJ to the ventricles (V), but is blocked by the refractory period and does not conduct in the reverse direction. In the segmented ladder diagram shown in Figure 10(b), the electrical signals transmitted from the atria (A) to partway up the atrioventricular junction (AVJ) are omitted, and only the ectopic spontaneous excitations generated at the AVJ and the downstream ventricular (V) excitations are depicted.

[0044] Figures 7(b), 8(b), 9(b), and 10(b) show the direction of electrical signal transmission between each segment, indicated by arrows. Identifying the direction of electrical signal transmission between each segment in this way makes it easier to understand the conduction of excitation in each part of the heart, which is useful for the electrophysiological diagnosis of arrhythmias.

[0045] Next, the generation processing unit 42 generates a linear ladder diagram (S240) based on the segment-type ladder diagram generated in step S220 and the direction of electrical signal transmission between segments identified in step S230. Specifically, the generation processing unit 42 identifies segments in which electrical signals are transmitted between adjacent parts in the vertical direction. The generation processing unit 42 then connects a point representing the estimated time of the start of excitation at the upper or lower end (hereinafter also referred to as the start point) and a point representing the estimated time of the completion of excitation at the lower or upper end (hereinafter also referred to as the completion point) with a line segment in the segment in which the electrical signal is transmitted. In this case, when the electrical signal is transmitted in the forward direction, the start point is located at the upper end and the completion point is located at the lower end. On the other hand, when the electrical signal is transmitted in the reverse direction, the start point is located at the lower end and the completion point is located at the upper end. Furthermore, the generation processing unit 42 identifies the location and time of excitation based on the segments in which electrical signals are transmitted between vertically adjacent parts and the direction of electrical signal transmission between segments, and marks a point indicating excitation between the upper and lower ends of the identified part. The generation processing unit 42 connects the point indicating excitation with the starting point of the part adjacent to the part where the excitation occurred.

[0046] Here, with reference to Figure 11, the method for creating a linear ladder graph will be explained in more detail. Figure 11 shows the case where electrical signals are transmitted in the forward direction. First, the generation processing unit 42 identifies the time of excitation in the atrium (A) (the beginning of the P wave in the electrocardiogram). The atrium (A) is drawn so that the time of excitation in the atrium (A) and the time of excitation completion are the same. The time of excitation in the atrium (A) and the time of excitation in the atrioventricular junction (AVJ) are assumed to be the same. As shown in Figure 11(b), in a segmented ladder graph, the time of excitation in the atrioventricular junction (AVJ) is slightly delayed (about 16 ms) compared to the time of excitation in the atrium (A), but in a linear ladder graph, this delay is considered negligible. Next, the generation processing unit 42 identifies the time of excitation completion in the atrioventricular junction (AVJ). The time of completion of excitation at the atrioventricular junction (AVJ) coincides with the time of excitation start of the ventricle (V) (the start of the QRS wave on the electrocardiogram). Next, the generation processing unit 42 draws a straight line connecting the excitation start time and the excitation completion time of the AVJ. This draws the AVJ. Next, the generation processing unit 42 identifies the excitation completion time of the ventricle (V) (the end of the QRS wave on the electrocardiogram) and draws a straight line connecting the excitation start time of the ventricle (V) (excitation completion time of the AVJ) and the excitation completion time of the ventricle (V). Alternatively, the generation processing unit 42 may assume an excitation source in the middle of the sinoatrial node (SA) (between the upper and lower ends), and allow the excitation to propagate from the sinoatrial node (SA) to the atria (A) over a period of about 40 ms. Next, a straight line is drawn from the excitation source in the sinoatrial node (SA) to the excitation start time in the atria (A). This creates the ladder diagram shown in Figure 11(c).

[0047] Furthermore, the generation processing unit 42 identifies candidates for estimated arrhythmia diagnoses (hereinafter also referred to as disease candidates) based on the segmented ladder graph generated in step S220 and the direction of electrical transmission between segments identified in step S230 (S250). Specifically, the generation processing unit 42 reads information regarding arrhythmia diagnosis candidates from the memory 14. The generation processing unit 42 then determines whether the generated segmented ladder graph contains the segmented ladder graph and the pattern of electrical signal transmission direction between segments contained in the information regarding arrhythmia diagnosis candidates. If the generated segmented ladder graph contains the ladder graph pattern contained in the information regarding arrhythmia diagnosis candidates, the generation processing unit 42 identifies the type of arrhythmia (disease name) associated with that ladder graph pattern as a disease candidate. This allows information to be provided to the physician to assist in diagnosis. For example, as shown in Figure 8(b), if electrical signals are transmitted to the sinoatrial node (SA) and atria (A) in a segmented ladder gram, but not to the atrioventricular junction (AVJ), it can be estimated that this is atrioventricular block without ectopic spontaneous excitation. Also, as shown in Figure 9(b), if electrical signals are transmitted to the sinoatrial node (SA) and atria (A) in a segmented ladder gram, but not to the atrioventricular junction (AVJ), and electrical signal transmission from the ventricle (V) is also observed, it can be estimated that this is atrioventricular block with ventricular escape contractions (or escape rhythms). Furthermore, as shown in Figure 10(b), in a segmented ladder diagram, if electrical signals are transmitted to the sinoatrial node (SA) and atria (A), but not to the atrioventricular junction (AVJ), and if electrical signals are generated at the AVJ and transmitted from the AVJ to the ventricles (V), it can be estimated that there is atrioventricular block with a rescue contraction (or rescue rhythm) from the AVJ. The generation processing unit 42 may also identify disease candidates from the diagram ladder diagram generated in step S240. In this embodiment, after generating the diagram ladder diagram in step S240, arrhythmia diagnosis candidates were identified in step S250, but the configuration is not limited to this.For example, after identifying candidate arrhythmia diagnoses in step S250, a ladder diagram may be generated in step S240, or the generation of the ladder diagram in step S240 and the identification of candidate regular heart diagnoses in step S250 may be performed simultaneously.

[0048] Next, the generation processing unit 42 outputs the ladder diagram generated in step S240, and, if a disease candidate was identified in step S250, the identified disease candidate (S260). For example, if the output unit 46 is a display device, the generation processing unit 42 displays the ladder diagram and disease candidate on the display device. If the output unit 46 is a printer, the generation processing unit 42 prints and outputs the ladder diagram and disease candidate. As described above, the ladder diagram is in the form that medical professionals have traditionally created manually. By outputting the ladder diagram, a ladder diagram in a form that medical professionals are familiar with can be presented to them. Therefore, medical professionals can easily grasp the excitation propagation of the conduction system in a specific electrocardiogram.

[0049] The generation processing unit 42 may output a ladder diagram along with a ladder diagram, consisting of a segment-type ladder diagram generated in step S220 and marks (e.g., arrows) indicating the direction of electrical signal transmission between segments identified in step S230 (hereinafter also referred to as a digital ladder diagram), or it may output a segment-type ladder diagram generated in step S220 (i.e., a segment-type ladder diagram that does not include marks indicating the direction of electrical signal transmission between segments). Furthermore, the generation processing unit 42 may output only a digital ladder diagram instead of a ladder diagram, or only a segment-type ladder diagram.

[0050] Furthermore, in this embodiment, the segmented ladder graph and the diagrammatic ladder graph had four levels (sinoatrial node (SA), atrium (A), atrioventricular junction (AVJ), and ventricle (V)), but the configuration is not limited to this. For example, the segmented ladder graph and the diagrammatic ladder graph may have five or more levels. Also, the levels of the segmented ladder graph and the diagrammatic ladder graph may be different. For example, the segmented ladder graph may be generated with five levels, while the diagrammatic ladder graph may be generated with four levels.

[0051] In this embodiment, machine learning was performed using electrocardiograms and segmented ladder graphs as training data to generate segmented ladder graphs from specific electrocardiograms. However, the system is not limited to this configuration. For example, machine learning may be performed using electrocardiograms and linear ladder graphs as training data to generate linear ladder graphs from specific electrocardiograms. In this case as well, since the linear ladder graph is automatically generated by machine learning, healthcare professionals can obtain a linear ladder graph from a specific electrocardiogram without spending time and effort.

[0052] However, as described in this embodiment, it is preferable to perform machine learning using electrocardiograms and segmented ladder graphs as training data, and to generate segmented ladder graphs from specific electrocardiograms. As mentioned above, a segmented ladder graph can show, for each region, the time from when it is estimated that one of the cells in that region was excited to when it is estimated that the excitation has spread to all the cells in that region, as quantitative time information. Therefore, by evaluating the segmented ladder graph, it is possible to judge the electrophysiological validity and consistency, and to grasp the excitation propagation of the conduction system of a specific electrocardiogram with greater accuracy.

[0053] The specific examples of the technology disclosed herein have been described in detail above, but these are merely illustrative and do not limit the scope of the claims. The technology described in the claims includes various modifications and changes to the specific examples described above. Furthermore, the technical elements described herein or in the drawings exhibit technical usefulness individually or in various combinations, and are not limited to the combinations described in the claims at the time of filing. In addition, the technology illustrated herein or in the drawings achieves multiple objectives simultaneously, and achieving even one of these objectives itself constitutes technical usefulness. [Explanation of Symbols]

[0054] 1: Ladder Gram System 10: Machine Learning Device 12: Arithmetic section 14: Memory 16: Learning data storage unit 18: Trained DNN memory unit 20:CPU 22: Learning Processing Unit 24: Communications Department 30: Ladder Gram Generator 32: Arithmetic section 34: Memory 36: Subject information storage unit 38: Trained DNN memory unit 40:CPU 42: Generation Processing Unit 44: Communications Department 46: Output section

Claims

1. An acquisition unit that acquires the subject's specific electrocardiogram, It includes a calculation unit, The calculation unit generates a ladder graph from a specific electrocardiogram based on machine learning using training data, which includes a training electrocardiogram and a training ladder graph corresponding to the training electrocardiogram.

2. The aforementioned learning ladder graph consists of time information related to the start and completion times of excitation in multiple parts of the heart, The ladder graph generation device according to claim 1, wherein the calculation unit generates a ladder graph in the specific electrocardiogram that is composed of specific time information related to specific times representing the start and completion of excitation of each of the multiple parts within the heart, based on the machine learning.

3. The ladder graph generation device according to claim 2, wherein the learning ladder graph is drawn using segments separated by the specific time based on the specific time information.

4. The ladder graph generation device according to claim 3, wherein the ladder graph is drawn using segments separated by the specific time based on the specific time information.

5. The ladder graph generating device according to claim 2, wherein the calculation unit determines the direction of transmission of electrical signals between the plurality of parts in the specific electrocardiogram based on the specific time information.

6. The system further includes an output unit that outputs the ladder graph generated by the calculation unit, The ladder graph generation apparatus according to claim 4, wherein the output unit outputs the ladder graph drawn using the segments.

7. The output unit further outputs a diagrammatic ladder graph, which shows the ladder graph as a line diagram. The aforementioned ladder diagram is a graph that shows the multiple parts along the order of electrical signal transmission on the vertical axis and time on the horizontal axis. Each of the aforementioned parts of the graph has an upper end and a lower end, The ladder graph generating apparatus according to claim 6, wherein the estimated start time of the portion on one of the upper and lower ends and the estimated completion time of the portion on the other of the upper and lower ends are connected by a line segment.

8. The ladder graph generating device according to any one of claims 1 to 7, wherein the calculation unit identifies candidate heart diseases based on the generated ladder graph.

9. A computer program for generating ladder diagrams, Computers, A computer program that functions as a generation unit that generates a ladder graph from a specific electrocardiogram of a subject, based on machine learning using training data including a training electrocardiogram and a training ladder graph corresponding to the training electrocardiogram.

Images