A paroxysmal atrial fibrillation attack position marking method and system

Abstract

The application relates to the technical field of electrocardio monitoring, in particular to a paroxysmal atrial fibrillation attack position marking method and system, which comprises the following steps: acquiring an electrocardio signal to be marked and inputting the electrocardio signal to a trained segmentation network model to obtain an output prediction sequence; traversing the prediction sequence and calculating a merging likelihood value and a segmentation total likelihood value of a current data point to perform sequence clustering; performing binary processing on the prediction sequence to screen potential atrial fibrillation points; traversing all the potential atrial fibrillation points, judging the continuity with a previous potential atrial fibrillation point, and generating a plurality of potential atrial fibrillation event segments based on the continuity; performing a processing action on the potential atrial fibrillation event segments based on the distance between the potential atrial fibrillation event segments and the size of the potential atrial fibrillation event segments to obtain actual atrial fibrillation event segments, wherein the processing action comprises merging and eliminating; and determining an atrial fibrillation starting position and an atrial fibrillation ending position based on wave information corresponding to the actual atrial fibrillation event segments. The application has the effect of improving the accuracy of atrial fibrillation event marking recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrocardio monitoring, and in particular to a paroxysmal atrial fibrillation attack position marking method and system. BACKGROUND

[0002] As a common arrhythmia disease, the accurate marking of the paroxysmal atrial fibrillation attack position (start and end points and the continuous region) is crucial for clinical diagnosis, treatment plan formulation (such as target point selection of catheter ablation) and disease prognosis evaluation. Accurate identification of the atrial fibrillation attack region can help doctors to judge the origin and development mode of atrial fibrillation and distinguish paroxysmal atrial fibrillation from persistent atrial fibrillation, thereby providing a basis for individualized diagnosis and treatment. However, the electrocardio signal is affected by individual differences (heart rate, heart rhythm baseline, etc.), acquisition noise (myoelectric interference, poor electrode contact, etc.) and the complex rhythm of atrial fibrillation, and there are many challenges in accurately marking the attack position.

[0003] Traditional atrial fibrillation region marking mostly relies on a semantic segmentation model combined with a simple threshold method (such as setting a fixed probability threshold to screen atrial fibrillation heartbeats) for marking. First, the electrocardio data to be marked is input into the model, and the model outputs an atrial fibrillation probability sequence based on the learning result. However, the output probability sequence has many scattered values, and there is ambiguity in the transition between atrial fibrillation heartbeats and sinus heartbeats, so a single hard threshold is easy to misjudge:

[0004] Event splitting problem: if a local low probability point (such as 0.4) appears in a segment of atrial fibrillation signal due to model prediction error or physiological fluctuation, the hard threshold method will split the continuous atrial fibrillation event into multiple segments. At the same time, when there is a missed or misidentified point in an atrial fibrillation event, the same atrial fibrillation event will be incorrectly divided into two atrial fibrillation events, increasing the number of atrial fibrillation attacks and reducing the duration of a single attack, which cannot reflect the real attack process;

[0005] Noise misjudgment problem: a short-term heart rhythm abnormality (such as atrial tachycardia) may be mislabeled as an atrial fibrillation event due to a probability value higher than the threshold, or a real short-duration atrial fibrillation event may be filtered out due to a value lower than the minimum heartbeat number threshold, resulting in missed diagnosis or misdiagnosis. SUMMARY

[0006] In order to improve the accuracy of atrial fibrillation event marking and identification, the present application provides a paroxysmal atrial fibrillation attack position marking method and system.

[0007] In a first aspect, the present application provides a paroxysmal atrial fibrillation attack position marking method, which adopts the following technical solution:

[0008] A paroxysmal atrial fibrillation attack position marking method, comprising the following steps:

[0009] Obtaining an electrocardio signal to be marked, and inputting the electrocardio signal into a trained segmentation network model to obtain an output prediction sequence, wherein the prediction sequence includes sequence values between 0 and 1 corresponding to a plurality of data points;

[0010] Traversing the prediction sequence and calculating a merging likelihood value and a segmentation total likelihood value of a current data point to perform sequence clustering;

[0011] Performing binary processing on the prediction sequence to screen potential atrial fibrillation points;

[0012] Traversing all the potential atrial fibrillation points and judging the continuity with a previous potential atrial fibrillation point, and generating a plurality of potential atrial fibrillation event segments based on the continuity;

[0013] Performing processing actions on each of the potential atrial fibrillation event segments based on the distance between the potential atrial fibrillation event segments and the size of the potential atrial fibrillation event segments to obtain actual atrial fibrillation event segments, wherein the processing actions include merging and eliminating;

[0014] Determining an atrial fibrillation start position and an atrial fibrillation end position based on wave information corresponding to the actual atrial fibrillation event segments.

[0015] In some embodiments, the training of the segmentation network model includes the following steps:

[0016] Obtaining an electrocardio training segment, locating a first sinus heartbeat and a second sinus heartbeat, and a first atrial fibrillation heartbeat and a second atrial fibrillation heartbeat in the electrocardio training segment;

[0017] Calculating an atrial fibrillation start position based on the first sinus heartbeat and the first atrial fibrillation heartbeat, and calculating an atrial fibrillation end position based on the second sinus heartbeat and the second atrial fibrillation heartbeat;

[0018] Labeling an atrial fibrillation region based on the atrial fibrillation start position and the atrial fibrillation end position, and labeling all regions outside the atrial fibrillation region in the electrocardio training segment as non-atrial fibrillation region labels;

[0019] Inputting the labeled electrocardio training segment as a training sample into the segmentation network model for training.

[0020] In some embodiments, traversing the prediction sequence and calculating a merging likelihood value and a segmentation total likelihood value of a current data point to perform sequence clustering includes the following steps:

[0021] Traversing each of the data points;

[0022] Calculating a log-likelihood value when the current data point is merged with each of the data points in the previous segment, and defining the log-likelihood value as a merging likelihood value;

[0023] The log likelihood value of the current data point and each data point in the latter section as a starting point of a new section when the new section is formed is calculated as a total segmentation likelihood value, wherein the length of the new section meets a preset length;

[0024] The difference between the merging likelihood value and the total segmentation likelihood value is used to determine whether the current data point is a clustering point or a segmentation point.

[0025] In some embodiments, the predicted sequence is binarized to screen potential atrial fibrillation points, including the following steps:

[0026] The numerical relationship between each sequence value in the predicted sequence and a set threshold value is determined;

[0027] If the sequence value is greater than the set threshold value, the sequence value is changed to 1, and the data point corresponding to the sequence value is defined as the potential atrial fibrillation point;

[0028] If the sequence value is not greater than the set threshold value, the sequence value is changed to 0.

[0029] In some embodiments, all the potential atrial fibrillation points are traversed to determine the continuity with the previous potential atrial fibrillation point, and based on the continuity, a plurality of potential atrial fibrillation event sections are generated, including the following steps:

[0030] The first potential atrial fibrillation point is traversed, and a starting index and an ending index are set synchronously;

[0031] It is determined whether the difference between the index of the potential atrial fibrillation point and the ending index is not greater than 1;

[0032] If yes, it is determined that the potential atrial fibrillation point is adjacent to the ending index and has continuity, the ending index is updated to the position of the potential atrial fibrillation point, and the next potential atrial fibrillation point is traversed;

[0033] If no, it is determined that the potential atrial fibrillation point is not adjacent to the ending index and the continuity is cut off, the current position of the ending index is recorded, and the potential atrial fibrillation event section is generated in combination with the starting index;

[0034] The potential atrial fibrillation point generates a new starting index and a new ending index at the current position to repeat the above steps until the predicted sequence is completely traversed.

[0035] In some embodiments, each potential atrial fibrillation event section is processed based on the distance between the potential atrial fibrillation event sections to obtain an actual atrial fibrillation event section, including the following steps:

[0036] when a continuity cut-off occurs in the traversal process, the number of r-wave peaks between the end index in the previous said potential atrial fibrillation event segment and the new start index in the next said potential atrial fibrillation event segment is calculated and recorded as a first value;

[0037] a value relationship between the first value and a predefined maximum interval heartbeat number is judged;

[0038] if the first value is less than the maximum interval heartbeat number, the two said potential atrial fibrillation event segments are merged into one said actual atrial fibrillation event segment;

[0039] if the first value is not less than the maximum interval heartbeat number, the two said potential atrial fibrillation event segments are each taken as an independent said actual atrial fibrillation event segment.

[0040] In some embodiments, the processing action on each said potential atrial fibrillation event segment to obtain an actual atrial fibrillation event segment based on the size of each said potential atrial fibrillation event segment includes the following steps:

[0041] the number of r-wave peaks between the start index and the end index in each said potential atrial fibrillation event segment is calculated and recorded as a second value;

[0042] a value relationship between the second value and a predefined minimum event r-wave number is judged;

[0043] if the second value is not less than the minimum event r-wave number, the said potential atrial fibrillation event segment is retained and taken as a final said actual atrial fibrillation event segment;

[0044] if the second value is less than the minimum event r-wave number, the values of all said potential atrial fibrillation points in the said potential atrial fibrillation event segment are updated to 0 or ignored.

[0045] In some embodiments, the atrial fibrillation start position and the atrial fibrillation end position are determined based on the wave information corresponding to the actual atrial fibrillation event segment, including the following steps:

[0046] M R peaks are respectively found forward and backward from the start index in the actual atrial fibrillation event segment, and the corresponding RR intervals are respectively calculated, and the last said RR interval with a regular interval is selected as a fuzzy start position region;

[0047] the T-wave end position of the heart beat corresponding to the previous R peak in the fuzzy start position region is taken as the atrial fibrillation start position;

[0048] M R peaks are respectively found forward and backward from the end index in the actual atrial fibrillation event segment, and the corresponding RR intervals are respectively calculated, and the last said RR interval with a regular interval is selected as a fuzzy end position region;

[0049] obtaining a P-wave starting position of a heart beat corresponding to a latter R-peak in the fuzzy ending position region as the atrial fibrillation ending position.

[0050] In some embodiments, the method further comprises the following steps:

[0051] establishing a parameter group between different maximum interval heart beat numbers and different minimum r-wave numbers;

[0052] performing performance index influence tests based on each parameter group, and setting a composite index based on the parameter groups according to test results;

[0053] selecting an optimal parameter group for the potential atrial fibrillation event segment based on the composite index.

[0054] In a second aspect, the application provides a paroxysmal atrial fibrillation attack position marking system, which adopts the following technical solution:

[0055] A paroxysmal atrial fibrillation attack position marking system is used to implement the paroxysmal atrial fibrillation attack position marking method.

[0056] The technical solution provided by the embodiments of the application has the following technical effects:

[0057] The technical solution provided by the embodiments of the application has the following technical effects: BRIEF DESCRIPTION OF DRAWINGS

[0058] Figure 1 FIG. 1 is a step schematic diagram of a paroxysmal atrial fibrillation attack position marking method provided by an embodiment of the application.

[0059] Figure 2 FIG. 3 is a structure schematic diagram of a segmentation network model provided by an embodiment of the application.

[0060] Figure 3 FIG. 5 is a schematic diagram of a training sample obtained by positioning and labeling provided by an embodiment of the application.

[0061] Figure 4 FIG. 7 is an example diagram for processing to obtain an actual atrial fibrillation event segment based on a maximum interval and a minimum r-wave number provided by an embodiment of the application.

[0062] Figure 5is a boundary diagram of the starting point and the ending point of the atrial fibrillation event provided by an embodiment of the present application.

[0063] Figure 6 is a composite index diagram obtained by different combinations of the maximum interval heartbeat number and the shortest event r wave number in the embodiment.

[0064] Figure 7 is a predicted sequence change diagram corresponding to different processing schemes in the embodiment. DETAILED DESCRIPTION

[0065] In order to more clearly understand the objects, technical solutions and advantages of the present application, the present application is described and explained in detail below in connection with the drawings and embodiments. However, it should be understood by those of ordinary skill in the art that the present application can be implemented without these details. In some cases, in order to avoid unnecessary description and make aspects of the present application obscure, well-known methods, processes, systems, components and / or circuits that have been described at a higher level will not be described in detail. It is obvious for those of ordinary skill in the art that various changes can be made to the embodiments disclosed in the present application, and the universal principles defined in the present application can be applied to other embodiments and application scenarios without deviating from the principles and scope of the present application. Therefore, the present application is not limited to the embodiments shown, but conforms to the broadest scope of the claimed range of the present application.

[0066] It should be noted that the description of the embodiments is used to help understand the present application, but does not constitute a limitation of the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as there is no conflict.

[0067] In the description of the present application, the meaning of one or more is one or more, the meaning of multiple is two or more, greater than, less than, more than, etc. are understood as not including the number, above, below, etc. are understood as including the number. If it is described as first, second, it is only used to distinguish technical features for the purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the sequence of indicated technical features.

[0068] In the description of the present application, the description of the terms "one embodiment", "some embodiments", "exemplary embodiment", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a combined manner.

[0069] As shown in Figure 1 The embodiment of the present application discloses a paroxysmal atrial fibrillation attack position marking method, comprising the following steps:

[0070] S100, acquiring a to-be-marked electrocardio signal and inputting the to-be-marked electrocardio signal into a trained segmentation network model to acquire an output prediction sequence, wherein the prediction sequence includes sequence values between 0 and 1 corresponding to a plurality of data points.

[0071] Firstly, the segmentation network model is trained based on a large number of electrocardio samples, and the trained model can output a prediction sequence of the electrocardio signal after inputting the to-be-marked electrocardio signal.

[0072] The prediction sequence is equal in length to the electrocardio signal data, and includes a plurality of sequence values between 0 and 1, wherein the sequence value represents a probability value of a corresponding electrocardio data point corresponding to atrial fibrillation, and the probability value increases from 0 to 1, and 0 is a determined non-atrial fibrillation point and 1 is a determined atrial fibrillation point.

[0073] Meanwhile, the prediction sequence output by the model is a plurality of scattered points with different probability values, and the purpose of further atrial fibrillation attack marking is to aggregate these scattered points into the same event, and each event corresponds to the occurrence of an atrial fibrillation.

[0074] S200, traversing the prediction sequence and calculating a merging likelihood value and a segmentation total likelihood value of a current data point to perform sequence clustering.

[0075] In the conventional technology, a hard threshold method is usually directly used to realize the collection of the atrial fibrillation region, and this method can quickly integrate the atrial fibrillation region with a large probability value together, however, if a missed recognition or misrecognized point appears in an atrial fibrillation event, the same atrial fibrillation event will be divided into two or more atrial fibrillation events.

[0076] Therefore, in order to reduce the occurrence of such a situation, the present application first uses a Bernoulli distribution likelihood ratio check adaptive change point detection method to divide the prediction sequence into a plurality of continuous segments.

[0077] The core logic is to traverse all sequence values in the predicted sequence, dynamically compare the likelihood values of the current data point and the previous data segment, the likelihood values of the current data point and the next segment, and analyze whether the data point belongs to the previous data segment or needs to be segmented as the beginning of a new data segment based on the numerical size of the likelihood values.

[0078] By the above method, the optimized clustering of the sequence with a probability value greater than 0.5 is realized, and the noise with a probability value less than or equal to 0.5 is allowed as an effective sequence point, avoiding the error of dividing a complete event into multiple events due to partial missing or wrong noise points in a complete event.

[0079] S300, binarizing the predicted sequence to screen potential atrial fibrillation points.

[0080] After the Bernoulli distribution likelihood ratio check, the predicted sequence is binarized by the hard threshold method. The operation process is to identify the data points corresponding to the probability values exceeding the threshold size as potential atrial fibrillation points, and the data points corresponding to the probability values below the threshold are considered as non-atrial fibrillation points.

[0081] S400, traversing all potential atrial fibrillation points and judging the continuity with the previous potential atrial fibrillation point, and generating a plurality of potential atrial fibrillation event segments based on the continuity.

[0082] After screening the candidate potential atrial fibrillation points, further positioning of the atrial fibrillation event region is performed. Specifically, since the atrial fibrillation event region must correspond to the atrial fibrillation point, this link only needs to traverse all potential atrial fibrillation points, and judge whether the current potential atrial fibrillation point is continuous with the previous potential atrial fibrillation point. If continuous, it is considered that the two selected potential atrial fibrillation points belong to the same atrial fibrillation event, and if not continuous, it is considered that the two selected potential atrial fibrillation points do not belong to the same atrial fibrillation event.

[0083] Among them, the continuity in the application represents whether the two atrial fibrillation points are adjacent or the same point.

[0084] After traversing all potential atrial fibrillation points, a plurality of atrial fibrillation event segments are generated based on the continuity between the potential atrial fibrillation points. The atrial fibrillation event segment is represented as a potential atrial fibrillation point that meets a certain number requirement and has continuity.

[0085] S500, processing the potential atrial fibrillation event segments based on the distance between the potential atrial fibrillation event segments and the size of the potential atrial fibrillation event segments to obtain actual atrial fibrillation event segments, the processing action including merging and eliminating.

[0086] After screening out the atrial fibrillation event segment, the distance between different adjacent atrial fibrillation event segments needs to be analyzed to analyze whether the two atrial fibrillation event segments belong to two independent events or belong to the same event. Secondly, the length of each potential atrial fibrillation event segment is analyzed to determine whether the event segment is a valid event segment.

[0087] Through the above two analysis works, further processing operations are performed on the potential atrial fibrillation event segments, including but not limited to merging of two or more atrial fibrillation events, and elimination of invalid events.

[0088] After completing the final processing, all potential atrial fibrillation event segments currently existing are regarded as final effective actual atrial fibrillation event segments, and each actual atrial fibrillation event segment corresponds to an episode of paroxysmal atrial fibrillation in a real environment.

[0089] S600, determining the atrial fibrillation start position and the atrial fibrillation end position based on the wave information corresponding to the actual atrial fibrillation event segment.

[0090] After confirming the event segment of the atrial fibrillation, in order to further let the patient and the doctor know the specific time point of the occurrence of the atrial fibrillation and the end of the atrial fibrillation, the start time position and the end time position of each episode of paroxysmal atrial fibrillation need to be further determined and marked through the corresponding electrocardio wave information in the actual atrial fibrillation event segment.

[0091] Through the above steps, the input to-be-labeled electrocardio information is quickly labeled by the prediction sequence through the training of the segmentation semantic model. At the same time, before binarization, the data is aggregated and segmented through Bernoulli distribution likelihood ratio check to accurately capture the probability distribution mode of the "continuous atrial fibrillation event", and further through the distance between the event segments, the length of the event segment and other indicators, the event segment is dynamically merged, eliminated and other processing actions to improve the accuracy of the event segment identification positioning and labeling. Finally, the wave signal accurately identifies the accurate start point position and end point position in an atrial fibrillation event segment.

[0092] In some embodiments, the segmentation network model can be a public semantic segmentation model, such as Unet, PPMpbileNet, or other semantic segmentation network model built by oneself. The specific model structure is as shown in Figure 2 The training of the segmentation network model includes the following steps:

[0093] S110, acquiring an electrocardio training segment, locating a first sinus heart beat and a second sinus heart beat in the electrocardio training segment, and a first atrial fibrillation heart beat and a second atrial fibrillation heart beat.

[0094] S120, calculating the atrial fibrillation start position based on the first sinus heart beat and the first atrial fibrillation heart beat, and calculating the atrial fibrillation end position based on the second sinus heart beat and the second atrial fibrillation heart beat.

[0095] S130, labeling an atrial fibrillation region tag based on the atrial fibrillation starting position and the atrial fibrillation ending position, and labeling all regions outside the atrial fibrillation region in the electrocardiogram training segment as non-atrial fibrillation region tags.

[0096] S140, inputting the labeled electrocardiogram training segment into the segmentation network model as a training sample for training.

[0097] First, a large amount of electrocardiogram data is sorted as an electrocardiogram training segment, and the atrial fibrillation region is positioned and labeled according to the position characteristics of different signals in the electrocardiogram training segment. The labeled electrocardiogram training segment is used as a training sample for subsequent model training.

[0098] As shown in Figure 3 , the positioning and labeling process is as follows:

[0099] First, the last sinus beat before the onset of atrial fibrillation is obtained and recorded as , the first atrial fibrillation beat after the onset of atrial fibrillation is obtained and recorded as , the last atrial fibrillation beat at the end of the onset of atrial fibrillation is recorded as , and the first sinus beat after the end of the onset of atrial fibrillation is recorded as .

[0100] The starting position of the onset of atrial fibrillation is defined as , which is the end position of the first T wave after , and the end position of the T wave is obtained by a general method, which will not be described here. If the end position of the T wave is not obtained due to abnormal conditions, the starting position is set to (( - ) / 2. The ending position of the onset of atrial fibrillation is defined as , which is the starting position of the first P wave after . Similarly, if it is not obtained, the ending position is set to (( - ) / 2.

[0101] Then the region between the atrial fibrillation starting position and the atrial fibrillation ending position is the region position of the onset of atrial fibrillation. After defining and labeling the region, non-atrial fibrillation region tags are added to other regions outside the region, so that the subsequent model learns the electrocardiogram features represented by the tag positions, and the model learns the probability of different electrocardiogram features corresponding to atrial fibrillation or non-atrial fibrillation.

[0102] In some embodiments, the predicted sequence is traversed and the combined likelihood value and the total likelihood value of the current data point are calculated for sequence clustering, including the following steps:

[0103] S210, traverse each data point.

[0104] S220, calculate the log-likelihood value of the current data point when merging with the previous data points and define it as the merging likelihood value.

[0105] S230, calculate the log-likelihood value of the current data point as the starting point of a new segment when forming a new segment with the subsequent data points as the new segment, wherein the length of the new segment meets the preset length.

[0106] S240, judge whether the current data point is a clustering point or a segmentation point based on the difference between the merging likelihood value and the segmentation total likelihood value.

[0107] When the i-th point is traversed, it is assumed that the data point is merged with the previously determined data segment as a whole, and the log-likelihood value of the complex Bernoulli distribution of the whole is calculated and defined as the merging likelihood value.

[0108] At the same time, it is assumed that the data point is the beginning of a new segment, and the log-likelihood value of the new segment composed of the data point and the subsequent data points is calculated and defined as the segmentation total likelihood value.

[0109] If the segmentation total likelihood value is greater than the merging likelihood value, and the difference is greater than 1, it means that the point "as the beginning of a new segment" is more complex than the data distribution "merging with the previous segment", so the point needs to be segmented, and the point does not belong to the part of the data distribution of the previous segment.

[0110] Further, it is also necessary to judge whether the length of the new segment obtained when the data point is the beginning of the new segment and the length of the previous segment meet the requirements. In the present application, there are at least 3 data points in the previous segment and the new segment, so as to avoid that a single noise point is misjudged as a segmentation point.

[0111] The present scheme allows a small number of low-probability points to exist within an event through dynamic comparison of likelihood values (as long as the overall likelihood value is higher, it will not be split), such as 1-2 points with p<0.5 in a segment of atrial fibrillation event. As long as the likelihood value after merging is higher than that after splitting into two segments, it is determined as the same event, which perfectly adapts to the reality that "there is physiological fluctuation in atrial fibrillation signal and there is local error in model prediction";

[0112] At the same time, from the clinical point of view, the onset of atrial fibrillation is a continuous electrophysiological abnormal process. The "segment length ≥ 3" condition of the present scheme avoids misjudging the segmentation point due to 1-2 point fluctuations; "likelihood value difference > 1.0" ensures that "true splitting" (which is indeed the end of atrial fibrillation / the beginning of new atrial fibrillation) and "false fluctuation" (noise within the event) can be distinguished - which conforms to the clinical judgment standard for "continuity of atrial fibrillation event".

[0113] Further,

[0114] The function definition of Bernoulli logarithmic distribution function in the present application is as follows:

[0115] ;

[0116] wherein, ) represents the logarithmic probability of observing the label sequence y in the case of the predicted probability p, reflecting the matching degree of the data and the probability distribution, and the greater the value represents the higher the matching degree; b represents the upper limit of the summation, representing the number of sequence data points participating in the calculation, corresponding to the total number of sample points used for analysis in a segment of electrocardio signal; a represents the data point currently traversed; The label variable sequence value is identified as the label, indicating the probability of the model predicting that the electrocardio signal at a certain time is atrial fibrillation, and the value is in the range of (0, 1); p represents the conditional probability value, which is a value before (0.5, 1). The label (0-1) of the atrial fibrillation probability prediction sequence naturally conforms to the Bernoulli test, and the hard threshold result in the conventional technology only exists “atrial fibrillation” and “non-atrial fibrillation”.

[0117] By comparing the likelihood value of Bernoulli distribution, the “continuous atrial fibrillation event probability distribution mode” can be accurately captured:

[0118] The real atrial fibrillation event segment: the label is mostly 1, p is close to 1, and the likelihood value is high;

[0119] Non-atrial fibrillation segment or noise segment: the label is mostly 0, p is close to 0, and the likelihood value is high;

[0120] If the label in a segment of sequence is 1 and 0 (for example, the atrial fibrillation event is split by noise points), the likelihood value will be significantly reduced – this is the underlying logic of “higher likelihood value after splitting” (the sum of the likelihood values of the two segments after splitting is higher than that of the combined segment).

[0121] In some embodiments, the prediction sequence is binarized to screen potential atrial fibrillation points, including the following steps:

[0122] S310, judging the numerical relationship between each sequence value in the prediction sequence and the set threshold value.

[0123] S320, if the sequence value is greater than the set threshold value, the sequence value is changed to 1, and the data point corresponding to the sequence value is defined as a potential atrial fibrillation point.

[0124] S330, if the sequence value is not greater than the set threshold value, the sequence value is changed to 0.

[0125] In the screening of the atrial fibrillation candidate point, first, a threshold thr1 is set, and the value greater than the threshold thr1 in the above-mentioned clustered prediction sequence is set to 1, and the value not greater than the threshold thr1 is set to 0, so that the data cluster point falling within the pre-defined threshold thr1 is considered as part of the potential atrial fibrillation event.

[0126] In this way, the sequence of different probability values is converted into an atrial fibrillation event parameter with only 0 or 1 through the binaryzation of the hard threshold, and the point with a value of 1 is identified as a potential atrial fibrillation point, and the point with a value of 0 is identified as a non-atrial fibrillation event point.

[0127] In some embodiments, all potential atrial fibrillation points are traversed and the continuity with the previous potential atrial fibrillation point is determined, and a number of potential atrial fibrillation event segments are generated based on the continuity, including the following steps:

[0128] S410, the first potential atrial fibrillation point is traversed, and the start index and the end index are set synchronously.

[0129] All potential atrial fibrillation points are traversed, and when the first potential atrial fibrillation point is traversed, the start index sIdx and the end index eIdx are set synchronously at the point, wherein the start index represents the address position identifier corresponding to the event start state, and the end index represents the address position identifier corresponding to the event end state.

[0130] Therefore, the first potential atrial fibrillation point at this time essentially has three attributes, which are the current checkpoint, the start index, and the end index.

[0131] S420, determine whether the difference between the index of the potential atrial fibrillation point and the end index is less than 1.

[0132] The distance between the current checkpoint and the end index is determined by the difference between the index of the current checkpoint and the end index (the same point or adjacent).

[0133] S430, if yes, it is determined that the potential atrial fibrillation point is adjacent to the end index and has continuity, the end index is updated to the position of the potential atrial fibrillation point, and the next potential atrial fibrillation point is traversed.

[0134] If it is not greater than 1, it is determined that the current potential atrial fibrillation point and the end index are adjacent and have continuity, and the end index is changed to the index position of the current checkpoint, and the current checkpoint is moved to the next potential atrial fibrillation point with the traversal process. It should be noted that the start index is the starting position mark of each atrial fibrillation event, and as long as the end index does not interrupt, the start index will not be updated with the traversal.

[0135] Case 1: The current checkpoint is the first potential AF point in the traversal process, at this time the current checkpoint (X1) and the end index (X1) are in the same position, the difference between the indexes is 0, which meets the requirement of not greater than 1, so the end index update process is X1→X1, and the index update process of the current checkpoint is X1→X2.

[0136] Case 2: The current checkpoint is the second potential AF point in the traversal process, at this time the index of the current checkpoint is X2, and the end index is X1, so the difference between the indexes is 1, which meets the requirement of not greater than 1, at this time the end index is updated to the index position of the current checkpoint, that is, X1→X2, and the index of the current checkpoint is updated to X2→X3.

[0137] As long as the data points in the traversal process and the end index always meet the above requirements, it is considered that there is always continuity, so these points are considered as part of the same AF event segment.

[0138] S440, if not, it is determined that the potential AF point is not adjacent to the end index and the continuity is cut off, the current position of the end index is recorded, and the potential AF event segment is generated in combination with the start index.

[0139] If the result is greater than 1, it is determined that the continuity between the current potential AF point and the end index is interrupted, and they are not in adjacent positions, at this time the position of the end index will not change and the position of the end index is recorded, and the area between the current position of the end index and the position of the start index is taken as the potential AF event segment.

[0140] For example, case 3:

[0141] The current checkpoint is the n-th potential AF point in the traversal process, and the current end index is X(n-5), and the index of the current checkpoint is Xn, and the result obtained after the difference calculation is 5, which does not meet the requirement of not greater than 1. Since the objects of traversal are all potential AF points, a difference of 5 indicates that the current potential AF point is separated from the previous potential AF point by 5 non-AF points, so the current checkpoint and the previous potential AF point are not adjacent.

[0142] At this time, based on the current end index X(n-5), and in combination with the current start index X1 and the end index X(n-5), a potential AF event segment with a length of n-4 is formed. The current checkpoint Xn is taken as the start data point of the new AF event segment, and a new start index and end index are generated at the same time, and the traversal labeling of the next event is started.

[0143] S450, the potential AF point generates a new start index and end index at the current position to repeat the above steps until the entire prediction sequence is traversed.

[0144] When the prediction sequence is traversed completely, the corresponding annotation event segment is obtained based on the result of the traversal analysis.

[0145] In some embodiments, the processing action on each potential atrial fibrillation event segment to obtain an actual atrial fibrillation event segment is based on the distance between the potential atrial fibrillation event segments, and includes the following steps:

[0146] S510, when continuity is interrupted in the traversal process, the number of r-peak values between the end index in the previous potential atrial fibrillation event segment and the new start index in the next potential atrial fibrillation event segment is calculated and recorded as a first value.

[0147] When continuity is interrupted in the traversal process, the number of r-peak values between the end index position in the previous potential atrial fibrillation event segment (i.e. the position of the end index at the moment of continuity interruption) and the start index position in the next potential atrial fibrillation event segment (i.e. the position of the current checkpoint at the moment of continuity interruption) is calculated and recorded as RC1.

[0148] S511, the value relationship between the first value and a predefined maximum interval heartbeat number is determined.

[0149] S512, if the first value is less than the maximum interval heartbeat number, the two potential atrial fibrillation event segments are merged into one actual atrial fibrillation event segment.

[0150] S513, if the first value is not less than the maximum interval heartbeat number, the two potential atrial fibrillation event segments are each regarded as an independent actual atrial fibrillation event segment.

[0151] A maximum interval heartbeat number ClustirDis is predefined. If RC1 is less than ClustirDis, the data points between the two potential atrial fibrillation event segments that originally belong to non-atrial fibrillation points are regarded as part of the same atrial fibrillation event, and thus the two potential atrial fibrillation event segments are merged (when merging, the non-atrial fibrillation points with a value of 0 between the two atrial fibrillation event segments are changed to 1 as atrial fibrillation points); otherwise, if RC1 is not less than ClustirDis, the non-atrial fibrillation points between the two potential atrial fibrillation event segments are not considered as part of the same or any atrial fibrillation event, and thus the two potential atrial fibrillation event segments are not merged, but are each regarded as an independent event segment.

[0152] In this way, whether the two events are considered as the same event based on the number of data points between the two event segments, so as to avoid incorrect splitting of a complete atrial fibrillation region due to small gaps caused by noise, misrecognition, etc.

[0153] In some embodiments, processing of potential atrial fibrillation event segments based on their size to obtain actual atrial fibrillation event segments includes the following steps:

[0154] S520, calculate the number of r-wave peaks between the start index and the end index in each potential atrial fibrillation event segment and record it as the second value.

[0155] During the traversal, once a potential atrial fibrillation event segment has been completely traversed (that is, at the moment of discontinuity interruption), it is necessary to further calculate the number of r-peaks between the start index (the position of the first data point in the event segment) and the end index (the position of the last data point in the event segment) in the potential atrial fibrillation event segment and define it as RC2.

[0156] S521, determine the numerical relationship between the second value and the predefined number of shortest event r waves.

[0157] S522, if the second value is not less than the number of shortest event r waves, then the potential atrial fibrillation event segment is retained and used as the final actual atrial fibrillation event segment.

[0158] S523, if the second value is less than the number of r waves in the shortest event, then the values ​​of all potential atrial fibrillation points in the potential atrial fibrillation event segment are updated to 0 or ignored.

[0159] A minimum number of r-waves in a potential atrial fibrillation event, MinAfDis, is predefined. RC2 is then checked to see if it is less than MinAfDis. If it is not less, all points in the potential atrial fibrillation event segment are considered atrial fibrillation points and treated as actual atrial fibrillation event segments. If it is less, all points in the potential atrial fibrillation event segment are marked as non-atrial fibrillation points. In this case, atrial fibrillation points that were originally 1 can be changed to 0, or the event segment can be ignored in subsequent atrial fibrillation event segment statistics.

[0160] Therefore, the candidate event segments are judged by length. When the length of an event segment (the number of effective atrial fibrillation points in the event segment) is short, it may belong to short noise events such as Pre-AFib3, and it needs to be removed. When the length of an event segment is relatively long and matches the occurrence of an atrial fibrillation event, it is regarded as a valid event.

[0161] Regarding the aforementioned determination process based on the maximum interval heart rate (ClustirDis) and the shortest event r-wave count (MinAfDis), to more intuitively demonstrate the effectiveness of the analysis process, this application employs... Figure 4 For example:

[0162] like Figure 3As shown, three potential paroxysmal atrial fibrillation events are predicted, defined as Pre-AFib1, Pre-AFib2, Pre-AFib3 respectively, while the black line represents the electrocardiogram (ECG) signal and the red line represents the predicted sequence.

[0163] Since the distance between Pre-AFib1 and Pre-AFib2 is small and the number of heartbeats is less than ClustrDis, it is considered that Pre-AFib1 and Pre-AFib2 are the same atrial fibrillation event, and at this time Pre-AFib1 and Pre-AFib2 are merged as an actual atrial fibrillation event segment; while the distance between Pre-AFib2 and Pre-AFib3 is large and the number of heartbeats is greater than ClustrDis, it is considered that Pre-AFib2 and Pre-AFib3 are two independent atrial fibrillation events and are not merged; at the same time, the length of Pre-AFib1 and Pre-AFib2 is longer and greater than MinAfDis, and the number of detected heartbeats is more, so it is considered to be an effective atrial fibrillation event; while the length of Pre-AFib3 is shorter, the number of detected signals is too small and less than MinAfDis, so it is not considered as an effective event and is directly ignored.

[0164] Then, after the above analysis process, the three potential atrial fibrillation event segments Pre-AFib1, Pre-AFib2, and Pre-AFib3 originally recorded will finally show that there is one actual atrial fibrillation episode (the region merged by Pre-AFib1 and Pre-AFib2), and the other positions are non-atrial fibrillation regions.

[0165] In some embodiments, the atrial fibrillation start position and the atrial fibrillation end position are determined based on the wave information corresponding to the actual atrial fibrillation event segment, including the following steps:

[0166] S610, find M R peaks forward and backward from the start index in the actual atrial fibrillation event segment and calculate the corresponding RR intervals, and select the last RR interval of the interval rule as the fuzzy start position region.

[0167] S620, obtain the T wave end position of the corresponding heartbeat of the previous R peak in the fuzzy start position region as the atrial fibrillation start position.

[0168] S630, find M R peaks forward and backward from the end index in the actual atrial fibrillation event segment and calculate the corresponding RR intervals, and select the last RR interval of the interval rule as the fuzzy end position region.

[0169] S640, obtain the P wave start position of the corresponding heartbeat of the next R peak in the fuzzy end position region as the atrial fibrillation end position.

[0170] In order to further accurately determine the start position and the end position of an atrial fibrillation episode, the application uses the method of first determining the fuzzy area and then determining the accurate point to locate and mark the two endpoints of the event.

[0171] First, the nearest M R peaks are found forward and backward from the start index in the actual atrial fibrillation event segment, and then the corresponding RR intervals are calculated. The position between the two R peaks of the last RR interval of the interval rule is the fuzzy start position of the atrial fibrillation event. In the same way, the forward and backward M R peaks are found, and then the RR intervals are calculated. The position between the two R peaks in the last RR interval of the interval rule is the fuzzy end position of the atrial fibrillation event.

[0172] Among them, M R peaks are found forward and backward in order to capture the rhythm change nodes of "transition from sinus rhythm to atrial fibrillation" and "transition from atrial fibrillation to sinus rhythm" through extension search. M can be set to 3, 5, etc.

[0173] The core determination standard of "interval rule" is that the RR interval of sinus rhythm has stability (small variation coefficient, the RR interval sequence of sinus rhythm may be "800 ms, 810 ms, 805 ms", and the maximum adjacent difference is only 10 ms), while the RR interval of atrial fibrillation has randomness (large variation coefficient, the RR interval sequence of atrial fibrillation may be "600 ms, 950 ms, 720 ms", and the maximum adjacent difference is up to 350 ms); Therefore, the "last rule interval" is the "critical interval" of the transition from sinus rhythm to atrial fibrillation or the transition from atrial fibrillation to sinus rhythm, and the area between the two R peaks corresponding to the "last rule interval" is the fuzzy range of the start or end of atrial fibrillation.

[0174] The RR interval sequence corresponding to the fuzzy start position presents a change from "regular (sinus) to irregular (atrial fibrillation)", and the "last rule interval" is the "last interval that still meets the stable characteristics of sinus" in this change.

[0175] The RR interval sequence corresponding to the fuzzy end position presents a change from "irregular (atrial fibrillation) to regular (sinus)", and the "last rule interval" is the "preliminary critical interval of the first rule interval" in this change.

[0176] Secondly, the accurate position is found according to the selected fuzzy position, specifically,

[0177] The precise starting point of atrial fibrillation is the T wave ending position of the preceding R peak in the RR interval corresponding to the ambiguous starting position. The T wave is a waveform generated during ventricular repolarization, usually appearing as a positive or negative wave after the QRS complex. The criterion for determining the T wave ending position is "the position where the T wave waveform returns to the baseline (isopotential line)" (i.e., the intersection of the T wave and the subsequent baseline). The physiological significance of its ending position is "completion of ventricular repolarization" and "the last physiological node of the sinus rhythm cycle".

[0178] The precise end point of atrial fibrillation is the P wave starting position of the heartbeat corresponding to the next R peak in the RR interval after the ambiguous end position. The P wave is a waveform generated during atrial depolarization, usually appearing as a small positive wave before the QRS complex. The criterion for determining the P wave starting position is "the position of the P wave waveform deviating from the baseline (isopotential line)" (i.e., the intersection of the P wave and the preceding baseline). The physiological significance of its starting position is "the beginning of regular atrial depolarization" and also "the first physiological node of the sinus rhythm cycle".

[0179] Among them, such as Figure 5 As shown, the black line represents the original ECG signal, the blue line represents the detected binary sequence corresponding to the original ECG, and the red line represents the binary sequence after being located by the above-mentioned fuzzy-precise start and interpretation boundary marking. In the binary sequence, the value 1 represents the atrial fibrillation region, and the value 0 represents the non-atrial fibrillation region. The eight sub-images below are magnified views of the areas marked by the green circles in the waveform above.

[0180] In some embodiments, the following steps are also included:

[0181] S710 establishes a parameter set between different maximum interval heart rate and different shortest event r-wave counts.

[0182] The S720 performs performance index impact tests based on each parameter group and establishes a composite index based on the test results and parameter groups.

[0183] S730 selects the optimal parameter set for potential atrial fibrillation event segments based on a composite index.

[0184] In this application, during the experimental testing phase, different values ​​for the maximum interval heart rate (ClustirDis) and the shortest event r-wave count (MinAfDis) resulted in different accuracy in the localization of atrial fibrillation events. The reason for this is as follows:

[0185] When the maximum interval heartbeat number ClustirDis is too small, short intervals caused by local noise in the same atrial fibrillation event will be misjudged as two independent events, leading to overestimation of the number of atrial fibrillation episodes and underestimation of the single duration, and the performance index Cmsitx is low; when the maximum interval heartbeat number ClustirDis is too large, two independent short atrial fibrillation events or atrial fibrillation events and normal heart rate fluctuations will be miscombined as one event, leading to underestimation of the number of atrial fibrillation episodes and overestimation of the single duration, and the performance index Cmsitx is also low.

[0186] When the minimum event r wave number MinAfDis is too small, smaller short noise events (such as short atrial tachycardia and ECG signal interference) cannot be filtered, and these noises will be misjudged as atrial fibrillation events, leading to a high false positive rate and a low performance index Cmsitx; when the minimum event r wave number MinAfDis is too large, real atrial fibrillation events will be misjudged, and the performance index Cmsitx will no longer improve.

[0187] Therefore, how to select the optimal combination of the maximum interval heartbeat number ClustirDis and the minimum event r wave number MinAfDis is an important reference for ensuring accuracy and stability in subsequent atrial fibrillation event labeling.

[0188] In this application, different combinations of the maximum interval heartbeat number ClustirDis and the minimum event r wave number MinAfDis are established, and the performance index Cmsitx corresponding to the labeling results is calculated, and a composite index is formed according to the results and parameter groups. Based on different composite index results, the optimal parameter threshold combination is selected in different scenarios.

[0189] As shown in Figure 6 , it lists the influence of different ClustirDis and MinAfDis on the final performance index. The horizontal axis represents the maximum interval heartbeat number ClustirDis value from 0-10, and the vertical axis represents the performance index Cmsitx value. Different line graphs correspond to different minimum event r wave number MinAfDis values. The line represents the change of the performance index Cmsitx with the change of the maximum interval heartbeat number based on the corresponding minimum event. The highest red dot represents the maximum value of the performance index Cmsitx.

[0190] As can be seen from the figure, the parameter combination corresponding to the maximum value of the performance index Cmsitx is (ClustirDis=4, MinAfDis=7), (ClustirDis=4, MinAfDis=8), and (ClustirDis=4, MinAfDis=9).

[0191] The performance index calculation can be changed and improved according to different scenes, and some conventional reference data indexes include accuracy, recall rate, the number of correctly identified atrial fibrillation events, the number of incorrectly identified atrial fibrillation events, and the like.

[0192] According to different electrocardio monitoring scenes, the requirements for the shortest event can be different, and then according to the obtained composite index, the maximum interval heartbeat number ClustirDis required for the optimal performance corresponding to the shortest event can be quickly obtained.

[0193] Finally, as shown in Figure 7 Fig. (a) shows the prediction sequence output by the model after inputting the electrocardio signal to be labeled, Fig. (b) shows the schematic diagram of the atrial fibrillation points and non-atrial fibrillation points classified after direct hard threshold binary processing, Fig. (c) shows the prediction sequence schematic diagram obtained after the Bernoulli logarithmic distribution function clustering operation, and Fig. (d) shows the final prediction sequence schematic diagram obtained after all the labeling and analysis steps of the present application, wherein the black line represents the ECG signal.

[0194] The present application also discloses a paroxysmal atrial fibrillation attack position marking system for implementing the paroxysmal atrial fibrillation attack position marking method.

[0195] The implementation principle is:

[0196] The segmented semantic model is trained to quickly label the prediction sequence of the input electrocardio information to be labeled, and the Bernoulli distribution likelihood ratio check is performed before binary to aggregate and segment the data to accurately capture the probability distribution mode of the continuous atrial fibrillation event, and further through the distance between the event segments, the length of the event segments and other indexes, the event segment merging, elimination and other processing actions are dynamically performed to improve the accuracy of the event segment identification and positioning labeling, and finally the wave signal accurately identifies the accurate start point position and end point position of an atrial fibrillation event segment.

[0197] It should be understood that although each step in the flowchart of the accompanying drawings is shown in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and they can be executed in other sequences.

[0198] The above are preferred embodiments of the present application, and do not limit the protection scope of the present application, therefore: all equivalent changes made according to the structure, shape, principle of the present application shall be covered within the protection scope of the present application.

Claims

1. A paroxysmal atrial fibrillation attack position marking method characterized by, The method comprises the following steps: Obtaining an electrocardio signal to be marked and inputting the electrocardio signal into a trained segmentation network model to obtain an output prediction sequence, wherein the prediction sequence comprises sequence values between 0 and 1 corresponding to a plurality of data points; Traversing the prediction sequence and calculating a merging likelihood value and a segmentation total likelihood value of a current data point to perform sequence clustering, specifically, Traversing each data point; Calculating a log-likelihood value of the current data point when merging with each data point in the front segment and defining the log-likelihood value as the merging likelihood value; Calculating a log-likelihood value of the current data point as a starting point of a new segment when forming the new segment with each data point in the rear segment as the segmentation total likelihood value, wherein the length of the new segment meets a preset length; Judging whether the current data point is a clustering point or a segmentation point based on the difference between the merging likelihood value and the segmentation total likelihood value; Performing binary processing on the prediction sequence to screen potential atrial fibrillation points; Traversing all the potential atrial fibrillation points and judging the continuity with a previous potential atrial fibrillation point, and generating a plurality of potential atrial fibrillation event segments based on the continuity; Performing processing actions on each potential atrial fibrillation event segment based on the distance between the potential atrial fibrillation event segments and the size of each potential atrial fibrillation event segment to obtain actual atrial fibrillation event segments, wherein the processing actions comprise merging and eliminating; Determining an atrial fibrillation starting position and an atrial fibrillation ending position based on wave information corresponding to the actual atrial fibrillation event segments.

2. The paroxysmal atrial fibrillation episode location marking method according to claim 1, characterized in that, The training of the segmentation network model comprises the following steps: Obtaining an electrocardio training segment, locating a first sinus heart beat and a second sinus heart beat and a first atrial fibrillation heart beat and a second atrial fibrillation heart beat in the electrocardio training segment; Calculating an atrial fibrillation starting position based on the first sinus heart beat and the first atrial fibrillation heart beat, and calculating an atrial fibrillation ending position based on the second sinus heart beat and the second atrial fibrillation heart beat; Labeling an atrial fibrillation region label based on the atrial fibrillation starting position and the atrial fibrillation ending position, and labeling a non-atrial fibrillation region label outside the atrial fibrillation region in the electrocardio training segment; Inputting the labeled electrocardio training segment into the segmentation network model as a training sample to perform training.

3. The paroxysmal atrial fibrillation episode location marking method according to claim 1, characterized in that, The binary processing on the prediction sequence to screen potential atrial fibrillation points comprises the following steps: Judging the numerical relationship between each sequence value in the prediction sequence and a set threshold value; If the sequence value is greater than the set threshold value, the sequence value is changed to 1, and the data point corresponding to the sequence value is defined as the potential atrial fibrillation point; If the sequence value is not greater than the set threshold value, the sequence value is changed to 0.

4. The paroxysmal atrial fibrillation episode location marking method according to claim 3, characterized in that, Traversing all the potential atrial fibrillation points and judging the continuity with a previous potential atrial fibrillation point, and generating a plurality of potential atrial fibrillation event segments based on the continuity, comprising the following steps: Traversing a first potential atrial fibrillation point and synchronously setting a starting index and an ending index; Judging whether the difference between the index of the potential atrial fibrillation point and the ending index is not greater than 1; If yes, it is determined that the potential atrial fibrillation point is adjacent to the ending index and has continuity, the ending index is updated to the position of the potential atrial fibrillation point, and the next potential atrial fibrillation point is traversed; If not, it is determined that the potential atrial fibrillation point is not adjacent to the end index and continuity cutoff, the current position of the end index is recorded, and the potential atrial fibrillation event segment is generated in combination with the start index; The potential atrial fibrillation point generates a new start index and end index at the current position to repeat the above steps until the entire predicted sequence is traversed.

5. The paroxysmal atrial fibrillation episode location marking method according to claim 4, characterized in that, Based on the distance between each potential atrial fibrillation event segment, a processing action is performed on each potential atrial fibrillation event segment to obtain an actual atrial fibrillation event segment, including the following steps: When a continuity cutoff occurs during the traversal process, the number of r-wave peaks between the end index in the previous potential atrial fibrillation event segment and the new start index in the subsequent potential atrial fibrillation event segment is calculated and recorded as a first value; A numerical relationship between the first value and a predefined maximum interval heartbeat number is determined; If the first value is less than the maximum interval heartbeat number, the two potential atrial fibrillation event segments are combined into one actual atrial fibrillation event segment; If the first value is not less than the maximum interval heartbeat number, the two potential atrial fibrillation event segments are each treated as an independent actual atrial fibrillation event segment.

6. The paroxysmal atrial fibrillation episode location marking method according to claim 5, characterized in that, Based on the size of each potential atrial fibrillation event segment, a processing action is performed on each potential atrial fibrillation event segment to obtain an actual atrial fibrillation event segment, including the following steps: The number of r-wave peaks between the start index and the end index in each potential atrial fibrillation event segment is calculated and recorded as a second value; A numerical relationship between the second value and a predefined minimum event r-wave number is determined; If the second value is not less than the minimum event r-wave number, the potential atrial fibrillation event segment is retained and treated as a final actual atrial fibrillation event segment; If the second value is less than the minimum event r-wave number, the values of all potential atrial fibrillation points in the potential atrial fibrillation event segment are updated to 0 or ignored.

7. The paroxysmal atrial fibrillation episode location marking method according to claim 6, characterized in that, Based on the wave information corresponding to the actual atrial fibrillation event segment, an atrial fibrillation start position and an atrial fibrillation end position are determined, including the following steps: In the actual atrial fibrillation event segment, M R peaks are found forward and backward from the start index, and the corresponding RR intervals are calculated, and the last RR interval with a regular interval is selected as a fuzzy start position region; The T-wave end position of the heart beat corresponding to the previous R peak in the fuzzy start position region is obtained as the atrial fibrillation start position; In the actual atrial fibrillation event segment, M R peaks are found forward and backward from the end index, and the corresponding RR intervals are calculated, and the last RR interval with a regular interval is selected as a fuzzy end position region; The P-wave start position of the heart beat corresponding to the subsequent R peak in the fuzzy end position region is obtained as the atrial fibrillation end position.

8. The paroxysmal atrial fibrillation episode location marking method according to claim 6, characterized in that, Further comprising the following steps: A parameter group is established between different maximum interval heartbeat numbers and different minimum event r-wave numbers; Based on each parameter group, a performance index influence test is performed, and a composite index is established based on the test results and the parameter groups; Based on the composite index, an optimal parameter group is selected for the potential atrial fibrillation event segment.

9. A paroxysmal atrial fibrillation episode location marking system, characterized by A method for implementing a paroxysmal atrial fibrillation episode location marker as claimed in any of claims 1-8.

Images