Atrial tachycardia analysis module that provides an evolving schedule of tachycardia events

Abstract

The invention described herein relates to systems and methods directed to a data-driven, continuous and adaptive learning approach for analyzing atrial tachycardia (AT) in the human body. The systems and methods can create an automatically evolving AT profile such that subsequent changes in AT are more accurately recognized and classified.

Description

[Technical field]

[0001] The present application is directed to systems and methods for data-driven, continuous and adaptive learning approaches to analyzing atrial tachycardia (AT) in patients. These approaches can be used during catheter ablation procedures and can provide information to medical professionals during ablation procedures regarding AT. [Background technology]

[0002] The human heart has four chambers: the right and left atria (atrial chambers) and the right and left ventricles (ventricular chambers). During normal operation, blood flows into the right atrium and is pumped through the right ventricle to the lungs, where the blood is oxygenated. The blood then returns from the lungs to the left atrium and is pumped out through the left ventricle, where the oxygenated blood is distributed to the body. Electrical signals cause the heart muscles to contract, allowing the heart to pump blood through the circulatory system. The pulmonary circuit is the pathway by which blood travels from the heart to the lungs and then back from the lungs to the heart. The systemic circuit is the pathway by which blood travels from the heart to the body and then back to the heart.

[0003] Atrial tachycardia (AT) is an electrical abnormality in which there is faster-than-normal activity in the atrial chambers of the heart. Activity during AT is not regulated by the sinus node, the part of the atrium that normally regulates the heart's electrical signals. Instead, activity is regulated by specific areas of the atria that either maintain an electrical feedback loop or spontaneously emit electrical signals. Electrical feedback loops in the heart are commonly referred to as "reentry mechanisms."

[0004] During AT, electrical signals disrupt the normal rhythm of atrial contraction. This can affect the patient's health and even threaten the patient's life. Symptoms of AT include chest palpitations, fainting, dizziness, sweating, chest pain, shortness of breath, fatigue, weakness, and even heart failure. Therefore, physicians aim to treat the patient's AT event to restore normal atrial contraction. Electrophysiologists (EPs) are typically physicians who treat AT, especially those involved in surgical procedures. However, other physicians can also treat AT. Thus, the use of the term EP in this disclosure can refer to both electrophysiologists and other physicians.

[0005] One of the procedures used to treat AT is catheter ablation of the area that is the source of the abnormal electrical signals. In a typical AT ablation procedure, the EP sedates the patient and inserts two catheters into the patient's heart through the patient's veins. An incision is made to access the patient's vein, typically in the upper part of the patient's right leg. One of the catheters is a coronary sinus catheter, which is a reference catheter that is placed in a vein inside the coronary sinus. This vein is important because it travels around the left atrium and therefore can be a good reference for recording the patient's atrial electrical activity. The second catheter is a catheter that the EP moves within the atrium during the ablation procedure to investigate the electrical signals. During the procedure, the EP attempts to locate the area involved in causing the AT. Such areas may be either areas that contain foci where pathological atrial activation spreads centrifugally, or circuits of reentry mechanisms. If the patient is not experiencing AT, the EP attempts to trigger AT via electrical stimulation. Once AT is triggered, the EP attempts to investigate the electrical signals to detect the area causing the AT.

[0006] The EP can locate the responsible region in at least two ways. First, the EP can use the electrical signal of the heart. According to this method, the EP tries to find the electrical signal in the middle of atrial diastole. Atrial diastole is the time interval between two successive contractions of the atrium. Contraction of the atrium is caused by the propagation of an electrical signal through the atrium. A typical atrial beat lasts about 50 microseconds, and the time between atrial beats is about 600-1000 microseconds. The electrical signal in the middle of atrial diastole is more likely to be associated with AT because it is separated in time from the electrical signal that caused the contraction of the atrium. If the EP finds an active region in the middle of the time period between atrial contractions, this region is more likely to be causing the AT. Software can be used to help the EP locate the middle of atrial diastole by visualizing the electrical signal of the heart. This software can visually show the electrical properties over a period of time. For example, the voltage and current of a region can be shown on the EP over a period of 10 seconds. Because atrial contractions typically produce an electrical signal pattern, this visualization can help the EP locate the area causing the AT. The display available to the EP to view the patient's heart's electrical signals can be transmitted from an EP recording system, which is typically a mobile cart or workstation with the necessary hardware connections to receive the signals during the ablation procedure.

[0007] Second, the EP can use a 3D mapping system that builds a visual map of the electrical signals propagating throughout the heart. The EP creates this 3D map by taking measurements of the heart with a catheter. The amount of measurements needed varies depending on the 3D mapping system, the desired accuracy, and the time the EP has available to perform the ablation procedure. If this 3D map is successfully created, the EP can then use the 3D map to analyze the electrical signals in order to identify areas that are propagating AT.

[0008] Once the EP identifies an area propagating AT, it ablates the area or performs a circuit disruption in the case of the reentry mechanism of AT with the goal of terminating the AT and re-establishing normal sinus rhythm in the heart. Summary of the Invention [Problem to be solved by the invention]

[0009] However, while the EP is performing the ablation procedure, another region may spontaneously take over the previous region, thus changing the characteristics of the AT. The characteristics of the AT may also change during the procedure due to another region in another part of the atrium taking over the target region while the EP is ablating it. Therefore, it is important for the EP to understand the ongoing profile of the AT to determine whether the previous information about the AT is still relevant or whether the AT has changed such that the EP has full access to information about different AT events. If the AT has changed significantly, the EP may need to reconsider the region to be ablated and reconsider the ablation procedure. This reconsideration traditionally leads to canceling the current visual map and redoing other procedures as the current visual map becomes irrelevant and may even lead to an incorrect diagnosis.

[0010] Therefore, there is a need to provide EPs with ongoing tachycardia analysis and planning suggestions that are automatically generated and updated. In particular, there is a need for systems and methods that automatically provide EPs information about ongoing and changing AT events during a procedure and allow EPs to easily determine whether AT discovered in new areas is related to previous AT, with minimal interaction from the EP or other medical professionals. Conventional methods and systems are generally based on triggers and thresholds that must be manually entered, such as the pulsatile acceptance criteria in Boston Scientific's RHYTHMIA acquisition system, and are therefore not easily adaptable to EPs, especially during surgery. [Means for solving the problem]

[0011] The present invention aims to improve the situation. To this end, the present invention provides systems and methods related to tracking, analyzing, and displaying information related to AT events in a patient. In particular, the systems and methods are directed to receiving information related to a recording of the cardiac electrical activity, analyzing the electrical activity recording by comparing the electrical activity with a determined plan for the current AT event, and generating an alert if a change in the AT is detected. Changes in the characteristics of the AT can be identified by changes in the electrical signal measured by a device such as an electrocardiogram (ECG) lead or a coronary sinus catheter lead. Such changes indicate that the mechanism generating the electrical signal is different and is probably not located in the same region.

[0012] The AT plan is built in real time based on AT electrical signal measurements from various leads and includes several different features. The various leads can include, for example, a 12-lead ECG and / or a coronary sinus catheter lead. The features can include, for example, delays between activations in different leads, electrical activation profiles in different leads, electrical activation profiles corresponding to the atria in each lead, and cycle length. Cycle length is the time between two waves resulting from the contraction of the atria. The plan is updated periodically by a training function. No prior data is required as all necessary data is captured during the procedure through automated signal analysis.

[0013] The systems and methods disclosed herein automatically create an evolving ongoing AT plan (or AT profile) so that subsequent changes in AT are more accurately recognized and classified. The systems and methods can use any or all data from various leads. However, typically not all data collected during a procedure is used when updating or evaluating a plan. Instead, only the most recently collected data may be used, as it is closest in time to the patient's current condition.

[0014] Various subsets of data can also be weighted more or less heavily. For example, data can be weighted such that more recently acquired data is given more weight than older data. In this way, the system and method can recognize changes in the AT profile and create an evolving plan. The system and method can then react quickly when something unusual occurs, which can be used as an alert for the EP that something is fundamentally different from the mechanism behind the current AT. Since AT can change in real time during the procedure, it is important for the EP to understand that when the AT profile changes during the ablation procedure, previous data is no longer relevant. The EP can also use recorded information about the patient's previous AT profile in the current or subsequent procedure to determine whether the patient is experiencing the previously measured AT.

[0015] The system and method includes and uses a computer system configured to store cardiac electrical activity data and a processor configured to implement a program to perform data analysis related to the received cardiac electrical activity data. As part of the data analysis, the computer system implements a data-driven, continuous, and adaptive learning approach to analyze the patient's AT. The system and method can generate an initial draft plan for AT without human intervention and / or manual initial configuration of an AT profile. Additionally, the system and method is adapted to track AT events and provide an evolving AT profile without the need for human intervention. The system and method can acquire various input data from the patient, such as data from an ECG or a coronary sinus catheter. For example, a conventional 12-lead ECG can be used in which ten electrodes are placed at different locations on the patient's body such that the magnitude of the cardiac electrical potential is measured from twelve different angles over the data acquisition period. A coronary sinus catheter can also be used to sense the cardiac electrical activity. The acquired data can then be processed, analyzed, and used to provide alerts to the EP. For example, to develop a proposed AT schedule and compare received electrical activity to the proposed schedule, the systems and methods may implement an idle function, a training function, and a detection function.

[0016] The present systems and methods allow EPs to more easily recognize when the AT changes during a procedure and when the AT matches previously recorded AT. Thus, the systems and methods provide lower human mortality rates, both intraoperative and extraoperative, easier and faster analysis of AT by EPs or other medical professionals, the ability to identify more complex AT configurations, increased success rates of medical procedures, better quality of life for patients, and lower medical costs.

[0017] Accordingly, the present invention is directed to a method for analyzing an atrial tachycardia signal by a computer, comprising the following operations: a) collecting a dataset representing a recording of electrical activity of a human heart from at least one input source; b) generating a current atrial tachycardia profile using said data set; c) collecting current data points of an atrial tachycardia signal from at least one input source; d) analyzing the current data point using the current atrial tachycardia profile to determine if the current data point is an outlier with respect to the current atrial tachycardia profile.

[0018] In various embodiments, the method may include one or more of the following features. - step b) comprises the steps of extracting selected features in the group comprising cycle length features, electrode activation sequence features and / or activation sequence morphological features, and training a feature-based anomaly detection algorithm using at least one and / or a combination of said extracted features, - step b) comprises augmenting the extracted features data before training the feature-based anomaly detection algorithm; - step d) comprises the steps of extracting features from the current data that are used to train a one class support vector machine (SVM), feeding the extracted features to the feature-based anomaly detection algorithm, and receiving in return a value indicative of whether the current data point is an outlier with respect to the current atrial tachycardia profile, The feature-based anomaly detection algorithm is a one-class support vector machine. - e) upon determining that the current data point is an outlier, issuing an outlier alert and repeating steps c) and d) using the current atrial tachycardia profile; - the method further includes the steps of: f) analyzing the current data point with the current atrial tachycardia profile to determine whether the current data point is a sustained outlier; and g) upon determining that the current data point is a sustained outlier, issuing a change alert indicating a change in the current atrial tachycardia profile; - the method further comprises the steps of: h) collecting a new set of data representative of a record of electrical activity of the human heart from at least one input source; i) creating a new current atrial tachycardia profile using said second set of data; and repeating steps c) and d) using said new current atrial tachycardia profile; - when operation d) determines that the current data point is not an outlier, step d) includes the step of: dl) adding the current data point to a memory buffer; - step d) further includes the steps of: d2) determining that a memory buffer threshold for data has been reached; and j) creating a new current atrial tachycardia profile using the data stored in the memory buffer and repeating operations c) and d) with said new current atrial tachycardia profile; and d3) determining that a memory buffer threshold for data has not been reached and repeating operations c) and d) with said current atrial tachycardia profile.

[0019] The invention also relates to a computer program comprising instructions for carrying out the method according to the invention, a data storage medium on which this computer program is recorded, and a computer system comprising a processor coupled to a memory on which this computer program is recorded.

[0020] Other features and advantages of the present invention will become readily apparent in the following description of the drawings illustrating illustrative embodiments of the invention. [Brief description of the drawings]

[0021] [Figure 1]FIG. 1 is an overall diagram showing a system implementing an AT analysis module according to the present invention. [Diagram 2] FIG. 2 is an overall diagram showing the functions performed by the AT analysis module of FIG. [Diagram 3] FIG. 3 shows an exemplary diagram of the operations performed in the detection function of FIG.

[0022] The drawings and the following description constitute most of the practical and clearly defined features, and as a result, they are not only useful for understanding the invention, but can also be used to contribute to its definition, if the need arises.

[0023] The embodiments described herein are directed to analyzing AT in a patient using input data related to the electrical activity of the patient's heart (hereinafter cardiac electrical activity data). The systems and methods implement a data-driven and continuous learning approach whereby received cardiac electrical activity data is processed and compared to a proposed AT to determine if an unexpected change in AT is occurring or has occurred.

[0024] First, the received cardiac electrical activity data may be used to construct a profile or proposed plan of the patient's cardiac electrical activity, including information about the existing AT. Based on this, a profile of the existing AT and a proposed plan may be generated. Complex AT configurations may be identified using techniques such as density estimation and data augmentation. The distribution of certain features of the measured cardiac electrical activity data signal is used in density estimation techniques to define a proposed plan feature value range.

[0025] For example, cycle length (which is the time interval between two electrical activations measured across two leads) can be used. If values ​​received at 200, 199, 201, and 202 ms are measured for cycle length, there may be a dense region of about 200 ms, with a lower limit of about 199 ms and an upper limit of about 202 ms. Anything outside this range may be considered in the non-dense region. This value range estimation process can be performed over several sets of features combined together.

[0026] An example of an algorithm that can be used to detect outliers in a data point cloud is the One-Class SVM (Support Vector Machine). One-Class SVMs are trained to learn a coarse, close boundary that defines the contours of the initial observation distribution. Any further observations that fall within this frontier are then considered to come from the same population as the first observation. Conversely, if they fall outside the frontier, they are considered anomalous with a level of confidence that can be derived from the training. For instance, in the above example, a cycle length of 250 ms falls outside the range of 199-202 ms.

[0027] Data augmentation techniques can be used to modify the input data to achieve a more robust and stable behavior of outlier detection. Examples of data augmentation techniques to improve AT detection include adding expected noise or tolerance to the measured data so that changes within this tolerance do not indicate a change in AT, increasing the statistical significance of data points within a specified range, and classifying data points that meet certain criteria as outliers so that they do not indicate a change in AT. Alternatively, it is possible to replace the SVM with another frontier estimation technique. For example, such a frontier for cycle length data points can be based on a tolerance of ±1 millisecond.

[0028] Using these and other techniques, outliers can be determined by comparing a data point to previous data points. In other words, it can be determined whether a data point is within the data point cloud of the previous data point. For example, a period length measurement of 203 milliseconds would not be detected as an outlier if the expected range was 199 to 202 milliseconds.

[0029] During the ablation procedure, cardiac electrical activity data is received and processed for comparison to the current proposed plan, so that cardiac electrical activity data that is outside of expected ranges can be identified.

[0030] 1 illustrates a system for implementing an AT analysis module according to the present invention during an ablation procedure. In a typical procedure, a patient is on an operating table 100. Various leads and sensors, such as ECG leads and electrodes and coronary sinus catheters, may be inserted or attached to the patient while the ablation procedure or other procedure is being performed.

[0031] As mentioned above, during an ablation procedure, catheter leads are typically inserted into the patient's heart and ECG leads are typically attached to the patient. The resulting catheter signal 110 and ECG signal 120 are transmitted from the operating table 100 to an EP recording system 140. A number of other signals 130 may also be transmitted from the operating table 100 to the EP recording system 140, such as signals containing data regarding blood pressure, temperature, and blood oxygen levels. The signals from the operating table 100 to the EP recording system 140 may be transmitted via a wired or wireless connection. The EP recording system 140 may be connected to an EP workstation 160 via a wired or wireless connection. The EP workstation may be in the same room or area as the EP recording system or may be remote from the EP recording system.

[0032] The EP recording system 140 may be a mobile cart or workstation with the necessary hardware connections to receive signals during the ablation procedure. The EP recording system may include a computer 141, an amplification system 142, one or more displays 143, a control system 144, and other systems 145 related to monitoring the patient, guiding the EP during the procedure, or providing information to or receiving information from the EP.

[0033] Signals received by the EP recording system 140 may be amplified using an amplification system 142 before being processed by the computer 141 for analysis. The computer 141 may include one computer or multiple computers and may include a processor, memory, a communication interface, and a user input interface. For example, the ECG signal 120 may be sent directly to a display 143 connected to a computer for specifically analyzing the ECG signal, while another computer may analyze the remainder of the incoming signal.

[0034] After analyzing the input signals, computer 141 can display information on the EP recording system and can transmit information to one or more displays 143 and / or EP workstation 160 where it can be displayed on one or more displays at EP workstation 160. Using the EP recording system's control system 144, an operator can control the EP recording system's computer 141, amplification system 142, and display 143. The operator can also use control system 144 to control computer 161 and various displays at EP workstation 160 or other systems related to the procedure.

[0035] For example, the operator may change the display period of the ECG signal from 10 seconds to 5 seconds on one of the EP workstation's ECG display 162 or the EP recording system's displays 143. Various other systems 145 may also be implemented in the EP recording system 140 to assist the operator and / or EP in performing the procedure. For example, a hospital paging system may be connected to the EP recording system 140.

[0036] The EP workstation 160 is where the EP performs the ablation procedure. The EP workstation 160 receives information from the EP recording system 140 and can have a variety of displays depending on the EP's preferences. The transmission of information 150 between the EP recording system 140 and the EP workstation 160 can be via a wired or wireless connection, or a combination thereof. For example, the EP can use an ECG display 162 and / or a 3D map display 163 to help locate the area to be ablated.

[0037] Various other displays 165 may also be used to assist the EP. For example, the patient's chart may be available on the display for the EP to refer to and take notes on during the procedure.

[0038] The computer 161 of the EP workstation can receive signals processed or unprocessed by the computer 141 in the EP recording system. The computer 161 further implements an AT analysis module according to the present invention and outputs results from the module to an AT analysis module display 164. However, the computer 141 in the EP recording system can also be used to implement the AT analysis module. The AT analysis module can be a program module contained in a computer readable memory and then executed by a computer processor. The corresponding processing can be performed using a set of instructions implemented in a programmable computer, which can include one or more non-transitory tangible computer readable storage media. For example, the programmable computer can include magnetic or optical storage media, solid-state electronic storage devices such as random access memory (RAM) or read-only memory (ROM), or any other physical device or medium used to store instructions for performing the desired processing. The AT analysis module can include a memory for storing input data received from the patient or can communicate with a memory containing such data. Furthermore, several computers can be used to implement the AT analysis module, e.g., a remote cloud system, as long as the output is made available to the EP in real time or near real time.

[0039] The systems of computers 141 and 161 may each include at least one processor, memory, at least one storage device, and input / output devices. Some or all of the components may be interconnected via a system bus. The processor may be single-threaded or multi-threaded and may use one or more cores. The processor may execute instructions, such as instructions stored in the memory and / or storage device. Information may be received and output using one or more I / O devices, including wired and wireless connections to other computer systems.

[0040] The above description is exemplary. Various other configurations are possible. For example, EP recording system 140 and EP workstation 160 may be combined, or various functions of EP recording system 140 and EP workstation 160 may be performed remotely in a cloud-based system.

[0041] The system and method may include a computer system configured to receive cardiac electrical activity data. This data may be obtained, for example, from a 12-lead ECG, a coronary sinus catheter, or other device that monitors electrical activity in the patient's heart. The recordings may be acquired by a computer used in the ablation procedure and then transmitted to a computer that includes a program module for performing the processing. The program module generates all the functions from the recordings and implements the various phases described below.

[0042] From the analysis of the recordings, different features of the electrical activity of the patient's heart can be determined that are characteristic and stable under tachycardia. Features are considered stable under tachycardia when they do not change to a specified tolerance during AT with the same conditions. Such features are stable because the same areas of the heart generate the electrical waves involved in the activity in the atria. The same electrical waves will have the same characteristics measured and also the same characteristics calculated based on the measured characteristics. These features may include, for example, cardiac cycle length, p-wave morphology, activation profile, coronary sinus activation sequence, other measurable data from ECG or measuring devices, and a number of features calculated from the measurable data. These features can be measured, for example, by ECG leads or coronary sinus catheter leads.

[0043] By analyzing the input data and comparing it to the AT profile, the system and method determine whether the received data matches the profile created from the current AT or if it is an outlier. The system can provide an output indication as to whether the received data matches the AT profile or if it is an outlier. This indication may be provided in real time and may be visual, audible, tactile, or any other method of notifying or alerting the EP.

[0044] For example, a visual indicator may be a window on a computer display that turns red when it would normally be white text on a black background; an auditory indicator may be a beep or click; a tactile indicator may be a vibration.

[0045] If several outliers are detected in a row, the system can send an alert that there is a change in the AT profile. This alert can be used by EPs or other medical professionals during surgery or other medical procedures to inform decisions about how to best treat the patient.

[0046] Outliers can be detected in a number of ways to filter out data points that are not consistent with previous data. One way to determine outliers is to use a data point cloud where the distance to the center of the data point cloud is measured. If the data point is within the cloud, it is not an outlier. If it is outside the cloud, it is an outlier. For example, the data point cloud may be centered at a period length of 150 milliseconds. If a data point at 300 milliseconds is measured, it may be considered an outlier, but a data point at 160 milliseconds may not. Other methods of detecting outliers may also be used.

[0047] In certain embodiments, the cardiac electrical activity data is used by an AT analysis module. The AT analysis module may be implemented as a software program executed by a computer having the received cardiac electrical activity data stored in a local memory, or may be configured to access cardiac electrical activity data stored on a remote computer or server. As shown in FIG. 2, the AT analysis module performs an idle function that collects and stores data, a training function that uses the received cardiac electrical activity data to generate density estimates that provide an AT profile, and a detection function that compares new received data to the AT profile to determine whether the new received data is an outlier.

[0048] The idle function (idle phase) 200 of the AT analysis module collects and stores input data representative of various characteristics of the patient's heart, such as electrical activity data from an ECG or catheter probe. The EP may or may not be performing an ablation procedure during the idle phase, depending on their preference. The idle function 200 typically runs for 10-15 seconds. Once a sufficient amount of input data has been collected, the AT analysis module has enough input data to create an AT profile. In alternative other embodiments, the idle function 200 may collect input data for a defined period of time until a predetermined number of data points have been collected or until the AT analysis module determines that the profile being generated is of sufficient quality. An example of a predetermined number of data points may be a specified number of p-waves. The operator may also manually instruct the analysis module when to start and stop collecting data. The operator may also manually command a separate module to collect data and automatically start the analysis module upon the occurrence of a specific condition, such as a measurement indicative of a potential AT switch. The operator may further configure a memory buffer to hold data for a period of time and execute a training function after the specified period of time. For example, an operator may configure the memory buffer to hold 3 minutes of data and then execute the training function after 20 seconds. After execution of the idle function 200, the AT analysis module executes the training function 210.

[0049] Execution of the training function (training phase) 210 may be triggered after either the idle function 200 or the detection function (detection phase) 220. In the training function 210, the AT analysis module uses the latest cardiac electrical activity data available to build a current draft plan. For example, a one-class SVM may be used to receive several types of features defined for the AT profile. Such features may be based on at least cycle length features, electrode activation sequence features (which are sequences of electrodes that are activated consecutively), and / or morphological features of the activation sequence. Before training the one-class SVM, the AT analysis module may augment data on the profile features as described above, so that a wider or narrower range of values ​​in the data point cloud may be considered within acceptable boundaries. A wider range of values ​​will result in a wider range in which the data points are non-outliers, while a narrower range of values ​​will result in a narrower range in which the data points are non-outliers. In contrast to some machine learning techniques, training of the one-class SVM is extremely fast and near real-time. This means that training the one-class SVM to adapt to changing conditions in the AT is not detrimental to the implementation of the present invention. Conversely, this feature is an advantage of one-class SVM, which is particularly well suited to adapt to potentially rapidly changing environments.

[0050] Various features may also be combined such that a data point may be considered an outlier even if its feature values ​​are individually within their respective expected ranges.

[0051] Combining features together makes it possible to detect cases where global changes matter despite the fact that changes in individual features appear regular. A one-class SVM is trained using the most recent input data available, i.e. all the data in the memory buffer, or a subset of it.

[0052] If the training function 210 is initiated after the execution of the idle function 200, the input data collected by the idle function 200 is used to create an AT profile. If the training function 210 is initiated after the detection function 220, the input data from the memory buffer is used to create a new AT profile, as described below.

[0053] In the training function 210, the AT analysis module uses data augmentation techniques to add variability to the data so that slight offsets from the expected profile do not trigger outlier alerts when the AT profile is used in the detection function 220. For example, the data may be augmented by adding a noise tolerance level to the data. After running the training function 210, the AT analysis module runs the detection function 220.

[0054] In the detection function 220, the AT analysis module uses the AT profile previously constructed by the training function 210 to detect whether the received input data are outliers and whether those outliers indicate a change in the current AT profile.

[0055] 3 illustrates an exemplary embodiment of the detection function. Input data is received in operation 300 and compared to the current AT profile in operation 310. Operation 310 determines whether the input data of operation 300 is an outlier, and the result is tested in operation 320.

[0056] As described above, density estimation techniques are used to determine whether a data point is an outlier. For example, a data point cloud can be used to determine whether a data point is an outlier. In other words, a data point is an outlier if it is outside the cloud, and is not an outlier if it is inside the cloud. When a one-class SVM is used to determine whether a data point is an outlier, the input data is first transformed to extract features used to train the one-class SVM, and these features are then fed to the one-class SVM, which returns a value indicating whether the data point is an outlier.

[0057] If the input data 300 is an outlier, the AT analysis module proceeds to operation 330 to determine whether the input data 300 is also a persistent outlier. The input data is a persistent outlier if a defined threshold is met. This persistent outlier threshold may be set manually by an EP or other medical professional, or may be set automatically by the AT analysis module. This persistent outlier threshold may be defined as a number of consecutive outliers, a percentage of outliers in a sample of the input data, a defined standard deviation away from the expected value, or any combination thereof. In a preferred embodiment, the number of consecutive outliers is used.

[0058] If the input data 300 is an outlier but not a persistent outlier, the AT analysis module outputs a warning that there was an outlier in operation 340 and waits for more input data.

[0059] If the input data 300 is both an outlier and a persistent outlier, operation 230 of Figures 2 and 3 is performed, which causes the AT analysis module to output a warning that there has been a switch in the AT profile, and the idle function 200 is executed anew with a cleared buffer. An indication that there has been a change in the AT profile, such as a visual or auditory signal, is also provided to the EP, which may cause the EP to change strategy or restart the procedure. For example, the EP may decide to create a new 3D map.

[0060] If operation 320 determines that the input data 300 is not an outlier, the input data 300 is sent to a memory buffer for storage in operation 350. Then, in operation 360, the AT analysis module determines whether a threshold amount of input data in the memory buffer has been reached. This memory buffer threshold may be set manually by an EP or other medical professional, or automatically by the AT analysis module. This memory buffer threshold may also be set according to the amount of input data, the duration of collection, or any combination thereof. The memory buffer limit itself may also be used to limit the amount of data stored, such that new incoming data replaces the oldest data stored in the memory buffer.

[0061] If the memory buffer threshold has been reached, the AT analysis module performs a new execution of training function 210, and the AT analysis module uses the input data in the memory buffer to update the current AT profile, as shown in operation 240 of Figures 2 and 3. If the memory buffer threshold has not been reached, the AT analysis module waits for more received input data in operation 370.

[0062] As is apparent from the above, the combination of the idle function 200, training function 210, and detection function 220 ensures that the AT analysis module is constantly analyzing the input data and continually alerts the EP when the AT profile changes and updates it under certain conditions.

[0063] One-class SVM provides the best results in our testing, but in alternative embodiments it can be replaced by another feature-based anomaly detection algorithm, such as: - Isolation forest: Liu, Fei Tony, Ting, Kai Ming, Zhou, Zhi-Hua. “Isolation forest” Data mining, 2008, ICDM'08. 8th IEEE International Conference on. - Localized outlier factors: Breunig, Kriegel, Ng, and Sander (2000) LOF: Identifying density based on localized outliers: Proc. ACM SIGMO. -Robust covariance:: Estimating support for high-dimensional distributions: Scholkopf, Bernhard, et al. Neural Comput 13.7(2001):1443-1471. -k‐nearest neighbor https: / / link.springer.eom / article / 10.1007 / s00778005006. - DBSCAN: Tran Manh Thang; untae Kim:. “Anomaly Detection Using DBSCAN Clustering with Multiple Parameters”, 2011 International Conference on Information Science and Applications. -CBLOF: He, Z.; Xu, X.; Deng,S. (2003). “Cluster-based local outlier detection”: Letters in Pattern Recognition. - Hierarchical Density-Based Cluster Analysis: Campello, RJGB; Moulavi, D.; Zime, A ; Sander, J. (2015): “Hierarchical Density Estimation for Data Clustering, Visualization, and Outlier Detection”: ACM Transactions on Knowledge Discovery from Data. - Gaussian mixtures: W. Liu, D. Cui, Z. Peng and J. Zhong: “Outlier detection algorithm based on mixture models”:2019 IEEE International Conference on Power, Intelligent Computing and Systems (ICPICS), 2019, pp.488-492, doi:10.1109 / ICPIC S47731.2019.8942474, or -Hidden Markov Models: A. Sultana, A. Hamou-Lhadi and M. Cputure: “Improved Hidden Markov Models for Anomaly Detection Using Frequent Common Patterns”: 2012 IEEE International Conference on Communications (ICC), 2012, pp.1113-1117, doi: 10.1109 / ICC.2012.6364527.

Claims

1. A method for analyzing atrial tachycardia signals by a computer, comprising the following operation steps: a) Collecting a dataset representing the electrical activity of a human heart from at least one input source; b) Creating a current atrial tachycardia profile using the dataset; c) Collecting current data points of the atrial tachycardia signal from at least one input source; d) Analyzing the current data points using the current atrial tachycardia profile and determining whether the current data points are outliers with respect to the current atrial tachycardia profile.

2. The method for analyzing atrial tachycardia signals according to claim 1, wherein step b) comprises extracting features selected from a group including features of cycle length, features of electrode activation sequences, and / or morphological features of activation sequences, and training a feature-based anomaly detection algorithm using at least one and / or a combination of the extracted features.

3. The method for analyzing atrial tachycardia signals according to claim 2, wherein step b) comprises enhancing the data of the extracted features before training the feature-based anomaly detection algorithm.

4. The method for analyzing atrial tachycardia signals according to claim 2 or 3, wherein step d) comprises extracting features used for training a one-class support vector machine from the current data points, supplying the extracted features to the feature-based anomaly detection algorithm, and receiving as a return a value indicating whether the current data points are outliers with respect to the current atrial tachycardia profile.

5. The method for analyzing atrial tachycardia signals according to claim 2 or 3, wherein the feature-based anomaly detection algorithm is a one-class support vector machine.

6. The method for analyzing atrial tachycardia signals according to claim 1, further comprising: e) when it is determined that the current data points are outliers, issuing an outlier warning and repeating steps c) and d) using the current atrial tachycardia profile.

7. f) Analyzing the current data points using the current atrial tachycardia profile and determining whether the current data points are persistent outliers; g) When it is determined that the current data point is a persistent outlier, issue a change warning indicating a change in the current atrial rate profile; The method for analyzing an atrial rate signal according to claim 1 or 2, further comprising. **Claim 8** h) newly collecting a second data set representing a record of the electrical activity of a human heart from at least one input source; i) creating a new current atrial rate profile using the second data set, and repeating steps c) and d) using the new current atrial rate profile; The method for analyzing an atrial rate signal according to claim 3, further comprising. **Claim 9** When it is determined in step d) that the current data point is not an outlier, dl) The method for analyzing an atrial rate signal according to claim 1, further comprising the step of adding the current data point to a memory buffer. **Claim 10** Step d) further comprises d2) determining that a memory buffer threshold for data has been reached, j) creating a new current atrial rate profile using the data stored in the memory buffer, and repeating steps c) and d) using the new current atrial rate profile. The method for analyzing an atrial rate signal according to claim 5, comprising the step of **Claim 11** Step d) further comprises d3) determining that a memory buffer threshold for data has not been reached, and repeating steps c) and d) using the current atrial rate profile. The method for analyzing an atrial rate signal according to claim 5, comprising the step of **Claim 12** A computer program comprising instructions for performing the method according to claim 1. **Claim 13** A data storage medium having recorded thereon the computer program according to claim 12. **Claim 14** A computer system including a processor coupled to a memory (4), wherein the computer program according to claim 12 is recorded in the memory (4).

Images