Description:[001] FIELD OF THE INVENTION
[002] The present invention relates to the cardiac monitoring of patients, and particularly, to a method and system for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient.

[003] BACKGROUND OF THE INVENTION
[004] Background description includes information that may be useful in understanding the present invention
[005] An electrocardiogram (ECG) is widely recognized as the standard parameter for diagnosing potential cardiovascular disease (CVD) in individuals. Traditionally, cardiologists utilize a 12-lead ECG to obtain essential cardiac data for the detection and diagnosis of CVD and its onset. However, the necessity of attaching multiple leads to the body for extended periods poses significant inconveniences.
[006] While clinical protocols require the recording of ECG signals from all 12 leads, most wearable ECG devices currently available on the market capture only a limited number of leads. In typical remote health monitoring scenarios, data is transmitted from the patient to either a physician or family members. Recording all 12 leads at the transmission point necessitates significant memory, bandwidth, and transmission power to relay the signals. To address these challenges, various signal compression techniques can be employed prior to wireless transmission to reduce data size, followed by decompression at the receiver’s end to reconstruct the original signal.
[007] However, research indicates that as the number of channels increases, the compression ratio and signal effectiveness decline. Consequently, despite the availability of advanced compression methods, the use of multichannel S12 techniques for telemonitoring applications is generally discouraged. Reduced lead techniques, such as the EASI configuration with three leads, which is similar to the Frank Vectorcardiography (FV) technique, require additional training for caregivers and may lack comprehensive diagnostic information. The S12 technique, which is the clinical standard, helps mitigate these challenges in remote health monitoring applications. Thus, there is a desired need for a technique that utilizes a minimal number of ECG leads while still providing complete 12-lead ECG data.

[008] OBJECTS OF THE INVENTION:
[009] Some of the objects of the present disclosure, which at least one embodiment herein satisfies, are listed herein below.
[0010] The primary objective of the present invention is to reconstruct the standard 12 leads ECG, from a single ECG lead by using seven distinct machine learning models corresponding to seven independent leads (I, V1, V2, V3, V4, V5, and V6) from ECG lead II.
[0011] These and other objects and advantages will become more apparent when reference is made to the following description and accompanying drawings.

[0012] SUMMARY OF THE INVENTION
[0013] This summary is provided to introduce concepts related to reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient. This summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter.
[0014] In an aspect of the present invention, a system for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient is described. The system includes a receiving module, at least one database, and a plurality of distinct machine learning models. The receiving module is configured to receive the new ECG data of the single ECG lead. The at least one database includes a first set of plurality of ECG leads’ data associated with a first group of patients, and a second set of plurality of ECG leads’ data associated with a second group of patients. The subject patient relates to the second group of patients. The plurality of distinct machine learning models correspond to a subset of a plurality of ECG leads, pre-trained with the first set of plurality of ECG leads’ data. The plurality of distinct machine learning models are configured to load the second set of plurality of ECG leads’ data, to additionally train, based on a transfer learning combined with a continual learning technique, the pre-trained plurality of distinct machine learning models.
[0015] Each of the plurality of distinct machine learning models includes at least one Convolutional Neural Networks (CNN) layer followed by at least one Max-Pooling layers after each CNN layer, at least one Long Short-Term Memory (LSTM) layer, at least one Flatten layer, and at least one Dense layer. Each of the pre-trained plurality of distinct machine learning models is additionally trained by re-training the LSTM layers of the pre-trained plurality of distinct machine learning models with the second set of plurality of ECG leads’ data specific to the subject patient. The additionally trained plurality of distinct machine learning models reconstruct the new plurality of ECG leads’ data associated with the subject patient from the new ECG data of the single ECG lead.
[0016] In an embodiment of the present invention, the plurality of ECG lead data is 12 ECG lead data.
[0017] In another embodiment of the present invention, the subset of plurality of ECG leads’ data is 7 ECG leads’ data.
[0018] In another embodiment of the present invention, the single ECG lead is an ECG lead II.
[0019] In another embodiment of the present invention, the subset of the plurality of ECG leads are ECG leads I, V1, V2, V3, V4, V5, and V6.
[0020] In another embodiment of the present invention, the remaining ECG leads’ data are reconstructed from the ECG leads I, II, V1, V2, V3, V4, V5, and V6 data.
[0021] In another embodiment of the present invention, each of the plurality of distinct machine learning model comprises 3 CNN layers, 3 max-pooling layers, one LSTM layer, one Flatten layer and one Dense layer.
[0022] In another embodiment of the present invention, the first set of the plurality of ECG lead data is a large sized and reusable plurality of ECG leads’ data.
[0023] In another embodiment of the present invention, the pre-trained plurality of distinct machine learning models are configured to reconstruct the plurality of ECG data, based on the transfer learning, associated with a subject patient of the first group of patients.
[0024] In another embodiment of the present invention, the second set of the plurality of ECG lead data comprises a small set of plurality of ECG leads’ data associated with the subject patient, recorded for a minimal time, in order to additionally train the pre-trained model to reconstruct a plurality of ECG data for the subject patient.
[0025] In another embodiment of the present invention, the plurality of distinct machine learning model are trained to reconstruct 1 second of other ECG lead data from 1 second of the single ECG lead data.
[0026] In another embodiment of the present invention, the LSTM layers of the additionally trained plurality of distinct machine learning model comprise the LSTM layers associated with the subject patient only without the pre-trained model’s LSTM layer, and subsequently, the LSTM layers associated with the subject patient is retained in the at least one database.
[0027] In another embodiment of the present invention, a new set of LSTM layers is added to the at least one database after the training of the models for newly added subject patient.
[0028] In another embodiment of the present invention, the LSTM layer is loaded from the at least one database according to the subject patient group.
[0029] In another embodiment of the present invention, the second set of plurality of ECG leads’ data loaded in the at least one LSTM layers is comparatively low.
[0030] In another embodiment of the present invention, the second set of plurality of ECG leads’ data comprises only 49.4 thousand parameters.
[0031] In another aspect of the present invention, a method of reconstruction of a plurality of electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient is described. The method includes the step of receiving the new ECG data of the single ECG lead. The step of receiving is performed by a receiving module. The method further includes the step of storing a first set of plurality of ECG leads’ data associated with a first group of patients, and a second set of plurality of ECG lead data associated with a second group of patients, in at least one database. The subject patient relates to the second group of patients. The method further includes the step of loading the plurality of ECG leads’ data associated with the subject patient, from the at least one database, to a pre-trained plurality of distinct machine learning models, to additionally train, based on a transfer learning combined with a continual learning techniques, the pre-trained plurality of distinct machine learning models specific to the subject patient. The plurality of distinct machine learning models correspond to a subset of the plurality of ECG leads and are pre-trained with the first set of plurality of ECG leads’ data.
[0032] Each of the plurality of distinct machine learning models includes at least one Convolutional Neural Networks (CNN) layer followed by at least one max-pooling layers after each CNN layer, the at least one Long Short-Term Memory (LSTM) layer, at least one Flatten layer, and at least one Dense layer. The additional training of the pre-trained plurality of distinct machine learning models includes the sub-steps of re-training the LSTM layers of the pre-trained plurality of distinct machine learning models with the second set of plurality of ECG leads’ data specific to the subject patient. The additionally trained plurality of distinct machine learning models reconstruct the plurality of ECG leads’ data associated with the subject patient from the new ECG data from the single ECG lead.
[0033] Various objects, features, aspects, and advantages of the inventive subject matter will become more apparent from the following detailed description of preferred embodiments, along with the accompanying drawing figures in which like numerals represent like components.

[0034] BRIEF DESCRIPTION OF DRAWINGS:
[0035] The illustrated embodiments of the subject matter will be understood by reference to the drawings, wherein like parts are designated by like numerals throughout. The following description is intended only by way of example and simply illustrates certain selected embodiments of devices, apparatus, systems, and methods that are consistent with the subject matter as claimed herein, wherein:
[0036] FIG. 1 illustrates a block diagram depicting a system configured for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient, in accordance with an exemplary embodiment of the present disclosure;
[0037] FIG. 2 illustrates a schematic block diagram of the system of Fig. 1, in accordance with an exemplary embodiment of the present disclosure;
[0038] FIG. 3 illustrates a schematic block diagram of a machine learning models of the system of Fig. 2, in accordance with an exemplary embodiment of the present disclosure;
[0039] FIG. 4 illustrates an architectural diagram depicting a system configured for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient, in accordance with an exemplary embodiment of the present disclosure;
[0040] FIG. 5 illustrates a schematic block diagram depicting a method for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient, in accordance with an exemplary embodiment of the present disclosure; and
[0041] FIG. 6 illustrates tabular performance results for 100 patients considered for pre-training, in accordance with an exemplary embodiment of the present disclosure.
[0042] The figures depict embodiments of the present subject matter for the purposes of illustration only. A person skilled in the art will easily recognize from the following description that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the disclosure described herein.

[0043] DESCRIPTION OF THE INVENTION:
[0044] The following is a detailed description of embodiments of the disclosure depicted in the accompanying drawings. The embodiments are in such detail as to clearly communicate the disclosure. However, the amount of detail offered is not intended to limit the anticipated variations of embodiments; on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the present disclosure as defined by the appended claims.
[0045] While the embodiments of the disclosure are subject to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the figures and will be described below. It should be understood, however, that it is not intended to limit the disclosure to the particular forms disclosed, but on the contrary, the disclosure is to cover all modifications, equivalents, and alternatives falling within the scope of the disclosure.
[0046] The terms “comprises”, “comprising”, or any other variations thereof used in the disclosure, are intended to cover a non-exclusive inclusion, such that a device, system, or assembly that comprises a list of components does not include only those components but may include other components not expressly listed or inherent to such system, or assembly, or device. In other words, one or more elements in a system or device proceeded by “comprises… a” does not, without more constraints, preclude the existence of other elements or additional elements in the system or device.
[0047] The present invention generates ECG data of 12 leads from a single lead ECG data as an input information. To construct the standard 12 lead ECG, the proposed approach employs seven distinct deep learning models corresponding to seven independent leads (I, V1, V2, V3, V4, V5, and V6). These models work for new input ECG data that has never been seen before and perform robust reconstruction. This approach is based on a type of continuous learning (CL) situation in which a model is trained with input that arrives incrementally without access to past data. Continuous learning, on the other hand, has long been a problem for deep learning models, as the continuous acquisition and training of incrementally available data from non-stationary data distributions usually results in catastrophic forgetting or interference. This constraint is a significant drawback for cutting-edge deep neural network models, which typically learn representations from stationary batches of training data, neglecting circumstances in which information becomes more available over time.
[0048] To address the above continuous learning issue, a new framework is proposed that employs a transfer learning (TL) technique using incrementally accessible data. The model is trained on the initial available data in the proposed framework and called a pre-training step. In this step, the model is trained with the data from a particular set of patients, and the trained model has performed accurate reconstruction for these sets of patients. As the new patients come, each new patient must provide 12 lead ECG data for a minimal time. This process can be regarded as a one-time registration process. Likewise, all the new patients’ 12 lead data will be saved. To perform reconstruction, a transfer learning approach will be applied to the pre-trained model with the newly available data. Transfer learning (TL) is a machine learning (ML) research challenge that focuses on storing knowledge learned while solving one problem and applying it to a different but related problem. In this approach, a model developed for a task is reused as the starting point for a model on a second task. This transfer learning approach is combined with the continual learning approach in the present invention. When the data is available from a new set of patients, a new model will be created for these patients. But here, instead of creating entirely a new model and saving all the weights again for a new set of patients, only a single last layer of weights will be saved. The pretrained model weights are taken as initial weights, and only the LSTM layer will be retrained with the new set of data. After the training process, only these LSTM layer weights will be saved in the database. Likewise, if the new patients are added incrementally, a new set of LSTM layers will be added to the database after the training stage for new patients.
[0049] To reconstruct 12 leads of ECG, all the layer weights will be loaded from the pre-trained model except for LSTM layer weights. These LSTM layer weights will be loaded from the database according to the patient set. For instance, to reconstruct 12 lead ECG for patient group 2, LSTM weights corresponding to patient group 2 will be loaded from the database, and reconstruction will be done accordingly using these weights. On the other hand, in the traditional approach, an entirely new model is created, and all the weights are saved in the database for every new patient group, because of which the memory requirements are very high since we must store all the 8.2 million parameters. In the prenset invention, only LSTM weights corresponding to each patient’s group will be saved in the database. The memory requirements for saving only LSTM weights will be low when compared to the overall model weights since there are only 49.4 thousand parameters to be stored. It should be emphasized that only 0.6% of the entire model weights will be saved in this approach as compared with the traditional approach.
[0050] For better understanding, one or more embodiments of the present invention shall be described with respect to the earlier-mentioned drawings.
[0051] FIG. 1 illustrates a block diagram depicting a system (100) configured for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data (104) from a new ECG data (102) of a single ECG lead positioned over a subject patient, in accordance with an exemplary embodiment of the present disclosure. FIG. 2 illustrates a schematic block diagram of the system (100) of Fig. 1, in accordance with an exemplary embodiment of the present disclosure. FIG. 3 illustrates a schematic block diagram of a machine learning models (206) of the system (100) of Fig. 2, in accordance with an exemplary embodiment of the present disclosure. FIG. 4 illustrates an architectural diagram depicting a system (100) configured for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data (406) from a new ECG data (402) of a single ECG lead positioned over a subject patient, in accordance with an exemplary embodiment of the present disclosure.
[0052] As illustrated, the system (100) includes a receiving module (202), at least one database (204), and a plurality of distinct machine learning models (206). The receiving module (202) is configured to receive the new ECG data (102) of the single ECG lead. The at least one database (204) includes a first set of plurality of ECG leads’ data associated with a first group of patients, and a second set of plurality of ECG leads’ data associated with a second group of patients. The subject patient relates to the second group of patients. The plurality of distinct machine learning models (206) correspond to a subset of a plurality of ECG leads, pre-trained with the first set of plurality of ECG leads’ data. The plurality of distinct machine learning models (206) are configured to load the second set of plurality of ECG leads’ data, to additionally train, based on a transfer learning combined with a continual learning technique, the pre-trained plurality of distinct machine learning models.
[0053] Each of the plurality of distinct machine learning models (206) includes at least one Convolutional Neural Networks (CNN) layer (302, 304, 306) followed by at least one Max-Pooling layers (308, 310, 312) after each CNN layer, at least one Long Short-Term Memory (LSTM) layer (314), at least one Flatten layer (316), and at least one Dense layer (318).
[0054] In one or more embodiments, each 1D CNN layer (302, 304, 306) has 128 learnable filters with a width of 5, which slide across the width of the input signal by five steps at each CNN layer. Further, each CNN layer (302, 304, 306) has 128 feature maps that correspond to 128 filters, with each feature map capturing different low-level features extracted from the input data. Each max pooling layer (308, 310, 312) has a pool length of 2 and a stride of 2 steps which works along the temporal dimension of extracted feature maps. This step reduces the temporal size of each feature map by half while keeping the dimension unchanged. Further, this step ensures in reducing the temporal size of the feature maps and subsequently the computational cost. Following that, one LSTM layer (314) with 64 memory cells was connected to the output of the previous max pooling layer (308, 310, 312). The LSTM layer (314) was chosen because of its ability to capture the signal’s temporal dependency. The LSTM layer (314) generates 64 feature maps, which are subsequently flattened to a single dimension with the help of the flattening layer (316). Finally, to find the final predicted target leads, a fully connected layer with 1000 output neurons are used. The fully connected layer refers to dense layer (318). Since the data was captured at a sampling rate of 1 kHz, 1000 output neurons correspond to one second of data. In addition, between the flattened layer (316) and the fully connected layer (i.e., dense layer (318)), a dropout layer with a 0.2 probability factor is applied to ensure that the machine learning model is not over fitted on training data.
[0055] Each of the pre-trained plurality of distinct machine learning models (206) is additionally trained by re-training the LSTM layers of the pre-trained plurality of distinct machine learning models with the second set of plurality of ECG leads’ data specific to the subject patient. The additionally trained plurality of distinct machine learning models reconstruct the new plurality of ECG leads’ data (106, 406) associated with the subject patient from the new ECG data (102, 402) of the single ECG lead.
[0056] In an embodiment of the present invention, the plurality of ECG lead data is 12 ECG lead data.
[0057] In another embodiment of the present invention, the subset of plurality of ECG leads’ data is 7 ECG leads’ data.
[0058] In another embodiment of the present invention, the single ECG lead is an ECG lead II.
[0059] In another embodiment of the present invention, the subset of the plurality of ECG leads are ECG leads I, V1, V2, V3, V4, V5, and V6.
[0060] In another embodiment of the present invention, the remaining ECG leads’ data are reconstructed from the ECG leads I, II, V1, V2, V3, V4, V5, and V6 data.
[0061] In another embodiment of the present invention, each of the plurality of distinct machine learning model comprises 3 CNN layers (302, 304, 306), 3 max-pooling layers (308, 310, 312), one LSTM layer (314), one Flatten layer (316) and one Dense layer (318).
[0062] In another embodiment of the present invention, the first set of the plurality of ECG lead data is a large sized and reusable plurality of ECG leads’ data.
[0063] In another embodiment of the present invention, the pre-trained plurality of distinct machine learning models (206) are configured to reconstruct the plurality of ECG data, based on the transfer learning, associated with a subject patient of the first group of patients.
[0064] In another embodiment of the present invention, the second set of the plurality of ECG lead data comprises a small set of plurality of ECG leads’ data associated with the subject patient, recorded for a minimal time, in order to additionally train the pre-trained model to reconstruct a plurality of ECG data for the subject patient.
[0065] In another embodiment of the present invention, the plurality of distinct machine learning model are trained to reconstruct 1 second of other ECG lead data from 1 second of the single ECG lead data.
[0066] In another embodiment of the present invention, the LSTM layers of the additionally trained plurality of distinct machine learning model comprise the LSTM layers associated with the subject patient only without the pre-trained model’s LSTM layer, and subsequently, the LSTM layers associated with the subject patient is retained in the at least one database.
[0067] In another embodiment of the present invention, a new set of LSTM layers is added to the at least one database after the training of the models for newly added subject patient.
[0068] In another embodiment of the present invention, the LSTM layer is loaded from the at least one database according to the subject patient group.
[0069] In another embodiment of the present invention, the second set of plurality of ECG leads’ data loaded in the at least one LSTM layers is comparatively low.
[0070] In another embodiment of the present invention, the second set of plurality of ECG leads’ data comprises only 49.4 thousand parameters.
[0071] In one or more embodiments, the system (100) may be part of a larger computer system (410) and/or maybe operatively coupled to a network (e.g., a second network 408) with the aid of a communication interface to facilitate the transmission of and sharing data and predictive results. The computer network may be a local area network (LAN), an intranet and/or extranet, an intranet and/or extranet that is in communication with the Internet, or the Internet. The network in some cases is a telecommunication and/or a data network, and may include one or more computer servers. In an example, the communication network includes, but not limited to, 2G network, 3G network, 4G network, LTE network, 5G network, 6G network, and so forth. The network, in some cases with the aid of a computer system, may implement a peer-to-peer network, which may enable devices coupled to the computer system to behave as a client or a server. In other examples, the system, the database, and the server may be integrated network node or a single integrated unit.
[0072] The system (100) may communicate with one or more other systems by the interfaces (e.g., network adapters). The memory (210) or memory locations may be, e.g., random-access memory, read-only memory, flash memory. The system may also include at least one electronic storage units (e.g., hard disks), and peripheral devices, such as cache, other memory, data storage, and/or electronic display adapters.
[0073] The system (100) may also include one or more IO Managers as software instructions that may run on the one or more processors (208) and implement various communication protocols such as User Datagram Protocol (UDP), Modbus, MQ Telemetry Transport (MQTT), Open Platform Communications Unified Architecture (OPC UA), Semiconductor's equipment interface protocol for equipment-to-host data communications (SECS/GEM), Profinet, or any other protocol, to access data in real-time from disparate data sources via any communication network, such as Ethernet, Wi-Fi, Universal Serial Bus (USB), Zigbee, Cellular or 5G connectivity, etc., or indirectly through a device’s primary controller, through a Programmable Logic Controller (PLC) or through a Data Acquisition System (DAQ), or any other such mechanism.
[0074] Further, the CPU(s) may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, logic circuitries, and/or any devices that manipulate data based on operational instructions. Among other capabilities, the one or more processor(s) (208) are configured to fetch and execute computer-readable instructions stored in the memory (210) of the system (100).
[0075] Further, the memory (210) may store one or more computer-readable instructions or routines, which may be fetched and executed to create or share data units over a network service. The memory (210) may include any non-transitory storage device including, for example, volatile memory such as RAM, or non-volatile memory such as EPROM, flash memory, and the like.
[0076] Further, the processing devices(s) may be implemented as a combination of hardware and programming device(s) (for example, programmable instructions) to implement one or more functionalities of the processing device(s). In examples described herein, such combinations of hardware and programming may be implemented in several different ways. In one example, the programming for the processing device(s) may be processor executable instructions stored on a non-transitory machine-readable storage medium and the hardware for the processing device(s) may include a processing resource (for example, one or more processors (208)), to execute such instructions. In other examples, the processing devices(s) may be implemented by electronic circuitry.
[0077] FIG. 5 illustrates a schematic block diagram depicting a method (500) for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient, in accordance with an exemplary embodiment of the present disclosure.
[0078] As illustrated, the method (500) includes the step of receiving (502) the new ECG data of the single ECG lead. The step of receiving is performed by a receiving module (202) of a system (100) as described in conjunction with Figs. 1-4. The method (500) further includes the step of storing (5020) a first set of plurality of ECG leads’ data associated with a first group of patients, and a second set of plurality of ECG lead data associated with a second group of patients, in at least one database. The subject patient relates to the second group of patients. The method (500) further includes the step of loading (506) the plurality of ECG leads’ data associated with the subject patient, from the at least one database, to a pre-trained plurality of distinct machine learning models, to additionally train, based on a transfer learning combined with a continual learning techniques, the pre-trained plurality of distinct machine learning models specific to the subject patient. The plurality of distinct machine learning models correspond to a subset of the plurality of ECG leads and are pre-trained with the first set of plurality of ECG leads’ data.
[0079] Each of the plurality of distinct machine learning models includes at least one Convolutional Neural Networks (CNN) layer followed by at least one max-pooling layers after each CNN layer, the at least one Long Short-Term Memory (LSTM) layer, at least one Flatten layer, and at least one Dense layer. The additional training of the pre-trained plurality of distinct machine learning models includes the sub-steps of re-training the LSTM layers of the pre-trained plurality of distinct machine learning models with the second set of plurality of ECG leads’ data specific to the subject patient. The additionally trained plurality of distinct machine learning models reconstruct the plurality of ECG leads’ data associated with the subject patient from the new ECG data from the single ECG lead.
[0080] FIG. 6 illustrates tabular performance results for a plurality of patients considered for pre-training, in accordance with an exemplary embodiment of the present disclosure.
[0081] As illustrated, the present invention is tested and validated using 148 patients from myocardial infarction category. In this work, 𝑅 statistics, the correlation coefficient (𝑟𝑥), and the regression coefficient (𝑏𝑥) are employed as evaluation measures for reconstruction performance. 𝑅22 statistics were utilized to assess the degree of relationship between the original and reconstructed signals. The 𝑅 value of a perfectly reconstructed signal will be 100%. The correlation coefficient estimates the amplitude differences between the original and reconstructed signals, whereas the regression coefficient estimates the similarity between the original and reconstructed signals. The achieved mean values of 𝑅22 statistics (averaged over all 11 leads), correlation, and regression coefficients are 93.62, 0.973,0.959 and 88.09, 0.952, 0.936 for the plurality of distinct machine learning models i.e. seven distinct machine learning models approach and single model approach, respectively. It can be observed that the plurality of distinct machine learning models approach outperformed the single model approach. In the seven models’ approach, the memory needed to store all the 57,506,904 parameters is 219.69 MB whereas the single model approach needed 214.51 MB to store 56,221,272 parameters. Thus, the seven distinct machine learning models approach needs 5.18 MB more memory than the single model approach, but the accuracy levels are high for the seven models’ approach.
[0082] The present invention can be utilized in multiple ways/places. For example:
• It can be used in hospitals for getting continuous ECG recordings of the patients without wearing those many wires.
• It can be used at home also for your regular ECG monitoring and get track your cardiac monitoring by wearing only one lead of the ECG, which would be much convenient to use.
• It can be used by the army to monitor their ECG with less bother about more tangled wires and track their cardiac health.
[0083] A few of the major advantages of the present invention over the conventional solutions:
• The present invention provides a neural network methodology for generating 12 Leads ECG data generation by just positioning only a single lead or electrode over a patient or person's body. Further, a transfer learning approach is used so a machine learning model can generate the 12 leads’ data from unknown person single ECD lead data which has not been used or seen by the model while it was trained or tested. The present invention with a minimum wire configuration on the body, improves patient comfort by recording fewer ECG leads, usually a single lead.
[0084] Those of skill in the art would appreciate that the various illustrative blocks, modules, elements, components, methods, and algorithms described herein may be implemented as electronic hardware, computer software, or combinations of both. To illustrate this interchangeability of hardware and software, various illustrative blocks, modules, elements, components, methods, and algorithms have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. Various components and blocks may be arranged differently (e.g., arranged in a different order, or partitioned in a different way) all without departing from the scope of the subject technology.
[0085] It is understood that any specific order or hierarchy of blocks in the processes disclosed is an illustration of example approaches. Based upon design preferences, it is understood that the specific order or hierarchy of blocks in the processes may be rearranged, or that all illustrated blocks be performed. Any of the blocks may be performed simultaneously. In one or more implementations, multitasking and parallel processing may be advantageous. Moreover, the separation of various system components in the implementations described above should not be understood as requiring such separation in all implementations, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.
[0086] It should be noted that the description and figures merely illustrate the principles of the present subject matter. It should be appreciated by those skilled in the art that conception and specific embodiment disclosed may be readily utilized as a basis for modifying or designing other structures for carrying out the same purposes of the present subject matter. It should also be appreciated by those skilled in the art by devising various systems that, although not explicitly described or shown herein, embody the principles of the present subject matter and are included within its spirit and scope.
[0087] Furthermore, all examples recited herein are principally intended expressly to be for pedagogical purposes to aid the reader in understanding the principles of the present subject matter and the concepts contributed by the inventor(s) to further the art and are to be construed as being without limitation to such specifically recited examples and conditions. The novel features which are believed to be characteristic of the present subject matter, both as to its organization and method of operation, together with further objects and advantages will be better understood from the following description when considered in connection with the accompanying figures.
[0088] Although embodiments for the present subject matter have been described in language specific to package features, it is to be understood that the present subject matter is not necessarily limited to the specific features described. Rather, the specific features and methods are disclosed as embodiments for the present subject matter. Numerous modifications and adaptations of the system/device of the present invention will be apparent to those skilled in the art, and thus it is intended by the appended claims to cover all such modifications and adaptations which fall within the scope of the present subject matter.
, Claims:We claim:
1. A system (100) for reconstruction of a plurality of new Electrocardiogram, ECG, leads’ data (104) from a new ECG data (102) of a single ECG lead positioned over a subject patient, the system comprising:
a. a receiving module (202) configured to receive the new ECG data (102) of the single ECG lead;
b. at least one database (204) comprising a first set of plurality of ECG leads’ data associated with a first group of patients, and a second set of plurality of ECG leads’ data associated with a second group of patients, wherein the subject patient relates to the second group of patients; and
c. a plurality of distinct machine learning models (206) corresponding to a subset of a plurality of ECG leads, pre-trained with the first set of plurality of ECG leads’ data, and configured to load the second set of plurality of ECG leads’ data, to additionally train, based on a transfer learning combined with a continual learning techniques, the pre-trained plurality of distinct machine learning models, each of the plurality of distinct machine learning models comprising at least one Convolutional Neural Networks, CNN, layer followed by at least one Max-Pooling layers after each CNN layer, at least one Long Short-Term Memory, LSTM, layer, at least one Flatten layer, and at least one Dense layer, wherein each of the pre-trained plurality of distinct machine learning models is additionally trained by re-training the LSTM layers of the pre-trained plurality of distinct machine learning models with the second set of plurality of ECG leads’ data specific to the subject patient,
wherein the additionally trained plurality of distinct machine learning models (206) reconstruct the new plurality of ECG leads’ data (104) associated with the subject patient from the new ECG data (102) of the single ECG lead.
2. The system (100) as claimed in claim 1, wherein the plurality of ECG lead data is 12 ECG lead data.
3. The system (100) as claimed in claim 1, wherein the subset of plurality of ECG leads’ data is 7 ECG leads’ data.
4. The system (100) as claimed in claim 1, wherein the single ECG lead is an ECG lead II.
5. The system (100) as claimed in claim 1, wherein the subset of the plurality of ECG leads are ECG leads I, V1, V2, V3, V4, V5, and V6.
6. The system (100) as claimed in claims 4-5, wherein the remaining ECG leads’ data are reconstructed from the ECG leads I, II, V1, V2, V3, V4, V5, and V6 data.
7. The system (100) as claimed in claim 1, wherein each of the plurality of distinct machine learning model (206) comprises 3 CNN layers (302, 304, 206), 3 max-pooling layers (308, 310, 312), one LSTM layer (314), one Flatten layer (316) and one Dense layer (318).
8. The system (100) as claimed in claim 1, wherein the first set of the plurality of ECG lead data is a large sized and reusable plurality of ECG leads’ data.
9. The system (100) as claimed in claim 1, wherein the pre-trained plurality of distinct machine learning models are configured to reconstruct the plurality of ECG data, based on the transfer learning, associated with a subject patient of the first group of patients.
10. The system (100) as claimed in claim 1, wherein the second set of the plurality of ECG lead data comprises a small set of plurality of ECG leads’ data associated with the subject patient, recorded for a minimal time, in order to additionally train the pre-trained model to reconstruct a plurality of ECG data for the subject patient.
11. The system (100) as claimed in claim 10, wherein the plurality of distinct machine learning model are trained to reconstruct 1 second of other ECG lead data from 1 second of the single ECG lead data.
12. The system (100) as claimed in claim 1, wherein the LSTM layers of the additionally trained plurality of distinct machine learning model comprise the LSTM layers associated with the subject patient only without the pre-trained model’s LSTM layer, and subsequently, the LSTM layers associated with the subject patient is retained in the at least one database.
13. The system (100) as claimed in claim 1, wherein a new set of LSTM layers is added to the at least one database after the training of the models for newly added subject patient.
14. The system (100) as claimed in claim 1, wherein the LSTM layer is loaded from the at least one database according to the subject patient group.
15. The system (100) as claimed in claim 1, wherein the second set of plurality of ECG leads’ data loaded in the at least one LSTM layers is comparatively low.
16. The system (100) as claimed in claim 1, wherein the second set of plurality of ECG leads’ data comprises only 49.4 thousand parameters.
17. A method (500) of reconstruction of a plurality of electrocardiogram, ECG, leads’ data from a new ECG data of a single ECG lead positioned over a subject patient, the method comprising:
a. receiving (502), by a receiving module of a system, the new ECG data of the single ECG lead;
b. storing (504) a first set of plurality of ECG leads’ data associated with a first group of patients, and a second set of plurality of ECG lead data associated with a second group of patients, in at least one database, wherein the subject patient relates to the second group of patients; and
c. loading (506) the plurality of ECG leads’ data associated with the subject patient, from the at least one database, to a pre-trained plurality of distinct machine learning models, to additionally train, based on a transfer learning combined with a continual learning techniques, the pre-trained plurality of distinct machine learning models specific to the subject patient, wherein the plurality of distinct machine learning models correspond to a subset of the plurality of ECG leads and are pre-trained with the first set of plurality of ECG leads’ data, each of the plurality of distinct machine learning models comprising at least one Convolutional Neural Networks, CNN, layer followed by at least one max-pooling layers after each CNN layer, the at least one Long Short-Term Memory, LSTM, layer, at least one Flatten layer, and at least one Dense layer, wherein the additional training of the pre-trained plurality of distinct machine learning models comprises re-training the LSTM layers of the pre-trained plurality of distinct machine learning models with the second set of plurality of ECG leads’ data specific to the subject patient, and
wherein the additionally trained plurality of distinct machine learning models reconstruct the plurality of ECG leads’ data associated with the subject patient from the new ECG data from the single ECG lead.