Description:A METHOD AND A SYSTEM FOR GENERATING ELECTROCARDIOGRAM DATA FROM PHOTOPLETHYSMOGRAM DATA

[001] The present disclosure relates to a method for generating Electrocardiogram (ECG) of a subject, particularly, the present subject matter relates to the method for generating the ECG data from Photoplethysmogram (PPG) data.
BACKGROUND OF THE INVENTION
[002] Background description includes information that may be useful in understanding the present invention.
[003] – Traditionally, Electrocardiograms (ECG) are widely regarded as the standard method for assessing cardiac activity. To detect even subtle abnormalities, long-term continuous monitoring is often necessary, as these irregularities may occur briefly after their onset. However, traditional ECG methods limit patients' mobility due to the need for multiple leads attached to the body. This has led to a growing demand for continuous, unobtrusive, and wearable ECG systems that require minimal attachments and can be comfortably used at home. While the Holter monitor is commonly used for continuous ECG monitoring in clinical settings, its bulkiness makes it impractical for everyday use. Newer devices aim to overcome this by reducing the size and number of leads, but they often either rely on skin-irritating chest patches or require the user to hold them with their fingers, making them unsuitable for long-term monitoring.
[004] ECG is the most prominently used heart monitoring signal that doctors rely on to get the heart condition of any person. But getting 10 leads on for longer to continuously monitor our heart function can restrict the person’s activity. Existing work on ECG data generation from PPG using GAN architecture uses a dual generator and dual discriminator which includes an increased model complexity.
[005] Therefore, there is a need for a solution to overcome the above-mentioned drawbacks.

SUMMARY OF THE INVENTION
[006] This summary is provided to introduce concepts related to a method for predicting a Crack Growth Rate (CGR) in a naval structure. The concepts are further described below in the detailed description. 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.
[007] The present subject matter discloses a system for generating Electrocardiogram (ECG) data from Photoplethysmogram (PPG) data of a subject. The system includes a receiving engine configured to receive a PPG signal associated with the subject. The system includes a pre-processing engine having a plurality of pass filters configured to pre-process the PPG data to remove a base line wandering and external noise data from the PPG data to generate pre-processed and re-sampled PPG data. The system includes a windowing engine configured to window the pre-processed and re-sampled PPG data to generate a plurality of non-overlapping data samples associated with the pre-processed and re-sampled PPG data, wherein the plurality of non-overlapping data samples is transmitted. The system includes a generator of Generative Adversarial Network (GAN) model configured to receive the plurality of non-overlapping data samples and a first model with a plurality of pre-trained weights associated with the first model. The generator of GAN model is configured to process the plurality of non-overlapping data samples with the plurality of pre-trained weights to generate the ECG data.
[008] The present subject matter discloses a method for generating Electrocardiogram (ECG) data from Photoplethysmogram (PPG) data of a subject. The method includes receiving, by a receiving engine, a PPG signal associated with the subject. The method includes pre-processing, by a pre-processing engine having a plurality of pass filters, the PPG data to remove a base line wandering and external noise data from the PPG data to generate pre-processed and re-sampled PPG data. The method includes windowing, by a windowing engine, the pre-processed and re-sampled PPG data to generate a plurality of non-overlapping data samples associated with the pre-processed and re-sampled PPG data, wherein the plurality of non-overlapping data samples is transmitted. The method includes receiving, by a generator of Generative Adversarial Network (GAN) model, the plurality of non-overlapping data samples and a first model with a plurality of pre-trained weights associated with the first model. The method includes processing, by the generator of GAN model, the plurality of non-overlapping data samples with the plurality of pre-trained weights to generate the ECG data.
[009] 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.

BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWINGS
[010] 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, systems, and methods that are consistent with the subject matter as claimed herein, wherein:
[011] Fig. 1 illustrates a diagram depicting a system for generating ECG data from PPG data of a subject, in accordance with an embodiment of the present subject matter;
[012] Fig. 2 illustrates a schematic block diagram of the system, in accordance with an embodiment of the present subject matter;
[013] Fig. 3 illustrates an operational flow diagram depicting a process for generating ECG data from PPG data of a subject, in accordance with an embodiment of the present subject matter; and
[014] Fig. 4 illustrates a schematic block diagram depicting a method for generating ECG data from PPG data of a subject, in accordance with an embodiment of the present subject matter.
[015] 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.

DETAILED DESCRIPTION
[016] 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.
[017] Fig. 1 illustrates a diagram 100 depicting a system 102 for generating ECG data from PPG data of a subject, in accordance with an embodiment of the present subject matter. The system 102 may be configured to utilize a GAN model to generate the ECG data from the PPG data. The PPG data and the ECG data may be related to a heart of the subject and may be used for cardiac health analysis of the subject. The ECG data may be 1 lead ECG data. The system 102 may be based on a deep convolution GAN based architecture that uses deep convolution layers for a generator and a discriminator portion. Upon training the GAN based architecture, a trained generator may be used to construct the 1 lead ECG data of the subject.
[018] In accordance with an embodiment of the present subject matter, the system 102 may be configured to receive a PPG signal associated with the subject.
[019] In response to receiving the PPG signal, the system 102 may be configured to pre-process the PPG data. The PPG data may be pre-processed to remove a base line wandering and external noise data from the PPG data. The base line wandering and external noise data may be removed from the PPG data to generate pre-processed and re-sampled PPG data.
[020] To that understanding, upon the generation of the pre-processed and re-sampled PPG data, the system 102 may be configured to perform a windowing operation on the pre-processed and re-sampled PPG data. The windowing operation may be performed to generate a number of non-overlapping data samples associated with the pre-processed and re-sampled PPG data.
[021] Subsequently, the system 102 may be configured to receive a first model with a number of pre-trained weights associated with the first model. Furthermore, the system 102 may be configured to process the number of non-overlapping data samples with the number of pre-trained weights to generate the ECG data.
[022] Fig. 2 illustrates a schematic block diagram 200 of the system 102, in accordance with an embodiment of the present subject matter. The system 102 may be configured to generate ECG data from PPG data of a subject. In an embodiment of the present subject matter, the subject may be a patient. The PPG data may be one lead PPG data and the ECG data may be one lead ECG data. The system 102 may be used in hospitals for getting continuous ECG recordings of the subjects without a use of wires to be worn by the subjects. The system 102 may also be used at home for regular ECG monitoring and track cardiac monitoring. The system 102 may be used in remote areas as the system 102is easily available, less costly, and easy to use for tracking the cardiac health.
[023] The system 102 may be configured to include an LSTM layer and may reduce a vanishing gradient problem while generating the ECG data from the PPG data. The system 102 may be light and may offer a less time consumption, less power consumption on a device having the system 102, and a less circuit complexity. The system 102 may use a single generator and a single discriminator with an inclusion of the LSTM layers to generate the ECG data from the PPG data.
[024] In an example, the system 102 may include a pre-processing system 202 and a generator of GAN model 204. The pre-processing system 202 may communicate with the generator of GAN model 204 to transmit data for generation of the ECG data by the generator of the GAN model 204.
[025] The pre-processing system 202 may include a processor 206, a memory 208, data 210, a receiving engine 212, a pre-processing engine 214, and a windowing engine 216. The generator of GAN model 204 may include an encoder 218, a multi head attention engine 220, a training engine 222, a lambda layer 224, and a decoder 226. In an example, the processor 206, the memory 208, data 210, the receiving engine 212, the pre-processing engine 214, and the windowing engine 216 may be communicatively coupled to one another. In an example, the encoder 218, the multi head attention engine 220, the training engine 222, the lambda layer 224, and the decoder 226 may be communicatively coupled to one another. Also, the pre-processing system 202 and the generator of the GAN model 204 may be communicatively coupled to one another.
[026] The pre-processing system 202 may be understood as one or more of a hardware, a configurable hardware, and the like. In an example, the processor 206 may be a single processing unit or a number of units, all of which could include multiple computing units. Among other capabilities, the processor 206 may be configured to fetch and/or execute computer-readable instructions and/or data stored in the memory 208.
[027] In an example, the memory 208 may include any non-transitory computer-readable medium known in the art including, for example, volatile memory, such as static random-access memory (SRAM) and/or dynamic random-access memory (DRAM), and/or non-volatile memory, such as read-only memory (ROM), erasable programmable ROM (EPROM), flash memory, hard disks, optical disks, and/or magnetic tapes. The memory 208 may further include the data 210.
[028] The data 210 serves, amongst other things, as a repository for storing data processed, received, and generated by the system 102.
[029] Continuing with the above embodiment, the receiving engine 212 may be configured to receive a PPG signal associated with the subject. The PPG signal may include PPG data related to heart of a subject. In response to a successful receipt of the PPG data by the receiving engine 212, the pre-processing engine 214 may be configured to pre-process and re-sample the PPG data. The pre-processing and re-sampling may be performed to remove a base line wandering and external noise data from the PPG data resulting in a generation of pre-processed and re-sampled PPG data. To that understanding, the pre-processing engine 214 may include a number of pass filters in a cascaded form that remove the base line wandering and external noise data from the PPG data. Examples of the number of pass filters may include, but are not limited to, butterworth filters. The pre-processing may include processing the PPG data with a high pass filter amongst the number of pass filters in the pre-processing engine 214. The high pass filter may be butterworth 4th order high filter. The pre-processing may further include processing the PPG data with a low pass filter amongst the number of pass filters upon processing with the high pass filter to generate pre-processed and re-sampled PPG data. The low pass filter may be butterworth 4th order low filter. The base line wandering and the external noise data may be filtered upon processing the PPG data with the number of pass filters.
[030] Upon a successful removal of the base line wandering and external noise data from the PPG data and the generation of the processed PPG data, the windowing engine 216 may be configured to window the pre-processed and re-sampled PPG data. The windowing may be performed to generate a number of non-overlapping data samples related with the pre-processed and re-sampled PPG data. The windowing may include resampling the pre-processed and re-sampled PPG data at a predetermined amount of frequency to generate the number of non-overlapping data samples. The number of non-overlapping data samples may be 4 second windows. To that understanding, the number of non-overlapping data samples may be transmitted by the pre-processing system 202 and received by the generator of the GAN model 204.
[031] In response to receiving the number of over-lapping data samples, the generator of the GAN model 204 may be configured to receive a first model with a number of pre-trained weights associated with the first model. Upon receiving the number of over-lapping data samples and the first model, the generator of the GAN model 204 may be configured to process the number of non-overlapping data samples with the number of pre-trained weights. The first model may be trained by the training engine 222 with pre-stored data associated with a number of subjects to generate the number of trained weights associated with the first model.
[032] Based on the processing of the non-overlapping data samples with the number of pre-trained weights, the ECG data may be generated. For generating the ECG data, the encoder 218 with one convolution layer within the generator of the GAN model 204 may be configured to process the number of non-overlapping data samples to generate a first output. The convolution layer may include a sin activation function to process the number of non-overlapping data samples. Further upon generation of the first output, the multi head attention engine 220 may be receive the first output. The multi head attention engine 220 may include a number of heads configured to process the first output to generate a second output. The number of heads may process only a number of unique parts of the first output to generate the second output. The multi head attention engine 220 and the LSTM layers may remove a vanishing gradient problem and generate the ECG data with a greater precision.
[033] To that understanding, the lambda layer 224 may be configured to perform a matrix multiplication of the first output and the second output to generate a third output. The third output may be transmitted to the decoder 226. The decoder 226 may include a number of neural networks that may be configured to process the third output to generate the ECG data. The number of neural network comprises a first convolution layer, a second convolution layer, a first LSTM layer, a second LSTM layer, and a third convolution layer. The LSTM layers may reduce vanishing gradients.
[034] The LSTM layers may be similar to an RNN network only with some extra memory features. For every time step, required pieces of information may be saved in the memory to be used later. While during backpropagation in an RNN the gradient of the error term is given as

[035] The error gradient on the ’a’ time step is given by:

after some ‘a’ time step

complete error gradient may become 0 and new weights of network may be

[036] In LSTM, c(t) is the state vector and may be shown in summary as:

[037] And, the derivative of Eq. (6) depends upon the above present four terms. Finally,

Where, 𝑃𝑡,𝑡,𝑅𝑡 and 𝑆𝑡 denote the four elements of the derivative of the cell state.
[038] Combining Eq.(7) and Eq.(3).

[039] Since LSTM has a forget gate, its vector of activation in the gradient term along the remaining additive structure always finds some parameter to update the weight at any time step.

[040] The gradient may vanish. Also, from above, it may be noticed that the gradient of LSTM’s cell state is the addition of four elements i.e., 𝑃𝑡,𝑡,𝑅𝑡 , and 𝑆𝑡. During a backpropagation process, the gradient values get balanced due to an additive feature of LSTM. In this way, all four element values may get updated properly. So, a proper balance between the values may cause the additive element not to vanish, and hence, the Vanishing gradient issue is resolved.
[041] The GAN model 204 may include a Discriminator portion having five layers of convolutions followed by Leaky ReLU layers and batch normalization layers, except the first Convolution layer. Lastly, one Dense layer may predict an input as either ’0’(fake) or ’1’(real). The discriminator may be a classifier that classifies its input into two real or fake classes. The generator portion may learn about data and try to reduce its error by updating its weights at every epoch of the training. The weights may get updated by calculating a loss of the discriminator and generator portion, and then a differential of the error may get reduced from the previous weights to get the updated weight. The generator of the GAN model 204 may have two loss functions. The losses being calculated may be one for a generator and the other for a discriminator training loss. The standard GAN loss function may be given by equation 10 a below. This is known as minimax loss.
𝐸[log(D(i))] + 𝐸𝑗[log (1 − D(G(j)))] equation-10
log(D(i)): shows the probability of rightly classified real images by the discriminator. G(j) shows the generator’s output for the input ‘j’. D(G(j)) denotes the probability of the discriminator’s label as real for fake data. 𝐸𝑖represents all real data expected values and all the random inputs fed to the generator.
[042] Fig. 3 illustrates an operational flow diagram depicting a process 300 for generating ECG data from PPG data of a subject, in accordance with an embodiment of the present subject matter. The subject may be a patient and the PPG data and the ECG data may be related to a heart of the patient. The PPG data may be one lead PPG data and the ECG data may one lead ECG data.
[043] At step 302, the process 300 may include receiving a PPG signal associated with the subject by the receiving engine 212 as referred in fig. 2. The PPG signal may include PPG data related to heart of the subject
[044] At step 304, the process 300 may include processing and re-sampling the PPG data with a high pass filter amongst a number of pass filters in the pre-processing engine 214. The high pass filter may be butterworth 4th order high filter.
[045] At step 306, the process 300 may include processing the PPG data with a low pass filter amongst the number of pass filters upon processing with the high pass filter to generate the pre-processed and re-sampled PPG data. The low pass filter may be butterworth 4th order low filter. Upon pre-processing the PPG data, a base line wandering and external noise data may be removed from the PPG data resulting in a generation of the pre-processed and re-sampled PPG data. The base line wandering and the external noise data may be filtered upon processing the PPG data with the number of pass filters. The number of pass filters may be in a cascaded form that removes the base line wandering and external noise data from the PPG data. Examples of the number of pass filters may include, but are not limited to, butterworth filters. The re-sampling may be performed after pre-processing, at a predetermined amount of frequency. The resampling may be performed with 128 Hz frequency.
[046] At step 308, the process 300 may include windowing of the pre-processed and re-sampled PPG data to generate the number of non-overlapping data samples. The windowing may be performed by the windowing engine 216 as referred in fig. 2. The number of non-overlapping data samples may be related with the pre-processed and re-sampled PPG data. The number of non-overlapping data samples may be 4 second windows (512 samples). To that understanding, the number of non-overlapping data samples may be transmitted by the pre-processing system 202and received by the generator of the GAN model 204. The generator of the GAN model 204 may find a relation with the PPG signal and an ECG signal.
[047] At step 310, the process 300 may include receiving a first model with a number of pre-trained weights associated with the first model by the generator of the GAN model 204.
[048] At step 312, the process 300 may include processing by the generator of the GAN model 204 the number of non-overlapping data samples with the number of pre-trained weights. The first model may be trained by the training engine 222 as referred in fig. 2 with pre-stored data associated with a number of subjects to generate the number of trained weights associated with the first model.
[049] At step 314, the process 300 may include processing by the encoder 218 with a convolution layer, the number of non-overlapping data samples to generate a first output. The convolution layer may include a sin activation function to process 300 the number of non-overlapping data samples. Further upon generation of the first output, the multi head attention engine 220 may be receive the first output.
[050] At step 316, the process 300 may include, processing the first output to generate a second output by the multi head attention engine 220 with a number of heads as referred in fig. 2. The number of heads may process 300 only a number of unique parts of the first output to generate the second output. The number of heads allow attending different parts of a same input sequence to generate the second output.
[051] At step 318, the process 300 may include performing by the lambda layer 224 as referred in fig. 2 a matrix multiplication of the first output and the second output to generate a third output. The third output may be transmitted to the decoder 226. The decoder 226 may include a number of neural networks that may be configured to process 300 the third output to generate the ECG data. The number of neural network comprises a first convolution layer, a second convolution layer, a first LSTM layer, a second LSTM layer, and a third convolution layer.
[052] Fig. 4 illustrates a schematic block diagram depicting a method 400 for generating ECG data from PPG data of a subject, in accordance with an embodiment of the present subject matter. The method 400 may be performed by the system 102 as referred in fig. 2 and components thereof.
[053] At block 402, the method 400 includes receiving, by a receiving engine 212, a PPG signal having PPG data associated with the subject.
[054] At block 402, the method 400 includes pre-processing and re-sampling, by a pre-processing engine 214 having a number of pass filters, the PPG data to remove a base line wandering and external noise data from the PPG data to generate pre-processed and re-sampled PPG data.
[055] At block 404, the method 400 includes windowing, by a windowing engine 216, the pre-processed and re-sampled PPG data to generate a number of non-overlapping data samples associated with the pre-processed and re-sampled PPG data, wherein the number of non-overlapping data samples is transmitted.
[056] At block 406, the method 400 includes receiving, by a generator of Generative Adversarial Network (GAN) model, the number of non-overlapping data samples and a first model with a number of pre-trained weights associated with the first model.
[057] At block 408, the method 400 includes processing, by the generator of GAN model 204, the number of non-overlapping data samples with the number of pre-trained weights to generate the ECG data.
[058] While the embodiments of the disclosure are subject to various modifications and alternative forms, specific embodiment 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 alternative falling within the scope of the disclosure.
[059] 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, 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.
[060] It will be further appreciated that functions or structures of a number of components or steps may be combined into a single component or step, or the functions or structures of one-step or component may be split among plural steps or components. The present invention contemplates all of these combinations. Unless stated otherwise, dimensions and geometries of the various structures depicted herein are not intended to be restrictive of the invention, and other dimensions or geometries are possible. In addition, while a feature of the present invention may have been described in the context of only one of the illustrated embodiments, such feature may be combined with one or more other features of other embodiments, for any given application. It will also be appreciated from the above that the fabrication of the unique structures herein and the operation thereof also constitute methods in accordance with the present invention. The present invention also encompasses intermediate and end products resulting from the practice of the methods herein. The use of “comprising” or “including” also contemplates embodiments that “consist essentially of” or “consist of” the recited feature.

, Claims:We Claim:

1. A system (102) for generating Electrocardiogram (ECG) data from Photoplethysmogram (PPG) data of a subject comprising:
a receiving engine (212) configured to receive a PPG signal having PPG data associated with the subject;
a pre-processing engine (214) having a plurality of pass filters configured to pre-process and re-sample the PPG data to remove a base line wandering and external noise data from the PPG data to generate pre-processed and re-sampled PPG data;
a windowing engine (216) configured to window the pre-processed and re-sampled PPG data to generate a plurality of non-overlapping data samples associated with the pre-processed and re-sampled PPG data, wherein the plurality of non-overlapping data samples is transmitted;
a generator of Generative Adversarial Network (GAN) model (204) configured to:
receive the plurality of non-overlapping data samples and a first model with a plurality of pre-trained weights associated with the first model; and
process the plurality of non-overlapping data samples with the plurality of pre-trained weights to generate the ECG data.

2. The system (102) as claimed in claim 8, wherein the GAN model (204) further comprises:
an encoder (218) with one convolution layer configured to process the plurality of non-overlapping data samples to generate a first output, wherein the convolution comprises a sin activation function to process the plurality of non-overlapping data samples;
a multi head attention engine (220) with a plurality of heads configured to process the first output to generate a second output, wherein the plurality of heads processes a plurality of unique parts of the first output to generate the second output;
a lambda layer (224) configured to perform a matrix multiplication of the first output and the second output to generate a third output, wherein the third output is transmitted;
a decoder (226) having a plurality of neural network configured to process the third output to generate the ECG data.
3. The system (102) as claimed in claim 1, wherein the plurality of neural network comprises a first convolution layer, a second convolution layer, a first LSTM layer, a second LSTM layer, and a third convolution layer.

4. The system (102) as claimed in claim 2, wherein the activation function is a sin function.

5. The system (102) as claimed in claim 1, comprising:
a training engine (222) configured to train the first model with pre-stored data associated with a plurality of subjects to generate a plurality of trained weights associated with the first model.

6. The system (102) as claimed in claim 1, wherein the PPG data is one lead PPG data and the ECG data is one lead ECG data.

7. The system (102) as claimed in claim 1, wherein pre-processing the PPG data comprises:
processing the PPG data with a high pass filter amongst the plurality of pass filters in the pre-processing engine (214); and
processing the PPG data with a low pass filter amongst the plurality of pass filters upon processing with the high pass filter to generate pre-processed and re-sampled PPG data, wherein the base line wandering and the external noise data is filtered upon processing the PPG data with the plurality of pass filters.
8. The method (400) as claimed in claim 1, wherein windowing the pre-processed data comprises:
the pre-processing engine (214) configured to resample the PPG data at a predetermined amount of frequency to generate the plurality of non-overlapping data samples.

9. A method (400) for generating Electrocardiogram (ECG) data from Photoplethysmogram (PPG) data of a subject comprising:

receiving, by a receiving engine (212), a PPG signal having PPG data associated with the subject;
pre-processing, by a pre-processing engine (214) having a plurality of pass filters, the PPG data to remove a base line wandering and external noise data from the PPG data to generate pre-processed and re-sampled PPG data;
windowing, by a windowing engine (216), the pre-processed and re-sampled PPG data to generate a plurality of non-overlapping data samples associated with the pre-processed and re-sampled PPG data, wherein the plurality of non-overlapping data samples is transmitted;
receiving, by a generator of Generative Adversarial Network (GAN) model (204), the plurality of non-overlapping data samples and a first model with a plurality of pre-trained weights associated with the first model; and
processing, by the GAN model (204), the plurality of non-overlapping data samples with the plurality of pre-trained weights to generate the ECG data.