System and methods for collecting and processing data on one or more physiological parameters of monitored subject
A method of collecting physiological parameter data of a monitored subject comprises measuring a biosignal from which the physiological parameter is deducible, including noise; converting the noisy measured biosignal to a vector having different frequency components with corresponding magnitude coefficients; discarding select frequency components with coefficients below a prescribed threshold; and communicating the reduced vector to a computing device for processing to deduce the physiological parameter. A method of processing physiological parameter data comprises receiving a measured biosignal with electromagnetic interference noise; obtaining from the noisy measured biosignal representative data using a machine learning algorithm; and determining the physiological parameter from the representative data. A system for monitoring a physiological parameter comprises a wearable sensor configured to measure a biosignal and to remove noise from the measured signal, and a portable computing device configured to receive a transmitted signal from the sensor and to determine the physiological parameter therefrom.
1 . A method of overcoming signal noise in measurement and wireless transmission of electrocardiogram (ECG) and photoplethysmography (PPG) biosignals of actively mobile subjects using non-invasive wearables for transmission to mobile devices over low power wireless communication to enable real-time evaluation and continuous monitoring of multiple physiological parameters of said actively mobile subjects, the method comprising:
(a) by non-invasive ECG and PPG sensors embodied among one or more wearable devices worn by an actively mobile patient:
(i) measuring a ECG and PPG biosignals of the mobile patient and performing real-time conversion thereof into digital ECG and PPG signals, in which there is embodied noise that is of detrimental relation to accurate measurement of the physiological parameters and includes both motion artifacts and electromagnetic interference;
(ii) converting said noisy digital ECG and PPG signals into digitally compressed ECG and PPG signals by performing, in real-time, inverse discrete cosine transformation of said noisy digital ECG and PPG signals and quantization of the transformed noisy digital ECG and PPG signals to derive vector representations thereof composed of frequency components and associated coefficients, and discarding a subset of said frequency components whose associated coefficients fall below a threshold such that the digitally compressed ECG and PPG signals comprise sparse vector data representative of both the ECG and PPG biosignals and said noise, and transmitting, via a low power wireless communication protocol, said digitally compressed noisy signals from the one or more wearable devices;
(b) in concurrent relation to step (a), and by one or more motion sensors embodied among said one or more wearable devices worn by the actively mobile patient, capturing motion data representative of body motion of the actively mobile patient during capture of the measured ECG and PPG biosignals and transmitting, via a low power wireless communication protocol, said motion data from the one or more wearable devices;
(c) by a mobile edge-computing device borne by the actively mobile patient, receiving, via said low power wireless communication protocol, of both the digitally compressed ECG and PPG signals and the motion data from the one or more wearable devices;
(d) by said mobile edge-computing device of the actively mobile subject, and through execution thereby of a machine learning algorithm trained to detect and classify patterns between sparse ECG and PPG signals and noise factors including at least said motion artifacts and said electromagnetic interference, using the digitally compressed ECG and PPG signals and the motion data from the one or more wearable devices as combined input to the machine learning algorithm to calculate, in real-time, estimated noiseless sparse vector ECG and PPG signals from which said motion artifacts are filtered out during classification using said motion data;
(e) by said mobile edge-computing device, decompressing said estimated noiseless sparse ECG and PPG signals, in real-time, to reconstruct source ECG and PPG biosignals therefrom;
(f) by said mobile edge-computing device, applying a multi-linear regression algorithm to features extracted from the reconstructed source ECG and PPG biosignals to calculate, in real-time, estimated noiseless quantifications of the physiological parameter;
(g) outputting, in real time and by said mobile edge-computing device, said estimated noiseless quantifications of the physiological parameters; and
(h) continuously repeating steps (a) through (g) for continuous real-time monitoring of the physiological parameters;
wherein the multiple physiological parameters comprise at least a plurality of skin temperature, oxygen saturation, blood pressure, heart rate and respiration rate, and the machine learning algorithm comprises a classifier trained using random initial guesses of weights and biases of neurons in a deep neural network, through which training samples were fed through layers thereof to calculate predicted sparse vector ECG-PPG signals to find a class label for ECG-PPG signals and noise interference, and a cost function used to measure difference between the predicted sparse vector ECG-PPG signals and desired outputs, with application of gradient descent and backpropagation over multiple training samples to minimize the cost function through adjustment of the weights and biases, arriving at weights that configure the classifier to eliminate the motion artifacts, and accurately derive the estimated noiseless sparse vector ECG and PPG signals, in step (d).
2 . The method of claim 1 wherein the one or more biosensors comprise a plurality of biosensors, and the noise data and the noise factors further comprise data overlap from the plurality of sensors.
3 . The method of claim 1 wherein, when the compressed noisy signal is wirelessly received from the one or more biosensors, and the noise data and the noise factors further comprise ambient noise.
4 . The method of claim 1 wherein the multiple physiological parameters comprise at least three of said skin temperature, oxygen saturation, blood pressure, heart rate and respiration rate.
5 . The method of claim 1 wherein the multiple physiological parameters comprise at least four of said skin temperature, oxygen saturation, blood pressure, heart rate and respiration rate.
6 . The method of claim 1 wherein the multiple physiological parameters comprise all five of said skin temperature, oxygen saturation, blood pressure, heart rate and respiration rate.
7 . The method of claim 1 wherein the one or more wearable devices comprise a smartwatch.
8 . The method of claim 1 wherein the one or more wearable devices comprise a smart patch.
9 . The method of claim 1 wherein the one or more wearable devices comprise a smartwatch and a smart patch.
10 . The method of claim 1 wherein said low power wireless communication protocol is Bluetooth Low Energy (BLE).
11 . The method of claim 1 wherein said mobile edge computing device is a smartphone.
12 . The method of claim 1 wherein the one or more wearable devices comprise at least one of a smartwatch and a smart patch, and said mobile edge computing device comprises a smartphone.
13 . The method of claim 12 wherein said low power wireless communication protocol is Bluetooth Low Energy (BLE).
14 . The method of claim 1 further comprising wireless transmission of the estimated physiological parameters by and from the mobile edge computing device to a remote patient-monitoring entity.
15 . The method of claim 1 further comprising monitoring said estimated noiseless quantifications of the physiological parameters for changes therein over time, and alerting the actively mobile patient upon detection of changes indicative of a health concern.
16 . The method of claim 14 further comprising monitoring said estimated noiseless quantifications of the physiological parameters for changes therein over time, and alerting the actively mobile patient upon detection of changes indicative of a health concern.
17 . The method of claim 14 further comprising monitoring, by said remote patient-monitoring entity, of said estimated noiseless quantifications of the physiological parameters for changes therein over time, and alerting the actively mobile patient upon detection of changes indicative of a health concern.
18 . The method of claim 15 wherein said health concern is a transmissible virus.
19 . The method of claim 16 wherein said health concern is a transmissible virus.
20 . The method of claim 17 wherein said health concern is a transmissible virus.