Deep neural network (DNN) assisted sensor for energy-efficient electrocardiogram (ECG) monitoring
View Patent ↗The present invention is directed to an energy-efficient method of monitoring a physiological signal while maintaining high accuracy. The method may comprise a Deep Neural Network (DNN) receiving an uncompressed sample of a continuous ECG signal from a sensor. The method may further comprise the DNN determining a first probability that the received sample is abnormal and a second probability that the received sample is normal. Finally, the method may further comprise the DNN determining to transmit the uncompressed sample if a threshold of abnormality is less than or equal to the difference between the first probability and the second probability. In some embodiments, the DNN may be a Convolutional Neural Network (CNN).
1. An energy-efficient method of monitoring a continuous physiological signal while maintaining high accuracy, the method comprises:
a) receiving, by a convolutional neural network (CNN), one or more uncompressed samples of the continuous physiological signal from a sensor;
wherein the sensor comprises a memory unit comprising the CNN;
b) determining, by the CNN, a first probability that the each uncompressed sample of the one or more uncompressed samples is abnormal and a second probability that the uncompressed sample is normal;
c) determining whether or not to transmit the one or more abnormal uncompressed samples using a threshold strategy that utilizes the first probability and the second probability; and
d) transmitting the one or more abnormal uncompressed samples based on said determination;
wherein the CNN has been trained with a dataset comprising at least 205 records;
wherein the CNN has 91,920 parameters;
wherein the CNN comprises a 64-channel 1-dimensional convolution filter having a filter length of 64, a batch normalization layer, a rectified linear activation layer, a dropout layer set to a 25% dropout rate, and a softmax layer.
2. The method of claim 1 , wherein the dataset comprises records of a plurality of heart cycles.
3. An energy-efficient system of monitoring an electrocardiogram (ECG) sensor while maintaining high accuracy, wherein the system comprises:
a) the ECG sensor comprising a memory unit comprising a convolutional neural network (CNN);
b) a transmitter;
c) a processor;
d) a Random Access Memory (RAM); and
e) an external memory, wherein the external memory comprises instructions for:
i) receiving, by the CNN, one or more uncompressed samples of a continuous ECG signal from the ECG sensor,
ii) determining, by the CNN, a first probability that the each uncompressed sample of the one or more uncompressed samples is abnormal and a second probability that the uncompressed sample is normal,
iii) determining whether or not to transmit the one or more abnormal uncompressed samples using a threshold strategy that utilizes the first probability and the second probability; and
iv) transmitting the one or more abnormal uncompressed samples based on said determination;
wherein the CNN has been trained with a dataset comprising at least 205 records;
wherein the CNN has 91,920 parameters;
wherein the CNN comprises a 64-channel 1-dimensional convolution filter having a filter length of 64, a batch normalization laver, a rectified linear activation layer, a dropout layer set to a 25% dropout rate, and a softmax layer.
4. The system of claim 3 , wherein the dataset comprises records of a plurality of heart cycles.
5. A non-transitory computer-readable storage medium having computer-executable instructions recorded thereon for an energy-efficient convolutional neural network (CNN)-driven electrocardiogram (ECG) monitor that, when executed by at least one processing circuit, perform a computer process, wherein the computer process comprises:
a) receiving, by a CNN, one or more uncompressed samples of a continuous ECG signal from a sensor;
wherein the sensor comprises a memory unit comprising the CNN;
b) determining, by the CNN, a first probability that the each uncompressed sample of the one or more uncompressed samples is abnormal and a second probability that the uncompressed sample is normal;
c) determining whether or not to transmit the one or more abnormal uncompressed samples using a threshold strategy that utilizes the first probability and the second probability; and
d) transmitting the one or more abnormal uncompressed samples based on said determination;
wherein the CNN has been trained with a dataset comprising at least 205 records;
wherein the CNN has 91,920 parameters;
wherein the CNN comprises a 64-channel 1-dimensional convolution filter having a filter length of 64, a batch normalization layer, a rectified linear activation layer, a dropout layer set to a 25% dropout rate, and a softmax layer.
6. The non-transitory computer-readable storage medium of claim 5 , wherein a duration of the uncompressed sample comprises a plurality of heart cycles.