The CSMwear patch was tested in human control subjects during periods of relaxation and stress induced through mental arithmetic, as shown in the video demonstration.
Figure a shows the derived stress from the heart’s kinetic energy for the CSMwear patch, which is contrasted with the inconclusive stress score shown in Figure b produced by the commercial Muse 2 device during the same time period.
Figure c shows a representative wavelet scalogram of the heart vibrations being fed into a residual convolutional neural network, which was used to classify periods of stress. The algorithm block diagram is shown in Figure d.
Neural activations for the deep learning algorithm are shown for the stress (e) and non-stress (f) classes, showing a high ability to distinguish between classes. Finally, Figure g shows the overall device accuracy to be 94% in classifying between periods of motion, stress, and relaxation.
