Development of Predictive CNN Based Model for Vital Signs Alerts
R. F. Gibadullin, N. S. Marushkai
Abstract
This article addresses the development of an automated system for reading and analyzing radar readings for discovering biological signals behind it. The idea behind that research is to utilize neural network technologies, and especially convolution neural networks (CNN) to make a process of biological signal analysis automatic and fault-tolerant. It was examined if the CNNs are applicable for such kind of task and their ability to be trained for the prediction of such vital signs. This involves heart-rate from raw radar readings data. To find this out a ResNet50V2 model with custom input and output layers was taken with its weights pre-trained on ImageNET. The model was fine-tuned using the real radar readings data. It was revealed that the CNN model can approximate radargrams to actual heart-rate readings. This research confirmed a guess that CNNs can be a very useful tool for radar reading analysis and prediction of an individual's vital signs according to this data and can be used to build systems based on such predictions.