Deriving Heart Rate and Respiratory Rate from Pulse Oximetry Using Neural Networks

The inductive belt placed around the abdomen and ribcage is vital at point of care testing to determine the respiratory rate (RR). Nowadays, extraction has become more convenient with the invention of portable and smart devices. Different algorithms have been explored in various studies; however, the downside is that the prototype devices tend to be sold by reputed companies at a higher price point when the actual electronics and resources used are affordable and readily available at the market. This led to the customization of pulse oximeters that provides heart rate (HR) and RR in one compact device. The extraction of RR from HR was done using an Artificial Neural Network (ANN) using a pulse oximeter programmed in Python language. The selected five features, namely, time of peaks, peaks, time of valleys, valley, and time since the last peak, provides a proper estimation of RR. The neural network helped in the classification of labels of valleys such as inspiration, expiration, and neutral. Overall, the MSE computed for the HR and respiration rate was 4.93 and 0.95, respectively, versus a medical-grade device.