Atrial Fibrillation and Sinus Rhythm Detection Using TinyML (Embedded Machine Learning)

Given the various technologies used to measure and detect cardiac arrhythmia, this project proposes using TinyML (Embedded Machine Learning) for atrial fibrillation and normal sinus rhythm classification. The machine learning model is going to be deployed in a microcontroller to bring a small, efficient, and straightforward prototype for the desired purpose. The proposed architecture of the neural model was composed of convolutional networks (CNN), where the input data from the PTB-XL database went through some pre-processing steps, such as filtering and dividing the temporal records into individual heartbeats. The prototype execution in the embedded environment was developed and carried out using the ESP32 development board. The results obtained verified that the model reached an overall accuracy of 94.1% and 94.04% in the training and test stages, respectively. In contrast, it got an overall accuracy of 99.33% in the microcontroller prototype inference, with data extracted from an advanced patient simulator that reproduces different cardiac signals.