Sleep Apnea Detection with Neural Networks

This work presents a system for detecting sleep apnea using data from a portable polygraph device, enabling studies to be conducted at the home of the patient. The system implements a hybrid neural network composed of a 1D Convolutional Neural Network (CNN-1D) and Long Short-Term Memory (LSTM) for apnea detection. Pre- and post-processing tasks are automated to correct potential errors based on deterministic medical criteria, leading to an estimated rate of respiratory events. This estimate serves as a preliminary diagnosis to assist doctors in evaluating the severity of each case. Additionally, the system incorporates an innovative temporal reconstruction process, generating detailed graphs that visualize the distribution of detected apneas throughout the night.