Enhanced Sleep Apnea Diagnosis Using Hybrid CNN-LSTM Framework and Machine Learning Classifier

The common sleep disorder known as sleep apnea is defined by frequent breathing pauses brought on by blockages in the airways. This condition disrupts sleep and increases the risk of serious health problems, such as cardiovascular illnesses. Traditional diagnostic approaches, such as polysomnography (PSG), are costly and involve complex procedures, often requiring patients to spend the night in specialized medical facilities which limits accessibility for many individuals. This study presents a novel deep learning framework for employing a hybrid architecture for sleep apnea detection. The model uses Long Short-Term Memory (LSTM) networks to describe time-dependent patterns associated with apnea episodes and Convolutional Neural Networks (CNN) to extract key features from ECG data. This combined approach leverages CNN's feature extraction capability and LSTM's strength in sequence learning to achieve improved detection performance. To further classify sleep apnea severity into four categories- normal, mild, moderate, and severe a Random Forest algorithm is employed. The model is tuned on a labelled ECG data, and its performance is evaluated using key metrics to ensure high accuracy and reliable generalization. To enhance usability, a web-based application has been developed, allowing users to upload ECG data and receive diagnostic insights in an accessible format. This hybrid approach of CNN and LSTM, coupled with the Random Forest classifier, offers a cost- effective and scalable alternative for the early detection and assessment of sleep apnea severity. By providing an efficient solution, this system can significantly improve health benefits, while alleviating the burden on healthcare infrastructure. Its versatile design ensures suitability for both clinical use and at- home monitoring, making it a practical tool for comprehensive sleep health management.