Apnea Detection in Polysomnographic Recordings Using Machine Learning Techniques.

Area of Science:

• Medical Technology
• Artificial Intelligence in Healthcare
• Sleep Medicine

Background:

• Polysomnography is the standard for diagnosing sleep disorders, but manual analysis of records is time-consuming.
• Accurate and efficient detection of sleep apnea and desaturation events is crucial for patient management.
• Existing automated methods often lack comprehensive analysis of key polysomnography channels.

Purpose of the Study:

• To develop an automated system for evaluating airflow and SpO2 signals in polysomnography records.
• To detect sleep apnea and desaturation events using machine learning techniques.
• To compare the performance of a convolutional neural network (CNN) against the k-nearest neighbors (k-NN) algorithm.

Main Methods:

• A convolutional neural network (CNN) was designed and optimized for hyperparameter tuning.
• The CNN model was trained and tested using a large database of nasal airflow and blood oxygen saturation (SpO2) signals.
• Performance was evaluated by comparing the CNN with the k-nearest neighbors (k-NN) classification method.

Main Results:

• The CNN achieved an 84% accuracy for apnea detection and 74% for desaturation detection.
• The k-NN classifier achieved 83% accuracy for apnea detection and 64% for desaturation detection.
• The proposed CNN demonstrated superior performance, particularly in desaturation detection, compared to k-NN.

Conclusions:

• An automated CNN-based system can effectively detect sleep apnea and desaturation from polysomnography data.
• This approach offers a more efficient alternative to manual record evaluation in sleep laboratories.
• Further development could enhance diagnostic capabilities for various sleep disorders.