Predi̇cti̇ng the severi̇ty of obstructi̇ve sleep apnea usi̇ng arti̇fi̇ci̇al intelli̇gence tools

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Sleep Medicine

Background:

• Obstructive sleep apnea syndrome (OSAS) poses significant health risks.
• Accurate prediction of OSAS severity is crucial for timely intervention.
• Current diagnostic methods can be resource-intensive.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) model for predicting OSAS severity.
• To identify key clinical and polysomnographic features associated with OSAS severity.
• To explore the potential of AI in enhancing OSAS diagnosis and risk assessment.

Main Methods:

• Utilized a dataset of 750 inpatients with 20 attributes, including demographic, medical history, anthropometric, and polysomnography (PSG) data.
• Preprocessed data using min-max scaling and Synthetic Minority Over-sampling Technique (SMOTE) to balance classes, increasing dataset to 1250.
• Developed a multilayer artificial neural network (ANN) model and evaluated performance with k-fold cross-validation; performed information gain analysis for feature importance.

Main Results:

• The ANN model demonstrated high accuracy in predicting OSAS severity (AUC: 0.966, CA: 0.880).
• Information gain analysis identified Epworth Sleepiness Scale, lowest nighttime oxygen saturation, percentage of sleep time with 80-90% oxygen saturation, and neck thickness as significant predictors.
• These features are critical risk factors for early OSAS detection and management.

Conclusions:

• AI-based models are effective tools for predicting OSAS severity.
• This research supports the development of advanced diagnostic tools for OSAS.
• AI can accurately assess OSAS severity using overnight pulse oximetry and other risk factors in suspected cases.