Machine learning approach for obstructive sleep apnea screening using brain diffusion tensor imaging.

Area of Science:

• Neuroimaging
• Machine Learning
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) is linked to significant health issues and mortality.
• Current diagnostic methods for OSA are complex, time-consuming, and involve lengthy waits.
• Early screening and intervention are crucial for managing OSA and improving patient outcomes.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models, specifically Support Vector Machine (SVM) and Random Forest (RF), in classifying OSA using brain Diffusion Tensor Imaging (DTI) data.
• To determine if faster and less complicated ML models can provide accurate OSA classification compared to traditional methods.
• To assess the potential of DTI-based ML models as a rapid screening tool for OSA.

Main Methods:

• Collected DTI data from 59 patients with OSA and 96 healthy controls using a 3.0-T MRI scanner.
• Calculated mean diffusivity maps from DTI data, realigned, averaged, and normalized to a common space.
• Trained and validated SVM and RF models using cross-validation to classify OSA from control subjects.

Main Results:

• The Random Forest (RF) model achieved 0.73 classification accuracy and an Area Under the Curve (AUC) of 0.85.
• The Support Vector Machine (SVM) model demonstrated comparable performance with 0.77 accuracy and 0.84 AUC.
• Both RF and SVM models showed similar statistical fitness to the DTI data, indicating comparable performance.

Conclusions:

• Machine learning models, particularly RF and SVM, can effectively classify obstructive sleep apnea using brain DTI data.
• These DTI-based ML models offer a potentially faster and less complex alternative for OSA screening.
• The findings suggest that DTI data combined with ML can aid in the early identification of OSA, potentially reducing diagnostic delays.