Prediction of Obstructive Sleep Apnea Using Hypothalamic Radiomics and Machine Learning.

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Sleep Medicine

Background:

• Obstructive sleep apnea (OSA) is a prevalent condition with significant health implications.
• Current diagnostic methods for OSA can be invasive or inconvenient.
• Identifying novel, non-invasive biomarkers for OSA prediction is crucial.

Purpose of the Study:

• To investigate hypothalamic radiomics from T1-weighted MRI as a potential biomarker for OSA prediction.
• To evaluate the performance of radiomics-based models, alone and combined with clinical factors, in identifying OSA patients.

Main Methods:

• 251 participants (127 OSA, 124 controls) underwent 3D T1-weighted MRI.
• Automated segmentation of hypothalamic subunits and extraction of radiomics features using PyRadiomics.
• Feature selection via statistical tests and regression, followed by training and validation of seven classifiers.

Main Results:

• Over 4000 radiomics features were extracted, with 52 selected for analysis.
• A Gradient Boosting Machine model combining radiomics and clinical data (BMI) achieved the highest AUC (0.808 internal, 0.777 external).
• Specific wavelet and first-order radiomics features from the posterior hypothalamus were identified as key predictors.

Conclusions:

• Hypothalamic radiomics, particularly when integrated with clinical data, presents a promising exploratory approach for OSA prediction.
• Radiomics analysis can reveal subtle hypothalamic changes associated with OSA, offering potential for improved diagnostic strategies.