Predicting Therapeutic Response to Hypoglossal Nerve Stimulation Using Deep Learning.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in medicine
• Sleep medicine

Background:

• Obstructive sleep apnea (OSA) is a common disorder.
• Hypoglossal nerve stimulator (HGNS) implantation is an effective treatment for moderate to severe OSA.
• Predicting HGNS efficacy is crucial for patient selection.

Purpose of the Study:

• To develop and validate machine learning (ML) and deep learning (DL) models.
• To predict the therapeutic efficacy of HGNS implantation using drug-induced sleep endoscopy (DISE) images.
• To identify optimal image subsets (base of tongue vs. velopharynx) for predictive modeling.

Main Methods:

• Trained six DL and five ML models on DISE images from 127 patients (responders vs. non-responders).
• Utilized images from the base of tongue (BOT) and velopharynx (VP).
• Evaluated model performance using precision, recall, F1 score, and overall accuracy.

Main Results:

• Models trained on VP images demonstrated higher accuracy than BOT or combined sets.
• The VCG-16 DL model achieved the highest performance on VP images (accuracy 0.833, F1 0.78, recall 0.883).
• Logistic regression was the top-performing ML model (accuracy 0.685, F1 0.813).

Conclusions:

• DL models show potential for predicting HGNS therapeutic efficacy from DISE images.
• Velopharynx imaging contains key features for predicting treatment success.
• Future multi-institutional datasets are needed for generalizable predictive models.