Analysis of upper airway CT-based radiomics in adult obstructive sleep apnea

TL;DR

This study shows that CT-based radiomics can effectively predict obstructive sleep apnea in adults using machine learning models.

Contribution

The study introduces a CT-based radiomics approach using machine learning for OSA prediction in adults.

Findings

The NaiveBayes algorithm achieved the best performance with AUCs of 0.819, 0.812, and 0.854 for different regions.

The Entire radiomic model outperformed other models in OSA prediction with an AUC of 0.854.

The model showed good predictive calibration and clinical application value in both training and test sets.

Abstract

Computed tomography (CT) is a key tool for evaluating the upper airway in adult patients with obstructive sleep apnea (OSA). This study aimed to assess the value of CT-based radiomics for OSA evaluation.

A total of 79 OSA patients and 19 healthy controls (HCs) were recruited between January 2023 and June 2024 and underwent upper airway CT scans. Radiomic features were extracted from CT images, and data processed using Python. Radiomic models were then developed to evaluate and predict OSA using ten machine learning algorithms. Model performance was assessed using area under the curve (AUC) values, calibration, and decision curve analysis (DCA).

The NaiveBayes machine learning algorithm based on radiomic features achieved the best result, and the AUCs for the Airway, Soft Tissue, and Entire in the test sets were 0.819, 0.812, and 0.854, respectively. In the test set, the Entire radiomic model was performed better than the other two models in OSA prediction with an AUC of 0.854 (95% CI, 0.674–1.000). The performance of Entire radiomic model was confirmed in the training and test set with satisfying predictive calibration and clinical application value.

Upper airway CT-based radiomics model appears to be a promising tool. This radiomics-based method may be convenient and efficient for OSA assessment and prediction.

Figures