Back to all papers

Machine learning identification of neuroimaging signatures in young and middle-aged male patients with obstructive sleep apnea: a multimodal MRI study.

June 20, 2026pubmed logopapers

Authors

Wang J,Qiu J,Wang Z,Ji L,Li Y,Wang Q,Chen R

Affiliations (7)

  • Department of Respiratory and Critical Care, The Second Affiliated Hospital of Soochow University, 1055 Sanxiang Road, Suzhou, China.
  • Department of Sleep Centre, The Second Affiliated Hospital of Soochow University, 1055 Sanxiang Road, Suzhou, China.
  • Oxford University and Oxford University Hospitals NHS Foundation Trust, Oxford, England.
  • Department of Radiology, The Second Affiliated Hospital of Soochow University, 1055 Sanxiang Road, Suzhou, China.
  • Department of Respiratory and Critical Care, The Second Affiliated Hospital of Soochow University, 1055 Sanxiang Road, Suzhou, China. [email protected].
  • Department of Sleep Centre, The Second Affiliated Hospital of Soochow University, 1055 Sanxiang Road, Suzhou, China. [email protected].
  • The Second Affiliated Hospital of Soochow University, 1055 Sanxiang Road, Suzhou, 215004, China. [email protected].

Abstract

This study aimed to develop and validate an integrated neuroimaging-based model for identifying severe obstructive sleep apnea (OSA) with severe oxygen desaturation (ODI ≥ 30 events/hour) in young and middle-aged male. A key goal was to characterize the neuroimaging signatures associated with the hypoxic burden of OSA. Data from 111 patients were utilized. The initial feature set integrated 50 Gy matter structural indices, seven dynamic functional connectivity (dFC) temporal metrics, and key demographic factors (age, BMI). A Random Forest algorithm was employed for feature selection based on variable importance measures (VIM), followed by Support Vector Machine (SVM) modeling to classify severe OSA patients (ODI ≥ 30 events/hour). The final model incorporated 28 features (2 demographic, 2 dFC, 24 structural). It achieved an accuracy of 70.91%, recall of 95.07%, precision of 72.25%, F1-score of 81.68%, and an AUC-ROC of 0.798. Feature selection at a VIM > 0.05 threshold (20 features) increased accuracy by 2.7% and achieved optimal recall (97.50%) and AUC (0.813). This study establishes a neuroimaging-based framework for identifying OSA patients with severe oxygen desaturation (ODI ≥ 30 events/hour), integrating gray matter structural and dynamic functional connectivity features. The features selected by the model overlap with brain regions and networks that support cognitive functions such as memory and attention, suggesting that the model captures neurobiologically meaningful patterns associated with the hypoxic burden of OSA.

Topics

Sleep Apnea, ObstructiveMagnetic Resonance ImagingMachine LearningNeuroimagingJournal Article

Ready to Sharpen Your Edge?

Subscribe to join 11k+ peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.