Identification of Difficult Laryngoscopy via Double-Pose Feature Clustering Network in Patients with Cervical Spondylosis
Intelligent Computing · 2026-01-01 · 5
DOI: 10.34133/icomputing.0321 ↗At a glance
Journal article in Intelligent Computing, published online on 2026-03-17. Written by Mingya Wang and 9 co-authors. 36 references. Funded by: Capital’s Funds for Health Improvement and Research, Innovation and Transformation Project, Peking University Third Hospital.
Abstract (original language)
Background
The identification of difficult laryngoscopy is a critical skill for anesthesiologists, particularly in high-risk procedures such as cervical spine surgery, which demands heightened accuracy in preoperative airway assessment. This study developed a deep learning algorithm, the double-pose feature clustering network (DPFCNet), designed to improve the identification of patients with difficult laryngoscopy through comprehensive analysis of neutral and extended cervical spine x-ray images.
Methods
In this prospective cohort study, 14,407 patients who underwent elective cervical spine surgery under general anesthesia (July 2016 to July 2023) were initially enrolled. Following rigorous screening, 1,568 participants were eligible and included in the analysis, comprising 319 difficult laryngoscopy cases and 1,249 easy laryngoscopy cases. The study evaluated the effectiveness of a deep learning model for identifying difficult laryngoscopy that was developed using a ResNet-based feature extraction module with dual-position image fusion technology and systematically compared it against conventional bedside assessments using area under the receiver operating characteristic curve (AUC) analysis.
Results
Conventional clinical indicators showed poor predictive capacity for difficult laryngoscopy: thyromental distance (AUC = 0.594), neck circumference (AUC = 0.662), inter-incisor gap (AUC = 0.607), and modified Mallampati test (AUC = 0.624). Their combination achieved AUC = 0.743. In contrast, DPFCNet considerably outperformed conventional methods with AUC = 0.866, demonstrating superior discriminative power.
Conclusion
This study employed a deep learning model for the comprehensive analysis of both neutral and extended cervical spine x-ray images, considerably improving the prediction accuracy of difficult laryngoscopy. DPFCNet provides an effective decision-support tool for preoperative risk stratification in cervical spondylosis patients.
Funding
- Capital’s Funds for Health Improvement and Research
- Innovation and Transformation Project, Peking University Third Hospital
Cite
APA
Wang, M., Jia, F., Yang, C., Fang, J., Li, J., Tian, Y., Li, M., Guo, X., Wei, C., & Han, Y. (2026). Identification of Difficult Laryngoscopy via Double-Pose Feature Clustering Network in Patients with Cervical Spondylosis. Intelligent Computing, 5. https://doi.org/10.34133/icomputing.0321
Vancouver
Wang M, Jia F, Yang C, Fang J, Li J, Tian Y, et al. Identification of Difficult Laryngoscopy via Double-Pose Feature Clustering Network in Patients with Cervical Spondylosis. Intelligent Computing. 2026;5. doi:10.34133/icomputing.0321
BibTeX
@article{Wang2026,
author = {Wang, Mingya and Jia, Fei and Yang, Cuiying and Fang, Jingchao and Li, Jiao and Tian, Yang and Li, Min and Guo, Xiangyang and Wei, Changwei and Han, Yongzheng},
title = {Identification of Difficult Laryngoscopy via Double-Pose Feature Clustering Network in Patients with Cervical Spondylosis},
journal = {Intelligent Computing},
year = {2026},
volume = {5},
doi = {10.34133/icomputing.0321}
}
EUVIMED