Difficult Airway Assessment Based on Multi-View Metric Learning.

Area of Science:

• Anesthesiology
• Medical Imaging
• Artificial Intelligence

Background:

• Preoperative assessment of difficult airways is critical for safe anesthesia intubation.
• Existing difficult airway recognition algorithms suffer from low accuracy and reliability.
• Multi-view facial imaging offers potential for improved difficult airway assessment.

Purpose of the Study:

• To develop and validate a novel deep learning network, DMF-Net, for accurate prediction of difficult airways using multi-view facial images.
• To enhance feature representation and information fusion from multi-view data for improved diagnostic performance.
• To provide a reliable AI-based auxiliary tool for clinical healthcare professionals.

Main Methods:

• Proposed a Dual-Path Multi-View Fusion Network (DMF-Net) utilizing multi-view metric learning.
• Employed a dual-path structure to extract features from frontal and lateral facial images.
• Incorporated Multi-Scale Feature Fusion and Hybrid Co-Attention Modules, along with consistency, complementarity, and Focal losses.

Main Results:

• DMF-Net achieved high performance metrics: 77.92% accuracy, 75.62% specificity, 82.50% sensitivity, and 71.35% F1-score.
• The method demonstrated superior accuracy and reliability compared to clinical bedside screening and existing AI-based approaches.
• Effectively avoided information bias through the use of specialized loss functions.

Conclusions:

• DMF-Net offers a significant advancement in the preoperative assessment of difficult airways.
• The proposed network can reliably identify patients at risk of difficult airways, aiding clinical decision-making.
• Implementation of DMF-Net can potentially reduce the incidence of anesthesia-related complications and improve patient outcomes.