Identifying Venous Insufficiency in Head and Neck Reconstruction Flaps Using Machine Learning and Deep Learning...

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Artificial Intelligence

Background:

• Venous insufficiency is a significant risk factor for flap failure in head and neck cancer reconstruction.
• Early detection of venous insufficiency is crucial for successful patient outcomes.
• Artificial intelligence (AI) presents a promising avenue for improving the early detection of flap complications.

Purpose of the Study:

• To evaluate the efficacy of machine learning and deep learning models in detecting venous insufficiency in head and neck flaps.
• To identify key image features indicative of venous insufficiency using AI-based visualization techniques.
• To establish AI as a valuable auxiliary tool for postoperative flap monitoring.

Main Methods:

• Retrospective analysis of clinical data and postoperative flap images from head and neck cancer patients (2018-2024).
• Development and comparison of eight machine learning classifiers and three deep learning models (ResNet, GoogleNet, Densenet).
• Utilized SHAP and Grad-CAM for feature analysis and model interpretability.

Main Results:

• A dataset of 2575 flap images from 576 patients was analyzed.
• The Random Forest model achieved 90.25% accuracy, with Hue_mean and Green_median identified as key features.
• The ResNet deep learning model demonstrated superior performance with 95.23% accuracy, 84.81% sensitivity, 97.27% specificity, and an AUC of 0.940.

Conclusions:

• Deep learning models show significant potential for accurately identifying flap venous insufficiency.
• AI-based tools can serve as effective auxiliary aids for postoperative monitoring of head and neck flaps.
• The findings support the integration of AI in clinical practice for enhanced patient care and reduced flap failure rates.