Interpreting Venous and Arterial Ulcer Images Through the Grad-CAM Lens: Insights and Implications in CNN-Based Wound...

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Wound diagnostics

Background:

• Chronic lower extremity wounds, including arterial and venous ulcers, pose significant clinical and economic challenges.
• Accurate differential diagnosis is crucial for effective treatment of these complex wounds.

Purpose of the Study:

• To evaluate advanced deep learning models for classifying arterial versus venous ulcers.
• To visualize the decision-making processes of these AI models using Grad-CAM.

Main Methods:

• Retrospective analysis of 607 lower extremity ulcer images (198 arterial, 409 venous).
• Training and evaluation of five convolutional neural networks: ResNet50, ResNeXt50, ConvNeXt, EfficientNetB2, and EfficientNetV2.
• Performance metrics included accuracy, precision, recall, F1-score, and ROC-AUC; Grad-CAM for visualization.

Main Results:

• High classification performance observed across models, with accuracy ranging from 72% to 98% (ResNeXt50).
• Strong discrimination between arterial and venous ulcers demonstrated by precision and recall metrics.
• EfficientNetV2 achieved the highest precision, indicating robust classification capabilities.

Conclusions:

• AI-assisted classification presents a valuable tool for improving diagnostic efficiency in differentiating venous and arterial ulcers.
• Deep learning models show promise in supporting clinical decision-making for chronic wound management.