Interpretable deep learning method to predict wound healing progress based on collagen fibers in wound tissue

Area of Science:

• Histopathology
• Computational Biology
• Dermatology

Background:

• Collagen fiber dynamics are vital for wound healing assessment, clinical treatment, and drug screening.
• Existing methods for analyzing collagen spatial patterns lack criteria for stratifying healing periods or detecting delays.
• A novel classification method for wound healing status based on collagen fibers is needed.

Purpose of the Study:

• To develop a deep learning method for classifying wound healing time points and delayed healing using histological images.
• To enhance model interpretability by identifying tissue regions driving predictions.

Main Methods:

• Fine-tuning a pre-trained VGG16 model for image classification.
• Employing an interpretable framework combining LayerCAM and Guided Backpropagation for visual analysis.
• Utilizing histological images of skin tissue for classification.

Main Results:

• Achieved 85% accuracy in a five-class classification task (normal skin, wound skin at 0, 3, 7, 10 days).
• Attained 78% accuracy in a three-class task (normal skin, wound skin at 0 days, diabetic wound skin at 10 days).
• The interpretable framework accurately localized collagen fibers without pixel-level annotations, confirming classification based on collagen regions.

Conclusions:

• The deep learning method accurately predicts wound healing time points and delayed healing using collagen fiber features.
• The model offers visual interpretability, increasing clinician trust in its predictions.
• This approach promises more precise and effective wound treatment practices.