Interpretation of Diabetic Foot Ulcer Image Classification Using Layer Attribution Algorithms

The rise of deep learning has led to the development of intricate neural networks with diverse layers for analyzing various data types. While these deep neural networks excel in tasks involving unstructured data such as images, text, and audio, their inherent complexity poses challenges in terms of interpretability. Unlike traditional machine learning models (e.g., decision trees, random forests) that are considered white-box models and are easy to interpret, deep neural networks often function as black-box models. Ensuring interpretability of these models is essential to comprehend the reasoning behind predictions and the contributing factors. This research employs Captum, a Python library, to elucidate predictions made by a PyTorch-based diabetic foot ulcer (DFU) image classification network.