The Effect of Image Preprocessing Algorithms on Diabetic Foot Ulcer Classification

Diabetic foot ulcers significantly affect patient health and healthcare costs, making an accurate diagnosis crucial. This research examines the impact of image preprocessing algorithms on accurately diagnosing and classifying diabetic foot ulcers using the ResNeXt50 classifier. We introduce new strategies for automatic detection of three conditions, i.e., out-of-focus, poor lighting conditions, and the existence of artifacts. For each condition, we identify suitable image preprocessing algorithms. Comparative analysis against baseline performance metrics revealed notable improvements with various preprocessing techniques. Canny Edge Detection notably enhanced the AUC of out-of-focus conditions, while Adaptive Histogram Equalisation and Gaussian Sharpening also showed positive outcomes for poor lighting conditions. Wavelength-based Denoising showed mixed results for artifacts. Overall, preprocessing algorithms improved diabetic foot ulcer classification performance, suggesting their potential integration into associated classification workflows. Recommendations include ongoing algorithmic evaluation and broader application in medical imaging. This study emphasises the vital role of image preprocessing in enhancing diabetic foot ulcer classification accuracy, with the potential to improve wound care and monitoring in real-world scenarios.