DFU Infection and Ischemia Classification: PSO-Optimized Deep Learning Networks

Abstract

The recognition of infection and local perfusion (i.e., ischemic) status of diabetic foot ulcer (DFU) on a regular and timely basis is crucial to promote wound healing and prevent the development of unwanted complications. The conventional DFU assessment method is limited to scheduled clinic visits, impeding close monitoring of foot lesion progression and its chronicity. This paper presents an efficient Particle Swarm Optimization (PSO)-incorporated framework for classifying DFU infection and ischemia conditions using three deep learning models: AlexNet, GoogleNet, and EfficientNet-B0. The optimized system performed well in all evaluation metrics, ranging between 0.82 and 0.92 and near-perfect scores of 0.97 - 1, respectively, indicating the high performance and robustness of the system for the DFU infection and ischemia classification tasks. These results are better than the recent related studies using the same datasets. This system performs competitively with the deeper and heavier Efficient-B5 model, suggesting the efficiency of the proposed strategy without demanding an extensive network exploration process or elaborative feature selection process. The future of this work includes transferring the technology for DFU management using a mobile-based technology platform to improve outpatient care delivery through rapid recognition of DFU infection and their perfusion to optimize limb salvage outcomes.

Keywords

• diabetic foot ulcer
• efficientnet
• infection
• ischemia
• pso