Abstract

Abstract- Diabetic Foot Ulcers (DFUs) pose significant healthcare challenges, often resulting in severe complications and amputations. Timely and accurate assessment of wound severity is crucial for effective treatment and improved patient outcomes.  This study presents an innovative approach to the automated detection of vascular disorders in diabetic feet utilizing infrared thermography. The study employs a dataset comprising 710 thermal images acquired from Al_Wafa Specialized Center for Diabetes and Endocrinology and Bartella General Hospital, which has been meticulously curated for research purposes. To address the specific challenges encountered in diabetic foot imaging and diagnosis, we introduce three tailored Convolutional Neural Network (CNN) architectures, each designed to identify different risk levels associated with diabetic foot complications. Low-level Diabetic Foot Infection Network (LL-DFI-Net): This architecture is aimed at the detection of low-risk groups, Medium-Level Diabetic Foot Infection Network (ML-DFI-Net) Designed for the medium-risk group, this model targets neuropathic individuals who do not exhibit signs of ischemia. High-Level Diabetic Foot Infection Network (HL-DFI-Net): Focused on the high-risk group, this architecture is specialized for detecting individuals with ischemia. The three networks have achieved high accuracy, as follows: LL-DFI-Net achieved 96.7213%, ML-DFI-Net recorded 97.916%, HL-DFI- reached 100%. These models aim to improve the detection and management of diabetic foot disorders, leading to better patient outcomes and healthcare strategies.

Statistics