Abstract:
Diabetes, one of the world's most common diseases, poses a significant threat to overall
health. Among its major health concerns is the depletion and weakening of bones, leading to
conditions like Charcot Foot or Diabetic Foot. Despite the global prevalence of diabetes and
its impact on millions of individuals, our understanding of its effects on bone health remains
limited. Addressing this knowledge gap becomes increasingly crucial which forms the basis
of this study. Diabetic foot cases often suffer from deteriorating bone health, resulting in
increased bone curvature, mechanical instability, and porosity.
This study seeks to investigate mechanics associated with diabetic foot, its analysis and
employs Artificial intelligence (AI) effectively to address early prognosis of this disease.
Generative Adversarial Networks (GANs) is used as a preferred methodology in this work to
produce synthetic data since obtaining rich data in these cases is challenging specially from
the hospitals in Pakistan. A total of 560 images were synthetically reproduced to assess
reliability. The reliability of this data was then evaluated using Fréchet’s Inception Distance
(FID) and Inception Score (IS) to address validation. These Images were used to train
multiple U-net models, which were further assessed and compared using the Intersection
over Union (IOU) metric.
Our study findings demonstrate the potential of AI-generated synthetic images in U-net
models for accurately identifying the advancement of bone damage in diabetic feet. Notably,
the Recurrent Residual (R2) U-Net model outperforms other models by effectively detecting
the progression of the disease on real x-ray images, achieving a significant average IOU
score of 0.75 which is in coherence with the literature published. These results hold valuable
implications for clinicians, as they can utilize our findings for early prognosis of this condition,
facilitating timely intervention and management strategies.
