Scalable Segmentation of Diabetic Foot Ulcers

The treatment of Diabetic Foot Ulcers (DFUs) is challenging due to their aggressive nature and slow healing, which can result in serious complications such as infection, gangrene, and amputation. Multiclass segmentation of tissues related to DFUs plays a crucial role in the early identification and effective treatment of these complications. Although Convolutional Neural Networks (CNNs) have been widely and successfully used in this context, they face challenges such as class imbalance and high computational demand. To resolve these issues, we propose Multiple Segmentation by Binary Networks (MsBNet), a modular approach based on combining multiple binary segmenters. We evaluate four different architectures: ShuffleNet, UNet, Attentio-nUNet, and PsPNet on our DFUs dataset. The results indicate that MsBNet outperforms DFUs multiclass segmentation models in terms of scalability and ability to deal with class imbalances. It is a flexible and promising solution for expanding medical applications, especially in environments with limited computational resources.