Automated Wound Segmentation using Attention Mechanism based on Enhanced MobileNetV2

Wound healing is a morphogenetic reaction to injury that restores physiological and anatomical function. It is crucial in fields like obesity treatment, cancer, trauma, burns, and diabetic foot ulcer (DFU) care. Wound care experts rely on affordable, high-resolution cell phone cameras for image documentation; however, lack of information can hinder proper wound management, documentation, and diagnosis, highlighting the need for improved therapeutic techniques and better automated image segmentation methods. Recent deep learning approaches for automatic feature extraction from images face challenges due to limited training data and low neural networks depths, which hinder accuracy by failing to capture complex patterns. This research introduces an improved MobileNetV2 model which utilized transfer learning, two additional dense layers, and fusion with spatial attention mechanism to improve the accuracy of the wound segmentation task. The proposed approach achieved a dice score of 92.40% on the Foot Ulcer Dataset and 96.92% on the Medetec Wound Dataset. In comparison to existing state-of- the-art deep learning-based VGG16, SegNet, U-Net, and MobileNetV2+CCL techniques, the proposed attention-based MobileNetV2 technique obtained the highest dice score.