A Unified Framework for Burn Area Segmentation Using Recurrent Attention and Memory-Augmented Networks

Abstract Burn injury region segmentation is a critical task in medical image analysis, especially for assessing the extent of injuries and aiding in treatment planning. Despite the advancement of convolutional neural networks in semantic segmentation, accurately delineating burn regions remains challenging due to the complex and variable nature of burn injuries. To address this, we propose an end-to-end deep learning framework that leverages a recurrent neural network (RNN) architecture integrated with an attention mechanism to sequentially identify and segment burn regions in medical images. The model is designed to process input images in a step-by-step manner, focusing on the most relevant areas of the image and progressively refining the segmentation. By simulating a human-like attention mechanism, the network is able to isolate burn regions while maintaining detailed boundaries, thus improving the precision of segmentation. This approach enables the automatic and robust identification of burn injury areas, facilitating more accurate diagnosis and personalized treatment recommendations.