DCA-U-Net: A Deep Learning Network for Segmentation of Laser-Induced Thermal Damage Regions in Mouse Skin OCT Images

Abstract Laser-induced thermal injury is a common form of skin damage in clinical treatment, and accurately assessing the extent of injury and treatment efficacy is crucial for patient recovery. In recent years, deep learning models have been increasingly applied to the automatic segmentation of skin injury regions. However, existing methods often suffer from a large number of parameters, leading to a significant decline in segmentation accuracy when reducing the number of model parameters, thus limiting their clinical applicability. To address this issue, we propose an efficient and lightweight segmentation model, Dilated ConvNeXT Attention U-Net (DCA-U-Net), based on U-Net. By incorporating the more efficient Dilated ConvNeXT Block (DCB) and Dual Module Attention Block (DMAB), DCA-U-Net significantly reduces the number of parameters while simultaneously improving feature extraction capability and segmentation accuracy. Compared to the standard U-Net, our model reduces the number of parameters by 33%. Experimental results on two different sections of mouse skin laser thermal damage Optical Coherence Tomography (OCT) datasets show that our model has better segmentation performance with insufficient or sufficient amount of data. These improvements not only enhance the model's ability to accurately identify skin thermal injury regions, but also substantially reduce computational costs while maintaining high segmentation accuracy, offering promising technical support for the precise diagnosis and treatment of skin laser thermal injuries.