ERCU-Net: Advancing Skin Lesion Segmentation Through Efficient and Accurate Deep Learning Model

Dermoscopy is a key technique for examining skin lesions, a primary risk factor for skin cancer, by providing detailed visual information. However, manual examination is often time-consuming and suffers from inter-observer variability, which is dependent on the dermatologist's skill. To overcome these limitations, automated segmentation using deep learning has emerged as a promising solution. Recent research has focused on enhancing the accuracy and efficiency of deep learning models for dermatological imaging. Developing an efficient, low-parameter model is particularly crucial for deployment on devices with limited computational resources, making the technology more accessible and cost-effective. This study addresses this need by proposing a novel model, the ERCU-Net, designed for high accuracy and efficiency. The methodology involved constructing the ERCU-Net architecture and training it on the ISIC 2018 dataset for 60 epochs, including dataset processing and hyperparameter optimization. Experimental findings demonstrate that the proposed model outperforms previous methods in skin lesion segmentation. This research presents a novel, effective, and efficient approach for automated detection of skin lesions. By aiding in accurate segmentation, the ERCU-Net model can support medical professionals in the early prevention and treatment of skin cancer, demonstrating significant potential for clinical application.