A Hybrid CNN-SEU-Net Approach for Improved Melanoma Detection in Dermoscopy Images : Combining CNN and SEU-Net for Robust Melanoma Detection

Melanoma is a highly aggressive form of skin cancer with high mortality rates if not diagnosed early. Accurate diagnosis poses significant challenges, particularly for nonspecialist healthcare providers, due to the complexity and variability of skin lesions. Current diagnostic methods often lack the precision needed for reliable early detection, highlighting the need for innovative solutions. This paper proposes a novel hybrid deep learning framework for melanoma detection that integrates SEU-Net and Convolutional Neural Networks (CNNs) to address these challenges. The framework begins with SEU-Net, an encoder-decoder segmentation model enhanced with skip connections and SE attention mechanisms, to generate a segmentation mask that identifies the region of interest (ROI) in dermoscopic images. The segmentation mask is multiplied with the original image to isolate the lesion area, creating a segmented image that minimizes irrelevant background information. This segmented image is then passed to a CNN, classifying the lesion as malignant or benign. The proposed approach was evaluated on the ISBI 2016 dataset, achieving an accuracy of 93.60\%, outperforming existing state-of-the-art methods. These results demonstrate the potential of the proposed framework to significantly improve the precision and reliability of melanoma diagnosis, enabling earlier detection and ultimately reducing mortality rates.