Melanoma skin cancer detection using deep learning-based lesion segmentation

Extreme caution should be exercised while dealing with any form of skin cancer, but especially malignant melanoma. It’s also on the rise, especially among whites who spend a lot of time outdoors in the sun. Because early diagnosis of melanoma may be beneficial and curative, it is crucial that it be detected at an early stage to improve survival rates. Accurate automatic skin lesion segmentation is in high demand due to the rapid proliferation of skin cancer. While deep learning models like CNN were widely utilized to enable proper segmentation, current encoder-decoder designs based on compactly connected networks (DenseNet) and residual networks (ResNet) were applied for skin lesion tasks. Complex parameter settings, a lack of multi-scale data, and an absence of appropriate information in pre-trained features all have an effect on the performance of skin lesion segmentation. This research proposes a system for segmenting skin lesions utilizing the UNet and Residual UNET (ResUNet) architectures to solve these issues. CNNs, which consist of encoders and decoders, form the basis of these designs. The architecture employs UNet and ResUNet to guarantee high-quality lesion segmentation at all times. The proposed models are evaluated on ISIC2018 and HAM10000 lesion pictures. Accuracy, dice coefficient, Jaccard index, sensitivity, and specificity are used to assess the performance of the models. The models’ efficacy is evaluated in light of state-of-the-art approaches.