Accurate Detection of Basal Cell Carcinoma Using Fuzzy U-Net and Deep Learning on Dermoscopic Images

Basal cell carcinoma (BCC) is a common kind of skin cancer that is distinguished by the presence of telangiectasias, which are tiny blood veins that resemble trees and are frequently seen inside skin lesions. Precise recognition of these characteristics is essential for prompt diagnosis and successful therapy. Deep learning (DL) models have advanced in skin cancer imaging in recent years, increasing the accuracy of diagnosis and feature segmentation. The ISIC 2019 dataset, a comprehensive collection of dermoscopic pictures covering a variety of skin lesions, including BCC, was employed in our suggested approach. Our approach started with applying a Wiener filter to denoise the images. This preprocessing step significantly improved image quality, making critical features more discernible and facilitating subsequent analysis. After denoising, we implemented the Fuzzy U-Net model for image segmentation. This model excels at accurately delineating lesions, providing precise boundaries that are essential for effective classification. A deep neural network (DNN) was then trained using the segmented pictures, enabling it to differentiate basal cell carcinoma from other skin diseases by identifying important characteristics. We used common assessment measures including precision of 97%, F1 score of 98%, and AUC of 0.99% and testing accuracy of 97.47% to assess our model's performance. The outcomes show that our method is reliable and successful, with a high level of accuracy in detecting basal cell cancer. In addition to expediting the diagnostic procedure, this approach may enhance patient outcomes by enabling earlier discovery and treatment.