Comparative Study of Deep Learning Methods through Improvement of Skin Image Quality of Basal Cell Carcinoma

Basal cell carcinoma (BCC) or commonly known as basal cell cancers is the most common non-melanoma skin disease in the world. BCC is the result of an interaction between a combination of genetic and environmental factors that are directly related to prolonged exposure to ultraviolet (UV) rays, such as the neck and head area. This study proposes to use the CLAHE (Contrast Limited Adaptive Histogram Equalization) method on skin images to improve image quality. CLAHE is an image quality improvement method that aims to obtain images with better contrast than the original image without reducing the quality of the image itself. The experimental results show that the CLAHE image are more accurate than the original image. For quantitative measurements, the performance is seen from the value of accuracy, precision, recall, specificity, and F-score. To be more specific, the experiments utilized the well-known public dataset called ISIC 2018 Dataset (for BCC image), while the normal skin used our own USK-Normal Skin dataset. The results for original image using VGG-16 architecture is 98,3%, ResNet-34 is 97,5%, EfficientNetB0 is 98,2%, and EfficientNet-B7 is 99,7%, respectively. Meanwhile, in the CLAHE image, the accuracy values obtained are higher, for the VGG-16 is 100%, ResNet-34 99.5%, EfficientNetB0 100% and EfficientNet-B7 100%, respectively. The normal skin dataset is available through this link: https://ademifi.github.io/USK-NORMAL-SKIN-DATASETI.