Impact of Image Pre-Processing on Skin Cancer Classification Melanoma Through Deep Learning

Melanoma is a type of cancer that originates from melanocytes, which begins when melanocytes grow uncontrollably and form malignant tumors. Although rare, melanoma is highly lethal, causing 75% of deaths from skin cancer, despite accounting for only 4% of skin cancer cases. Therefore, an automated approach is needed to classify this disease. In melanoma classification, hair on the skin can interfere with image analysis by obstructing the details and shapes of the skin areas that need to be identified. To address this issue, a preprocessing scheme using the skin hair removal method is employed, which utilizes morphological features to recognize and remove hair contours, then restore the lost details with inpainting techniques. However, this process typically reduces the image quality and makes it slightly blurry. Therefore, this study also proposes the use of CLAHE (Contrast Limited Adaptive Histogram Equalization) preprocessing to enhance the sharpness of image details. The test results show that the developed model can effectively distinguish between normal skin and melanoma. The accuracy of the images after preprocessing is higher compared to the original images. The testing was evaluated based on accuracy, precision, recall, specificity, and F-score. For the original images, the accuracy for the ResNet-34 architecture is 94% and for EfficientNet-B0 it is 98.6%. Meanwhile, for the images with CLAHE preprocessing, the accuracy for both ResNet-34 and EfficientNet-B0 architectures reached 100%.