Detection of Pre-Processed Skin Cancer Images into Benign/Malignant Class with Lightweight Deeplearning

Timely recognition and handling of skin cancer (SC) is crucial to reduce its severity. Clinical-level detection of SC is conducted through manual examination, and succeeded by Digital Dermoscopy Image (DDI) assessment. This study proposes a solution utilizing a Lightweight Deep-Learning (LDL) scheme to automate the DDI inspection procedure. The phases of this study encompass; DDI collection and resizing, feature extraction utilizing the LDL model, and classification and performance validation by 3 -fold cross-validation. This study analyzed 8000 images, comprising 4000 benign and 4000 malignant specimens. Initially, unprocessed images are analyzed, resulting in classification accuracy over 89 %. All images are subsequently pre-processed utilizing Shannon's entropy and the Brownian Butterfly Algorithm (BBA) to enhance the image pixels. The experimental inquiry utilizing Shannon's enhanced images attained accuracy over 96 % with the selected DDI database. This study verifies that the preprocessed image enhances accuracy in the DDI.