Skin Cancer Identification using CNN: A Lightweight Model with Fine-Grained Classification

Melanoma has a mortality rate that is far greater than that of other types of cancer, it is often regarded as the most dangerous form of skin cancer that can be found anywhere in the world. While there have been numerous advances in the field of computer-aided melanoma detection, the scarcity of images poses a formidable obstacle for robust deep learning models to detect melanoma. To distinguish and isolate melanoma from nervous disease, this study introduces an intelligent Region of Interest framework utilizing a transfer learning approach to detect and identify melanoma. The enhanced k-means algorithm is used to identify ROIs from images, allowing for the identification of discriminating elements by training only on images containing melanoma cells, thus allowing for the identification of ROIs. Using a "Convolutional Neural Network based transfer learning model" that incorporates data enhancement methods, the purpose of this research is to construct a model that is capable of processing ROI pictures from the DermIS and DermQuest datasets. Notable accuracy rates are shown by the framework that has been presented. These rates are 70.4% and 91.4%, respectively, for "DermIS and DermQuest", respectively. Particularly noteworthy is the fact that the ROI-based transfer learning technique outperforms other approaches that are already in use and depend on entire pictures for categorization.