Recurrent Neural Network-Based Concept for the Skin Cancer Classification Model

Abstract

Identification and categorization of skin lesions are crucial steps in the diagnosis of skin
cancer. Current deep learning-oriented CAD techniques continue to struggle with difficult
skin diseases with complicated characteristics such fuzzy borders, the existence of artifacts,
low background contrast, and sparse training datasets. They also depend strongly on the
proper tuning of millions of variables, which frequently results in over-fitting, subpar
generalization, and significant resource consumption. The data collecting, pre-processing,
feature extraction, and classification steps make up the developed skin cancer
classification system. The information is initially acquired through web resources.
Additionally, mean filtering is used to complete the pre processing. The features are then
taken from CNN's pooling layer during the feature extraction stage. Additionally, RNN does
the classification in the last stage, classifying the output into Angioma, Nevus, Lentigo NOS,
Solar Lentigo, Melanoma, Seborrheic Keratosis, and BCC.