A MULTI-CLASS SKIN CANCER CLASSIFICATION USING OPTIMIZED CNN FOR DEEP SKIN HEALTHCARE SYSTEM

Abstract

Skin cancer is the most common type of cancer worldwide, and early and precise detection is important
for patient survival. Clinical examination of skin lesions is critical, but it is fraught with difficulties such as extended
wait times and subjective interpretations. Every year, around a million people are diagnosed with the condition in
India alone. As the disease advances, the rate of survival falls drastically. To address these issues and assist
dermatologists in making more accurate diagnoses, deep learning algorithms have been created. Deep learning
algorithms can enhance diagnosis speed and accuracy, resulting in early detection and treatment. This study's
objective was to develop strong deep learning (DL) prediction models for the categorization of skin cancer by first
addressing a common severe class imbalance problem and other preprocessing work, then building an improved
model for prediction, and finally proposing an end-to-end smart healthcare system via a web application. We tackle
this issue by utilizing ISIC's HAM10000 dataset. It comprises 10015+ dermoscopic images that are freely available
to the public. In this study, we will use convolutional neural networks (CNN) to detect and categorize seven different
types of skin cancer using historical clinical imaging data. Building an optimized CNN model to accurately diagnose
skin cancer, lowering the false negative prediction rate, and performing data visualization are some of our main
goals for this study. The study will demonstrate how CNNs have the capacity to accurately categorize various skin
cancers, which can aid in the detection and improvement of a patient's prognosis.