Skin Cancer Detection Using Convolutional Neural Networks

Abstract

Skin cancer remains one of the most prevalent and potentially fatal forms of cancer worldwide, highlighting the urgent need for early, accurate, and scalable diagnostic methods. This project proposes a deep learning-based solution for automated skin cancer classification using Convolutional Neural Networks (CNNs) trained on the HAM10000 dataset—a benchmark collection of dermatoscopic images representing seven distinct skin lesion types, including melanoma, basal cell carcinoma, and benign nevi. The framework incorporates robust image preprocessing techniques and a customized CNN architecture designed to optimize feature extraction and classification performance across diverse lesion categories. To further enhance model generalization and address potential class imbalance, the project explores data augmentation strategies tailored for medical imagery. A user-friendly interface, developed using Streamlit, enables real-time inference and accessibility for both clinical and non-specialist use. Experimental results demonstrate high classification accuracy and strong differentiation between malignant and benign lesions, supporting the system’s utility as a reliable, cost-effective, and accessible diagnostic aid. This work underscores the significant role of AI-powered tools in augmenting dermatological decision-making, especially in resource-constrained environments where timely diagnosis can substantially impact patient outcomes.

Keywords: deep learning, Streamlit Interface, Convolutional Neural Networks (CNN), Image Classification, Skin Cancer Detection, HAM10000, Melanoma, Dermatoscopic Images, Medical Image Analysis, Data Augmentation, Automated Diagnosis, Artificial Intelligence in Healthcare, Early Detection, Lesion Classification, Medical AI Tools