Skin Cancer Detection Through Images Using Deep Convolutional Neural Networks

Abstract

Skin cancer is a prevalent and potentially life-threatening disease. Early detection and accurate diagnosis are crucial for effective treatment and improved patient outcomes. This research focuses on developing a skin cancer detection system using Deep Convolutional Neural Networks (CNNs) and achieving high accuracy. The study utilizes a dataset of dermoscopic images, specifically the ISIC 2020 dataset containing 10,015 images of skin lesions. Pre-processing techniques, including resizing, cropping, enhancement, and normalization, are applied to improve the dataset’s quality and consistency. The proposed CNN model is built upon the InceptionV3 architecture and leverages transfer learning for feature extraction. The training process involves dividing the pre-processed dataset into training, validation, and testing subsets. The model’s parameters are iteratively optimized using a loss function and backpropagation algorithm to maximize accuracy. Performance evaluation of the model is conducted through rigorous comparisons with existing techniques using standard evaluation metrics. The primary focus is on achieving accurate skin cancer detection, particularly in the early stages. The study also explores the scalability of the model to larger datasets and assesses its performance against advanced techniques. The results demonstrate that the developed CNN-based skin cancer detection system achieves an impressive accuracy of 92%. This research contributes to the advancement of automated skin cancer detection systems, offering the potential for improved patient outcomes and reduced healthcare burdens associated with delayed diagnoses.