Abstract

Background: Melanoma, though rare, is highly fatal due to rapid metastasis. Early detection improves survival rates significantly. Higher Human Development Index (HDI) correlates with better detection and outcomes. Advances in artificial intelligence (AI) and deep learning offer improved accuracy and efficiency in diagnosing melanoma, addressing the need for better methods. Methods: This study utilized a dataset from the International Skin Imaging Collaboration (ISIC), comprising over 23,000 dermatoscopic images. A subset of 640 images (512 for training and 128 for testing) was used to evaluate a proposed convolutional neural network (CNN) architecture. The images were pre-processed to enhance features and remove artifacts, followed by lesion segmentation and classification using the CNN. Performance was compared with k-nearest neighbors (KNN) and support vector machines (SVM). Results: The CNN demonstrated superior performance in detecting and classifying melanoma compared to KNN and SVM. The architecture, consisting of convolutional and pooling layers followed by fully connected layers, achieved high accuracy in distinguishing between benign and malignant lesions. Pre-processing steps, including artifact removal and color enhancement, were crucial in improving detection accuracy. Conclusion: This study highlights the effectiveness of deep learning techniques, particularly CNNs, in melanoma detection. The findings support the integration of AI-driven methods in clinical practice to aid dermatologists in early and accurate detection of melanoma. Future research should focus on refining these techniques and expanding their application to broader datasets.

Keywords: Melanoma, Deep Learning, Convolutional Neural Networks (CNN), Image Processing, Artificial Intelligence (AI)