A Deep Convolutional Neural Network Framework for Multiclass Skin Lesion Classification Using Dermoscopic Images

Abstract

Early identification of malignant skin lesions can support timely clinical assessment and treatment. Dermoscopic images contain subtle variations in color, texture, border structure, and lesion morphology that are difficult to characterize consistently using handcrafted image features. This study investigates a convolutional neural network (CNN)-based framework for automated multiclass skin-lesion classification. The original study uses dermoscopic images, image resizing and normalization, data augmentation, convolutional feature extraction, pooling, dropout, and fully connected classification. The manuscript reports an overall test accuracy in the mid-80% range. A major objective of the revised study is to make the experimental protocol reproducible and the reported performance internally consistent. In particular, the dataset identity, class distribution, train/test protocol, CNN architecture, hyperparameters, and evaluation metrics must be stated explicitly. The proposed framework is intended as a computer-vision decision-support tool rather than a replacement for dermatological diagnosis. The revised presentation emphasizes class-wise precision, recall, F1-score, confusion-matrix analysis, reproducibility, limitations, and responsible clinical interpretation. The work is positioned within the image-processing and computer-vision scope of IJCNIS and is designed to provide a transparent baseline for future lightweight and explainable skin-lesion analysis systems.