TITLE:
Automated Classification of Dermoscopic Images Using a Convolutional Neural Network: Improving Diagnostic Accuracy in Skin Cancer Detection
AUTHORS:
Eiman Abniah Alamrani, Mohammad Hujooj
KEYWORDS:
Skin Neoplasms, Dermoscopy, Convolutional Neural Networks, Artificial Intelligence, Image Interpretation, Computer-Assisted
JOURNAL NAME:
Journal of Computer and Communications,
Vol.14 No.6,
June
18,
2026
ABSTRACT: Skin cancer remains a major global health concern with rising incidence. Early diagnosis is essential for improved prognosis; however, existing diagnostic methods such as dermoscopy are dependent on clinician expertise and are subject to inter-observer variability. This study aimed to develop and evaluate a convolutional neural network (CNN) for automated classification of dermoscopic images into benign and malignant lesions, and to assess its diagnostic performance in comparison with conventional methods and established deep learning models. A quantitative experimental design was conducted using the International Skin Imaging Collaboration dataset. Images were preprocessed (224 × 224 resizing, normalization, color correction) and augmented through rotation, flipping, zoom, brightness variation, and noise injection. Data were divided into training (70%), validation, and testing sets, with five-fold cross-validation. The CNN architecture included three convolutional layers (32, 64, 128 filters), max pooling, 50% dropout, and a fully connected layer. Training used the Adam optimizer (learning rate 0.001), batch size of 32, and early stopping. Our model achieved an accuracy of 90.3%, sensitivity of 92.1%, specificity of 88.5%, and AUC-ROC of 0.95. The confusion matrix indicated 276 true positives, 245 true negatives, 35 false positives, and 28 false negatives. The model outperformed conventional dermoscopic accuracy (75% - 84%) and demonstrated superior performance compared to VGG-16 and VGG-19 architectures. The proposed CNN demonstrated high diagnostic performance and improved classification consistency. It has potential as a decision-support tool for early skin cancer detection, particularly in resource-constrained settings.