Deep Learning for Dermatology: An Innovative Framework for Approaching Precise Skin Cancer Detection

A research team led by Mohammad Tahmid Noor has published a new framework for precise skin cancer detection using deep learning models, with their DenseNet201 architecture achieving 93.79% accuracy on a binary classification task. The paper, accepted for the 16th International IEEE Conference on Computing, Communication and Networking Technologies (ICCCNT 2025), compares two prominent convolutional neural network architectures—VGG16 and DenseNet201—for differentiating benign from malignant skin lesions.

The technical approach involved processing 3,297 skin lesion images resized to 224x224 pixels through both models. While both architectures provided excellent performance, DenseNet201 demonstrated superior accuracy in this binary classification task. The researchers specifically focused on evaluating model accuracy and computational efficiency, with implications for early detection systems that could assist dermatologists in clinical workflows.

This work builds on growing applications of computer vision in medical diagnostics, where AI-assisted tools can help address the global challenge of skin cancer detection. With millions diagnosed annually worldwide, such frameworks could potentially improve screening efficiency and accessibility. The authors note there's still room for improvement and plan to work with new datasets to achieve even greater accuracy in future iterations of their research.