AUTHOR=Leema A. Anny , Balakrishnan P. , Gopichand G. , Rajarajan G. 

TITLE=LMS-ViT: a multi-scale vision transformer approach for real-time smartphone-based skin cancer detection

JOURNAL=Frontiers in Artificial Intelligence

VOLUME=Volume 8 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2025.1612502

DOI=10.3389/frai.2025.1612502

ISSN=2624-8212

ABSTRACT=Skin cancer is the abnormal growth of skin cells. It occurs mostly in skin exposed to sunlight. To prevent the occurrence of skin cancer, avoid exposing skin to ultraviolet radiation. Skin cancer can be very harmful if found very late. Traditional convolutional neural networks (CNNs) face challenges in fine-grained lesion classification due to their limited ability to extract detailed features. To overcome such limitations, we introduced a novel approach in the form of a lightweight multi-scale vision transformer (LMS-ViT) application for the automated detection of skin cancer using dermoscopic images and the HAM10000 dataset. Unlike CNNs, LMS-ViT employs a multi-scale attention mechanism to capture both global lesion structures and fine-grained textural details, improving classification accuracy. This study combines skin images from the HAM10000 dataset with pictures taken using a smartphone. It uses a compact method to mix important features, which makes the system faster and suitable for real-time use in medical apps. The proposed system enables real-time skin cancer classification via a smartphone camera, making it portable and platform-independent. Experimental results show that LMS-ViT surpasses CNN-based models across all skin lesion categories, achieving 90% accuracy, an 18% improvement over CNN, while reducing computational cost by 30%. LMS-ViT also improves precision, recall, and F1-score, particularly in complex categories such as Vasc (0.96 to 1.01) and Nv (0.94 to 1.01), demonstrating superior classification power. With real-time android implementation, LMS-ViT offers accessible, mobile-friendly diagnostics for early skin cancer detection.