Deep Learning Architectures for Skin Lesion Classification: A Comparative Study of CNN, Transformer and Hybrid Models

View at publisher Summarise with VIRAT
Authors: Darshan P, Karthika. R, Rajith K. S, Sriharish V J, Vishal Srivatsan S, Gurukarthi K C, Gurukumar K C
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Skin cancer is a highly prevalent and potentially lethal cancer, demanding precise and timely diagnosis. This study presents a comparative evaluation of four deep learning architectures applied for the classification of seven classes of dermoscopic skin lesions in the HAM10000 dataset: an InceptionV4-DenseNet201 ensemble, a PatchNet-ResNet50 ensemble, a Vision Transformer (ViT), and a MaxViT-Hybrid model. Model performance was compared according to class precision, recall, F1-score and overall accuracy. The MaxViTHybrid achieved the highest macro-averaged F1-score (0.85) and an overall accuracy of 0.90, demonstrating a good performance over all types of lesions including challenging classes like DF, VASC and AKIEC. The InceptionV4DenseNet201 ensemble achieved the highest overall accuracy$(0.91)$and weighted-average metrics$(0.91)$. These results indicate that hybrid transformer-based models with spatial feature fusion hold promise for reliable and generalizable automated diagnosis of skin cancer.

Fields of Study & Subjects

Cutaneous Melanoma Detection and Management AI in cancer detection Advanced Neural Network Applications

Publication details

Journal / source
International Conference on Intelligent Systems for Communication, IoT and Security (ICISCoIS)
Pages
1–6
Publication date
8 January 2026
Language
EN