Related Experiment Videos
Hybrid Vision Transformer-CNN Architecture with Optimized Feature Selection for Skin Cancer Classification
Abrar Almjally1, Munazza Aziz2, Shaheryar Najam3
1Information Technology Department, College of Computer and Information Sciences, Imam Mohammad Ibn Saud Islamic University (IMSIU), Riyadh 13318, Saudi Arabia.
Diagnostics (Basel, Switzerland)
|August 13, 2026
Summary
This study introduces a unified framework for accurate automated skin cancer classification, significantly improving early melanoma detection. The novel approach integrates preprocessing, segmentation, and hybrid feature learning for enhanced diagnostic reliability.
Area of Science:
- Dermatology
- Medical Imaging
- Computer Science
Background:
- Melanoma is a dangerous skin cancer requiring early diagnosis for better outcomes.
- Automated skin lesion classification is difficult due to image variations and artifacts.
Purpose of the Study:
- To develop a unified framework for robust multi-class skin cancer classification.
- To enhance early skin cancer detection through improved automated analysis.
Main Methods:
- An end-to-end architecture integrating artifact removal, lesion segmentation (CutisNet), and hybrid feature learning (MobileNetV2 + handcrafted descriptors).
- Gray Wolf Optimization for feature refinement and a hybrid GNN-CNN classifier for classification.
Main Results:
- The framework achieved a maximum classification accuracy of 96.7% on the PH2 dataset.
- Outperformed existing state-of-the-art methods on multiple benchmark dermoscopic datasets.
Conclusions:
- The proposed framework enhances the robustness and accuracy of automated skin cancer classification.
- It shows potential for reliable computer-aided diagnosis and early skin cancer detection support.