Related Experiment Videos
Deep learning-based multi-class classification of cutaneous lesions for dermatological assessment
1Department of Information Technology, Chennai Institute of Technology, Chennai, India.
Cutaneous and Ocular Toxicology
|July 19, 2026
Summary
This study introduces a deep learning framework for skin cancer detection, achieving high accuracy in classifying cutaneous lesions. The AI model aids in early diagnosis, potentially improving patient outcomes.
Area of Science:
- Dermatological Oncology
- Artificial Intelligence
- Medical Imaging Analysis
Background:
- Skin cancer is a prevalent malignancy globally, necessitating early detection for better treatment outcomes.
- Current diagnostic methods, including visual and dermoscopic analysis, can be time-consuming and vary between observers.
- Artificial intelligence (AI) offers potential for developing computer-aided diagnostic systems to support dermatologists.
Purpose of the Study:
- To propose a deep learning framework for multi-class classification of cutaneous lesions from dermoscopic images.
- To enhance the accuracy and reliability of AI-driven skin cancer diagnosis.
Main Methods:
- A deep learning model integrating Vision Transformer (ViT) for global features and SE-ResNet for local features.
- An adaptive attention-based fusion mechanism to combine complementary representations.
- Hyperparameter optimization using an Improved Crocodile Optimization Algorithm (ICOA) for enhanced stability.
Main Results:
- The framework achieved high classification accuracies: 99.05% on HAM10000, 98.31% on ISIC 2019, and 99.17% on PH2 datasets.
- Demonstrated robustness and generalizability across multiple benchmark datasets.
Conclusions:
- The proposed AI framework shows significant potential as a clinical decision-support tool.
- It can aid in the early detection and improved diagnosis of skin cancer.
- The integration of ViT, SE-ResNet, and ICOA offers a promising approach for dermatological AI applications.
