KEY TAKEAWAYS
- The study aimed to evaluate ViT-based deep learning for better skin lesion detection and classification based on their spread ratio.
- Results indicated that ViTs and GradCAM significantly enhanced the accuracy of skin lesion detection and classification.
Skin cancer is a prevalent disease, and early detection significantly improves treatment outcomes. Deep learning techniques have emerged as valuable tools for assisting clinicians in identifying and classifying skin lesions. Vision transformers (ViT), with their multiclass prediction capabilities, offer a promising approach for accurate skin lesion detection.
Muhammad Shafiq and the team aimed to develop a novel ViT Gradient-Weighted Class Activation Mapping (GradCAM) based architecture, named ViT-GradCAM, for the detection and classification of skin lesions by analyzing the lesion’s spread ratio on the skin surface area.
Researchers used the HAM 10000 dataset, comprising 10,015 dermatoscopic images of 7 different skin lesion types, to train and validate the ViT-GradCAM model. Preprocessing and data augmentation techniques were employed to address class imbalance issues and enhance model performance.
The ViT-GradCAM model achieved an accuracy of 97.28%, precision of 98.51%, recall of 95.2%, and an F1 score of 94.6% in classifying the dermatoscopic images into the 7 skin lesion categories.The study demonstrated that ViT-GradCAM surpasses the performance of other state-of-the-art deep learning models in skin lesion detection and classification. The architecture also allowed for visualization of the specific pixels within regions associated with skin pathologies.
The study proposed that using ViTs and GradCAM together provided a viable solution for accurate skin lesion detection and classification. The ViT-GradCAM model showed promise as a tool to support clinical experts, enabling earlier diagnosis and potentially improving patient outcomes.
This study was conducted independently, without support from any sponsors or funders.
Source: https://pubmed.ncbi.nlm.nih.gov/39221858/
Shafiq M, Aggarwal K, Jayachandran J, et al. (2024). “A novel Skin lesion prediction and classification technique: ViT-GradCAM.” Skin Res Technol. 2024;30(9):e70040. doi:10.1111/srt.70040
