Related Experiment Video
Updated: Jun 28, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.8K
Optimizing vitiligo diagnosis with ResNet and Swin transformer deep learning models: a study on performance and
Fan Zhong1, Kaiqiao He2, Mengqi Ji1
1College of Electrical Engineering, Sichuan University, Chengdu, China.
Scientific Reports
|April 21, 2024
Summary
Deep learning models can significantly improve vitiligo diagnosis. The Swin Transformer Large model demonstrated superior accuracy and interpretability, aiding dermatologists in detecting this skin condition.
Area of Science:
- Dermatology
- Medical Imaging
- Artificial Intelligence
Background:
- Vitiligo is a hypopigmented skin disease causing melanin loss, requiring expert dermatological assessment.
- A shortage of specialized dermatologists challenges timely and accurate vitiligo diagnosis.
- Deep learning models offer a potential solution to enhance diagnostic accuracy and efficiency.
Purpose of the Study:
- To develop and evaluate deep learning models for improved vitiligo detection.
- To compare the performance of ResNet and Swin Transformer models for vitiligo classification.
- To assess model interpretability by visualizing highlighted diagnostic regions.
Main Methods:
- Comparative analysis of five deep learning models: ResNet (ResNet34, ResNet50, ResNet101) and Swin Transformer (Base, Large).
- Uniform testing conditions were applied to all models.
- Class activation maps and feature maps were used to evaluate model interpretability.
Main Results:
- The Swin Transformer Large model achieved the highest classification performance with an AUC of 0.94, accuracy of 93.82%, sensitivity of 94.02%, and specificity of 93.5%.
- Visualizations confirmed that highlighted regions corresponded to vitiligo lesions, indicating diagnostic relevance.
- Feature map visualization provided insights into model mechanisms, aiding interpretability and potential performance tuning.
Conclusions:
- Deep learning, particularly the Swin Transformer Large model, shows significant potential to enhance vitiligo diagnostic accuracy and efficiency.
- The model's interpretability features support clinical decision-making in dermatology.
- Further research is warranted to fully leverage deep learning in medical diagnostics.
