AI-Driven Enhancement of Skin Cancer Diagnosis: A Two-Stage Voting Ensemble Approach Using Dermoscopic Data
Tsu-Man Chiu1,2, Yun-Chang Li2, I-Chun Chi3
1School of Medicine, Chung Shan Medical University, Taichung 402, Taiwan.
Cancers
|January 11, 2025
Summary
This study developed an AI diagnostic model to improve skin cancer detection. The two-stage AI approach significantly reduced misclassifications of malignant lesions, enhancing diagnostic accuracy for melanoma and other skin cancers.
Area of Science:
- Dermatology
- Medical Imaging
- Artificial Intelligence
Background:
- Skin cancer, including deadly melanoma, is a global health concern.
- Current detection methods are effective but costly and time-consuming.
- Artificial intelligence (AI) shows promise in improving skin cancer diagnosis.
Purpose of the Study:
- To develop and validate an AI diagnostic model for skin cancer detection.
- To enhance diagnostic accuracy and reduce false negatives for malignant lesions.
- To assess the model's performance across different ethnic datasets.
Main Methods:
- Utilized datasets from ISIC and CSMU Hospital for model development.
- Fine-tuned eight pre-trained AI models, including CNNs and vision transformers.
- Implemented a two-stage classification strategy with an ensemble model for improved accuracy.
Main Results:
- Significantly reduced false negatives for malignant lesions in the ISIC dataset.
- Completely eliminated false negatives for malignant cases in the CSMUH dataset.
- Demonstrated improved diagnostic precision and reduced misclassification rates.
Conclusions:
- The AI model successfully distinguishes between melanoma, non-melanoma skin cancer, and benign cases.
- A two-stage classification strategy effectively minimizes false negatives in malignant lesions.
- AI holds potential as a valuable clinical tool to improve skin cancer diagnosis and reduce mortality, especially in resource-limited settings.
