To develop a deep learning-ensemble model to enhance the performance of dermatoscopic differential diagnosis between malignant melanoma and other malignant skin lesions, thereby improving diagnostic accuracy.
Key Findings:
All nine models significantly distinguished between malignant melanoma and other malignant skin lesions (p<0.001), indicating strong diagnostic capability.
AUCs on the test dataset ranged from 0.921 to 0.967, suggesting high accuracy in model predictions.
The XGBoost ensemble model achieved an AUC of 1.00 on the training dataset and 0.988 on the test dataset, demonstrating excellent performance.
Interpretation:
The integration of deep learning and ensemble learning strategies significantly improves the accuracy of dermatoscopic melanoma differential diagnosis, indicating potential for clinical application.
Limitations:
Existing models may suffer from training data bias, which can affect generalizability.
Limited external validation on large-sample, multi-center datasets poses challenges for clinical applicability.
Potential challenges in real-world clinical application due to variability in dermoscopy images must be addressed.
Conclusion:
The study demonstrates that combining deep learning and ensemble learning can enhance melanoma detection accuracy, providing a valuable tool for clinical practice and improving patient outcomes.
