To develop an AI framework that enhances differential diagnosis of melanocytic neoplasms using a multi-modal dataset, which combines imaging and expert knowledge.
Top-1 accuracy of 0.699 for forty-class classification, indicating robust performance across multiple categories.
0.915 ROC AUC for few-shot whole-slide image tasks, showcasing effectiveness in limited data scenarios.
0.925 AUPRC for fully supervised whole-slide image tasks, reflecting high precision in predictions.
Improvements of up to 13.8% over linear and fully finetuned methods, underscoring the framework's superiority.
Interpretation:
The findings indicate that integrating a query database with a knowledge-enhanced AI framework can significantly enhance the performance of existing pathology models without the need for fine-tuning, potentially transforming diagnostic practices.
Limitations:
The dataset is proprietary and may limit broader accessibility, hindering widespread adoption and validation.
Performance may vary with different datasets or clinical settings, necessitating further validation in diverse environments.
Conclusion:
Melan-Dx represents a significant advancement in the differential diagnosis of melanocytic neoplasms, demonstrating the potential of knowledge-enhanced AI in pathology.
