To implement a privacy-sensitive federated learning architecture for skin cancer classification in decentralized clinical environments, emphasizing the critical need for patient confidentiality.
Key Findings:
MobileNetV2 with the ring-based FL topology achieved an accuracy of 98.88% and F1-score of 98.80%, indicating a significant advancement in diagnostic accuracy (p < 0.01).
This performance was significantly better than the centralized baseline (p < 0.01), underscoring the effectiveness of federated learning in clinical settings.
MobileNetV2 demonstrated higher convergence rates and reduced communication overhead compared to VGG16, making it a more efficient choice for federated learning applications.
Interpretation:
Federated learning can effectively classify skin cancer while preserving patient privacy, with lightweight architectures like MobileNetV2 being particularly suited for this purpose, potentially transforming clinical practices.
Limitations:
The study is based on simulated data distributions which may not fully represent real-world clinical scenarios, suggesting a need for further validation in actual clinical settings.
Performance may vary with different datasets and clinical environments, indicating the necessity for future research to explore these variables.
Conclusion:
The proposed federated learning framework offers a scalable and privacy-compliant solution for AI-aided dermatological diagnosis in distributed healthcare settings, reinforcing the importance of patient privacy in healthcare AI applications.
