Clinical Deployment Limitations

The experiments rely on handcrafted handwriting features rather than raw signal learning. The DARWIN dataset is small for complex deep architectures, which limits generalization and increases overfitting risk. The paper identifies inter-subject variability and task-specific noise as challenges for handwriting-based diagnosis. Deployment to other devices, populations, clinical settings, or acquisition protocols is not directly established. The article presents larger multi-center datasets, multimodal biomarkers, adaptive routing, and automated rank selection as future directions.