Evaluating Explainable AI Methods in Deep Learning Models for Early Detection of Cerebral Palsy

XAI methods can reliably identify which body movements matter for medical AI predictions: Both CAM and Grad-CAM successfully pinpointed key body points that influenced the deep learning model's decisions about cerebral palsy risk, suggesting these explanation tools can help doctors understand what the AI is "looking at" when making clinical recommendations.

Explanation stability varies depending on the specific XAI method and what you're measuring: Grad-CAM proved more stable when tracking movement velocity, while CAM performed better at identifying stable bone positions and maintaining consistent internal representations—indicating that choosing the right explanation method matters for different clinical contexts.

Using multiple AI models together may provide more reliable explanations than single models: The ensemble approach (combining multiple models) offered more balanced and representative results compared to individual models, which showed inconsistent performance, suggesting that medical AI systems should consider using multiple models to generate more trustworthy explanations.

Explanation robustness was tested realistically for medical applications: The researchers validated their findings using perturbations that mimicked real-world variations in infant movement without distorting the actual dynamics, making their conclusions more applicable to actual clinical use rather than artificial testing scenarios.

Current XAI methods show promise but require careful evaluation for medical use: The findings demonstrate that explainable AI can work in pediatric medical settings, but the variability in performance across different metrics emphasizes the need for thorough, context-specific validation before deploying these systems in clinical practice.