AI is coming for Parkinson's imaging, however we have some serious quality issues we must be ready for along the path. Dzialas and colleagues in
@MDJ_Journal walk us through most of them.
Key Points:
- This about the challenges for AI and imaging in the domains of diagnosis, prognosis, and intervention.
- 810 studies published and 244 were pertinent to these 3 domains.
- They looked at outcomes and rated them using five minimal quality criteria (MQC) including "data splitting, data leakage, model complexity, performance reporting, and indication of biological plausibility."
- The majority of published studies tried to distinguish PD from controls (54%) or atypicals (25%).
- Prognostic or interventional studies were "sparse."
- Only about 1/5 passed all five MQC's.
- Few used external test sets (8%) for validation.
My take: Look folks AI is coming, however to be useful in Parkinson's disease and beyond, we don't need more publications, we need higher quality publications which will address the 5 minimal quality criteria. I love that early in our imaging journey this group of investigators has started the dialogue.
movementdisorders.onlinelibr… #Parkinsons