May 16, 2026

@michaelokun

Could AI help uncover hidden Parkinson’s subtypes years earlier? Multimodal means combining many different types of information such as genetics, brain scans, spinal fluid markers and cognitive testing into one analysis. Reyes and colleagues describe in a new paper in npj Artificial Intelligence how a new AI framework called SPARROW may help identify Parkinson’s disease subtypes using multimodal reasoning and interpretable AI. Key points: - SPARROW combined genetics, MRI brain scans, spinal fluid biomarkers and cognitive testing into a single reasoning framework for Parkinson’s disease subtyping. - The AI system achieved strong accuracy in distinguishing slow, moderate and fast progressing Parkinson’s disease subtypes using baseline patient data. - The system generated step by step reasoning traces, allowing health care providers to better understand how the AI arrived at each subtype decision. My take: Parkinson’s disease is not one disease. It is many diseases hiding under one name. This study is exciting because it moves beyond simply identifying Parkinson’s and instead asks whether AI can help us understand who may progress faster and why. The interpretability piece is critical. Black box AI will struggle in neurology clinics. Transparent reasoning may be the bridge that builds trust and utility. The performance of this approach may not be as good as a clinician. Also, the scientists turned the imaging data into text. Here are 5 points that resonated w/ me: 1- Parkinson’s disease likely contains multiple biological subtypes and these differences may explain why symptoms and progression vary so widely between folks. 2- Cognitive testing still emerged as one of the strongest signals in predicting subtype and progression. 3- MRI and genetic information alone were weak predictors, however when combined w/ other data they improved accuracy substantially. 4- The AI system was designed to tolerate missing data, which mirrors real world neurology practice where every test is not always available. 5- The future of Parkinson’s care may involve AI systems that help health care providers personalize prognosis, treatment selection and monitoring much earlier #ai