An interpretable machine learning model can predict the severity of Parkinson's disease motor symptoms from brain imaging alone, according to a new study posted on arXiv.
The research
Aixa X. Andrade at an unspecified institution (the paper lists affiliations under arXiv:2607.02553) analyzed data from 28 participants, including 24 people with Parkinson's disease and 4 controls. The goal was to predict motor severity as measured by the MDS-UPDRS Part III, a standard clinical rating scale.
The researchers extracted two types of imaging features: quantitative susceptibility mapping (QSM), which measures iron and myelin content in brain tissue, and resting-state fMRI-derived regional homogeneity (ReHo), which reflects how synchronized activity is within local brain regions. They tested 13 different combinations of imaging, clinical, and multimodal features using four machine learning models: support vector regression, Elastic Net, Random Forest, and XGBoost, all trained with nested cross-validation to avoid overfitting.
Imaging-only models carried meaningful predictive signal, while clinical-only models performed weakly. The strongest global fit came from combining full fMRI, full QSM, and clinical variables, explaining 45.4% of the variance in motor severity. The best clinically close predictions came from selected QSM plus clinical variables, which predicted 75.0% of participants within plus or minus 5 MDS-UPDRS Part III points and had the lowest mean absolute error among top-performing models. SHAP analysis highlighted cerebellar, thalamic, striatal, insular, and motor cortical features as most important.
Why it matters
Parkinson's disease affects movement, but symptom severity is usually assessed through clinical observation, which can miss subtle brain changes. This study shows that structural (QSM) and functional (ReHo) imaging contribute differently depending on the prediction goal. For someone curious about their own cognition, it reinforces that brain health is multidimensional: iron content, neural synchrony, and clinical presentation each tell a different part of the story. The fact that imaging-only models outperformed clinical-only models suggests that objective brain markers may one day help track disease progression or treatment response more precisely than symptom checklists alone.
What you can do
- Stay physically active: exercise is one of the few interventions with evidence for supporting motor and cognitive function in Parkinson's.
- Challenge your brain with novel learning tasks, which may help maintain neural synchrony.
- If you notice changes in movement, coordination, or balance, consult a neurologist early.
- Consider participating in research studies that use imaging to study brain health.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.