A clinically interpretable model for Parkinson's disease: insights from complementary MRI modalities

Accurate identification of Parkinson's disease may remain challenging in certain clinical scenarios, highlighting the need for a clinical decision-support model based on complementary MRI modalities. The case-control study included 462 participants, split into training [247 Parkinson's disease, 122 normal controls (NC)] and testing (60 Parkinson's disease, 33 NC) cohorts to construct and validate a clinical decision-support model. We employed least absolute shrinkage and selection operator and logistic regression analysis to develop the model utilizing quantitative susceptibility mapping, arterial spin labelling and T1-weighted images. The model was assessed for calibration, discrimination and clinical utility, with individual predicted scores correlated to Parkinson's disease clinical symptoms. The model's performance was also evaluated independently in multiple system atrophy, progressive supranuclear palsy and essential tremor, with DeLong test comparisons. The model was trained in the training cohort (Parkinson's disease: 61.39 ± 8.06 years old, 134 males; NC: 61.29 ± 7.56 years old, 54 males) with an area under the ROC curve (AUC) of 0.842 [95% confidence interval (CI) 0.798-0.883] and 0.771 (95% CI 0.657-0.870) in the testing cohort, whose clinical utility was confirmed. The predicted scores were positively correlated with Unified Parkinson's Disease Rating Scale II (P = 0.034) and III (P = 0.015), while negatively correlated with the Purdue Pegboard total (P = 0.010) and assembly scores (P = 0.007). When discriminating each disease group from NC, the model's performance to identify multiple system atrophy and progressive supranuclear palsy with AUC values of 0.881 and 0.895 was superior than essential tremor with an AUC of 0.709 before false discovery rate correction (multiple system atrophy-NC versus essential tremor-NC: P = 0.041; progressive supranuclear palsy-NC versus essential tremor-NC: P = 0.010). The model based on complementary MRI modalities was an objective clinical decision-support measure for Parkinson's disease with potential applications in other parkinsonisms (multiple system atrophy, progressive supranuclear palsy).