Parkinson’s disease (PD) pathology is characterized by neuromelanin loss and iron accumulation in substantia nigra pars compacta (SNc), as well as profound locus coeruleus (LC) degeneration. We have shown previously that neuromelanin-sensitive and iron-sensitive MRI sequences can robustly detect and quantify these phenomena (Huddleston DE, et al, 2017, PMID:28240402; Langley J, et al, 2017, PMID:27687624). No validated biomarkers for PD exist, and clinical diagnostic accuracy is sub-optimal in clinico-pathologic correlation studies. Neuroimaging tools are urgently needed to confirm PD neuropathology, assist clinical diagnosis, and enhance clinical trial recruitment.

To develop a multivariable signature to assist Parkinson’s disease diagnosis.

67 individuals (36 PD, 31 control) were studied at the Emory Movement Disorders Clinic under an IRB approved protocol with informed, written consent. MRI data were acquired using a Siemens Prisma-Fit 3T scanner. Neuromelanin-sensitive MRI (NM-MRI) and R2* (iron-sensitive MRI) acquisition and automated image processing to extract relevant features in SNc and LC were done using published methods (Langley J, et al, 2017, PMID: 27687624; Huddleston DE, et al, 2017, PMID: 28240402). The Movement Disorders Society UPDRS (MDS-UPDRS) Parts I and II questionnaires, the REM Sleep Behavior Disorder Questionnaire (RBDQ) and the Non-Motor Symptoms Questionnaire (NMSQ) were collected. With this dataset of extracted MRI and clinical features, logistic regression with elastic net (with 10-fold cross-validation) was used to train a classifier to accurately differentiate PD from controls. Sensitivity analysis was performed to identify informative features.

In a model including 6 MRI features of neuromelanin and iron pathology and 6 clinical features, receiver operating characteristic analysis revealed 94.2% diagnostic accuracy. Sensitivity analysis identified SNc volume, SNc R2*, MDS-UPDRS-II total score, and RBDQ item 2 (aggressive, action-packed dreams? yes/no) as the most informative features.

This multivariable profile exhibits diagnostic accuracy in a potentially useful range for clinical and translational applications, and warrants further development.