In this paper we present a machine-learning tool to estimate the cognitive scores of Parkinson's Disease (PD) patients undergoing pre-DBS evaluation. 

 Low pre-surgical cognitive scores are are often an exclusion criteria while planning DBS surgery. The University of Florida DBS-Cognitive Rating Scale (DBS-CRS) is a tool that condenses the results of a 3-h exam into a five-point scale ranging from 1 (least) to 5 (most) cognitive concern for surgery. It is effective for communicating pre-DBS cognitive concerns to an interdisciplinary team. Data from structural brain MRI has been used to study cognitive decline and dementia in PD patients. However to this date it has not been used to estimate the neurocognitive scores of  DBS candidates with PD.  

We conducted a retrospective review of patients undergoing pre-surgical workup for deep brain stimulation at the University of Florida. We collected data on demographics, imaging, and neuropsychological testing. Whole brain morphometry was performed using Freesurfer’s recon-all function, and the resulting brain volumetric data were used as features for machine learning to predict the CRS from neuropsychological testing. The models were evaluated using 5-fold cross-validation with a hold-out test set to ensure generalizability.

We analyzed data from 300 patients, collecting whole brain morphometry features using the DKT40 classifier atlas. Based on our clinical experience, we created a binary classifier task: a CRS score of 1-2 indicated “proceed forward without hesitation” for DBS surgery, while a score of 3 or higher meant “requires full consideration” before proceeding. We developed a Lasso-based classifier, achieving a mean accuracy of 0.94, a precision of 0.94, and a recall of 0.93.

Structural brain MRI data can be used to estimate PD patient’s cognitive scores and assist in the decision to proceed or not with DBS surgery.