Abstract
Objectives: This work attempted to assess the feasibility of deep-learning based method in detecting the alterations of diffusion kurtosis measurements associated with Parkinson's disease (PD). Methods: A group of 68 PD patients and 77 healthy controls (HCs) were scanned on the scanner-A (3T Skyra) (DATASET-1). Meanwhile, an additional 5 healthy travelling volunteers were scanned with both the scanner-A and an additional scanner-B (3T Prisma) (DATASET-2). Diffusion kurtosis imaging (DKI) of DATASET-2 has an extra b shell than that of DATASET-1. In addition, a 3D convolutional neural network (CNN) was trained from Dataset 2 to harmonize the quality of scalar measures of scanner-A to a similar level of scanner-B. Whole-brain unpaired t-test and Tract-Based Spatial Statistics (TBSS) was performed to validate the differences between PD and control groups with model fitting method and CNN method respectively. We further clarified the correlation between clinical assessments and DKI results.Results: In the left substantia nigra (SN), an increase of mean diffusivity (MD) was found in PD group. In the right SN, fractional anisotropy (FA) and mean kurtosis (MK) values were negatively correlated with Hoehn & Yahr (H&Y) scales. In the putamen, FA values was positively correlated with H&Y scales. It is worth noting that, these findings were only observed with the deep-learning method. There was neither group difference, nor correlation with clinical assessments in the SN or striatum exceeding the significant level by using the conventional model fitting method.Conclusions: CNN method improves the robustness of DKI and can help to explore PD-associated imaging features.