Automated Pipeline for Brain ROI Analysis with Results Comparable to Previous Freehand Measures in Clinical Settings

Abstract

Diffusion tensor imaging (DTI) has become a relatively common MR imaging technique in only 10 years. DTI can provide important information of brain microstructure in vivo. Many quantitative DTI analysis methods utilize either region of interest (ROI) or voxel-wise whole-brain methods. ROI methods do not require potentially bias-inducing image data altering, e.g., resampling and smoothing, and are the preferred method in clinical settings. We present an automated pipeline for quantitative ROI analysis of brain DTI data. The pipeline includes pre-processing, registrations, and calculation of mean (and SD) DTI scalar values from the automated ROIs. In addition to atlas regions, the pipeline accepts freehand ROIs, as long as the frame of reference is also provided. By the uniquely designed pipeline, we ensure that the results can be retrospectively compared to previously conducted manual freehand ROI measurement results, if desired. We validated the feasibility of the pipeline by comparing manual freehand ROI measurement results from 40 subjects against the results obtained from automated ROIs. A single set of freehand ROIs (drawn similarly to the original freehand manual ROIs in the population) was input to the pipeline, and the resulting scalar values from the automated ROIs were compared to the manual freehand ROIs’ data. Adopting a limit for goodness of fit of z = ± 1.6 resulted in 94 % success rate for the pipeline’s automated ROI registrations in the whole population. The pipeline can reduce the time taken in clinical ROI measurements.