This study examines the sensitivity of DTI and quantitative histology measures to detect microstructural damage following experimental traumatic brain injury (TBI) in mice.

Contribution: Contributed to the statistical analysis and machine learning of histology and MRI data of experimental model of TBI. Performed a random forest analysis (RFA) to identify the important regions and metrics affected by injury.

Findings: Using histology metrics, RFA was able to predict correct classification of all the injured animals, but wrongly classified one of the shams as injured, reaching an overall classification accuracy of 90%. Using a different combinations of DTI metrics (FA/TR, AD/RD, and WL/WP/TR), the same sham was consistently misclassified as injured and one injured sample was misclassified as sham reaching an overall classification accuracy of 80%. Multiple DTI metrics as well as GFAP and IBA-1 of the optic tract and brachium of superior colliculus were consistently ranked as important for classification. Hippocampus was also ranked as an important ROI for classification using DTI metrics.

Papers/abstracts: Haber et al., eNeuro 2017, and abstract will be presented at Society for Neuroscience 2017 (Sadeghi et al.).