MO-FG-CAMPUS-IeP3-04: Evaluation of Reproducibility of Diffusion Tensor Imaging Metrics Using Automatically Generated Regions of Interest

Authors


Abstract

Purpose:

To establish diffusion tensor imaging (DTI) as a valid diagnostic tool, it is essential to first establish the reproducibility of DTI metrics, such as fractional anisotropy (FA) and mean diffusivity (MD). In this study, we developed two fully automated pipelines for obtaining FA and MD in multiple ROIs, and used them to evaluate intra- and inter-session reproducibility.

Methods:

Twenty subjects were scanned at 3.0T in two identical sessions separated by 1 week. During each session, the subjects were scanned twice with identical protocols, which comprised a 3D T1 anatomical scan and a DTI scan of 7 minutes. Two pipelines were developed for image analysis. In the first approach, DTI images were linearly registered to anatomical images, which were then non-linearly registered to the MNI152 brain template. And DTI tensor fitting was performed in the MNI152 space. In the second pipeline, the tensor fitting was performed in the native space and the resulting FA maps were non-linearly registered to the FMRIB58 FA template. The reproducibility of FA and MD were evaluated for 48 white matter ROIs using coefficients of variation (CV) and intraclass correlation coefficients (ICC).

Results:

The second pipeline was found to have higher reproducibility, with 85% of ROIs having inter-session CV <=2% for FA and <=1.6% for MD, and inter-session ICC >=0.82 for FA and >=0.79 for MD. Large variations were only found in small or narrow ROIs which are more prone to partial volume effect and misalignment. Most ROIs had higher inter-session CV than intra-session CV, though the difference was not statistically significant.

Conclusion:

Our results indicate that current DTI technology can achieve high enough reproducibility for many clinical applications, and that a fully automated image registration pipeline can be a valid and efficient approach for extracting FA and MD information from a large number of ROIs.

Ancillary