Meeting Banner
Abstract #3028

Voxel Misassignments and Their Consequences in DTI Skeleton-Based Group Analyses

Michael Bach1, 2, Bram Stieltjes1, Frederik B. Laun, 12, Alexander Leemans3, Klaus Hermann Fritzsche, 14

1Quantitative Imaging-based Disease Characterization, German Cancer Research Center (DKFZ), Heidelberg, Germany; 2Dpt. of Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany; 3Image Sciences Institute, University Medical Center Utrecht, Utrecht, Netherlands; 4Medical Imaging and Biological Informatics, German Cancer Research Center (DKFZ), Heidelberg, Germany

Considering the widespread and growing use of tract-based spatial statistics (TBSS), it is important to fully understand the potential limitations of this approach. In this work the voxels of the cingulum bundle (CB) and corpus callosum (CC) are followed up through the TBSS-process. Our results reveal that CB voxels are assigned to the CC-skeleton and vice versa. This strongly influences the statistical inference, as shown here for a group comparison between 15 Alzheimers diseases patients and 15 healthy controls. In conclusion the potential misassignment of adjacent white matter tracts must be born in mind when interpreting TBBS results.

Abstract PDF Presentation Video

Keywords

adjacent alignment amount analyses another applied arrow arrows assigned assignment axial basis became belonged binary bins biological black blue bottom brain bundle cancer characterization citations classifications clearly colors compensating completely concordance consequences contribution controls coronal corpus corrected counting datasets diffusion disappeared disease either enabled evaluated evaluation even example exemplary expected extent false findings flirt fraction full fully gradient green highly histogram improve improved incorporates independence indicated inference influence influences informatics inherent initial integrity interestingly interpreting introduce invariance limitation major maps masked measure medical mistaken mixed mixture motion must native occurs original originating package patients perfect permutation physics pipeline potential powerful previous procedure projected projection proximity purely quantitative radiology recommended refocused registration repetitions respectively retrieved rotational sciences section segmentation segmentations significantly since skeleton slices software space spatial statistical statistics steps strongly structure studies subject subjects substantial tensor theoretically thick threshold tract tracts true unknown versa vice view white whose wrong wrongly yellow