Supplemental material for:
Quantitative mouse brain phenotyping based on single and multispectral MR protocols
Alexandra Badea, Sally Gewalt, Brian Avants, James Cook, G Allan Johnson
NeuroImage 63:1633-1645, 2012 PMCID: PMC3604698
Sophisticated image analysis methods have been developed for human brains, but such tools still need to be adapted and optimized for quantitative small animal imaging. We propose a framework for quantitative assessment of anatomical phenotypes in mouse models of neurological and psychiatric conditions. The framework for automated segmentation encompasses an MR-based atlas space; and a suite of software tools to spatially normalize images into this space, and to derive quantitative morphometric phenotypes. We show that a suite of segmentation tools (Avants, Epstein et al. 2008) designed for human neuroimaging can be adapted and streamlined for the segmentation of mouse brain images acquired with multispectral MR protocols. We present a flexible approach for segmenting such hyperimages, and identify optimal combinations of image channels. Brain imaging with T1, T2* and T2w contrast yielded accuracy in the range of 83% for large structures such as hippocampus and caudate putamen (Hc and CPu), but only 54% in white matter tracts, and 44% for the ventricles. The addition of diffusion tensor imaging (DTI) images improved accuracy for large gray matter structures (by 5%), white matter (10%), and for the ventricles (15%). The use of Markov Random Field based segmentation improved overall accuracy in the C57BL/6 strain by 5.8%, so that Dice coefficients for Hc and CPu were 93%, for white matter 79%, for ventricles 68%, and for substantia nigra 80%. We demonstrate the pipeline by segmenting two test strains (BXD29, APP/TTA). Such an approach looks promising for characterizing temporal changes in mouse models of human neurological and psychiatric conditions. It may also provide anatomical constraints for functional imaging such as fMRI, and help evaluate putative therapies.
Figure S1. The use of non-uniformity correction filter (N4ITK, Tustison et al, 2010) increased segmentation accuracy relative to segmentation based on uncorrected images.
Figure S2. A combination of T1w and T2w channels with different weighting factors (0.7, 0.3) yielded highest Dice coefficients in test strains.
Figure S3. The use of cross-correlation metric implementation in ANTs was favored because of its relative speed over MI and apparent benefits on accuracy in several brain structures (ns).
CIVMSpace is designed to work on the Microsoft Windows and Mac OS X platforms.
Supported browsers on Windows are Internet Explorer versions 6.0 and above, and Mozilla Firefox versions 1.0 and above.
On the Mac, Safari versions 1.3 and above and Mozilla Firefox versions 1.0 and above are supported.
VoxStation requires a working Java installation. It has been tested with Java 1.4, 1.5 and 1.6 on Windows, and Java 1.4 and 1.5 on Mac OS X.
We thank Dr. Yi Jiang for supplying the DTI mouse brain images, Yi Qi for help with specimen preparation, Gary Cofer for his MR expertise, and Darin Clark for the initial implementation of the bilateral filter. We thank Sally Zimney for editorial assistance. The study was supported by the Duke Center for In Vivo Microscopy (grants to GAJ: NIH/NIBIB national Biomedical Technology Resource Center [P41 EB015897] and Small Animal Imaging Resource Program [U24 CA092656]).