Multi-Study Longitudinal Tensor-Based Morphometry Processing Pipelines for Mapping Neurodegeneration
Poster No:
1839
Submission Type:
Abstract Submission
Authors:
Sunanda Somu1, Alyssa Zhu2, Siddharth Narula1, Iyad Ba Gari3, Shruti Gadewar2, Talia Nir4, Neda Jahanshad5
Institutions:
1University of Southern California, Los Angeles, CA, 2USC, Marina Del Rey, CA, 3University of Southern California, Marina Del Rey, CA, 4University of Southern California Keck School of Medicine, Marina del Rey, CA, 5Imaging Genetics Center, Keck School of Medicine of University of Southern California, Los Angeles, California
First Author:
Co-Author(s):
Neda Jahanshad, PhD
Imaging Genetics Center, Keck School of Medicine of University of Southern California
Los Angeles, California
Imaging Genetics Center, Keck School of Medicine of University of Southern California
Los Angeles, California
Introduction:
Longitudinal studies offer deep insights into the rate of brain volume changes and their associations with neurodegenerative disease progression, such as Alzheimer's disease (AD). Tensor-Based Morphometry (TBM) quantifies structural brain changes by analyzing deformation field gradients derived from image registrations [1], offering higher regional specificity compared to regional volumetric measurements. TBM results can be influenced by preprocessing and image registration methods [2], necessitating careful workflow consideration in study designs. Longitudinal processing pipelines often use a subject-specific midspace template to minimize bias from any single time point [3], [4]. The evaluations of such pipelines have largely focused on the method of calculating the midspace rather than the optimal pre-processing steps, which may vary by dataset. We assessed the sensitivity of four TBM pre-processing pipelines, run in two AD datasets, to detect associations with clinical impairment.
Methods:
Baseline (BL) and follow-up (FU) Siemens, GE, and Philips 3T T1-weighted brain MRI (T1w) from ADNI3(N=373) and Prevent-AD(N=213) were used to compare 4 longitudinal TBM pipelines outlined in Fig 1. Two T1w preprocessing options were compared: 1) the default FreeSurfer preprocessing pipeline (FS) or 2) a combination of existing preprocessing tools includingT1w NLM denoising (Denoised) [5]–[7] (Fig 1B). For each subject, a midspace template was created from their preprocessed BL and FU T1w using ANTs [8]. The BL and FU T1w were warped to the subject template using 2 approaches: 1) the BL, FU, and template were all skull-stripped (Masked) or 2) all images retained the skull (SkullOn). The resulting BL log Jacobian was then subtracted from the FU to quantify net changes, and divided by the scan time interval to obtain the annual rate of volumetric change (Fig 1C). Each subject's BL T1w image was nonlinearly warped to a common BL study-specific minimal deformation template (MDT) using ANTs. Each BL MDT was warped to a common multi-site MDT [9], [10](Fig 1D). These transformations were applied to the Jacobian maps to spatially normalize them for pooled statistics across studies. Participants that failed QC in any pipeline (e.g. mis-registration) were excluded from all analyses (i.e. matched N across pipes). Voxel-wise mixed effects linear regressions were run to test for associations between rates of change in brain volume and 1) clinical dementia ratings (CDR), or 2) diagnosis conversion (i.e., stable CN vs cognitive decline). Fixed effects included BL age, sex, and the time interval; scan site was a random effect. FDR was used for multiple comparisons correction across voxels. We compared each model's root mean squared error (RMSE), and the extent and distribution of significant effects across the four pipelines: Denoised+Masked, Denoised+SkullOn, FS+Masked, FS+SkullOn.
Results:
Both higher CDR and disease progression were significantly associated with greater rates of CSF expansion and tissue atrophy across all pipelines. Some of the largest effects were consistently found in the hippocampus, amygdala, ventricles and temporal lobe. Denoised pipelines exhibited lower overall RMSE, with Denoised+SkullOn demonstrating the lowest RMSE (Fig 2). The most widespread associations were detected with Denoised+Masked. In contrast, the largest effect sizes were detected with FS+Masked.
Conclusions:
Longitudinal TBM revealed detailed maps of neurodegeneration and rates of decline. However, no pipeline consistently outperformed the others across comparisons. The Denoised preprocessing pipeline outperformed default FS preprocessing in terms of RMSE and number of passing voxels but not effect sizes. Differences between Denoised SkullOn and Masked pipelines were more subtle. Future work will extend testing to 3 and more timepoints.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Lifespan Development:
Aging
Modeling and Analysis Methods:
Image Registration and Computational Anatomy 1
Motion Correction and Preprocessing
Neuroinformatics and Data Sharing:
Workflows
Keywords:
Aging
Design and Analysis
Morphometrics
Spatial Normalization
Spatial Warping
Workflows
1|2Indicates the priority used for review
Provide references using author date format
[2] I. Yanovsky, A. D. Leow, S. Lee, S. J. Osher, and P. M. Thompson, “Comparing registration methods for mapping brain change using tensor-based morphometry,” Med. Image Anal., vol. 13, no. 5, pp. 679–700, Oct. 2009.
[3] N. J. Tustison et al., “The ANTs Longitudinal Cortical Thickness Pipeline,” bioRxiv, p. 170209, Aug. 16, 2018. doi: 10.1101/170209.
[4] M. Reuter, N. J. Schmansky, H. D. Rosas, and B. Fischl, “Within-subject template estimation for unbiased longitudinal image analysis,” Neuroimage, vol. 61, no. 4, pp. 1402–1418, Jul. 2012.
[5] F. Isensee et al., “Automated brain extraction of multisequence MRI using artificial neural networks,” Hum. Brain Mapp., vol. 40, no. 17, pp. 4952–4964, Dec. 2019.
[6] Y. Zhang, M. Brady, and S. Smith, “Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm,” IEEE Trans. Med. Imaging, vol. 20, no. 1, pp. 45–57, Jan. 2001.
[7] J. V. Manjón, P. Coupé, L. Martí-Bonmatí, D. L. Collins, and M. Robles, “Adaptive non-local means denoising of MR images with spatially varying noise levels,” J. Magn. Reson. Imaging, vol. 31, no. 1, pp. 192–203, Jan. 2010.
[8] B. B. Avants, C. L. Epstein, M. Grossman, and J. C. Gee, “Symmetric diffeomorphic image registration with cross-correlation: evaluating automated labeling of elderly and neurodegenerative brain,” Med. Image Anal., vol. 12, no. 1, pp. 26–41, Feb. 2008.
[9] M. Jenkinson, P. Bannister, M. Brady, and S. Smith, “Improved Optimization for the Robust and Accurate Linear Registration and Motion Correction of Brain Images,” Neuroimage, vol. 17, no. 2, pp. 825–841, Oct. 2002.
[10] N. Jahanshad et al., “Multi-Site Meta-Analysis of Morphometry,” IEEE/ACM Trans. Comput. Biol. Bioinform., vol. 16, no. 5, Sep. 2019, doi: 10.1109/TCBB.2019.2914905.
Acknowledgments:
This work is supported in part by NIH grants: R01AG059874, P41EB015922, U01AG068057.