Can shape be as informative as microstructure for cognitive performance assessment?
Poster No:
2192
Submission Type:
Abstract Submission
Authors:
Yui Lo1,2,3, Yuqian Chen1,2, Leo Zekelman2,4, Dongnan Liu3, Wan Liu5, Fan Zhang6, Yogesh Rathi1,2, Nikos Makris1,7, Alexandra Golby1,2, Weidong Cai3, Lauren O'Donnell1,2
Institutions:
1Harvard Medical School, Boston, MA, 2Brigham and Women’s Hospital, Boston, MA, 3The University of Sydney, Sydney, NSW, 4Harvard University, Boston, MA, 5Beijing Institute of Technology, Beijing, Beijing, 6University of Electronic Science and Technology of China, Chengdu, Sichuan, 7Massachusetts General Hospital, Boston, MA
First Author:
Yui Lo
Harvard Medical School|Brigham and Women’s Hospital|The University of Sydney
Boston, MA|Boston, MA|Sydney, NSW
Harvard Medical School|Brigham and Women’s Hospital|The University of Sydney
Boston, MA|Boston, MA|Sydney, NSW
Co-Author(s):
Introduction:
The connectivity and microstructure of the brain's connections are known to relate to individual cognitive performance, including language performance (Zekelman et al., 2022). Deep learning methods have shown that the connectivity of fiber tracts is predictive of language proficiency in children with epilepsy (Jeong et al., 2021). However, the potential of shape as a feature for predicting individual cognitive performance remains largely unexplored. Shape is an important descriptor of fiber tracts (Corouge, Gouttard and Gerig, 2004) and is known to change with aging (Schilling et al., 2022). This study investigates the relevance of fiber tract shape in predicting neurocognitive language performance.
Methods:
We employed measures of fiber tract shape and individual language performance in 1065 subjects from the Human Connectome Project Young Adult (HCP-YA) dataset. Whole brain tractography was generated from HCP-YA diffusion MRI (dMRI) using a two-tensor unscented Kalman filter method (Malcolm, Shenton and Rathi, 2010), followed by the parcellation of tractography into 953 fiber clusters using an anatomically curated atlas (Zhang et al., 2018). We studied 12 fiber tract shape features, including cluster length, diameter, elongation, span, curl, volume, trunk volume, branch volume, total surface area, total radius of end regions, total area of end regions, and irregularity (Yeh, 2020). For comparison, we extracted traditional fiber tract microstructure features of fractional anisotropy (FA), mean diffusivity (MD), and the traditional connectivity feature of the number of streamlines (NoS). In total, 15 features were computed for each fiber cluster. We utilized a convolutional neural network (CNN) (Liu et al., 2023) to predict subject-specific language performance given each input feature. We predicted two neurocognitive language assessments: the NIH-TB Oral Reading Recognition Test (TORRT), and the NIH-TB Picture Vocabulary Test (TPVT) (Weintraub et al., 2013). The Pearson correlation coefficient (r) was employed to assess the prediction performance of each CNN model.
Results:
Certain shape features demonstrated equivalent or higher performance in comparison to microstructure features (Figure 1). For example, the total surface area shape feature outperformed all microstructure and connectivity features for both TORRT and TPVT language performance prediction. The total surface area feature measures the surface area of the 3D volume occupied by a fiber cluster. In the TORRT evaluation, 7 out of 12 shape features surpassed the performance of either FA, MD, or NoS. Similarly, in the TPVT evaluation, the same proportion of shape features-7 out of 12-exceeded the effectiveness of either FA, MD, or NoS.
·Comparative Analysis of TORRT and TPVT Prediction Across Microstructure, Connectivity, and Shape Features
Conclusions:
Our experimental results show that measures of the shape of fiber tract connections are informative for the prediction of individual, subject-specific language performance. This suggests that shape-related features can serve as an alternative and potentially superior source of features for predicting and evaluating various cognitive abilities, potentially outperforming microstructural and connectivity features in certain scenarios. This also indicates that the shape of the white matter fiber tracts may relate to important functions of the human brain.
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
White Matter Anatomy, Fiber Pathways and Connectivity 1
Novel Imaging Acquisition Methods:
Diffusion MRI 2
Keywords:
Data analysis
Language
Machine Learning
Tractography
White Matter
1|2Indicates the priority used for review
Provide references using author date format
Jeong, J.-W. et al. (2021) ‘Prediction of baseline expressive and receptive language function in children with focal epilepsy using diffusion tractography-based deep learning network’, Epilepsy & behavior: E&B, 117, p. 107909.
Liu, W. et al. (2023) ‘Fiber Tract Shape Measures Inform Prediction of Non-Imaging Phenotypes’, arXiv [cs.CV]. Available at: http://arxiv.org/abs/2303.09124.
Malcolm, J.G., Shenton, M.E. and Rathi, Y. (2010) ‘Filtered multitensor tractography’, IEEE transactions on medical imaging, 29(9), pp. 1664–1675.
Schilling, K.G. et al. (2022) ‘Aging and white matter microstructure and macrostructure: a longitudinal multi-site diffusion MRI study of 1218 participants’, Brain structure & function, 227(6), pp. 2111–2125.
Weintraub, S. et al. (2013) ‘Cognition assessment using the NIH Toolbox’, Neurology, 80(11 Suppl 3), pp. S54–64.
Yeh, F.-C. (2020) ‘Shape analysis of the human association pathways’, NeuroImage, 223, p. 117329.
Zekelman, L.R. et al. (2022) ‘White matter association tracts underlying language and theory of mind: An investigation of 809 brains from the Human Connectome Project’, NeuroImage, 246, p. 118739.
Zhang, F. et al. (2018) ‘An anatomically curated fiber clustering white matter atlas for consistent white matter tract parcellation across the lifespan’, NeuroImage, 179, pp. 429–447.