Brigham Research Institute Poster Session Site logo-1

Tengfei Xue



Job Title

POI Research Trainee

Academic Rank




Tengfei Xue, Fan Zhang, Chaoyi Zhang, Yuqian Chen, Suheyla Cetin-Karayumak, Leo R. Zekelman, Steve Pieper, William M. Wells, Yogesh Rathi, Nikos Makris, Weidong Cai, and Lauren J. O’Donnell

Principal Investigator

Lauren J. O'Donnell

Research Category: Neurosciences


Neurocognitive prediction with tractography fiber cluster microstructure: A large-scale diffusion MRI study across ~8k participants in the Adolescent Brain Cognitive Development (ABCD) dataset

Scientific Abstract

Neuroimaging-based prediction of neurocognition is valuable for studying how the brain’s structure relates to cognitive function. This prediction is performed using regression analysis methods. However, the prediction accuracy of popularly used linear regression models is relatively low. We propose a novel deep learning method for regression that leverages neurocognitive performance measures and diffusion MRI neuroimaging data. We analyze a large-scale dataset that is composed of diffusion MRI and neurocognitive data from 8735 participants in the Adolescent Brain Cognitive Development Study. Using a fine-scale parcellation of white matter tractography into fiber clusters, we extract cluster-specific measures of brain tissue microstructure. To measure higher-order cognition, we use three scores related to neurocognitive components of general cognitive ability, executive function and learning/memory. We then propose a permutation importance method to identify important fiber clusters for prediction of these scores across the whole brain. The proposed deep learning regression method improves the prediction accuracy of neurocognitive performance using diffusion MRI data. We find that clusters within the middle longitudinal fasciculus are crucial for predicting performance across all three neurocognitive components. Also, clusters belonging to the superficial white matter are found to be important in neurocognitive prediction, especially for executive function.

Lay Abstract

Adolescence is understood as an important time for cognitive and brain white matter development. However, the precise white matter brain structures that enable healthy cognitive performance are not fully understood. In this study, we propose a deep learning method to predict measures of cognitive performance from finely parcellated white matter connections derived from brain imaging data in a large sample of over 8000 participants. We find that our deep learning method is able to make accurate predictions of cognitive performance. We identify particular brain white matter pathways that may be crucial for healthy cognitive performance during adolescence.

Clinical Implications

In this study we identify white matter pathways that may be crucial for the development of higher-order cognition in adolescents.