Predicting brain dynamic behaviours from DTI and wearable sensors
Poster No:
1158
Submission Type:
Late-Breaking Abstract Submission
Authors:
Jacob Mathew1, Christina Leong1, Maryam Tayebi2, Eryn Kwon3, Joshua McGeown2, Justin Fernandez4, Samantha Holdsworth5, Vickie Shim3
Institutions:
1University of Auckland, Auckland, New Zealand, 2Mātai Medical Research Institute, Gisborne, Outside America, 3Auckland Bioengineering Institute, University of Auckland, Auckland, Outside America, 4Auckland Bioengineering Institute, University of Auckland, Auckland, New Zealand, 5Mātai Medical Research Institute, Gisborne, New Zealand
First Author:
Co-Author(s):
Vickie Shim, PhD
Auckland Bioengineering Institute, University of Auckland
Auckland, Outside America
Auckland Bioengineering Institute, University of Auckland
Auckland, Outside America
Late Breaking Reviewer(s):
Rosanna Olsen
Rotman Research Institute, Baycrest Academy for Research and Education
Toronto, Ontario
Rotman Research Institute, Baycrest Academy for Research and Education
Toronto, Ontario
Introduction:
Sport- and recreation-related head injuries are the second most common cause of hospitalization for traumatic brain injury (TBI), affecting over 50 million people worldwide and leads to long-term health impairments(Zhong et al., 2025). Up to 50% of mild TBI (mTBI) cases in sports go undetected due to the lack of reliable real-time detection methods(Brush et al., 2018). This study aims to develop subject-specific machine learning (ML) models based on subjects' diffusion tensor images (DTI) to predict time-dependent brain strains from head impact profiles, enabling early detection of concussions and optimising rehabilitation pathways for mTBI patients.
Methods:
Twenty male high school rugby players (n = 20; age = 16 – 18 years) were recruited. DTI and T1 structural scans were acquired using a 3.0T GE MRI scanner advanced MRI scans(Tayebi et al., 2024). Subject-specific computational head models were generated using T1 and DTI images where T1 was used to obtain shape and DTI material properties. In particular, brain tissue was modelled using a hyper-viscoelastic, fibre-reinforced anisotropic formulation which uses FA values from DTI to determine subject-specific material properties(Shim et al., 2022).
Kinematic data were recorded using instrumented mouthguards (iMGs), capturing angular acceleration traces for each head impact profile. The head impacts were simulated in FEBio (www.febio.org) to compute 3D brain strains from head impacts using high-performance computing resources (www.nesi.org.nz).
Strain and head impact data were used to train non-linear regression machine learning (ML) models using Random Forest (RF), with an 80% training and 20% testing split. The top 100 features were extracted from the head impact profile using the TSfresh Python library(Christ et al., 2018) to predict time-dependent strain profiles. Intra-subject RF models were trained on strain and impact data from the same athlete, while inter-subject RF models used data from all athletes. Peak maximum principal strain (MPS) values were compared to simulated results from FEBio and to a pre-trained convolutional neural network (CNN) from Worcester Polytechnic Institute, which was trained on impacts simulated with a standardized head model(Wu et al., 2019).
Kinematic data were recorded using instrumented mouthguards (iMGs), capturing angular acceleration traces for each head impact profile. The head impacts were simulated in FEBio (www.febio.org) to compute 3D brain strains from head impacts using high-performance computing resources (www.nesi.org.nz).
Strain and head impact data were used to train non-linear regression machine learning (ML) models using Random Forest (RF), with an 80% training and 20% testing split. The top 100 features were extracted from the head impact profile using the TSfresh Python library(Christ et al., 2018) to predict time-dependent strain profiles. Intra-subject RF models were trained on strain and impact data from the same athlete, while inter-subject RF models used data from all athletes. Peak maximum principal strain (MPS) values were compared to simulated results from FEBio and to a pre-trained convolutional neural network (CNN) from Worcester Polytechnic Institute, which was trained on impacts simulated with a standardized head model(Wu et al., 2019).
Results:
The intra-subject RF model predicted time-dependent strain with a normalized root mean square error (NMRSE) from 0.06 to 0.18, shown in Figure 1. The predicted traces closely followed the target strain patterns from FEBio, capturing the oscillatory motion of brain tissue. For peak MPS predictions, significant differences (p<0.05) between the three ML models were observed in Figure 2. The CNN model had the largest spread, and a median error of 0.122, whereas the intra-subject RF model achieved a significantly lower error of just 0.03-over four times more accurate. The inter-subject RF model performed moderately, with an absolute error of 0.07.
Conclusions:
This study presents the first application of machine learning to predict time-dependent brain strain from DTI and head impacts with high accuracy, which has implications for early concussion detection strategies for mTBI patients. For peak MPS predictions, the intra-subject RF model (using DTI based subject-specific head models) outperformed the pre-trained CNN which used a standardized head model by a factor of four and surpassed the inter-subject model trained on multiple athlete datasets. This underscores the importance of obtaining subject-specific material properties and geometries from advanced MRI scans in predicting brain strains.
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Diffusion MRI Modeling and Analysis 2
Other Methods
Novel Imaging Acquisition Methods:
Diffusion MRI
Keywords:
Machine Learning
Modeling
MRI
STRUCTURAL MRI
Trauma
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Christ, M., Braun, N., Neuffer, J., & Kempa-Liehr, A. W. (2018). Time Series FeatuRe Extraction on basis of Scalable Hypothesis tests (tsfresh – A Python package). Neurocomputing, 307, 72-77. https://doi.org/https://doi.org/10.1016/j.neucom.2018.03.067
Shim, V. B., Tayebi, M., Kwon, E., Guild, S., Scadeng, M., Dubowitz, D., Mcbryde, F., Rosset, S., Wang, A., Fernandez, J. W., Li, S., & Holdsworth, S. (2022). Combining advanced magnetic resonance imaging (MRI) with fintie element (FE) analysis for characterising subject-specific injury patterns in the brain after traumatic brain injury. Engineering with Computers.
Tayebi, M., Kwon, E., McGeown, J., Potter, L., Taylor, D., Condron, P., Qiao, M., McHugh, P., Maller, J., Nielsen, P., Wang, A., Fernandez, J., Scadeng, M., Shim, V., & Holdsworth, S. (2024). Characterizing the Effect of Repetitive Head Impact Exposure and mTBI on Adolescent Collision Sports Players’ Brain with Diffusion Magnetic Resonance Imaging. Journal of Neurotrauma. https://doi.org/10.1089/neu.2024.0064
Wu, S., Zhao, W., Ghazi, K., & Ji, S. (2019). Convolutional neural network for efficient estimation of regional brain strains. Scientific Reports, 9(1), 17326. https://doi.org/10.1038/s41598-019-53551-1
Zhong, H., Feng, Y., Shen, J., Rao, T., Dai, H., Zhong, W., & Zhao, G. (2025). Global Burden of Traumatic Brain Injury in 204 Countries and Territories From 1990 to 2021. American Journal of Preventive Medicine. https://doi.org/https://doi.org/10.1016/j.amepre.2025.01.001