Traumatic Brain Injury

Quantifying Morphology-Related Deviations in Brain Strain Using an Automated Mesh Morphing Method

Deformation pipeline for subject-specific finite element head model.

Abstract

Finite element head models (FEHMs) have been widely used to study the biomechanics in traumatic brain injury (TBI). Most FEHMs are constructed to reflect the average head shape, which inevitably leads to the omission of individual brain morphology. In this study, an automated mesh morphing method based on radial basis function-thin plate spline (RBF-TPS) with automated landmark extraction and projection was developed. Five representative subject-specific head models and the baseline model were subjected to head kinematics from six datasets covering diverse impact scenarios. Results showed that morphology-related deviations increased with loading severity, reaching up to 0.21 for MPS95 and 0.14 s^-1 for MPSR95. Logistic regression indicated that TBI risk thresholds varied by approximately 19.4% for MPS95 and 11.4% for MPSR95 across representative models. These findings indicate that subject-specific morphology affects strain response beyond size scaling alone, underscoring the importance of incorporating individual morphology into brain injury prediction models.

DOI Source Document
Quantifying Morphology-Related Deviations in Brain Strain Using an Automated Mesh Morphing Method
AI-based identification of head impact locations, speeds, and force based on head kinematics simulations

Head impact parameter identification from helmeted impact simulations.

Abstract

Objective: With the development of wearable sensors, head kinematics data have become widely available. However, key impact information—such as impact direction, speed, and force—which is crucial for helmet development, is still not directly measured. This study presents a deep learning model designed to accurately predict these head impact parameters from head kinematics during helmeted impacts. Methods: Leveraging a dataset of 16,000 simulated helmeted head impacts using the Riddell helmet finite element model, a Long Short-Term Memory (LSTM) network was implemented to process the head kinematics: linear accelerations and angular velocities. Results: In simulations, the model accurately predicted impact direction, speed, and the force profile with R² exceeding 70% for all tasks. Validation on 79 on-field impacts recorded by instrumented mouthguards and videos demonstrated that the model significantly outperformed traditional methods in identifying impact locations, achieving 79.7% accuracy compared to the previous highest of 49.4%. Conclusion: This work highlights the potential of deep learning to enhance helmet design and head impact sensing by recovering key impact parameters from wearable kinematic data. Future work should evaluate model performance across helmet types and sports, and apply transfer learning for broader applicability.

DOI Source Document
AI-based identification of head impact locations, speeds, and force based on head kinematics simulations
Differences between two maximal principal strain rate calculation schemes in traumatic brain analysis with in-vivo and in-silico datasets

Comparison of two maximal principal strain rate calculation schemes.

Abstract

Brain deformation caused by a head impact leads to traumatic brain injury (TBI). The maximum principal strain (MPS) was used to measure the extent of brain deformation and predict injury, and recent evidence has indicated that incorporating the maximum principal strain rate (MPSR) and the product of MPS and MPSR, denoted as MPS × SR, enhances the accuracy of TBI prediction. However, ambiguities have arisen about the calculation of MPSR. Two schemes have been utilized: one is to use the time derivative of MPS (MPSR1), and another is to use the first eigenvalue of the strain rate tensor (MPSR2). To quantify the discrepancies between these two methodologies, we compared them across eight in-vivo and one in-silico head impact datasets and found that 95MPSR1 was slightly larger than 95MPSR2 and 95MPS × SR1 was 4.85% larger than 95MPS × SR2 on average. Across every element in all head impacts, the average MPSR1 was 12.73% smaller than MPSR2, and MPS × SR1 was 11.95% smaller than MPSR2. Logistic regression models trained to predict TBI showed no significant difference in predictability between the two schemes. The consequence of misuse of MPSR and MPS × SR thresholds was also examined, showing false decision rates around 1%, suggesting that the two methodologies are not significantly different in detecting TBI.

DOI Source Document
Differences between two maximal principal strain rate calculation schemes in traumatic brain analysis with in-vivo and in-silico datasets
Padded Helmet Shell Covers in American Football: A Comprehensive Laboratory Evaluation with Preliminary On-Field Findings

Laboratory impact configurations for padded helmet shell covers.

Abstract

S.I. : Concussions II Padded Helmet Shell Covers in American Football: A Comprehensive Laboratory Evaluation with Preliminary On-Field Findings NICHOLAS J. C ECCHI ,1 ASHLYN A. C ALLAN,1 LANDON P. W ATSON,1 YUZHE LIU,1 XIANGHAO ZHAN,1 RAMANAND V. V EGESNA,2 COLLIN PANG,1 ENORA LE FLAO,1 GERALD A. G RANT,3,4,5 MICHAEL M. Z EINEH,6 and D AVID B. C AMARILLO1,3,7 1Department of Bioengineering, Stanford University, Stanford, CA 94305, USA; 2Department of Biomedical Engineering, University of Southern California, Los Angeles, CA 90089, USA; 3Department of Neurosurgery, Stanford University, Stanford, CA 94305, USA; 4Department of Neurology, Stanford University, Stanford, CA 94305, USA; 5Department of Neurosurgery, Duke University, Durham, NC 27710, USA; 6Department of Radiology, Stanford University, Stanford, CA 94305, USA; and 7Department of Mechanical Engineering, Stanford University, Stanford, CA 94305, USA (Received 29 November 2022; accepted 8 February 2023) Associate Editor Stefan M. Du

DOI Source Document
Padded Helmet Shell Covers in American Football: A Comprehensive Laboratory Evaluation with Preliminary On-Field Findings
AI-Based Denoising of Head Impact Kinematics Measurements With Convolutional Neural Network for Traumatic Brain Injury Prediction

Convolutional neural network architecture for head impact kinematics denoising.

Abstract

Wearable devices used for measuring head impact kinematics are inherently noisy due to imperfect coupling with the human body. This study developed one-dimensional convolutional neural network (1D-CNN) models to denoise tri-axial linear acceleration and angular velocity signals recorded from instrumented mouthguards. Using 163 dummy head impacts for training, validation was performed at three levels: kinematics, brain injury criteria, and tissue-level strain and strain rate. Denoising reduced pointwise RMSE by 36% and peak absolute error by 56%, while six brain injury criteria errors decreased by 82% on average. Maximum principal strain and strain-rate errors were reduced by 35% and 69%, respectively. Blind testing on 118 college football impacts and 413 PMHS impacts demonstrated similar improvements. The 1D-CNN approach provides an effective method for reducing measurement noise in mouthguard-derived head kinematics and supports future real-world TBI monitoring.

DOI Source Document
AI-Based Denoising of Head Impact Kinematics Measurements With Convolutional Neural Network for Traumatic Brain Injury Prediction
Brain Deformation Estimation With Transfer Learning for Head Impact Datasets Across Impact Types

Transfer learning workflow for machine-learning head models.

Abstract

Objective: The machine-learning head model (MLHM) to accelerate the calculation of brain strain and strain rate, which are the predictors for traumatic brain injury (TBI), but the model accuracy was found to decrease sharply when the training/test datasets were from different head impacts types (i.e., car crash, college football), which limits the applicability of MLHMs to different types of head impacts and sports. Particularly, small sizes of target dataset for specific impact types with tens of impacts may not be enough to train an accurate impact-type-specific MLHM. Methods: To overcome this, we propose data fusion and transfer learning to develop a series of MLHMs to predict the maximum principal strain (MPS) and maximum principal strain rate (MPSR). Results: The strategies were tested on American football (338), mixed martial arts (457), reconstructed car crash (48) and reconstructed American football (36) and we found that the MLHMs developed with transfer learning are significantly more accurate in estimating MPS and MPSR than other models, with a mean absolute error (MAE) smaller than 0.03 in predicting MPS and smaller than 7 s⁻¹ in predicting MPSR on all target impact datasets. High performance in concussion detection was observed based on the MPS and MPSR estimated by the transfer-learning-based models. Conclusion: The MLHMs can be applied to various head impact types for rapidly and accurately calculating brain strain and strain rate. Significance: This study enables developing MLHMs for the head impact type with limited availability of data, and will accelerate the applications of MLHMs.

DOI Source Document
Brain Deformation Estimation With Transfer Learning for Head Impact Datasets Across Impact Types
Brain strain rate response: Addressing computational ambiguity and experimental data for model validation

Brain strain rate response under experimental and computational loading conditions.

Abstract

Traumatic brain injury (TBI) is an alarming global public health issue with high morbidity and mortality rates. Although the causal link between external insults and consequent brain injury remains largely elusive, both strain and strain rate are generally recognized as crucial factors for TBI onsets. With respect to the flourishment of strain-based investigation, ambiguity and inconsistency are noted in the scheme for strain rate calculation within the TBI research community. Furthermore, there is no experimental data that can be used to validate the strain rate responses of finite element (FE) models of the human brain. The current work presented a theoretical clarification of two commonly used strain rate computational schemes: the strain rate was either calculated as the time derivative of strain or derived from the rate of deformation tensor. To further substantiate the theoretical disparity, these two schemes were respectively implemented to estimate the strain rate responses from a previous-published cadaveric experiment and an FE head model secondary to a concussive impact. The results clearly showed scheme-dependent responses, both in the experimentally determined principal strain rate and model-derived principal and tract-oriented strain rates. The results highlight that cross-scheme comparison of strain rate responses is inappropriate, and the utilized strain rate computational scheme needs to be reported in future studies. The newly calculated experimental strain rate curves in the supplementary material can be used for strain rate validation of FE head models. Statement of significance: - Delineates a theoretical clarification of two algorithms for strain rate computation. - Highlights the strain rate responses directly depends on the computational schemes. - Presents experimental strain rate curves, serving as references for strain rate validation of finite element head models. 1.

DOI Source Document
Brain strain rate response: Addressing computational ambiguity and experimental data for model validation
Translational models of mild traumatic brain injury tissue biomechanics

Comparison of mTBI pathology thresholds across translational models.

Abstract

T raumatic brain injury (TBI) is a global health concern. Mild TBI (mTBI) which accounts for the majority of TBI cases, is hard to detect since often the imaging is normal but can still cause brain damage and long-term sequelae. Physiologically, acute primary damage to the brain is thought to be caused by tissue deformation from the inertial movement of the brain after rapid head rotation. Respecting tissue biomechanics, animal models are often used to understand the pathophysiology of mTBI. We have reviewed the literature focusing on connecting biome- chanics with mTBI pathologies at the tissue scale using neuroimaging, neurobehavioral tests, and pathologies across species, particularly studies using strain and strain rate. These studies have found strain and strain rate predict mTBI pathol- ogy and strain is generalizable across species, including small animals, large animals, and humans. We propose that re- searchers can leverage tissue-level strain and strain rate to bridge biomechanics and mTBI pathology . Addresses 1 Department of Bioengineering, Stanford University, Stanford, CA, USA 2 Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of T echnology and Emory University, Atlanta, GA, USA 3 Department of Radiology, Stanford University, Stanford, CA, USA 4 Department of Neurosurgery , Duke University, Durham, NC, USA

DOI Source Document
Translational models of mild traumatic brain injury tissue biomechanics
Piecewise Multivariate Linearity Between Kinematic Features and Cumulative Strain Damage Measure (CSDM) Across Different Types of Head Impacts

Piecewise distributions of cumulative strain damage measures.

Abstract

Building on previous findings that no single global linear model can describe the relationship between brain strain and kinematic features across diverse head impact types, this study examines whether piecewise multivariate linearity exists. Using K-means clustering, 3161 impacts from simulations, college football, mixed martial arts, and car crashes were partitioned into data-driven clusters. Within clusters, cumulative strain damage measure (CSDM, threshold 0.15) was regressed on kinematic features. K-means-based partitioning significantly improved regression accuracy compared to models without partitioning or those based solely on impact type. Additional analyses showed that partitioning by maximum angular acceleration at 4706 rad/s² yielded particularly strong piecewise linearity. These results support the existence of piecewise multivariate linear relationships and suggest improved strategies for rapid CSDM prediction.

DOI Source Document
Piecewise Multivariate Linearity Between Kinematic Features and Cumulative Strain Damage Measure (CSDM) Across Different Types of Head Impacts
Finding the Spatial Co-Variation of Brain Deformation With Principal Component Analysis

Spatial co-variation patterns of brain deformation across impact types.

Abstract

This study applies principal component analysis (PCA) to investigate spatial co-variation patterns of traumatic brain injury metrics, including maximum principal strain (MPS), MPS rate (MPSR), and their product, across four types of head impacts: simulations, football, mixed martial arts, and car crashes. PCA was used to decompose element-wise injury metrics into principal components, with the first principal component (PC1) explaining over 80% of the variance in all datasets. High PC1 coefficients were consistently observed in regions such as the corpus callosum and midbrain. A PCA-based machine learning head model (PCA-MLHM) was then developed to predict PC1 and inverse-transform to reconstruct whole-brain injury metrics. Compared with the previous MLHM, PCA-MLHM achieved similar MPS estimation accuracy while reducing model parameters by 74%, improving interpretability and computational efficiency.

DOI Source Document
Finding the Spatial Co-Variation of Brain Deformation With Principal Component Analysis