This study focuses on developing a data-driven brain dynamics decomposition approach based on principal component analysis (PCA) to analyze the spatial distribution of brain deformation and extract representative patterns of injury metrics in different types of head impacts. The aim is to address the limitations of kinematics-based models that suffer from significant differences in both kinematics and injury metrics across head impact types. The study applies PCA to decompose the patterns of injury metrics, including maximum principal strain (MPS), MPS rate (MPSR), and MPSXMPSR, for all impacts in each impact type and investigates their distributions among brain regions using the first principal component (PC1). The study also develops a deep learning head model (DLHM) to predict PC1 and then inverse-transform to predict for all brain elements. The DLHM reached mean absolute errors of <0.018 for MPS, <3.7 (1/s) for MPSR, and <1.1 (1/s) for MPSXMPSR, which are much smaller than the injury thresholds. The brain injury metric in a dataset can be decomposed into mean components and PC1 with high explained variance, enabling better interpretation of the patterns in brain injury metrics and sensitivity across impact types while reducing the dimensionality of DLHM. However, there are several limitations worth mentioning. Firstly, although KTH FE model has been validated with experimental data on brain-skull relative motion, intracranial pressure, and brain strain, it is relatively limited when compared with more advanced FE models recently developed that consider gyri and sulci as well as cerebral vasculature modeling that significantly affects shear stress modeled by FE models. Therefore, future studies can adopt more advanced FE models to improve fidelity in brain dynamics modeling. Secondly, while this study focuses on modeling PC1 only, other components' influences should not be ignored as they may include more detailed information closely related to the injury outcome warranting further validation. Additionally, the mean components are extracted based on the mean over the training impacts which may lead to a significant decrease in prediction accuracy if test impacts are sampled from a significantly different data distribution. Therefore incorporating more higher-order principal components may enable more accurate DLHMs based on brain dynamics decomposition approach with loss function weighing PC1 more during DLHM training. Overall, this study provides readers with a novel data-driven brain dynamics decomposition approach that enables better interpretation of injury metrics patterns and sensitivity across impact types while reducing DLHM dimensionality. The limitations mentioned above can be addressed in future studies to improve fidelity in brain dynamics modeling and enable even more accurate DLHMs based on brain dynamics decomposition approach.
Error: needs to be re-run
Data-Driven Brain Dynamics Decomposition Approach for Head Impact Analysis
Head impacts are a major cause of traumatic brain injury (TBI) and can have serious long-term consequences. To better understand the effects of head impacts, researchers have developed kinematics-based models to analyze the spatial distribution of brain deformation and extract representative patterns of injury metrics in different types of head impacts. However, these models suffer from significant differences in both kinematics and injury metrics across head impact types.
In this study, researchers proposed a data-driven brain dynamics decomposition approach based on principal component analysis (PCA) to address the limitations of existing kinematics-based models. The aim is to analyze the spatial distribution of brain deformation and extract representative patterns of injury metrics in different types of head impacts.
Principal Component Analysis
The study applied PCA to decompose the patterns of injury metrics, including maximum principal strain (MPS), MPS rate (MPSR), and MPSXMPSR, for all impacts in each impact type and investigated their distributions among brain regions using the first principal component (PC1). The PC1 was used as an indicator for predicting overall brain deformation due to its high explained variance compared with other components.
Deep Learning Head Model
The study also developed a deep learning head model (DLHM) to predict PC1 and then inverse-transform it to predict for all brain elements. The DLHM reached mean absolute errors of <0.018 for MPS, <3.7 (1/s) for MPSR, and <1.1 (1/s) for MPSXMPSR which are much smaller than the injury thresholds set by regulatory bodies such as National Operating Committee on Standards For Athletic Equipment (NOCSAE). This indicates that DLHM is able to accurately predict overall brain deformation even when there are significant differences between training data sets used by kinematic models and test data sets used by DLHM due to sample size or sampling method variations.
Conclusion
Overall, this study provides readers with a novel data-driven brain dynamics decomposition approach that enables better interpretation of injury metrics patterns and sensitivity across impact types while reducing DLHM dimensionality. However, there are several limitations worth mentioning such as limited fidelity when compared with more advanced FE models recently developed that consider gyri and sulci as well as cerebral vasculature modeling that significantly affects shear stress modeled by FE models; ignoring other components' influences may include more detailed information closely related to the injury outcome warranting further validation; mean components extracted based on mean over training impacts may lead to decrease in prediction accuracy if test impacts sampled from significantly different data distribution; incorporating higher order principal components may enable more accurate DLHMs based on brain dynamics decomposition approach with loss function weighing PC1 more during DLHM training etc.. Future studies should address these issues in order improve fidelity in brain dynamics modeling enabling even more accurate DLHMs based on this approach .
Strain and strain rate are effective traumatic brain injury predictors. Kinematics-based models estimating these metrics suffer from significant different distributions of both kinematics and the injury metrics across head impact types. To address this, previous studies focus on the kinematics but not the injury metrics. We have previously shown the kinematic features vary largely across head impact types, resulting in different patterns of brain deformation. This study analyzes the spatial distribution of brain deformation and applies principal component analysis (PCA) to extract the representative patterns of injury metrics (maximum principal strain (MPS), MPS rate (MPSR) and MPSXMPSR) in four impact types (simulation, football, mixed martial arts and car crashes). We apply PCA to decompose the patterns of the injury metrics for all impacts in each impact type, and investigate the distributions among brain regions using the first principal component (PC1). Furthermore, we developed a deep learning head model (DLHM) to predict PC1 and then inverse-transform to predict for all brain elements. PC1 explained >80% variance on the datasets. Based on PC1 coefficients, the corpus callosum and midbrain exhibit high variance on all datasets. We found MPSXMPSR the most sensitive metric on which the top 5% of severe impacts further deviates from the mean and there is a higher variance among the severe impacts. Finally, the DLHM reached mean absolute errors of <0.018 for MPS, <3.7 (1/s) for MPSR and <1.1 (1/s) for MPSXMPSR, much smaller than the injury thresholds. The brain injury metric in a dataset can be decomposed into mean components and PC1 with high explained variance. The brain dynamics decomposition enables better interpretation of the patterns in brain injury metrics and the sensitivity of brain injury metrics across impact types. The decomposition also reduces the dimensionality of DLHM.
