Machine Learning Classification of Traumatic Brain Injury Patients and Healthy Controls Using Multiple Indices of Diffusion Tensor Imaging

doi:10.21203/rs.3.rs-42800/v1

Download PDF

Research article

Machine Learning Classification of Traumatic Brain Injury Patients and Healthy Controls Using Multiple Indices of Diffusion Tensor Imaging

https://doi.org/10.21203/rs.3.rs-42800/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Background: Diffusion tensor imaging (DTI) indices provide quantitative measures of white matter microstructural changes following traumatic brain injury (TBI). However, there is still insufficient evidence for their use as predictive measures. Recently, there has been growing interest in using machine learning (ML) approaches to aid the diagnosis of many neurological and psychiatric illnesses including TBI. The aim of this study is to examine the potential of using multiple DTI indices in conjunction with ML to automate the classification of healthy subjects and patients with TBI across a spectrum of TBI severity.

Methods: Participants were adult patients with chronic TBI (n=26) and age and gender-matched healthy controls (n=26). DTI images were obtained from all the participants. Tract-based spatial statistics (TBSS) analysis was applied to the DTI images. Classification models were built using principle component analysis (PCA) and support vector machines (SVM). Receiver operator characteristic (ROC) curve analysis and area under the curve (AUC) were used to assess the classification performance of the different classifiers.

Results: The whole-brain white matter TBSS analyses showed significantly decreased FA, as well as increased MD, AD, and RD in TBI patients compared with healthy controls (all p-value < 0.01). The PCA and SVM-based ML classification using combined DTI indices classified TBI patients and healthy controls with the accuracy of 90.5% with an area under the curve (AUC) of 93 +/- 0.09.

Conclusion: This study demonstrates the potential of a joint DTI and ML approach for objective classification of TBI patients and healthy controls.

Nuclear Medicine & Medical Imaging

Traumatic Brain Injury

Diffuse axonal injury

Diffusion Tensor Imaging

Machine learning

Support vector machine

Classification model

Traumatic brain injury (TBI) is a leading cause of mortality and morbidity worldwide[1]. Individuals with TBI have an elevated risk of developing numerous neurocognitive and psychiatric illnesses [2, 3]. Diffuse axonal injury (DAI) is one of the most common and important pathologic features of TBI [4, 5]. DAI following TBI occurs as a result of acceleration/deceleration trauma to the brain, which results in a shear disruption of the white matter axons [6, 7].

DAI may include extensive microscopic axonal damage, even in the absence of abnormal findings on conventional CT and MR imaging [8]. On the other hand, especially in the case of mild TBI (mTBI), a traumatic accident may result in a psychologically traumatic experience, which can lead to severe stress reactions. In such a case it is often difficult to observe any indication of structural damage on conventional CT and MRI, and it may therefore be difficult to differentiate patient symptoms caused by DAI from those caused by the psychological impacts of the traumatic experience [9]. Unlike conventional imaging techniques, diffusion tensor imaging (DTI) can identify and quantify white matter microstructural changes following DAI, even in the case of mTBI [10–12]. Therefore, DTI has been considered one of the most promising techniques for the study and diagnosis of DAI following TBI.

DTI measures the directional coherence of water diffusion along white matter axons [13, 14]. A variety of parameters can be obtained from a DTI scan (DTI indices), such as fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD). DTI indices reflect the integrity of white matter microstructure and have been extensively applied as neuroimaging biomarkers to study a range of clinical conditions such as multiple sclerosis, Alzheimer’s disease, and major depressive disorders. FA measures anisotropic water diffusion within white matter fibers and can be used to indicate axonal structural integrity. Reduced FA is believed to reflect a loss of axonal integrity, indicating possible damage to myelin or the axon membrane and/or decreased axonal coherence. MD describes the average magnitude of water diffusion, regardless of the diffusion direction. Increased MD are thought to reflect overall structural disintegration. FA and MD provide sensitive but not specific measures of white matter microstructural alterations. AD is defined as the diffusion of water parallel to white matter fibers, while RD is the diffusion of water perpendicular to white matter fibers [15]. Both AD and RD are called directionality indices, and they can be used to specifically differentiate axonal injury from demyelination in white matter tracts [16, 17].

Even though many previous studies have used DTI to study TBI, there are currently three main shortcomings that need to be addressed. First, while FA and MD were extensively studied in previous literature, AD and RD have been much less commonly studied [18–20]. Second, to date, few studies have used four DTI indices to investigate whole-brain, white matter changes in TBI patients. Most previous DTI studies focused on studying the main white matter tracts known to be susceptible to damage following TBI (e.g. corpus callosum), rather than studying the whole brain [21–23]. While focusing on susceptible white matter tracts may provide more specific results, it might lead to the loss of information, which might potentially be fatal considering the heterogeneity in the distribution of lesions among patients. The third shortcoming, which may in part be due to the other two shortcomings, is the modest diagnostic accuracy of previous DTI studies; for use as a clinical biomarker of TBI, high diagnostic accuracy is essential [24].

To overcome the aforementioned shortcomings, we used multiple DTI indices to conduct a comprehensive study of white matter alterations in the whole brains of patients with DAI following mild, moderate, or severe chronic TBI. We applied the tract-based spatial statistics (TBSS) method for the DTI indices (FA, MD, AD, and RD) to systematically study DAI-associated changes in white matter tracts across the whole brain. We also used a machine learning (ML) approach to maximally use the DTI indices to classify TBI patients and healthy controls with high accuracy. Recently, there has been growing interest in using Machine Learning (ML) to analyze neuroimaging data to find clinical biomarkers for many neurological and psychiatric conditions [24, 25]. Unlike traditional univariate analysis methods such as voxel-based analyses, supervised ML techniques applying multivariate analyses allow for better predictions and inferences at the level of individuals [26].

When applying ML to neuroimaging data we have to consider the nature of the data and what ML technique is suitable for it. In many neuroimaging studies, the number of observations (subjects) is very low (< 1000) in comparison with the huge number of predictor variables per observation (e.g., over 100 000 voxels in a typical brain scan) [27, 28]. Building a ML or prediction model using high dimensional data may lead to overfitting, resulting in performance degradation and poor generalizability. Therefore, dimensionality reduction is an essential step to improve the prediction accuracy [27]. Principal component analysis (PCA) is a commonly used data-driven non-supervised dimensionality reduction technique. PCA reduces redundant features by linearly transforming correlated variables (e.g. voxels in a neuroimaging scan) into a lower number of uncorrelated variables known as principal components (PCs). High-dimensional neuroimaging data can be transformed into relatively few PCs that maximally explain the variance of the data [29]. PCA has been successfully used for dimensionality reduction in previous neuroimaging classification studies involving schizophrenia, Alzheimer’s disease, and major depressive disorder [30].

A support vector machine (SVM) is a supervised ML technique popular in the neuroimaging field, as it can provide a valid classification from high dimensional data [31, 32]. The SVM employs a maximum margin classification algorithm designed to derive the optimal separation between two classes of data (TBI patients group and healthy control group in this study) by identifying a “hyper plane” that crosses the n-dimensional space to separate the training dataset into two pre-defined labels. Each sample within the training dataset is referred to as a vector. The support vectors are defined as the samples that are critical to the positioning of the hyper plane inside the n-dimensional feature space.

In this study, we demonstrate the potential of a joint DTI and ML approach for objective classification of TBI patients and healthy controls. We propose a classification model using combined DTI indices, PCA and SVM that can be used to identify the complex patterns of white matter damage and classify TBI patients and healthy controls with high accuracy.

Participants

Participants were recruited from the outpatient clinic of the neuropsychology unit at the Department of Psychiatry and the Department of Neurosurgery, Kyoto University Hospital.

Inclusion criteria were: 1) age more than 18 years, 2) an injury sustained through significant trauma; 3) a brain MRI or CT scan showing possible diffuse pathology (without large focal lesions (>10 mm3); 4) the injury occurred ≥ 6 months before the study; 5) ability to give informed consent for participation; 6) ability to undergo MRI. Exclusion criteria were: 1) history of another TBI with altered consciousness; 2) history of drug or alcohol abuse; 3) history of neurological or psychiatric disorder before TBI onset; 4) contraindications to MRI (e.g., implanted metal, claustrophobia). Neuropsychiatrists (UK, TM) specialized in the neuropsychiatric aspects of TBI assessed patients MRI and CT scans and confirmed the information concerning the clinical history and residual symptoms related to the inclusion and exclusion criteria mentioned above.

Twenty-six patients with TBI (20 males, mean age of 40.15 years, standard deviation [SD] 14.93) were recruited to the study. According to the Glasgow Coma Scale (GCS) or Japan Coma Scale (JCS; a measure of the severity of impaired consciousness used in Japan), 5 patients (19.2%) had mild TBI, 2 (7.7%) had moderate TBI, and 19 (73.1%) had severe TBI. The relationship between the JCS score and severity of injury has been explored previously [33]. For comparison purposes, twenty-six age- and sex matched healthy controls (20 males, mean age of 40.08 years, SD 12.82) were recruited to the study.

Imaging acquisition

For DTI, diffusion-weighted volumes were acquired on a 3.0-T whole body scanner (MAGNETOM Tim Trio; Siemens, Erlangen, Germany) using a 40-mT/m gradient and a receiver-only eight-channel phased-array head coil. The scanning parameters were: echo time (TE) = 96 ms, repetition time (TR) = 10 500 ms, matrix = 96 × 96, field of view = 192 × 192 mm, 70 contiguous axial slices of 2.0-mm thickness, 81 non-collinear axis motion-probing gradients, and b = 1500 s/mm2. The b = 0 images were acquired before each set of nine diffusion-weighted images, thus giving 90 volumes in total.

Image processing

DTI data were processed using the FMRIB Software Library (FSL) version 5.2 (FSL; Oxford Centre for Functional MRI of the Brain; www.fmrib.ox.ac.uk/fsl) [34,35]. DTI images were registered to the b = 0 image by affine transformations to minimize distortion due to head motion and eddy currents [35]. Images were then brain-extracted using the Brain Extraction Tool [36]. Using the DTIFIT program, the diffusion tensors were calculated for whole brain volumes and fractional anisotropy (FA) maps, axial diffusivity maps (AD [λ1]), radial diffusivity maps (RD [(λ2+λ3)/2]), and mean diffusivity maps (MD [(λ1+ λ2+λ3)/3]) were generated [37- 40]. MD, AD and RD are in units of mm²/s, whereas FA is unitless. Tract-Based Spatial Statistics (TBSS) was used for voxel-wise analysis of the DTI indices maps [40]. TBSS is a fully automated whole-brain analysis technique for applying voxel-wise statistics to diffusion indices while minimizing the effects of misalignment that may occur using a conventional voxel-based analysis method [40].

The TBSS procedure included nonlinear registration of all subjects FA images into the common FMRIB58_FA template space [39]. TBSS performs a non-linear registration to align each FA image to every other one and then calculates the amount of warping needed for the images to be aligned. The most representative image is determined as the one needing the least warping for all other images to be aligned to it [41]. The aligned FA images were then averaged to create a 4D mean FA image, which was then subjected to thinning to create a mean FA skeleton representing the center of all white matter tracts, thereby removing partial-volume confounds. The FA skeleton was then thresholded at an FA value of 0.2 to limit the effects of poor alignment across subjects, to exclude areas with extremely low mean FA, and to ensure that grey matter and CSF voxels were excluded from the skeleton. The same non-linear transformation steps were applied to the MD, AD, and RD maps. For statistical analysis, the randomize tool in FSL was used to conduct a non-parametric permutation-based statistics using the Threshold-Free Cluster Enhancement (TFCE) method with 10,000 permutations to investigate group differences in FA, MD, AD and RD.Voxelwise maps were thresholded at p < 0.05 and corrected for multiple comparisons with family-wise error rate (FWE). The significant white matter clusters were identified with reference to the atlas tool JHU ICBM-DTI-81 white matter labels. To qualitatively assess structural difference between TBI patients and healthy controls groups, mean DTI indices for each subject were extracted by averaging FA, MD, AD, and RD for the significant white matter clusters using fslstats tool in FSL.

Dimensionality reduction and feature extraction

PCA was applied to FA, MD, AD, and RD skeletonized maps, in which each voxel represented a variable in the cross validation training dataset, and the same transformation was then applied to the test dataset. A voxel-wise approach was used to combine FA, MD, AD, and RD into one dataset named “ALL”, and PCA was applied to this ALL dataset in the same manner.

Support Vector Machines

A support vector machine (SVM) was used to perform ML analysis using the PCs of the DTI indices. For each DTI index, an SVM classification task was trained to distinguish TBI patients from healthy controls. PCs of the skeletonized maps of each DTI index and the ALL dataset were evaluated. Five-fold cross validation was used to yield an unbiased assessment of the classification method and prevent overestimation. A linear kernel SVM was chosen as a classifier and the hyperparameter (C) of the linear kernel SVM was fixed to 1.0. To evaluate the different classification tasks, the mean accuracy rate and its SD was calculated for the five-fold cross validation for each classification task (Fig.1). To validate the robustness of the classification results, 1000 times permutation tests were conducted to assess the statistical significance of the classification accuracy scores. To further estimate the performance of the different classification task, receiver operating characteristic curves (ROC) were plotted for each classifier and the areas under the curves (AUC) were obtained. The AUC quantifies the overall ability of the classifier to distinguish between the TBI and the healthy subjects. The ML analysis including feature extraction was conducted in Python using the Python library scikit-learn.

Whole brain TBSS analysis of multiple DTI indices

Whole-brain TBSS analysis of the DTI diffusion indices showed a major disruption in the white matter microstructure in the TBI patients group. Significantly decreased FA and increased MD, AD, and RD were present in several major white matter tracts in the TBI group compared with the healthy control group (corrected p value < 0.01) (Fig. 2 & Fig.3). Fig.4 showed boxplots graphical representation for the mean values and standard deviations of FA, MD, AD, and RD for TBI patients and healthy control groups.

Dimensionality reduction and feature extraction

PCA was applied to FA, MD, AD, RD and ALL during cross validation in both training and test datasets. Table 1. Shows average number of PCs that explained 90% of the variance between the TBI and healthy controls group across the five cross validation folds.

SVM Classification

The ML multivariate analysis using SVM with five-fold cross validation was able to classify the TBI patients and healthy control groups using PCs of the skeletonized maps of each DTI index (number of voxels = 97698) and the ALL dataset (number of voxels = 390792). High accuracy (90.5%) was achieved by the classifier trained on the combined indices dataset (ALL) and (86.5%) the classifier trained on the FA maps (all p-value < 0.01) (Table 1).Table 1 below shows the average number of PCs and the mean accuracy score for each classifier.

Table 1: Mean accuracy score for each classification task

Classification task	Average number of PCs	Accuracy (+/- SD)
FA maps	32	0.865 (+/- 0.052)
MD maps	29.4	0.829(+/- 0.073)
AD maps	33.6	0.845(+/- 0.114)
RD maps	29.4	0.845(+/- 0.114)
ALL maps	31.6	0.905(+/- 0.096)

FA: Fractional anisotropy; MD: mean diffusivity; AD: axial diffusivity; RD: radial diffusivity; ALL: voxel-wise combination of FA, MD, AD, and RD in one dataset; PCs: Principle components; SD: standard deviation

ROC curves for each classifier are shown in Fig. 5. The AUCs were 93% and 89% for the ALL and FA classifier tasks respectively.

DTI indices provide quantitative measures of white matter microstructural changes following TBI. However, there is still insufficient evidence for their use as predictive measures. In the current study we studied the potential of using whole-brain multiple DTI indices in conjunction with ML algorithms as a predictive measure to automate the classification of TBI patients and healthy controls. We applied DTI to 26 subjects with chronic TBI and 26 age- and gender-matched healthy controls to obtain values for multiple diffusion indices within the white matter tracts of each subject.

The TBSS analysis of the DTI indices showed significantly decreased FA and increased MD, AD, and RD in the TBI patients in comparison with the healthy control subjects. These results are consistent with previous studies [42-48]. Many studies have reported reduced FA in chronic moderate to severe TBI, even in the absence of visible lesions on conventional CT and structural MRI [8,46,47]. Although FA and MD changes have been extensively reported in TBI, only a limited number of studies compared AD and RD between chronic TBI patients and healthy controls. Kinnunen et.al. conducted a TBSS analysis on 28 TBI patients (8 mTBI and 20 moderate/severe TBI) and 26 healthy controls. They found that TBI patients showed large areas of reduced FA and increased MD, as well as increased AD and RD [48]. Perez et al. studied AD and RD in 16 TBI patientswith chronic moderate to severe TBI. Their results demonstrated disproportionally high AD and RD in TBI patients [49]. Cubon et al. studied a total of 39 subjects with chronic TBI (22 mTBI and 17 moderate/severe TBI), and in both a whole-brain analysis and one using specific regions of interest, they found that moderate to severe TBI patients had increased AD and RD [50].

DTI indices derived from the diffusion images were used together with PCA and SVMs to classify the TBI patients and healthy controls. Five classification models were examined, and the model using the combined indices (ALL) was found to be the most accurate with 90.5% accuracy and a high AUC (93%, SD 0.09). Previous studies that have used ML in conjugation with DTI images to differentiate TBI patients from healthy controls have obtained different approaches. These studies used different ML statistical techniques and different DTI parameters and obtained variable levels of accuracy [51,52]. Mitra et al. applied a decision ensemble ML technique with 10-fold cross validation to DTI data (FA and Network-Based Statistics [NBS]) and revealed an overall classification accuracy of 68% [53]. Lui et al. used other MRI-based imaging parameters in addition to DTI (T1 and rs-fMRI) and tested several different classification approaches on 23 mild TBI patients and 25 healthy controls. They obtained an accuracy of 86% with a multilayer perceptron (neural network) using only relevant variables, and 80% with a Bayesian network using all variables [54]. Fagerholm et al. applied linear SVM analysis to DTI graph metrics of white matter connectivity data from moderate/severe TBI patients and revealed an overall model accuracy of 93.4% [55]. Our model was able to classify TBI patients and healthy controls with 90.5% accuracy. Although Fagerholm et al. obtained higher accuracy than our model, a practical advantage of our method is that it uses well-known DTI indices (FA, MD, AD, and RD), rather than graph metrics of white matter connectivity.

The findings of this study should be considered in the context of certain limitations. First, in the case of inaccurate registration of individual images into standard space, partial volume effects can lead to abnormal DTI indices. To eliminate any likelihood of such an error, we applied TBSS preprocessing to minimize this issue and visually checked the registration accuracy. There was a strong correlation between the different DTI indices, and the high classification performance may reflect the underlying variance shared across the different diffusion measures. DTI measures change over time after TBI, and our results are therefore likely to be specific to the chronic phase (> 6 months) post-injury. Investigating TBI in the chronic phase might potentially be problematic, as patients frequently show some degree of brain atrophy. If this is the case, changes in DTI indices such as lower FA may be due to partial volume effects such as contamination of measurements by cerebrospinal fluid. Again, using the TBSS preprocessing approach should have minimized this potential issue, as the white matter was skeletonized and only the central points of tracts were studied, greatly reducing the risk of partial volume effects [28, 52].

Second, MD is mathematically related to AD and RD and combining them together in one dataset (ALL) might affect the classification performance. However, we combined them together with FA as different DTI indices convey different aspects of white matter microstructure which are potentially complementary for the discrimination power of the classification model [56-59]. Moreover, applying PCA prior to classification removed any redundant information and intercorrelation in the combined DTI indices dataset [59].

Third, our sample size was small, which may have affected the generalizability of our findings, especially regarding mTBI. Therefore, future work with larger patient groups is needed to build a more generalizable predictive model. It is important to note that the sample studied were predominately moderate to severe TBI as recruitment was limited to TBI patients who attend the outpatient clinic of the neuropsychology unit. Although these patients may have detectable abnormalities on clinical MRI, it is important to validate an automated method that can be used as a predictive tool for a wide spectrum of TBI severity.

Fourth, our model needs to be validated against traditional methods; therefore, future work will focus on the statistical validation of the model. Our model did not include many important clinical and social outcomes, and future work should explicitly model other clinical measures such as acute imaging changes, injury severity, neurocognitive consequences, and measures of gray matter atrophy. Finally, our study is cross-sectional, and a longitudinal study would provide better predictions for a clinical application.

We conducted a comprehensive DTI study of patients with mild, moderate, or severe TBI in the chronic stage. Our study results suggest that ML can be used with DTI indices as a predictive tool for the automated classification of TBI.

Ethics approval and consent to participate

This study was approved by the Committee on Medical Ethics of Kyoto University and all procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. Written informed consent was obtained from all individual participants included in the study.

Consent for publication

Not applicable.

Availability of data and materials

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare that they have no competing interests.

Funding

This study did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Authors' contributions

All authors contributed to the study conception and methodology. Conceptualization, methodology, investigation, formal analysis and visualization were performed by Hiba Abuelgasim Fadlelmoula Abdelrahman. Gaku Fujimoto, Ubukata Shiho, Ueda Keita and Toshiya Murai have contributed to the conceptualization, methodology and investigation. Naoya Oishi and Toshihiko Aso have contributed to the validation and formal analysis. The first draft of the manuscript was written by Hiba Abuelgasim Fadlelmoula Abdelrahman and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Acknowledgements

Not applicable

TBI: Traumatic brain injury

DAI: Diffusion axonal injury

DTI: Diffusion tensor Imagining

FA: Fractional anisotropy

MD: Mean diffusivity

AD: Axial diffusivity

RD: Radial diffusivity

ALL: Voxel-wise combination of FA, MD, AD, and RD in one dataset

TBSS: Tract-based spatial statistics

TFCE: Threshold-Free Cluster Enhancement

FEW: Family-wise error rate

ML: Machine Learning

PCA: Principle component analysis

SVM: Support vector machine

PCs: Principal components

GCS: Glasgow Coma Scale

JCS: Japan Coma Scale

ROC: Receiving operating curve

AUC: Area under the curve

Hyder AA, Wunderlich CA, Puvanachandra P, Gururaj G, Kobusingye OC. The impact of traumatic brain injuries: a global perspective. NeuroRehabilitation. 2007;22:341–53. https://doi.org/10.3233/nre-2007-22502.
Koponen S, Taiminen T, Portin R, Himanen L, Isoniemi H, Heinonen H, Hinkka S, Tenovuo O. Axis I and II psychiatric disorders after traumatic brain injury: a 30-year follow-up study. Am J Psychiatry. 2002;159:1315–21. https://doi.org/10.1176/appi.ajp.159.8.1315.
Holsinger T, Steffens DC, Phillips C, Helms MJ, Havlik RJ, Breitner JC, Guralnik JM, Plassman BL. Head injury in early adulthood and the lifetime risk of depression. Arch Gen Psychiatry. 2002;59:17–22. https://doi.org/10.1001/archpsyc.59.1.17.
Smith DH, Meaney DF. Axonal damage in traumatic brain injury. Neuroscientist. 2000;6:483–95. https://doi.org/10.1177/107385840000600611.
Browne KD, Chen X-H, Meaney DF, Smith DH. Mild traumatic brain injury and diffuse axonal injury in swine. J Neurotrauma. 2011;28:1747–55. https://doi.org/10.1089/neu.2011.1913.
Gennarelli TA, Thibault LE, Adams JH, Graham DI, Thompson CJ, Marcincin RP. Diffuse axonal injury and traumatic coma in the primate. Ann Neurol. 1982;12:564–74. https://doi.org/10.1002/ana.410120611.
Werner C, Engelhard K. Pathophysiology of traumatic brain injury. Br J Anaesth. 2007;99:4–9. https://doi.org/10.1093/bja/aem131.
Hulkower M, Poliak D, Rosenbaum S, Zimmerman M, Lipton ML. A decade of DTI in traumatic brain injury: 10 years and 100 articles later. AJNR Am J Neuroradiol. 2013;34:2064–74. https://doi.org/10.3174/ajnr.A3395.
Klimova A, Korgaonkar MS, Whitford T, Bryant RA. Diffusion tensor imaging analysis of mild traumatic brain injury and posttraumatic stress disorder. Biol Psychiatry Cogn Neurosci Neuroimaging. 2019;4:81–90. https://doi.org/10.1016/j.bpsc.2018.10.004.
Assaf Y, Pasternak O. Diffusion tensor imaging (DTI)-based white matter mapping in brain research: a review. J Mol Neurosci. 2008;34:51–61. https://doi.org/10.1007/s12031-007-0029-0.
Kinnunen KM, Greenwood R, Powell JH, Leech R, Hawkins PC, Bonnelle V, Patel MC, Counsell SJ, Sharp DJ. White matter damage and cognitive impairment after traumatic brain injury. Brain. 2010;134:449–63. https://doi.org/10.1093/brain/awq347.
Niogi SN, Mukherjee P. Diffusion tensor imaging of mild traumatic brain injury. J Head Trauma Rehabil. 2010;25:241–55. https://doi.org/10.1097/HTR.0b013e3181e52c2a.
Alexander AL, Lee JE, Lazar M, Field AS. Diffusion tensor imaging of the brain. Neurotherapeutics. 2007;4:316–29. https://doi.org/10.1016/j.nurt.2007.05.011.
Douglas DB, Iv M, Douglas PK, Ariana A, Vos SB, Bammer R, Zeineh M, Wintermark M. Diffusion tensor imaging of TBI: potentials and challenges. Top Magn Reson Imaging. 2015;24:241. https://doi.org/10.3390/medsci7010002.
Aung WY, Mar S, Benzinger TL. Diffusion tensor MRI as a biomarker in axonal and myelin damage. Imaging Med. 2013;5:427.
Liu Y, Duan Y, He Y, Yu C, Wang J, Huang J, Ye J, Parizel PM, Li K, Shu N. Whole brain white matter changes revealed by multiple diffusion metrics in multiple sclerosis: a TBSS study. Eur J Radiol. 2012;81:2826–32. https://doi.org/10.1016/j.ejrad.2011.11.022.
Shenton ME, Hamoda H, Schneiderman J, Bouix S, Pasternak O, Rathi Y, Vu M-A, Purohit MP, Helmer K, Koerte I. A review of magnetic resonance imaging and diffusion tensor imaging findings in mild traumatic brain injury. Brain Imaging Behav. 2012;6:137–92. https://doi.org/10.1007/s11682-012-9156-5.
Witwer BP, Moftakhar R, Hasan KM, Deshmukh P, Haughton V, Field A, Arfanakis K, Noyes J, Moritz CH, Meyerand ME. Diffusion-tensor imaging of white matter tracts in patients with cerebral neoplasm. J Neurosurg. 2002;97:568–75. https://doi.org/10.3171/jns.2002.97.3.0568.
Inglese M, Makani S, Johnson G, Cohen BA, Silver JA, Gonen O, Grossman RI. Diffuse axonal injury in mild traumatic brain injury: a diffusion tensor imaging study. J Neurosurg. 2005;103:298–303. https://doi.org/10.3171/jns.2005.103.2.0298.
Kraus MF, Susmaras T, Caughlin BP, Walker CJ, Sweeney JA, Little DM. White matter integrity and cognition in chronic traumatic brain injury: a diffusion tensor imaging study. Brain. 2007;130:2508–19. https://doi.org/10.1093/brain/awm216.
Chang MC, Kim SH, Kim OL, Bai DS, Jang SH. The relation between fornix injury and memory impairment in patients with diffuse axonal injury: a diffusion tensor imaging study. NeuroRehabilitation. 2010;26:347–53. https://doi.org/10.3233/NRE-2010-0572.
Kasahara K, Hashimoto K, Abo M, Senoo A. Voxel-and atlas-based analysis of diffusion tensor imaging may reveal focal axonal injuries in mild traumatic brain injury—comparison with diffuse axonal injury. Magn Reson Imaging. 2012;30:496–505. https://doi.org/10.1016/j.mri.2011.12.018.
Kumar R, Gupta RK, Husain M, Chaudhry C, Srivastava A, Saksena S, Rathore RK. Comparative evaluation of corpus callosum DTI metrics in acute mild and moderate traumatic brain injury: its correlation with neuropsychometric tests. Brain Inj. 2009;23:675–85. https://doi.org/10.1080/02699050903014915.
Orru G, Pettersson-Yeo W, Marquand AF, Sartori G, Mechelli A. Using support vector machine to identify imaging biomarkers of neurological and psychiatric disease: a critical review. Neurosci Biobehav Rev. 2012;36:1140–52. https://doi.org/10.3389/fpsyt.2013.00187.
Linden DE. The challenges and promise of neuroimaging in psychiatry. Neuron. 2012;73:8–22. https://doi.org/10.1016/j.neuron.2011.12.014.
Zarogianni E, Moorhead TW, Lawrie SM. Towards the identification of imaging biomarkers in schizophrenia, using multivariate pattern classification at a single-subject level. Neuroimage Clin. 2013;3:279–89. https://doi.org/10.1016/j.nicl.2013.09.003.
Mwangi B, Tian TS, Soares JC. A review of feature reduction techniques in neuroimaging. Neuroinformatics. 2014;12:229–44. https://doi.org/10.1007/s12021-013-9204-3.
:13 – 5

https://doi.org/10.1016/S2215-0366(15)00549-0

Passos IC, Mwangi B, Kapczinski F. Big data analytics and machine learning: 2015 and beyond. Lancet Psychiatry 2016;3:13 – 5. https://doi.org/10.1016/S2215-0366(15)00549-0.
Guyon I, Elisseeff A. An introduction to variable and feature selection. J Mach Learn Res. 2003;3:1157–82. https://psycnet.apa.org/doi/10.1162/153244303322753616.
Joliffe I. Principal component analysis. NY: Springer.2; 2002.
O'Dwyer L, Lamberton F, Bokde AL, Ewers M, Faluyi YO, Tanner C, Mazoyer B, O'Neill D, Bartley M, Collins DR. Using support vector machines with multiple indices of diffusion for automated classification of mild cognitive impairment. PloS one. 2012;7:e32441. https://doi.org/10.1371/journal.pone.0032441.
Hellyer PJ, Leech R, Ham TE, Bonnelle V, Sharp DJ. Individual prediction of white matter injury following traumatic brain injury. Ann Neurol. 2013;73:489–99. https://doi.org/10.1002/ana.23824.
Namiki J, Yamazaki M, Funabiki T, Suzuki M, Fujishima S, Hori S, Aikawa N. Difficulty and inaccuracy of assessment of the consciousness level by the Glasgow Coma Scale: comparison with the Japan Coma Scale. J Jpn Soc Emerg Med. 2007;10:20–5.
Smith SM. Fast robust automated brain extraction. Hum Brain Mapp. 2002;17:143–55. https://doi.org/10.1002/hbm.10062.
Smith SM, Jenkinson M, Woolrich MW, Beckmann CF, Behrens TE, Johansen-Berg H, Bannister PR, De Luca M, Drobnjak I, Flitney DE. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage. 2004;23:208-S19. https://doi.org/10.1016/j.neuroimage.2004.07.051.
Woolrich MW, Jbabdi S, Patenaude B, Chappell M, Makni S, Behrens T, Beckmann C, Jenkinson M, Smith SM. Bayesian analysis of neuroimaging data in FSL. Neuroimage. 2009;45:173-S86.
Ennis DB, Kindlmann G. Orthogonal tensor invariants and the analysis of diffusion tensor magnetic resonance images. Magn Reson Med. 2006;55:136–46. https://doi.org/10.1002/mrm.20741.
Behrens TE, Woolrich MW, Jenkinson M, Johansen-Berg H, Nunes RG, Clare S, Matthews PM, Brady JM, Smith SM. Characterization and propagation of uncertainty in diffusion‐weighted MR imaging. Magn Reson Med. 2003;50:1077–88. https://doi.org/10.1002/mrm.10609.
Elsevier

MRI atlas of human white matter
Mori S, Wakana S, Van Zijl PC, Nagae-Poetscher L. MRI atlas of human white matter: Elsevier; 2005.
Smith SM, Jenkinson M, Johansen-Berg H, Rueckert D, Nichols TE, Mackay CE, Watkins KE, Ciccarelli O, Cader MZ, Matthews PM. Tract-based spatial statistics: voxelwise analysis of multi-subject diffusion data. Neuroimage. 2006;31:1487–505. https://doi.org/10.1016/j.neuroimage.2006.02.024.
Andersson JL, Jenkinson M, Smith S. Non-linear registration aka Spatial normalisation FMRIB Technial Report TR07JA2. FMRIB Analysis Group of the University of Oxford. 2007.
Arfanakis K, Haughton VM, Carew JD, Rogers BP, Dempsey RJ, Meyerand ME. Diffusion tensor MR imaging in diffuse axonal injury. AJNR Am J Neuroradiol. 2002;23:794–802.
Aoki Y, Inokuchi R, Gunshin M, Yahagi N, Suwa H. Diffusion tensor imaging studies of mild traumatic brain injury: a meta-analysis. J Neurol Neurosurg Psychiatry. 2012;83:870–6.
Mayer AR, Ling JM, Yang Z, Pena A, Yeo RA, Klimaj S. Diffusion abnormalities in pediatric mild traumatic brain injury. J Neurosci. 2012;32:17961–9. https://doi.org/10.1523/JNEUROSCI.3379-12.2012.
Mannell MALJ. Gasparovic C. Phillips JP, Doezema D, Reichard R. Yeo RA A prospective diffusion tensor imaging study in mild traumatic brain injury. Neurology. 2010;74:643–50. https://doi.org/10.1212/WNL.0b013e3181d0ccdd.
Wallace EJ, Mathias JL, Ward L. Diffusion tensor imaging changes following mild, moderate and severe adult traumatic brain injury: a meta-analysis. Brain Imaging Behav. 2018;12:1607–21. https://doi.org/10.1007/s11682-018-9823-2.
Nakayama N, Okumura A, Shinoda J, Yasokawa Y, Miwa K, Yoshimura S, Iwama T. Evidence for white matter disruption in traumatic brain injury without macroscopic lesions. J Neurol Neurosurg Psychiatry. 2006;77:850–5.
Xu J, Rasmussen I-A, Lagopoulos J, Håberg A. Diffuse axonal injury in severe traumatic brain injury visualized using high-resolution diffusion tensor imaging. J Neurotrauma. 2007;24:753–65. https://doi.org/10.1089/neu.2006.0208.
Perez AM, Adler J, Kulkarni N, Strain JF, Womack KB, Diaz-Arrastia R. Marquez de la Plata CD. Longitudinal white matter changes after traumatic axonal injury. J Neurotrauma. 2014;31:1478–85. https://doi.org/10.1089/neu.2013.3216.
Cubon VA, Putukian M, Boyer C, Dettwiler A. A diffusion tensor imaging study on the white matter skeleton in individuals with sports-related concussion. J Neurotrauma. 2011;28:189–201. https://doi.org/10.1089/neu.2010.1430.
Martinez BI, Stabenfeldt SE. Current trends in biomarker discovery and analysis tools for traumatic brain injury. J Biol Eng. 2019;13:16. https://doi.org/10.1186/s13036-019-0145-8.
Mateos-Perez JM, Dadar M, Lacalle-Aurioles M, Iturria-Medina Y, Zeighami Y, Evans AC. Structural neuroimaging as clinical predictor: A review of machine learning applications. Neuroimage Clin. 2018;20:506–22. https://doi.org/10.1016/j.nicl.2018.08.019.
Mitra J, Shen K-k, Ghose S, Bourgeat P, Fripp J, Salvado O, Pannek K, Taylor DJ, Mathias JL, Rose S. Statistical machine learning to identify traumatic brain injury (TBI) from structural disconnections of white matter networks. Neuroimage. 2016;129:247–59. https://doi.org/10.1016/j.neuroimage.2016.01.056.
Lui YW, Xue Y, Kenul D, Ge Y, Grossman RI, Wang Y. Classification algorithms using multiple MRI features in mild traumatic brain injury. Neurology. 2014;83:1235–40. https://doi.org/10.1212/WNL.0000000000000834.
Fagerholm ED, Hellyer PJ, Scott G, Leech R, Sharp DJ. Disconnection of network hubs and cognitive impairment after traumatic brain injury. Brain. 2015;138:1696–709. https://doi.org/10.1093/brain/awv075.
Wee CY, Yap PT, Li W, Denny K, Browndyke JN, Potter GG, Welsh-Bohmer KA, Wang L, Shen D. Enriched white matter connectivity networks for accurate identification of MCI patients. Neuroimage. 2011;54(3):1812–22. https://doi.org/10.1016/j.neuroimage.2010.10.026.
Zhang YT, Liu SQ. Individual identification using multi-metric of DTI in Alzheimer’s disease and mild cognitive impairment. Chin Phys B. 2018;8:088702.
Zhao M, Liu J, Cai W, Li J, Zhu X, Yu D, Yuan K. Support vector machine based classification of smokers and nonsmokers using diffusion tensor imaging. Brain Imaging and Behavior. 2019 Aug 19:1–9. https://doi.org/10.1007/s11682-019-00176-7.
Doan NT, Engvig A, Persson K, Alnæs D, Kaufmann T, Rokicki J, Córdova-Palomera A, Moberget T, Brækhus A, Barca ML, Engedal K. Dissociable diffusion MRI patterns of white matter microstructure and connectivity in Alzheimer’s disease spectrum. Scientific reports. 2017;7:45131.

Download PDF

Editorial decision: Major revision
23 Jan, 2021
Review #4 received at journal
16 Jan, 2021
Reviewer #4 agreed at journal
24 Dec, 2020
Review #3 received at journal
07 Sep, 2020
Review #2 received at journal
07 Sep, 2020
Reviewer #3 agreed at journal
17 Aug, 2020
Reviewer #2 agreed at journal
16 Aug, 2020
Review #1 received at journal
14 Aug, 2020
Reviewer #1 agreed at journal
11 Aug, 2020
Reviewers invited by journal
23 Jul, 2020
First submitted to journal
14 Jul, 2020
Editor assigned by journal
14 Jul, 2020
Submission checks completed at journal
13 Jul, 2020
Editor invited by journal
13 Jul, 2020

You are reading this latest preprint version

Machine Learning Classification of Traumatic Brain Injury Patients and Healthy Controls Using Multiple Indices of Diffusion Tensor Imaging

Status:

Version 1

Abstract

Figures

Introduction

Material And Methods

Results

Discussion

Limitations

Conclusions

Declarations

Abbreviations

References

Status:

Version 1