Quantitative multimodal imaging in traumatic brain injuries producing impaired cognition

Abstract

Purpose of Review:

Cognitive impairments are a devastating long-term consequence following traumatic brain injury (TBI). This review provides an update on the quantitative muti-modal neuroimaging studies that attempt to elucidate the mechanism(s) underlying cognitive impairments and their recovery following TBI.

Recent Findings:

Recent studies have linked individual specific behavioral impairments and their changes over time to physiological activity and structural changes using EEG, PET and MRI. Multimodal studies that combine measures of physiological activity with knowledge of neuroanatomical and connectivity damage have also illuminated the multifactorial function-structure relationships that underlie impairment and recovery following TBI.

Summary:

The combined use of multiple neuroimaging modalities, with focus on individual longitudinal studies, has the potential to accurately classify impairments, enhance sensitivity of prognoses, inform targets for interventions, and precisely track spontaneous and intervention-driven recovery.

Introduction:

Traumatic brain injury (TBI) is a leading cause of death and long-term disability in the United States and worldwide[1]. In addition to 50,000 TBI-related civilian deaths in the US each year, there are over 1 million Americans treated in emergency departments and 235,000 hospital admissions annually[2]. Cognitive impairment is a central determinant of vocational, social and emotional functioning, and approximately 70% of TBI survivors experience chronic cognitive impairments (TBI Model Systems National Data and Statistical Center). Currently, there are more than 5.3 million persons in the US alone with chronic cognitive dysfunction as a result of TBI[3].

Despite this high disease burden, mechanisms underlying impaired cognitive function have not been characterized fully, limiting clinicians’ ability to accurately diagnose, prognose, track, and identify mechanisms of impairment. Current prognostic models rely heavily on injury severity, lack sensitivity[4] and account for only a moderate portion of variance in clinical outcomes[5]. Unsurprisingly, to date, there are no effective therapeutic interventions for cognitive impairment following TBI[6]. The failure of most TBI clinical trials has been attributed to insufficient knowledge of the pathophysiology of deficits which precludes development of targeted therapies for an inherently heterogeneous disease[7]. Here, we discuss and summarize efforts to overcome these barriers using quantitative multimodal neuroimaging techniques.

Electroencephalography (EEG)

The electrophysiological monitoring of voltage fluctuations resulting from neuronal activity has long been used to objectively study cognitive functioning following TBI because of its excellent time resolution[8–10]. Specifically, the amplitude and latency of Event Related Potentials (ERPs), which is recorded during cognitive behavioral tasks and represents stimulus detection and evaluation, have been extensively used to diagnose [11,12], understand, and predict cognitive impairment[8]. Early perceptual discrimination of stimulus (N100, a negative peak ~ 100 msecs post stimulus[13]) has been shown to be mostly preserved following TBI[14,15]. Conversely, the classic P300, a positive peak ~300 msecs post-stimuli associated with attentional and working memory resources[16], has reliably been shown to have smaller amplitudes and longer latencies in response to high-conflict stimuli in individuals with TBI compared to healthy controls (HCs) [14,15,17–19] and individual amplitudes have been associated with attention, executive functioning, and processing speed[18,19]. In one study, P3A and P3b, both components of the P300, were found to be associated with executive function impairment, with the former relating to sub-acute impairments and the latter with chronic impairments[20**]. In another recent study, individual longitudinal changes in P300 amplitude and latency following TBI were related to changes in psychological outcomes, suggesting the utility of P300 as an objective biomarker for tracking of recovery[21**]. A neural component of the failure to evaluate and regulate (conflict processing), the N450, in HCs reveals higher amplitude for incongruent trials but shows no discrimination between incongruent and congruent processing in individuals with TBI[22,23].

In contrast to ERP analyses, spectral analyses of oscillatory activity offer a plethora of possible variables (power, frequency, coherence, etc.) during both resting and during-task EEG that can be related to cognitive behavior. Decreased alpha power, increased theta power and decreased beta synchronization during resting EEG have been reported in individuals with TBI but were not associated with global cognition[24,25]; however, delta and theta coherence were positively correlated with working memory and resting beta synchronization was related to language processing and executive functioning[26]. Resting delta/alpha power ratio, which was significantly associated with confusion and attention symptoms in individuals remaining in post-traumatic confusional state also successfully predicted cognitive recovery at 1, 2, and 5 years post-injury[27*] and theta synchronization in acute stages successfully predicted recovery of executive functioning 2 months later[28**]. Analyses of spectral power during cognitive tasks have showed reduced theta synchronization and phase coherence and no change in alpha synchronization preceding high-conflict stimuli in individuals with TBI compared to HCs[19,29]. An increase in midline theta power during conflict processing (and not during rest) was positively correlated with performance on an executive attention task in individuals with TBI[30*, see Figure 1] and lower peak alpha power during an attention task was related to better performance on cognitive assessments of attention and executive function[31*]. In clinical trials of transcranial direct current stimulation (tDCS) which demonstrated improved cognition, spectral EEG was successfully used as an outcome assessment: 1) a concomitant increase in coherence of alpha and theta bands correlated with improvement in minimally conscious patients[32**] and 2) a decrease in theta power correlated with improvements in processing speed, attention, and memory in individuals with acute to subacute TBI[33].

Figure 1:

Multimodal imaging derivations demonstrate associations with cognitive impairment following TBI: A. Event Related Potentials (ERPs) and spectral power derived from EEG recorded during performance of Attention Network Test [108,109]. The spectral power significantly associates with executive attention behavior [30]. Structural and functional connectome measures, and measures of the relationship between the two correlates with B) longitudinal cognitive recovery in mild TBI [71] and C) cross-sectional measures of consciousness in severe brain injury [70] (anatomical brain injury (ABI). D. Flumazenil PET shows differential activations in TBI (compared to age-matched HC). Activity within the anterior forebrain mesocircuit [110] predict executive attention. Unpublished data courtesy of Shah & Kuceyeski.

While EEG offers a clinically feasible, inexpensive, task-specific activity report of cognitive behavior, able to dissociate when/where cognitive processing fails, the signal to noise ratio (neuronal activity filtered through the scalp) is low, and spatial resolution and sources can be difficult to decipher. However, by relating EEG to individual behavior and tracking its changes as cognitive function recovers, EEG has the potential to provide valuable insight into the pathophysiology that can be exploited for biomarkers and treatment planning.

Positron Emission Tomography (PET)

PET imaging is a powerful emerging technique that allows for non-invasive insight into cellular or subcellular brain mechanisms associated with cognition, depending on the ligand used[34].

Fluorodeoxyglucose (FDG) PET, used to measure tissue glucose metabolism, illustrates the chronic effects of TBI both globally[35] and within specific brain regions[36–38]. Following TBI, reduced metabolism in several regions has been related to cognition: 1) decreased FDG signal in temporal and frontal lobes[37] correlates with worse memory and executive functioning[35,39], 2) glucose metabolism in cingulate gyrus and medial frontal gyrus[36,37] is negatively associated with IQ[37], and 3) reduced middle orbitofrontal cortex metabolism negatively associates with episodic memory[38]. FDG-PET has also proven to be sensitive to recovery of consciousness in TBI, with thalamic metabolism grading with level of recovery: lowest in individuals in a minimally conscious state (MCS) [38], increasing in those in Post-traumatic Amnesia (PTA), and even more in those who have emerged from PTA. An increase in glucose metabolism in the left prefrontal and medial temporal regions accounted for more than 80% of variance in improvement of executive functioning following pharmacological intervention of Amantadine in individuals with TBI[40].

More recently, other PET ligands have been used to study cognition in TBI. Amyloid and Flortaucipir PET have offered evidence of increased amyloid deposition and tau accumulation, respectively, in individuals with TBI compared to HCs, but no associations were found between amyloid and tau pathology and cognitive functioning[41,42]. Flumazenil (FMZ), which binds to the GABA inhibitory neurons and can serve as a proxy for cellular integrity, is a potentially sensitive marker of cognitive impairment[38,43, see Figure 1]. Differential patterns of FMZ binding, reflecting altered integrity, have been found in individuals with TBI compared to HCs[38,43]. Decreased FMZ binding in the right thalamus has been associated with worse full-scale and performance IQ, decreased FMZ binding in the left medial frontal gyrus with worse full-scale, performance, and verbal IQ[43], and decreased FMZ binding in the left angular gyrus with poorer episodic memory[38]. Lastly, functional PET collected during a memory retrieval task showed a different pattern of regional blood flow compared to HCs, despite similar task performance between groups, suggesting functional reorganization in TBI[44].

PET is expensive and not as widely available as other imaging modalities. However, PET provides a unique insight into the subcellular and cellular process which can potentially serve as tools for appropriate stratification and studies of mechanism[45,46].

Magnetic Resonance Imaging (MRI)

The noninvasive, non-contrast, and versatile nature of MRI makes it especially suited for studies of pathology-related changes in brain anatomy, white matter connectivity, and physiological activation patterns that may contribute to cognitive impairment after TBI[47].

Hemorrhage, petechiae, and diffuse axonal injury (DAI) associated with TBI can be identified easily on anatomical MRI (T2, T2-FLAIR)[48,49**], along with neurodegenerative effects of injury[50–53]. A recent study showed that MRI-based evidence of DAI following acute TBI, particularly in the corpus callosum and brainstem, was related to worse global cognitive functioning at 12 months post-injury[49]. Whole brain and regional (frontal and tempo-parietal) volumes have also been shown to be correlated with global cognitive impairment in TBI[48]. Regional reductions in and increased rates of atrophy of grey and white matter have been found in individuals with TBI compared to HCs. From the acute to chronic phase of TBI, total brain volume and white matter volume reductions have been related to measures of attention, processing speed, and executive functioning[51], whereas the rate of atrophy has been correlated to memory performance[54].

DAI, a common pathological mechanism in TBI, results in damage of white matter connections via a shearing effect, which can impact the transmission of neuronal signals and lead to cognitive impairments. DAI is most sensitively detected via diffusion MRI (dMRI), which uses the shape of diffusion of water molecules in the brain to reconstruct underlying white matter fiber architectures, and can lead to decreased fractional anisotropy (FA) or increased mean diffusivity (MD). Numerous studies have found relationships between dMRI-derived measures of axonal injury and cognitive functioning[55–64]. Of note, one study found reduced FA in individuals with TBI compared to HCs, and showed a relationship between abnormal FA in at least one white matter region on acute imaging and executive functioning 6 months later[58]. Additional studies have provided further evidence of decreased FA in the white matter of individuals with TBI compared to HCs, which was associated with poorer processing speed and verbal fluency[65,66]. A meta-analysis of dMRI in 9 mild TBI studies reported that higher FA in particular tracts, including the longitudinal fasciculus, sagittal striatum, cerebellum, and internal capsule, was related to better global cognitive functioning[64]. Interestingly, in this study higher FA in the cerebellar peduncles correlated with worse global cognitive functioning and higher FA in the anterior corona radiata correlated with worse attention, working memory and processing speed. In another meta-analysis of 20 studies, Wallace et al, (2018) found that increased FA and reduced MD associated with better cognitive outcomes, with attention and memory having the strongest relationships with dMRI metrics, particularly in the corpus callosum, fornix, internal capsule, arcuate and uncinate fasciculi[67**].

Tractography, a process by which the diffusion MRI signal is used to reconstruct likely white matter pathways in the brain, allows for graph-based representation of the white matter connectivity network, or structural connectome (SC). Connectomics has been applied to various neurological disorders, including TBI [57,68,69,70*,71*]. Small-worldness is a common property of biological networks, and represents a balance of numerous short-range and few long-range connections, enabling efficient transfer of information with a relatively low wiring cost[72]. A recent meta-analysis of 15 studies found that TBI populations may be shifted away from small-worldness, with longer characteristic path length and increased clustering coefficient, toward a regular lattice[73*]. Limited studies have investigated relationships between SC metrics and cognitive functioning in TBI. Kuceyeski et al., (2016) showed increased SC modularity, degree, and efficiency in individuals in disorders of consciousness (DOC) compared to HCs, but did not find correlations between SC metrics and Coma Recovery Scale-Revised scores, CRS-R[70]. Two studies in mild TBI found a correlation between reduced SC network efficiency and assessments of executive functioning and verbal learning[74], while another study showed an association between increased SC betweenness centrality values in language areas and poorer verbal memory[75*].

Measurement of the blood oxygenation level dependent (BOLD) signal in the brain via functional MRI (fMRI) allows for noninvasive, whole brain monitoring of activation patterns over time. The BOLD signal has been shown to be tightly coupled to neuronal activation[76]. The temporal resolution of fMRI is coarse compared to EEG, but allows for higher resolution spatial sampling. fMRI data can be analyzed using i) Independent Component Analysis (ICA), which allows for identification of networks associated with various functions, ii) seed-based correlations which allow investigations of spatial relationships to areas of interest or iii) functional connectomes (FCs), which represent similarity of activation patterns of regions over time and allow the application of graph theoretical approaches as described above. Numerous studies have shown changes in the task-free (resting-state) fMRI patterns in TBI patients, most indicating increases of default-mode network (DMN) activation compared to HCs. Bernier et al., 2017 found DMN hyperconnectivity during rest and task in individuals with TBI compared to HCs, which was positively correlated with working memory. FC hyperconnectivity in the frontal and middle temporal gyrus at rest was positively correlated with attention, processing speed, and executive functioning[77]. Increased FC within the sensory-motor regions at rest have been found to be correlated with better attention in individuals with TBI, a relationship absent in HCs[78]. Similar findings were reported by Grossner et al., 2019, which found that FC was positively correlated with better monitoring and error-awareness in individuals with TBI, but negatively correlated in HCs[79]. Correlations have also been shown between FC measures and recovery of consciousness post-injury[80–84]. These observations have lent further support to the “less wiring, more firing” hypothesis that has been suggested in multiple studies of neurological disorders[77,82,85–89]. However, increased FC in the precuneus has recently been related to worse global cognitive functioning in individuals with TBI[89]. Degree centrality, a graph theory metric that represents the number of connections between a node of interest and its neighboring nodes, in the left middle frontal gyrus was found to be positively correlated with global cognitive functioning and attention in individuals with mild TBI[90]. Until recently, static FC, or FC calculated using the entire fMRI time series, has predominated the literature. To capture how the brain’s FC can change over the duration of a scan, Van der Horn et., al 2019 utilized dynamic, or sliding-window, FC. They demonstrated that individuals with TBI who had poor recovery, characterized by sensorimotor and higher level cognitive deficits, spent less time in segregated dynamic FC and exhibited fewer transitions between states at 3 months post-injury[91].

Multimodal Imaging

To date, limited studies utilizing multimodal neuroimaging to identify and predict post-TBI cognitive impairment have been conducted. Multimodal MRI studies investigating the SC and FC networks jointly are most prevalent in the literature[92,93,94**]. One multimodal connectomic study showed that individuals with TBI with more SC injury exhibited less FC in the DMN, and higher DMN FC was associated with better processing speed[82]. In addition, Palacios et al., (2013) found that increased FC with the DMN in the frontal lobes was associated with worse SC in the cingulum[86]. A more recent study showed that SC between regions active during working memory and reasoning tasks on fMRI were dissociable and correlated with the respective cognitive domain in individuals with TBI[94]. A study using dynamic, sliding-window FC showed that patients with disorders of consciousness had less transitions between dynamic FC states and spent significantly more time in dynamic FC states that were closely coupled with SC, a finding they replicated in anaesthesia[95**]. Recent work has focused on implementing mathematical models to better codify the relationship between SC and FC in both healthy and pathological populations[96–103]. Hellyer et al., (2015) used a neural mass model to show that TBI-related decreases in SC were related to reduced metastability of FC, the latter being correlated with poorer cognitive flexibility, processing speed, and memory[104]. Network Diffusion (ND) modeling has also been adopted to study the relationship between SC and FC[105]. Increased ND model propagation time, a measure reflecting the “amount” of SC used to reconstruct observed FC, was significantly related to i) better CRS-R scores in a cross-sectional sample of individuals in DOC and ii) greater improvements in attention and memory from subacute to chronic mild TBI[70,71, see Figure 1].

Data obtained from EEG and PET have also been related to MRI to understand the relationship between function and structure post-TBI. EEG coherence, which has been associated with cognitive functioning, was related to greater T2 relaxation time in white and grey matter in individuals with TBI[26]. Decreased theta band synchrony in the first week of TBI predicted executive functioning 2 months later, and voxels within the theta band activity on MRI had less FA[20]. FA in the corpus callosum was also found to be correlated with beta coherence[24]. Alpha power and thalamic atrophy projecting to the temporo-occipital areas predicted functional recovery from acute to chronic status in severe TBI[106]. Reduced FA in the corpus callosum on MRI and reduced FDG metabolism in the cingulate gyrus on PET was found in individuals with TBI and cognitive impairments and not in HCs[107].

Conclusion:

The plethora of non-invasive imaging techniques and analyses presented here allow for increased insight into the biological substrate underlying heterogeneous cognitive impairments following structural brain injury. However, to date, validated, sensitive biomarkers remain unidentified. This is unsurprising given the advantages and disadvantages of each imaging modality and the previous focus on cross-sectional group level analyses.

More recent studies have related functional and structural changes to individual specific behavior, allowing for tracking of natural and intervened recovery and insight into the biological mechanisms. Furthermore, the few studies that have utilized multimodal imaging have a clear advantage over any single modality because they provide a more comprehensive quantification of the relationship between function and structure. Future studies focused on deriving biomarkers that are sensitive at the individual level with validation via studies of longitudinal change are key to converting on the promise of improving diagnosis of impairments, refining prognostic models, informing interventions, and precisely tracking spontaneous and intervention-driven recovery.

Key Points:

• Neuroimaging studies that associate with behavior at the individual level are useful for biomarker studies.

• Very few biomarkers are currently validated in studies of longitudinal change or intervention.

• Multimodal studies are valuable to disentangle complex interactions between function and structure following heterogeneous brain injury.