3.1 Nature of the injury
Detail on the nature of the injury is provided in the prior report (16). In brief, TG was attacked on March 4, 2006 (at the time, 41 years old). He is right-handed, university-educated, and a soldier/journalist/writer. Research ethics approval was obtained from Simon Fraser University and the National Research Council. Captain Greene and his wife Debbie participate as full investigators in all aspects of the research (Note: They are both authors on this paper). The open severe TBI resulted from an attack with a crude axe. TG was leading a goodwill meeting with elders in the village of Shinkay, Kandahar, Afghanistan. As a sign of respect, the soldiers removed their helmets and laid down their weapons. A young male struck TG with the axe into the crown of his head with full strength, as a signal for a larger pending attack from the Taliban. Immediately after the engagement, TG’s vitals were stabilized through emergency care and he survived medivac extraction to Kandahar Air Field for advanced care. He transferred to the US Army Landstuhl Regional Medical Centre in Germany for neurosurgical treatment and induced into a medical coma. Once medically stable, TG was transported home to Vancouver General Hospital (Vancouver, Canada). Initial prognosis anticipated permanent vegetative state, but TG emerged from coma and recovered full consciousness after approximately 18 months. Following acute care, he was admitted to the Halvar Jonson Centre for Brain Injury Centre (Alberta, Canada) for a 14-month intensive rehabilitation program. Since then, TG has continued daily home-based rehabilitation with the main long-term objective of recovering ambulatory walking abilities and resumed an active writing career, which included publishing the book: “March Forth: An Inspiring True Story of a Canadian Soldier’s Journey of Love, Hope and Survival” (24).
The injury involved both penetration and rotational impact. The fracture to TG’s skull was approximately along the midsagittal plane, extending from the frontal bone posteriorly along the sagittal suture. There was both gray and white matter cortical tissue damage, extending laterally to the right frontal and left parietal lobes away from the midline and inferiorly to the lateral ventricle. The injury affected primary motor and premotor areas along with primary somatosensory and superior parietal areas. The injury depth affected the anterior cingulate gyri, corpus callosum (body and genu), and surrounding white matter tissue. See D’Arcy et al. (16), for a more detailed description along with MRI and fMRI imaging results. Visualization of the injury together with the summary analysis of the areas of greatest fMRI activation change over the three-year study is available in Supplemental Figure 1.
3.2 Experimental design
Similar to the prior study, we used a longitudinal design to evaluate motor activation changes over time. Clinical, EEG, and MEG data were acquired at regular intervals approximately every 3 months from July 2018 until August 2019. There were five time points in total, with three baseline (B1-3) and two treatment time points (T4-5). The treatment time points were collected halfway through the PT+TLNS program (7 weeks) and at the end of the program (14 weeks).
Research Timeline
|
Date
|
Clinical Milestones
|
|
|
2006/03
|
Initial Injury. Prognosis: permanent vegetative state
|
|
|
2007/09
|
Fully Conscious
|
|
|
2007-2010
|
Intensive Physical Therapy
|
Phase 1:
Motor fMRI Study
(D'Arcy et al. 2016)
|
MRI 1
|
2010/05
|
Stands at wall-mounted bar without safety harness
|
MRI 2
|
2010/08
|
Takes steps inside parallel bar with harness and assistance
|
MRI 3
|
2010/11
|
No longer needed lift to get into MRI machine
|
MRI 4
|
2011/02
|
Stands and pivots with assistance
|
MRI 5
|
2011/05
|
Stands for 2 min with knee blocks
|
MRI 6
|
2011/08
|
Stands for 6 min with knee blocks and assistance
|
MRI 7
|
2011/11
|
Stands for 10 min with knee blocks assistance
|
MRI 8
|
2012/02
|
Stands for 30 s without knee blocks or assistance
|
MRI 9
|
2012/05
|
Sits without support
|
MRI 10
|
2012/08
|
Stands with walker
|
MRI 11
|
2012/11
|
Takes steps inside parallel bar with assistance
|
|
MRI 12
|
2013/02
|
Takes steps with walker with assistance
|
|
|
2012-2013
|
PT with Lokomat device
|
|
|
2014-2015
|
PT with ReWalk exoskeleton device
|
|
|
2016-2018
|
Plateau in Recovery
|
Phase 2: Current Study
|
|
2018/04
|
Intensive Physical Therapy treatment begins
|
1st Baseline (B1)
|
2018/07
|
2 Minute Timed Stand - Moderate Support
FIST Score - 13
|
2nd Baseline (B2)
|
2018/10
|
1 Minute Timed Stand - Moderate Support
FIST Score - 19
|
3rd Baseline (B3)
|
2019/04
|
FIST Score – 12
Continued plateau in recovery coupled with lack of motivation and intensity
|
|
2019/04
|
Physical Therapy + Translingual Neurostimulation begins
|
1st Treatment (T1)
|
2019/05
|
20 Minute Timed Stand - Moderate Support
FIST Score - 21
|
2nd Treatment (T2)
|
2019/07
|
20 Minute Timed Stand - Minimal Support
FIST Score - 33
Deep breathing improved – was able to blow up a balloon for the first time
Renewed motivation and intensity to engage in therapy activities
|
|
|
2019+
|
Ongoing Physical Therapy + Translingual Neurostimulation
|
Table 1: Timeline and milestones for clinical recovery, including date of initial injury and prior longitudinal studies with TG
The baseline time points involved intensive PT alone and treatment time-points involved continued intensive PT+TLNS stimulation. Experimental parameters were kept constant across all time points. The timeline of clinical milestones relative to experimental assessments is indicated in Table 1.
3.3 Neuromodulation
The PoNS® device consists of a light-weight portable controller worn around the neck and a stimulator with 143 gold-plated electrodes positioned to electrically stimulate the anterior dorsal tongue (1.5 mm diameter electrodes in a hexagonal pattern 2.2 mm apart; Kaczmarek, 2017). The patient holds the stimulator in place by applying upward pressure from the tongue. The stimulation level is adjustable and user-dependent, with increasing subjective intensity in discrete values from 1 to 60, which increase stimulation pulse length (μs) without any increase in electrical voltage levels (i.e., ensuring safe dosage control). PoNS® stimulation delivers pulses in triplets at 5 ms intervals (i.e., 200 Hz) every 20 ms (50 Hz), with a 17.5 V operating voltage and 440 μA current for each pulse. In accordance with a device level setting procedure, the user increases the stimulation levels to a comfortable sensation between the minimum perceptible level and the maximum tolerable level. The PoNS® procedure was developed from empirical psychophysical studies of optimal tactile sensation for comfortable long-term use.
3.4 Study protocol
3.4.1 Clinical treatment and movement scores
The treatment program consisted of physiotherapy exercises for one-year (PT alone; B1-B3), followed by the same program paired with the PoNS® device for 14 weeks (PT+ TLNS; T4-T5). Both the PT alone and PT+TLNS programs included six training days a week, with three training sessions a day. During in-clinic sessions, the therapist worked directly with TG. The PT alone program included three in-clinic days during each of the baseline testing visits and in-home training sessions between visits. The first week of the PT+TLNS training involved two out of three daily training sessions in-clinic to help familiarize TG with usage of the PoNS® device. The second week involved one out of three daily training session in-clinic. During weeks 3-14, TG took the device home and followed a physiotherapist-outlined program. Monthly check-in occurred to assess program goals and download usage data from the PoNS® device. Daily training sessions were divided into morning, afternoon and evening. Morning sessions consisted of a warm up (about 5 exercises working on upper body movements such as chin tucks, shoulder rolls, thoracic movements without the PoNS® device), balance (20 minutes of sitting balance work with the PoNS® device), gait (20 minute sessions of standing exercises with the PoNS® device) and Breathing and Awareness training (BAT, 20 minutes of mindfulness/meditation with the PoNS® device). The afternoon session consisted of balance training, movement control (20 minutes of physiotherapy exercises without the PoNS® device) and gait. The evening session involved BAT.
Clinical movement scores included timed stand (up to 20 minutes) and the Function In Sitting Test (FIST) (25). The timed stand was measured as the time TG stood independently in one place with assistance. The FIST is a 14 item, performance-based, clinical examination of sitting balance with demonstrated test-retest and intra- and interrater reliability (26). The FIST bridges the gaps between simple observations about sitting balance/trunk control and balance measures more heavily weighted towards standing balance or gait ability. TG was asked to perform basic, everyday activities in a seated position with an examiner scoring his performance using a 0-4 point ordinal scale, with a maximum possible score of 56.
Upper limb motor control improvements were also a clinical priority. As described in the prior report (16), upper arm function has recovered to a higher functional level than lower limb abilities. For example, TG performs many common daily upper limb tasks independently. However, on-going rehabilitation has also focused on improved upper limb function as spasticity has continued to limit functional recovery. With respect to spasticity, motor control, and movement abilities, TG has a clear right>left functional asymmetry. While previously right hand dominant, he performs most functional tasks with his left hand. Given the technical constraints of MEG and the established pattern of upper limb motor control impairment, basic left and right responses were selected to investigate corresponding changes in both EEG and MEG activation over time.
3.4.1. EEG – Motor function
Using a modified pre-established protocol (27), motor EEG data were also recorded at five time points: B1, B2, B3, T4, T5. Data were collected using a 32-channel recording system (g.Nautilus g.LADYbird, g.tec medical engineering, Graz, Austria) at a sampling frequency of 500 Hz. The design of the motor task mirrored MEG data collections, with TG responding using a custom designed button pad at a self-guided pace (approximately every 2-4 seconds) with all four fingers (in sequence, digits 1-4). At each visit, TG performed three x 2.5-minute motor sessions with each hand.
For each EEG recording session, data were manually cleaned to reduce artifacts and to ensure task compliance. Data segments containing major artifacts (muscle activity and large movement artifacts in particular) were discarded. Button press events that occurred out of sequence (for example, multiple buttons pressed simultaneously) were also discarded. After identifying clean EEG segments, an average of approximately 70 click events per hand (~66% of available events) remained for each recording session.
Following artifact rejection, data were processed to extract event-related desynchronization/synchronization (ERD/ERS) activity (28) for the motor task. To identify the frequencies of interest for TG, time-frequency analysis (time locked to the click event and averaged across trials) was performed from 5 to 35 Hz for each recording session. For TF analysis, a Laplacian spatial filter was applied to the electrodes in the contra-lateral motor area. (Laplacian for left hemisphere centered at C3 with [FC5 FC1 CP5 CP1] as the surrounding electrodes; mirrored electrodes for the right hemisphere.) TF analysis identified the high beta range as demonstrating robust ERD/ERS activity. Subsequently, all electrodes were filtered from 24-30 Hz, corresponding to the most active frequencies in TF analysis.
Data were further processed with an optimal spatial filter technique (29) which uses gradient descent to find the linear combination of channels that maximizes the power ratio between two conditions (in this case, the post-movement period where high power (ERS) is expected and the pre-movement period where low power (ERD) is expected). This technique, while not traditional, was chosen to maximize ERD/ERS signal quality by providing some targeted source localization. For each recording session, EEG electrodes in the contra-lateral hemisphere were used to calculate the channel weights that maximized beta power in the post-movement period (from 0 to 0.5 s after button press) relative to the pre-movement period (from -1 to 0 s prior to button press). (For the left hemisphere, input channels were [AF3 F7 F3 FC5 FC1 T7 C3 CP5 CP1 P7 P3]; mirrored electrodes for the right hemisphere. The Laplacian described above was used as the ‘initial guess’.)
To quantify ERD/ERS activity for comparison between sessions, beta rebound (magnitude of beta ERS following the movement event) was selected as the relevant output measure. Beta rebound is typically measured relative to resting activity, but due to the sequential nature of the finger movement task, there was no baseline rest period for each event. Instead, the period immediately prior to button press (during movement, while ERD is occuring) was used as the baseline. Beta rebound was calculated as (max ERS – min ERD) / (min ERD).
To determine the reliability of the beta rebound output measures for each session, 5000 bootstrap iterations were calculated using random sampling of epochs with replacement. To test the significance of changes during treatment, the difference between each treatment session and the overall baseline was compared using a weighted contrast: T - ⅓ (B1 + B2 + B3). Bootstrap outputs were resampled (100,000 iterations) to determine the overall distribution for the contrast hypothesis, then the significance of the result was determined using a percentile test.
3.4.2 MEG
MEG data were recorded at 5 time points: B1, B2, B3, T4, T5. Data were collected at a sampling frequency of 1200 Hz in a magnetically shielded room using a 275-channel MEG system (CTF systems; Coquitlam, Canada), and the head position was continuously tracked. The data recording was performed while TG was in a seated position during four 5-minute motor task data acquisition (2 per hand in a randomly assigned order). In the motor task, TG was instructed to press a button (Lumitouch, Photon Control Inc., Burnaby, Canada) at a self-guided pace, approximately every 2 to 4 seconds in sequence with digits 1 to 4 alternating between both high right and left hands (Note: M/EEG resting state, with eyes open- and closed- 10 minute sessions, were also recorded and analyzed in a separate study).
Prior to the MEG data collection, the head shape was digitized using a Polhemus FASTRAK digitizer for co-registration of MEG data with his anatomical MRI. The anatomical MRI was segmented using Freesurfer. The gray/white matter boundary mesh was down-sampled to 4K vertices and brain activity was estimated for each vertex.
For each recording, noise segments with muscular artifacts or head motion exceeding 5 mm from the median head position during the recording were rejected. Then, independent component analysis (ICA) was computed and artifactual components were discarded. Trials free from noise segments spanning from -.5 to 1.5 seconds relative to button press onset were grouped for each session and condition.
To estimate brain activity during the motor task, an event-related Dynamical Imaging of Coherent Sources (DICS) beamformer (30) was calculated using FieldTrip (31) to localize the motor signal. Similar to the EEG analysis, the post-movement ERS ‘rebound’ period (from 0.25 to 1 s after button press) was contrasted against the pre-movement ERD period (from .75 to 0 s prior to button press). First, the complex Fourier spectra were calculated in the same high beta range (centralized at 27 Hz with a 5 Hz taper parameter) where motor activity was most prominent in the EEG analysis. Next, the inverse filter was computed using data from both the ‘rebound’ and ‘active’ periods, then applied to the two conditions separately. Finally, the contrasted motor activity map was calculated for each trial according to the formula (post-movement ERS – pre-movement ERD)/(pre-movement ERD).
To test the significance of differences between sessions, Partial Least Squares (PLS) analysis was used (32,33). PLS is a multivariate statistical approach based on singular value decomposition. In this study, we used both Non-Rotated Task PLS (referred to here as ‘contrast-driven’ PLS) and Mean Centering Task PLS (referred to here as ‘data-driven’ PLS). Multiple analyses were used to confirm common MEG activation changes for right hand (RH) and left hand (LH).
The contrast-driven method enables testing of specific hypotheses about the contrast between conditions by setting a hypothesis driven design matrix. A specific contrast pattern is provided as an input and PLS identifies the singular value and salience of the input contrast. To test significance, a series of permutations was run by permuting trials across sessions. A single p-value is rendered to mitigate against multiple comparisons. In our case, to identify whether contrast-driven changes were robust against variation in the contrast pattern, the contrast-driven analysis was run three separate times using timed stand, FIST, and EEG beta rebound (see above) as the input contrasts.
The goal of the data-driven method is to automatically decompose the multidimensional data into ‘latent variables,’ which describe the maximum variance between sessions. Each latent variable is composed of three elements: a contrast pattern (describing change between sessions); a singular value (describing the strength of the change); and a vector of saliences (describing the brain vertices expressing the change). As with data-driven PLS, permutation was used to test the significance. A single p-value per latent variable is rendered to mitigate against multiple comparisons.
PLS also includes bootstrapping, which enables visualization of robustness of change across brain vertices. The output can be interpreted as z-scores that demonstrate the spatial distribution of differences between sessions. For all contrast- and data-driven cases, bootstrapping was used to generate spatial visualizations of the activity changes.