Improved Mild Closed Head Traumatic Brain Injury Outcomes With a Brain-Computer Interface Amplified Cognitive Remediation Training

Abstract

This study is a retrospective chart review of 200 clients who participated in a non-verbal restorative cognitive remediation training (rCRT) program between 2012 and 2020. Each client participated in the program for about 16 weeks, and the study as a whole occurred over a five-year period. The program was applied to effect proper neural functional remodeling needed to support resilient, flexible, and adaptable behaviors after encountering a mild closed head traumatic brain injury (mTBI). The rCRT program focused on improving functional performance in executive cognitive control networks as defined by fMRI studies. All rCRT activities were delivered in a semi-game-like manner, incorporating a brain-computer interface (BCI) that provided in-the-moment neural network performance integrity metrics (nPIMs) used to adjust the level of play required to properly engage long-term potentiation (LTP) and long-term depression (LTD) network learning rules. This study reports on t-test and Reliable Change Index (RCI) changes found within individual cognitive abilities' performance metrics derived from the Woodcock-Johnson Cognitive Abilities III Test. We compared pre- and post-scores from seven cognitive abilities considered dependent on executive cognitive control networks against seven non-executive control abilities. We observed significant improvements (p < 10^-4) with large Cohen's deffect sizes (0.78-1.20) across 13 of 14 cognitive ability domains with a medium effect size (0.49) on the remaining one. The mean percent change for the pooled trained domain was double that observed for the pooled untrained domain, at 17.2% versus 8.3%, respectively. To further adjust for practice effects, practice effect RCI values were computed and further supported the effectiveness of the rCRT (trained RCI 1.4-4.8; untrained RCI 0.-08-0.75).

Keywords: bci; brain computer interface; cognitive remediation training; executive cognitive control; mtbi; qeeg; quantitative electroencephalography; traumatic brain injury.

Figure 1. Neurocognitive functions of interest and canonical networks

Large-scale intrinsic and task-evoked circuits. Fronto-parietal network (FPN): DLPFC = dorsolateral prefrontal cortex; IPL = inferior parietal lobule; DFC = dorsal frontal cortex; IPS = intraparietal sulcus; Precuneus; MCC = middle cingulate cortex. Cingulo-opercular network (CON): APFC = anterior prefrontal cortex; AI/FO = anterior insula/frontal operculum; dACC/MFSC = dorsal anterior cingulate cortex and medial superior frontal cortex; Thalamus*. Salience network (SN): dACC = dorsal anterior cingulate cortex; aI = anterior insula; TP = temporal pole; SLEA = seblenticular extended amygdala*. Default mode network (DMN): aMPFC = anterior medial prefrontal cortex; AG = angular gyrus; PCC = posterior cingulate cortex (includes precuneus). Working memory network (WMN): SPL = superior parietal lobule; DLPFC = dorsolateral prefrontal cortex; MPFC = medial prefrontal cortex; VAC = ventral anterior cingulate area; AR = agranular retrolimbic area; DPC = dorsal posterior cingulate area; dACC = dorsal anterior cingulate cortex; pregenual area (ACC): SG = supramarginal gyrus; Hip = hippocampus*. *Subcortical areas are unlikely to be reliably measured using electrocardiography (EEG) and may be excluded from our analysis.

Figure 2. Histogram of pooled changes

Histograms show the distribution of percent changes observed for domains that received explicit training (blue) and for domains that did not receive training (orange). Overlap between the two distributions is visible in brown.

Figure 3. Percent changes observed across each area