Directed functional connectivity analysis of bimanual motor skill learning under acute stress

Introduction: : Stress is a pervasive factor in high-stakes emergency medicine environments. Understanding its impact on the learning and retention of bimanual motor skills is critical for developing effective training protocols. This study examines how acute stress modulates directed functional connectivity (dFC) between brain regions during the acquisition and retention of endotracheal intubation (ETI), a vital emergency procedure. Functional near-infrared spectroscopy (fNIRS) provides a non-invasive method for monitoring task-evoked cortical activity, offering valuable insights into the neural processes associated with learning. We apply causal connectivity analysis to fNIRS signals to investigate how stress alters learning-related brain network dynamics, and whether these changes negatively affect behavioral performance.

Materials and

Methods: : dFC was computed from brain activity measurements obtained while participants performed ETI procedure following Institutional Review Board (IRB) and Army Human Research Protection Office (AHRPO) approved protocol. A continuous-wave fNIRS device was used to measure brain hemodynamics as a surrogate for neural activity. Forty medical trainees were recruited and evenly assigned to either a stressor or non-stressor group. The stressor group was exposed to aversive noise and a countdown timer designed to induce a stress response. Each participant completed ten repetitions of the ETI procedure on an airway manikin across three consecutive training days, followed by six retention trials conducted after an eight-week interval. Each trial lasted up to three minutes, with a two-minute rest period between trials.

dFCs between five brain regions involved in psychomotor tasks were estimated using non-linear Granger causality framework. This model replaces the vector autoregressive (VAR) component of linear Granger causality with an attention-based Long Short-Term Memory (LSTM) recurrent autoencoder, enabling it to capture long-term temporal dependencies between brain regions. Granger causality was computed by ablating fNIRS signals from each brain region and evaluating the effect on signal prediction, yielding 25 directional connectivity scores for each repetition of the ETI task. The dFC measures were then used as biomarkers to investigate learning process by comparing changes in connectivity between pre- and post-training trials, as well as between post-training and retention trials. Multivariate comparisons were conducted using support vector machine (SVM) classifier with five-fold cross-validation.

Results, Conclusions, and Discussions:: The dFC-based analysis revealed greater connectivity changes between pre- and post-training trials in the stressor group (Accuracy: 0.91, MCC: 0.91) compared to the non-stressor group (Accuracy: 0.88, MCC: 0.87). However, this trend reversed for post-training to retention comparisons, where the non-stressor group exhibited larger changes (Accuracy: 0.94, MCC: 0.96) than the stressor group (Accuracy: 0.91, MCC: 0.94), see Table 1. Recursive feature elimination identified RPMC-to-RPFC and SMA-to-LPFC as the dominant connections for distinguishing pre- vs. post-training trials in the non-stressor and stressor groups, respectively. For post-training vs. retention, LPFC-to-RPMC and RPFC-to-SMA emerged as most discriminative in the non-stressor and stressor groups, respectively.

The non-stressor group included 17 participants with high success rate; all 8 who returned for the retention phase demonstrated learning progression. The stressor group comprised 11 participants with high success rates; however, 3 of the 7 returning participants did not exhibit consistent improvement. The threshold success rates for both groups were estimated using the cumulative sum (CUSUM) score. Among returning participants with high success rates (learners) in both groups, learning was associated with a gradual decline in dominant dFC during training, followed by partial recovery during the retention phase (Figure 1). In contrast, returning participants in the stressor group with high failure rates (non-learners) showed relatively stable dFC during training (Figure 1c–d), followed by a decline during retention. Notably, high-performing individuals in the stressor group exhibited greater dFC variability (Figure 1c–d) and a lower average success rate (73.91%) compared to those in the non-stressor group (87.54%). Among non-learners, the non-stressor group showed a higher average failure rate (93.33%) than the stressor group (74.07%).

This study demonstrates that stress alters dFC during bimanual motor skill learning, particularly impairing long-term skill retention. Although both the stressor and non-stressor groups exhibited learning-related changes in dFC, the stressor group showed greater variability and poorer retention performance. These findings highlight the potential of dFC-based biomarkers for assessing the impact of stress on learning and underscore the importance of stress management in preparing providers for high-stakes tasks in emergency medicine.

Acknowledgements and/or References (Optional):: The authors gratefully acknowledge the support of this work through the U.S. Army Futures Command, Combat Capabilities Development Command Soldier Center STTC cooperative research agreement # W912CG2120001.