Early Identification of Recovery Potential After Acute Brain Injury Using Functional Near-Infrared Spectroscopy
Matthew Kolisnyk1, Karnig Kazazian2,3, Garima Gupta2,3
1Western Institute of Neuroscience, Western University, London, ON, Canada. mkolisny@uwo.ca.
Background:
Accurate early prognostication in patients with acute brain injury remains a major challenge in neurocritical care. Conventional bedside assessments provide limited insight into long-term outcomes and may not fully capture preserved brain function that supports recovery. Functional neuroimaging can detect brain activity not evident at the bedside, but its use in intensive care remains constrained by cost, logistics, and the need for stronger evidence supporting its value. Functional near-infrared spectroscopy (fNIRS) offers a scalable, bedside-compatible approach for assessing brain function in critically ill patients, but its value for early prognostication has yet to be established.
Methods:
In this prospective observational cohort study, 33 patients with acute brain injury in the intensive care unit (ICU) underwent fNIRS recording while listening to two audio-only movie clips. Functional connectivity features were used to train a machine learning model to classify 6-month functional outcome, defined by the Glasgow Outcome Scale-Extended (favorable ≥ 4, unfavorable < 4). Model performance was assessed using balanced accuracy and statistically evaluated using permutation testing. Performance was compared with validated behavioral assessments and clinical variables. Secondary analyses evaluated prediction of behavioral responsiveness (observable command-following after testing) and covert awareness (neural command-following).
Results:
A total of 26 patients had an unfavorable outcome and 7 had a favorable outcome. The fNIRS-based model predicted 6-month outcome with a balanced accuracy of 81.3% (sensitivity = 85.7%, specificity = 76.9%; p = 0.006), outperforming clinical models (balanced accuracy = 67.6%; p = 0.018). The fNIRS-based model also predicted recovery of behavioral responsiveness (balanced accuracy 78.5%; p = 0.008) but not covert awareness (70.4%; p = 0.109).
Conclusions:
Bedside fNIRS provides objective neural measures associated with later functional recovery and behavioral responsiveness in patients with acute brain injury. These findings suggest that fNIRS may capture clinically relevant brain function not detected by conventional assessments. With further validation in larger, multicenter cohorts, such approaches may complement existing methods for early prognostication.
