Objective: There is a lack of reliable tools used to predict functional recovery in unresponsive patients following a severe brain injury. The objective of the study is to evaluate the prognostic utility of resting-state functional magnetic resonance imaging for predicting good neurologic recovery in unresponsive patients with severe brain injury in the intensive-care unit.
Methods: Each patient underwent a 5.5-min resting-state scan and ten resting-state networks were extracted via independent component analysis. The Glasgow Outcome Scale was used to classify patients into good and poor outcome groups. The Nearest Centroid classifier used each patient's ten resting-state network values to predict best neurologic outcome within 6 months post-injury.
Results: Of the 25 patients enrolled (mean age = 43.68, range = [19-69]; GCS ≤ 9; 6 females), 10 had good and 15 had poor outcome. The classifier correctly and confidently predicted 8/10 patients with good and 12/15 patients with poor outcome (mean = 0.793, CI = [0.700, 0.886], Z = 2.843, p = 0.002). The prediction performance was largely determined by three visual (medial: Z = 3.11, p = 0.002; occipital pole: Z = 2.44, p = 0.015; lateral: Z = 2.85, p = 0.004) and the left frontoparietal network (Z = 2.179, p = 0.029).
Discussion: Our approach correctly identified good functional outcome with higher sensitivity (80%) than traditional prognostic measures. By revealing preserved networks in the absence of discernible behavioral signs, functional connectivity may aid in the prognostic process and affect the outcome of discussions surrounding withdrawal of life-sustaining measures.
Keywords: Brain injury; Coma; Consciousness; Machine learning; Resting state; fMRI.
© 2023. The Author(s), under exclusive licence to Springer-Verlag GmbH Germany.