AI-assisted high-density near-infrared spectroscopy system for cerebral oxygen saturation measurement in a porcine ischemia model

Abstract Stroke is one of the leading causes of mortality and is characterized by a sudden interruption of cerebral perfusion, resulting in neurological impairment. Especially in large-vessel occlusion (LVO), the absence of timely reperfusion therapy can lead to severe ischemic injury, causing permanent disability or death. Near-infrared spectroscopy (NIRS) is a non-invasive technique that estimates regional cerebral oxygen saturation ( $$\hbox {rSO}_2$$ ) by analyzing near-infrared light transmitted through biological tissue. Despite its portability and cost-effectiveness, commercial NIRS systems that employ spatially resolved spectroscopy (SRS) algorithms lack sufficient reliability to distinguish between normal and stroke-affected conditions because of oversimplified assumptions. In this study, we propose an artificial intelligence–assisted NIRS system that analyzes high-density optical measurements to estimate cerebral oxygenation beyond the simplified assumptions of conventional SRS algorithms. Trained on an MRI-based synthetic dataset using cortical $$\hbox {rSO}_2$$ as the target label, the proposed model leverages measurements obtained at multiple source–detector distances to exploit depth-dependent spatial information and improve sensitivity to cortical oxygenation signals. The proposed system was validated through a porcine common carotid artery occlusion experiment and compared with a conventional SRS-based NIRS system. Both systems showed decreases during occlusion and recovery after reperfusion, but statistically significant baseline-relative changes after multiple-comparison correction were observed only for NIRSIT-X. Furthermore, we compared the measured oxygen saturation against jugular venous oxygen saturation ( $$\hbox {SjvO}_2$$ ) as a physiological reference. In conclusion, based on linear mixed-effects modeling, the proposed system demonstrated a substantially stronger association with $$\hbox {SjvO}_2$$ than the conventional system, achieving a markedly higher marginal $$R^2$$ (0.561 vs. 0.177), indicating stronger physiological relevance for cerebral oxygenation monitoring.