Ischemic stroke is a major cause of mortality with complicated pathophysiological mechanisms, and hematoxylin and eosin (HE) staining is a histochemical diagnosis technique heavily relying on subjective observation. In this study, we developed a noninvasive assay using Raman spectroscopy for in vitro diagnosis and visualization of cerebral ischemia/reperfusion injury and protective effects of ferulic acid. By establishing a middle cerebral artery occlusion (MCAO) model in Sprague-Dawley male rats, we found effective interventions by ferulic acid using the neurological function score and HE staining. Raman spectra of neuronal and neuroglial cells exhibited significant intensity changes of protein, nucleotide, lipid, and carbohydrate at 780, 814, 1002, 1012, 1176, 1224, 1402, 1520, 1586, 1614, and 1752 cm-1. Cluster vector analysis highlighted the alterations at 1002, 1080, 1298, 1430, 1478, 1508, 1586, and 1676 cm-1. To evaluate the levels of neuron injury and intervention performance, a random forest model was developed on Raman spectral data and achieved satisfactory accuracy (0.9846), sensitivity (0.9679-0.9932), and specificity (0.9945-0.9989), ranking peaks around 1002 cm-1 as key fingerprint for classification. Spectral phenylalanine-to-tryptophan ratio was the biomarker to visualize neuronal injury and intervention performance of ferulic acid with a resolution of 1 μm. Our results unravel the biochemical changes in neuronal cells with cerebral ischemia/reperfusion injury and ferulic acid treatment, and prove Raman spectroscopy coupled with machine learning as a power tool to classify neuron viability and evaluate the intervention performance in pharmacological research.
Keywords: Ischemic stroke; machine learning; vibrational spectroscopy.