Project Details

Development of neuroimaging biomarkers for 2-year neurocognitive outcomes in mild neonatal brain injury

Rina Bao, PhD

Summary

Abstract - One in two hundred newborns suffer from brain injuries related to the lack of oxygen and blood supply, a neonatal brain dysfunction known as hypoxic-ischemic encephalopathy (HIE). We aim to use rigorous image analysis and artificial intelligence (AI) to address a decades-long mystery: why do a sixth of HIE patients still end up developing neurocognitive impairments by 2 years of age, even though their brain Magnetic Resonance Imaging (MRI) scans present little or no explicit injuries in the newborn stage? Computer-extracted whole-brain patterns developed by our team have shown significant group differences (p<0.05, Wilcoxon Test) between normal outcomes vs adverse outcomes in N=39 HIE patients with MRIs from Massachusetts General Hospital (MGH) read clinically as normal. In this study, we are going to test if the preliminary findings from N=39 single-site HIE patients with MRIs read clinically as normal will hold true in the proposed larger-scale multi-site tests with N=172 mild HIE patients. Specifically, we will test if we can differentiate between HIE cases with MRIs read as normal with normal versus adverse 2-year outcomes using computer-extracted whole-brain patterns in groups and at the individual levels. Aim 1 will use single-variate statistics to test the hypothesis that the two sub-cohorts of HIE patients with clinically normal MRIs (121 with normal and 51 with adverse outcomes) are statistically differentiable by computer-extracted MRI features. Aim 2 will use multivariate machine learning to combine computer-extracted MRI features to predict normal versus adverse outcomes for each individual HIE patient with a clinically normal MRI. BACKGROUND: One in two hundred newborns suffer from brain injuries related to the lack of oxygen and blood supply, a neonatal brain dysfunction known as hypoxic-ischemic encephalopathy (HIE). We aim to use rigorous image analysis and artificial intelligence (AI) to address a decades-long mystery: why do a sixth of HIE patients still end up developing neurocognitive impairments by 2 years of age, even though their brain Magnetic Resonance Imaging (MRI) scans present little or no explicit injuries in the newborn stage? GAP: There is no biomarker to target the mild HIE patients (1/3-1/2 of all HIE patients), the majority of which have normal MRIs. HYPOTHESIS: Computer-extracted whole-brain patterns can identify, among HIE patients with clinically normal MRIs, those who will develop adverse 2-year neurologic outcomes, with statistical significance. METHODS: Aim 1 will use single-variate statistics to test the hypothesis that the two sub-cohorts of HIE patients with clinically normal MRIs (121 with normal and 51 with adverse outcomes) are statistically differentiable by computer-extracted MRI features. Aim 2 will use multivariate machine learning to combine computer-extracted MRI features and to predict normal versus adverse outcomes for each individual HIE patient with a clinically normal MRI. RESULTS: Presented on Pediatric Academic Society Meeting 2021, we have found that the 2-year normal-vs-adverse HIE outcomes are significantly associated with texture features in whole-brain regions in HIE cases clinically read as normal. Computer-extracted whole-brain patterns developed by our team have shown significant group differences (p<0.05, Wilcoxon Test) between normal outcomes vs adverse outcomes in 39 HIE patients from Massachusetts General Hospital (MGH) with MRIs clinically read as normal. IMPACT: While current prognostic biomarkers are mostly on moderate and severe HIE patients, our work, if successful, will be the first biomarker to target the mild HIE patients (1/3-1/2 of all HIE patients) who typically have normal MRIs to stratify risk, suggest differential treatment, and evaluate therapeutic effects early.