Quantitative Diffusion MRI Characterization of Prognosis in Neonatal Hypoxic Ischemic Injury

University Of California, San Francisco, San Francisco CA

Investigators

Abstract

Project Summary/Abstract Despite advances in treatment over the past decade for neonatal hypoxic ischemic encephalopathy, approximately 50% of infants continue to have long-term cognitive, motor and language impairments at 2 years of age. There remains an unmet need to identify subpopulations of infants at birth who are at highest risk for specific neurodevelopmental impairments and who will most likely benefit from new, targeted, delayed therapies. Expert MRI scoring systems, currently the standard evaluation method of MRIs obtained for clinical trials, demonstrate limited accuracy at identifying specific impairments. We hypothesize that compared to expert reader scoring of MRI, quantitative characterizations of the severity and location of brain injury on neonatal brain diffusion MRI are more significantly correlated with motor and language outcomes at 2 years of age. We plan to use diffusion MRI analysis techniques on the recently completed High-dose Erythropoietin for Asphyxia and Encephalopathy (HEAL) trial, and the just-released, first-of-its-kind public Boston Neonatal Brain Injury Dataset (BONBID) data. The HEAL data is unique in that (a) it is the largest modern HIE cohort in the US (N=500); (b) MRIs were acquired at 4-5 days of age using a harmonized protocol across 17 sites; and (c) detailed 2-year neurobehavioral tests included Bayley Scales of Infant Development-III (BSID-III). Our specific aims are: 1) Correlate severity of brain injury with 2-year motor and language BSID-III scores in the HEAL cohort. We will define brain injury severity by computer-extracted continuous ADC metrics: injury lesion volume at ADC z-score percentiles, intra-lesion ADC heterogeneity (histogram statistics), and lesion geometry (sphericity, elongation, regularity). We will correlate these injury severity metrics in specific brain regions known to correspond with motor and language function with BSID-III motor and language scores, respectively. 2) Correlate location of brain injury with 2-year motor and language BSID-III scores in the HEAL cohort. We will define brain injury location at both the voxel and white matter tract (connectivity) levels. Using voxel-wise lesion symptom mapping (v-LSM), we will identify voxels where presence of injury is significantly correlated with BSID-III scores. Subsequently, using connectome-based LSM (c-LSM), we will identify fiber tracts where the presence of injury anywhere along the tract is significantly associated with BSID-III scores. 3) To test the accuracy and generalizability of our injury severity and location metrics, we will use advanced statistical modeling and machine learning methods to find the optimal combination of severity and location metrics that produces the highest association with 2-year outcomes. We will further test our hypotheses in a separate HIE cohort (BONBID) to evaluate the generalizability of our techniques. We anticipate this multipronged quantitative diffusion MRI study will offer a mechanistic understanding of how quantitative measures of HIE injury severity and location impact neurodevelopmental outcomes. Our findings will inform future development of individualized quantitative MRI biomarkers, and ultimately serve as an early, reliable secondary endpoint to facilitate and expedite future therapeutic innovations.