35 - Prematurity-induced cerebellar injury in human: unbiased, rigorous neuropathology using cloud-based artificial intelligence
Monday, May 1, 2023
9:30 AM – 11:30 AM ET
Poster Number: 35 Publication Number: 35.435
Abhya P. Vij, Boston Children's Hospital and Boston Medical Center, Boston, MA, United States; Georgios Sanidas, School of Medicine, University of Patras, Athens, Attiki, Greece; Javid Ghaemmaghami, Children's National Health System, Washington, D.C., DC, United States; Adriana R. Valenzuela, Children's National Health Systems, Manassas Park, VA, United States; Rodolfo Cardenas Trejo, Children's National Health System, Washington, DC, United States; Vittorio Gallo, Children's National Health System, Washington, DC, United States; Panagiotis Kratimenos, Children's National, Washington, DC, United States
Pediatric Intern Boston Children's Hospital and Boston Medical Center Boston, Massachusetts, United States
Background: Survivors of preterm birth have higher rates of neurodevelopmental disorders compared to those born at term. The human cerebellum is subject to a critical period of development in the final trimester of pregnancy. The precise pattern of cerebellar injury in preterm-born human neonates is unclear as adequately powered postmortem studies are lacking. Objective: By analyzing postmortem human cerebellar tissues, we aimed to assess the morphological differences in cerebellar development between term and preterm infants. We sought a scalable, accurate, context-aware, and cloud-based method to segment whole-slide tissue scans and obtain unbiased quantification of the cerebellar cortex. Design/Methods: We created an artificial intelligence model trained on Amazon Sagemaker using the context-aware PSP deep learning algorithm to classify whole slide scans into their layer composition, with over 95% pixel validation accuracy and segmentation of whole slides from term (n=34) and preterm (n=54) human subjects. To quantify our results, we analyzed both the area of the layer throughout each slide, and the local thicknesses of each layer along its length. After matching subjects using corrected gestational age, we performed linear regression analysis to compare the total surface area of each cerebellar layer in term and preterm infants, and t-testing for significance. Results: At term-equivalent age, the preterm cerebellum demonstrated a decreased depth of foliation, delayed disappearance of the EGL, increased area of IGL, and decreased area of ML. Further, the rate of decline in EGL area is related to gestational age at birth, with extremely preterm infants exhibiting the slowest thinning of EGL. Preterm infants demonstrated a larger surface area of IGL compared to those born term. Despite the larger surface area, preterm IGL is depopulated with increased gliosis (GFAP+) compared to term indicating that the increased IGL surface in preterm neonates is due to glia-neuropil expansion.
Conclusion(s): Our data demonstrate indicate that preterm birth is associated with delayed and atypical cerebellar development. Our artificial intelligence and cloud-based approach to quantification of cerebellar layer thickness offers exceptional accessibility, efficiency and objectivity in the evaluation of human tissues, and allowed us to capture subtle differences in cerebellar morphology among preterm and term neonates. This methodology offers a pipeline for a collaborative repository for quantified brain tissue samples across sites that can be used for higher-powered studies.
