A deep learning approach to quantitatively capture cerebellar overgrowth in pediatric NF1 patients.

Abstract

e14001 Background: Macrocephaly is one of the most common clinical manifestations of NF1 patients. As compared to normals, NF1-associated brains show enlarged gray and whiter matter volumes with varying degrees of regional specificity. Despite this knowledge, little is known about NF1-specific cerebellar volume changes in childhood, particularly relative to their age-matched and gender-matched controls. We sought to quantitatively capture cerebellar enlargement in pediatric NF1 patients against a large cohort of age and gender-matched, control patients. Methods: Clinical data and 3D T1 MPRAGE sequences (either pre or post contrast) were acquired from 61 pediatric NF1 patients, between the ages of 1 and 21, seen at the University of Minnesota Masonic Children’s Hospital (UMMCH) and satellite clinics. Imaging and clinical data from 956 control patients was obtained from a variety of public sources. SynthSeg, an open-source, whole brain segmentation model was used to segment all intracranial structures including the cerebellum from all available imaging. Results: Males and females with NF1 had, on average, 17.1% and 12.9% larger cerebellar volumes, respectively, compared to age-matched peers. Median percentile for cerebellar volume in males and females with NF1 was 96.4 and 94.6, respectively. Cerebellum-to-intracranial volume ratio was overall maintained in males with NF1 (median: 52nd percentile), but not in females with NF1 (24th percentile). Conclusions: Pediatric NF1 cerebellar volumes can be captured by automated means. Pediatric NF1 patients appear to have significantly enlarged cerebellar volumes as compared to their peers. Despite overall cerebellar enlargement in both genders, females had a relatively larger overgrowth of total intracranial volume compared to their cerebellums.