A total of 600 pediatric brain magnetic resonance imaging (MRI) images (400 normal, 200 hydrocephalus) was divided into a training set (n=420), validation set (n=90) and test set (n=90). All normal scans were obtained from patients between the age of 0 and 18 seen at our institution between 2011-2017. Hydrocephalus cohort was selected among patients with posterior fossa tumors who had a radiological evidence of ventriculomegaly, Scans from this cohort was composed of 50 MRIs from our institution and 150 scans from three outside centers. Expert segmentation of ventricles was performed on all T2-weighted and thin-slice T1-weighted spoiled gradient recalled acquisition in the steady state (SPGR) scans. Our encoder-decoder convolutional neural network architecture consisted of a UNet with a pre-trained ResNet50 as the encoder.

In this multi-institutional study, we present a deep-learning model that automatically segments ventricles and outputs volumetric information. This clinically applicable and externally validated tool may enhance our current understanding of ventricular development and improve accuracy of ventricular volumetric measurement.