Deep Learning Mixture-of-Experts for Cytotoxic Edema Assessment in Infants and Children

Abusive Head Trauma (AHT) is the most important source of morbidity and mortality for abused children. Cytotoxic Edema (CE) has been suggested to be a sign of poor outcome for children with AHT, but this has not been tested. We propose a Mixture-of-Experts (MoE) deep learning system that includes two 3D network architectures optimized to learn patterns of CE from two types of clinical MRI data - a trace Diffusion Weighted Image (DWI) and the calculated Apparent Diffusion Coefficient map (ADC). We devise a novel approach based on volumetric analysis of 3D images (using pixels from time slices) and 3D convolutional neural network (CNN). Experiments on a dataset curated from a Children's Hospital Colorado (CHCO) [1] patient registry show a predictive performance F1 score of 0.93 at distinguishing CE patients from children with severe neurologic injury without CE. In addition, we perform ablation studies to determine the association between CE and AHT, and overall functional outcome and in-hospital mortality of infants and young children.