Microstructural neuroimaging using spherical convolutional neural networks

We present a novel framework for estimating microstructural parameters of compartment models using recently developed orientationally invariant spherical convolutional neural networks and efficiently simulated training data. The networks were trained to predict the ground-truth parameter values from simulated noisy data and applied on imaging data acquired in a clinical setting to generate microstructural maps. Our network could estimate model parameters more accurately than conventional non-linear least squares or a multi-layer perceptron applied on powder-averaged data (i.e., the spherical mean technique).