Neural net systems have been developed to process magnetic resonance imaging (MRI) data from sensor space into real space and then to analyze the images for diagnosis. These systems can handle large 3d or higher dimensional data sets and perform correlations that are beyond human abilities. A critical requirement is that these systems be validated and to understand how their image reconstruction relates to ground truth.  This opportunity involves developing neural net systems that have built in explainability.  Both publicly available systems such as Tensor Flow and in-house code will be used.  Training and evaluation data sets will be obtained using NIST MRI phantoms, imaged on both the NIST preclinical and University of Colorado 3 T clinical scanners.  Emphasis will be put on going beyond simple feedforward networks to allow cognitive association with patient/system history and on novel side chains to allow monitoring of the reconstruction and diagnostic processes. Additional projects may include creating large brain/head virtual data sets using simple growth algorithms to mimic brain tissue as well as various lesions. Bloch solvers with built in stochastic uncertainties, noise, and artifacts will be used to convert these images into sensor space for use in neural net validation.