Numerical Uncertainty of Convolutional Neural Networks Inference for Structural Brain MRI Analysis

This paper investigates the numerical uncertainty of Convolutional Neural Networks (CNNs) inference for structural brain MRI analysis. It applies Random Rounding-a stochastic arithmetic technique- to CNN models employed in non-linear registration (SynthMorph) and whole-brain segmentation (FastSurfer), and compares the resulting numerical uncertainty to the one measured in a reference imageprocessing pipeline (FreeSurfer recon-all). Results obtained on 32 representative subjects show that CNN predictions are substantially more accurate numerically than traditional image-processing results (nonlinear registration: 19 vs 13 significant bits on average; whole-brain segmentation: 0.99 vs 0.92 Sorensen-Dice score on average), which suggests a better reproducibility of CNN results across execution environments.