Layer Selection in Feature-Based Losses Affects Image Quality and Microstructural Consistency in Deep Learning Super-Resolution of Brain Diffusion MRI

2026-05-15

AI Summary

This study addresses the challenges of high-resolution diffusion MRI, which is constrained by hardware limitations and prolonged scan times. Existing deep learning–based super-resolution methods often introduce artifacts and compromise microstructural consistency. Leveraging 7T human connectome data, the authors employ a UNet architecture for 2D super-resolution reconstruction and systematically evaluate the impact of feature loss derived from different layers of VGG16 on image fidelity and diffusion signal consistency. They find, for the first time, that deeper-layer feature losses induce grid-like artifacts and bias diffusion parameter estimation, whereas the shallowest-layer feature loss best preserves microstructural integrity. Experiments demonstrate that this strategy effectively suppresses artifacts even at up to 9× super-resolution, yielding reconstructions highly consistent with ground-truth high-resolution data, with both image signal-to-noise ratio and VGG layer depth jointly modulating artifact manifestation.

Abstract

Clinical application of high-resolution diffusion MRI is hindered by hardware limitations and prohibitive scan times, motivating computational super-resolution. This study investigates the efficacy of a feature-based loss function in preserving diffusion signal consistency in deep learning super-resolution. Using 7T data from the human connectome project to generate pairs of low- and high-resolution diffusion weighted images (DWI), we trained UNets for 2D super-resolution. Ablation and isolation studies evaluated different VGG16-layers for feature-based losses against an image-based L1 baseline. Deeper layers and combinations thereof resulted in grid-like artifacts in super-resolution DWIs, which persisted in diffusion parameters like quantitative and fractional anisotropy. No such artifacts were present when using the shallowest layer. Downstream analysis for this layer showed great consistency with the ground truth, even for 9-fold super-resolution. Image SNR and used VGG16-layer depths modulated artifact appearance and severity, mandating careful selection of contributing layers for application in and beyond diffusion MRI.

Problem

• super-resolution
• diffusion MRI
• feature-based loss
• microstructural consistency
• image artifacts

Innovation

• feature-based loss
• super-resolution
• diffusion MRI
• VGG16 layer selection
• microstructural consistency