Diffusion-Driven Generation of Minimally Preprocessed Brain MRI

TL;DR

Despite the negative results in statistical testing, the presented DDPMs are capable of generating high-resolution 3D $T_1$-weighted brain images and are the first 3D non-latent diffusion model for brain data without skullstripping or registration.

Abstract

The purpose of this study is to present and compare three denoising diffusion probabilistic models (DDPMs) that generate 3D $T_1$-weighted MRI human brain images. Three DDPMs were trained using 80,675 image volumes from 42,406 subjects spanning 38 publicly available brain MRI datasets. These images had approximately 1 mm isotropic resolution and were manually inspected by three human experts to exclude those with poor quality, field-of-view issues, and excessive pathology. The images were minimally preprocessed to preserve the visual variability of the data. Furthermore, to enable the DDPMs to produce images with natural orientation variations and inhomogeneity, the images were neither registered to a common coordinate system nor bias field corrected. Evaluations included segmentation, Frechet Inception Distance (FID), and qualitative inspection. Regarding results, all three DDPMs generated coherent MR brain volumes. The velocity and flow prediction models achieved lower FIDs than the sample prediction model. However, all three models had higher FIDs compared to real images across multiple cohorts. In a permutation experiment, the generated brain regional volume distributions differed statistically from real data. However, the velocity and flow prediction models had fewer statistically different volume distributions in the thalamus and putamen. In conclusion this work presents and releases the first 3D non-latent diffusion model for brain data without skullstripping or registration. Despite the negative results in statistical testing, the presented DDPMs are capable of generating high-resolution 3D $T_1$-weighted brain images. All model weights and corresponding inference code are publicly available at https://github.com/piksl-research/medforj .

Figures (4)

• Figure 1: Parasagittal slices from representative pass and fail images. (a) Images that passed manual QA. (b) Images that failed manual QA. Failure reasons include Pathology ($\bullet$), Limited FOV ($\blacksquare$), Aggressive defacing ($\blacktriangle$), Motion artifacts ($\blacksquare$), and Noise ($\blacktriangle$).
• Figure 2: Uncurated synthetic volumes from each prediction type (triplanar display). All volumes were generated with 64 DDIM steps.
• Figure 3: Triplanar views of synthetic images and their nearest neighbors. The first group of three columns shows the synthetic image, and the remaining groups of columns show the first and second nearest neighbors in the real data training set, respectively. Nearest neighbor is computed as mean squared error in voxel space.
• Figure 4: Raincloud plots showing the distributions of volumes as computed by SynthSeg.