FONDUE: Robust resolution-invariant denoising of MR Images using Nested UNets

Recent human neuroimaging studies tend to have increased magnetic resonance image (MRI) acquisition resolutions, seeking finer levels of detail and more accurate brain morphometry. However, higher-resolution images inherently contain greater amounts of noise contamination, leading to poorer quality brain morphometry if not addressed adequately. This study proposes a novel, robust, resolution-invariant deep learning method to denoise structural human brain MRIs. We explore denoising of T1-weighted (T1w) brain images from varying field strengths (1.5T to 7T), voxel sizes (1.2mm to 250µm), scanner vendors (Siemens, GE, and Phillips), and diseased and healthy participants from a wide age range (young adults to aging individuals). Our proposed Fast-Optimized Network for Denoising through residual Unified Ensembles (FONDUE) method demonstrated stable denoising capabilities across multiple resolutions with performance comparable to the state-of-the-art methods. FONDUE was capable of denoising 0.5mm3 isotropic T1w images in under 3 minutes on an NVIDIA RTX 3090 GPU using less than 8GB of video memory. We have also made the repository of FONDUE as well as its trained weights publicly available on: . ### Competing Interest Statement The authors have declared no competing interest.