Resection cavities, tumors, and other lesions can fundamentally alter brain structure and present as abnormalities in brain MRI. Specifically, quantifying subtle neuroanatomical changes in other, not directly affected regions of the brain is essential to assess the impact of tumors, surgery, chemo/radiotherapy, or drug treatments. However, only a limited number of solutions address this important task, while many standard analysis pipelines simply do not support abnormal brain images at all. In this paper, we present a method to perform sensitive neuroanatomical analysis of healthy brain regions in the presence of large lesions and cavities. Our approach called “FastSurfer Lesion Inpainting Tool” (FastSurfer-LIT) leverages the recently emerged Denoising Diffusion Probabilistic Models (DDPM) to fill lesion areas with healthy tissue that matches and extends the surrounding tissue. This enables subsequent processing with established MRI analysis methods such as the calculation of adjusted volume and surface measurements using FastSurfer or FreeSurfer. FastSurfer-LIT significantly outperforms previously proposed solutions on a large dataset of simulated brain tumors (N = 100) and synthetic multiple sclerosis lesions (N = 39) with improved Dice and Hausdorff measures, and also on a highly heterogeneous dataset with lesions and cavities in a manual assessment (N = 100). Finally, we demonstrate increased reliability to reproduce pre-operative cortical thickness estimates from corresponding post-operative temporo-mesial resection surgery MRIs. The method is publicly available at https://github.com/Deep-MI/LIT and will be integrated into the FastSurfer toolbox.

A robust, fast, and accurate segmentation of neonatal brain images is highly desired to better understand and detect changes during development and disease, specifically considering the rise in imaging studies for this cohort. Yet, the limited availability of ground truth datasets, lack of standardized acquisition protocols, and wide variations of head positioning in the scanner pose challenges for method development. A few automated image analysis pipelines exist for newborn brain Magnetic Resonance Image (MRI) segmentation, but they often rely on time-consuming non-linear spatial registration procedures and require resampling to a common resolution, subject to loss of information due to interpolation and down-sampling. Without registration and image resampling, variations with respect to head positions and voxel resolutions have to be addressed differently. In deep learning, external augmentations such as rotation, translation, and scaling are traditionally used to artificially expand the representation of spatial variability, which subsequently increases both the training dataset size and robustness. However, these transformations in the image space still require resampling, reducing accuracy specifically in the context of label interpolation. We recently introduced the concept of resolution-independence with the Voxel-size Independent Neural Network framework, VINN. Here, we extend this concept by additionally shifting all rigid-transforms into the network architecture with a four degree of freedom (4-DOF) transform module, enabling resolution-aware internal augmentations (VINNA) for deep learning. In this work, we show that VINNA (i) significantly outperforms state-of-the-art external augmentation approaches, (ii) effectively addresses the head variations present specifically in newborn datasets, and (iii) retains high segmentation accuracy across a range of resolutions (0.5–1.0 mm). Furthermore, the 4-DOF transform module together with internal augmentations is a powerful, general approach to implement spatial augmentation without requiring image or label interpolation. The specific network application to newborns will be made publicly available as VINNA4neonates.

Diffusion-weighted magnetic resonance imaging (dMRI) permits a detailed in-vivo analysis of neuroanatomical microstructure, invaluable for clinical and population studies. However, many measurements with different diffusion-encoding directions and possibly b -values are necessary to infer the underlying tissue microstructure within different imaging voxels accurately. Two challenges particularly limit the utility of dMRI: long acquisition times limit feasible scans to only a few directional measurements, and the heterogeneity of acquisition schemes across studies makes it difficult to combine datasets. Left unaddressed by previous learning-based methods that only accept dMRI data adhering to the specific acquisition scheme used for training, there is a need for methods that accept and predict signals for arbitrary diffusion encodings. Addressing these challenges, we describe the first geometric deep learning method for continuous dMRI signal reconstruction for arbitrary diffusion sampling schemes for both the input and output. Our method combines the reconstruction accuracy and robustness of previous learning-based methods with the flexibility of model-based methods, for example, spherical harmonics or SHORE. We demonstrate that our method outperforms model-based methods and performs on par with discrete learning-based methods on single-, multi-shell, and grid-based diffusion MRI datasets. Relevant for dMRI-derived analyses, we show that our reconstruction translates to higher-quality estimates of frequently used microstructure models compared to other reconstruction methods, enabling high-quality analyses even from very short dMRI acquisitions.

The hypothalamus plays a crucial role in the regulation of a broad range of physiological, behavioral, and cognitive functions. However, despite its importance, only a few small-scale neuroimaging studies have investigated its substructures, likely due to the lack of fully automated segmentation tools to address scalability and reproducibility issues of manual segmentation. While the only previous attempt to automatically sub-segment the hypothalamus with a neural network showed promise for 1.0 mm isotropic T1-weighted (T1w) magnetic resonance imaging (MRI), there is a need for an automated tool to sub-segment also high-resolutional (HiRes) MR scans, as they are becoming widely available, and include structural detail also from multi-modal MRI. We, therefore, introduce a novel, fast, and fully automated deep-learning method named HypVINN for sub-segmentation of the hypothalamus and adjacent structures on 0.8 mm isotropic T1w and T2w brain MR images that is robust to missing modalities. We extensively validate our model with respect to segmentation accuracy, generalizability, in-session test-retest reliability, and sensitivity to replicate hypothalamic volume effects (e.g., sex differences). The proposed method exhibits high segmentation performance both for standalone T1w images as well as for T1w/T2w image pairs. Even with the additional capability to accept flexible inputs, our model matches or exceeds the performance of state-of-the-art methods with fixed inputs. We, further, demonstrate the generalizability of our method in experiments with 1.0 mm MR scans from both the Rhineland Study and the UK Biobank—an independent dataset never encountered during training with different acquisition parameters and demographics. Finally, HypVINN can perform the segmentation in less than a minute (graphical processing unit [GPU]) and will be available in the open source FastSurfer neuroimaging software suite, offering a validated, efficient, and scalable solution for evaluating imaging-derived phenotypes of the hypothalamus.