Diffusion magnetic resonance imaging (diffusion MRI) is widely employed to probe the diffusive motion of water molecules within the tissue. Numerous diseases and processes affecting the central nervous system can be detected and monitored via diffusion MRI thanks to its sensitivity to microstructural alterations in tissue. The latter has prompted interest in quantitative mapping of the microstructural parameters, such as the fiber orientation distribution function (fODF), which is instrumental for noninvasively mapping the underlying axonal fiber tracts in white matter through a procedure known as tractography. However, such applications demand repeated acquisitions of MRI volumes with varied experimental parameters demanding long acquisition times and/or limited spatial resolution. In this work, we present a deep-learning-based approach for increasing the spatial resolution of diffusion MRI data in the form of fODFs obtained through constrained spherical deconvolution. The proposed approach is evaluated on high quality data from the Human Connectome Project, and is shown to generate upsampled results with a greater correspondence to ground truth high-resolution data than can be achieved with ordinary spline interpolation methods. Furthermore, we employ a measure based on the earth mover’s distance to assess the accuracy of the upsampled fODFs. At low signal-to-noise ratios, our super-resolution method provides more accurate estimates of the fODF compared to data collected with 8 times smaller voxel volume.

Goal: Structural brain graphs are conventionally limited to defining nodes as gray matter regions from an atlas,with edges reflecting the density of axonal projections between pairs of nodes. Here we explicitly model the entire set of voxels within a brain mask as nodes of high-resolution, subject-specific graphs. Methods: We define the strength of local voxel-to-voxel connections using diffusion tensors and orientation distribution functions derived from diffusion MRI data. We study the graphs’ Laplacian spectral properties on data from the Human Connectome Project. We then assess the extent of inter-subject variability of the Laplacian eigenmodes via a procrustes validation scheme. Finally, we demonstrate the extent to which functional MRI data are shaped by the underlying anatomical structure via graph signal processing. Results: The graph Laplacian eigenmodes manifest highly resolved spatial profiles, reflecting distributed patterns that correspond to major white matter pathways. We show that the intrinsic dimensionality of the eigenspace of such high-resolution graphs is only a mere fraction of the graph dimensions. By projecting task and resting-state data on low frequency graph Laplacian eigenmodes, we show that brain activity can be well approximated by a small subset of low frequency components. Conclusions: The proposed graphs open new avenues in studying the brain, be it, by exploring their organisational properties via graph or spectral graph theory, or by treating them as the scaffold on which brain function is observed at the individual level.

Classifying subjects as healthy or diseased using neuroimaging data has gained a lot of attention during the last 10 years, and recently, different deep learning approaches have been used. Despite this fact, there has not been any investigation regarding how 3D augmentation can help to create larger datasets, required to train deep networks with millions of parameters. In this study, deep learning was applied to derivatives from resting state functional MRI data, to investigate how different 3D augmentation techniques affect the test accuracy. Specifically, resting state derivatives from 1112 subjects in ABIDE (Autism Brain Imaging Data Exchange) preprocessed were used to train a 3D convolutional neural network (CNN) to classify each subject according to presence or absence of autism spectrum disorder. The results show that augmentation only provide minor improvements to the test accuracy.

Background: Stereotactic radiosurgery (SRS) can be an effective primary or adjuvant treatment option for intracranial tumors. However, it carries risks of various radiation toxicities, which can lead to functional deficits for the patients. Current inverse planning algorithms for SRS provide an efficient way for sparing organs at risk (OARs) by setting maximum radiation dose constraints in the treatment planning process.Purpose: We propose using activation maps from functional MRI (fMRI) to map the eloquent regions of the brain and define functional OARs (fOARs) for Gamma Knife SRS treatment planning.Methods: We implemented a pipeline for analyzing patient fMRI data, generating fOARs from the resulting activation maps, and loading them onto the GammaPlan treatment planning software. We used the Lightning inverse planner to generate multiple treatment plans from open MRI data of five subjects, and evaluated the effects of incorporating the proposed fOARs.Results: The Lightning optimizer designs treatment plans with high conformity to the specified parameters. Setting maximum dose constraints on fOARs successfully limits the radiation dose incident on them, but can have a negative impact on treatment plan quality metrics. By masking out fOAR voxels surrounding the tumor target it is possible to achieve high quality treatment plans while controlling the radiation dose on fOARs.Conclusions: The proposed method can effectively reduce the radiation dose incident on the eloquent brain areas during Gamma Knife SRS of brain tumors.

Magnetic resonance imaging (MRI) is one of the pillars of modern medical imaging, providing a non-invasive means to generate 3D images of the body with high soft-tissue contrast. Furthermore, the possibilities afforded by the design of MRI sequences enable the signal to be sensitized to a multitude of physiological tissue properties, resulting in a wide variety of distinct MRI modalities for clinical and research use. 

This thesis presents a number of advanced brain MRI applications, which fulfill, to differing extents, two complementary aims. On the one hand, they explore the benefits of a multimodal approach to MRI, combining structural, functional and diffusion MRI, in a variety of contexts. On the other, they emphasize the use of advanced mathematical and computational tools in the analysis of MRI data, such as deep learning, Bayesian statistics, and graph signal processing. 

Paper I introduces an anatomically-adapted extension to previous work in Bayesian spatial priors for functional MRI data, where anatomical information is introduced from a T1-weighted image to compensate for the low anatomical contrast of functional MRI data. 

It has been observed that the spatial correlation structure of the BOLD signal in brain white matter follows the orientation of the underlying axonal fibers. Paper II argues about the implications of this fact on the ideal shape of spatial filters for the analysis of white matter functional MRI data. By using axonal orientation information extracted from diffusion MRI, and leveraging the possibilities afforded by graph signal processing, a graph-based description of the white matter structure is introduced, which, in turn, enables the definition of spatial filters whose shape is adapted to the underlying axonal structure, and demonstrates the increased detection power resulting from their use. 

One of the main clinical applications of functional MRI is functional localization of the eloquent areas of the brain prior to brain surgery. This practice is widespread for various invasive surgeries, but is less common for stereotactic radiosurgery (SRS), a non-invasive surgical procedure wherein tissue is ablated by concentrating several beams of high-energy radiation. Paper III describes an analysis and processing pipeline for functional MRI data that enables its use for functional localization and delineation of organs-at-risk for Elekta GammaKnife SRS procedures. 

Paper IV presents a deep learning model for super-resolution of diffusion MRI fiber ODFs, which outperforms standard interpolation methods in estimating local axonal fiber orientations in white matter. Finally, Paper V demonstrates that some popular methods for anonymizing facial data in structural MRI volumes can be partially reversed by applying generative deep learning models, highlighting one way in which the enormous power of deep learning models can potentially be put to use for harmful purposes. 

In this work, we leverage the Laplacian eigenbasis of voxelwise white matter (WM) graphs derived from diffusionweighted MRI data, dubbed WM harmonics, to characterize the spatial structure of WM fMRI data. Our motivation for such a characterization is based on studies that show WM fMRI data exhibit a spatial correlational anisotropy that coincides with underlying fiber patterns. By quantifying the energy content of WM fMRI data associated with subsets of WM harmonics across multiple spectral bands, we show that the data exhibits notable subtle spatial modulations under functional load that are not manifested during rest. WM harmonics provide a novel means to study the spatial dynamics of WM fMRI data, in such way that the analysis is informed by the underlying anatomical structure.

Brain activation mapping using functional magnetic resonance imaging (fMRI) has been extensively studied in brain gray matter (GM), whereas in large disregarded for probing white matter (WM). This unbalanced treatment has been in part due to controversies in relation to the nature of the blood oxygenation level-dependent (BOLD) contrast in WM and its detachability. However, an accumulating body of studies has provided solid evidence of the functional significance of the BOLD signal in WM and has revealed that it exhibits anisotropic spatio-temporal correlations and structure-specific fluctuations concomitant with those of the cortical BOLD signal. In this work, we present an anisotropic spatial filtering scheme for smoothing fMRI data in WM that accounts for known spatial constraints on the BOLD signal in WM. In particular, the spatial correlation structure of the BOLD signal in WM is highly anisotropic and closely linked to local axonal structure in terms of shape and orientation, suggesting that isotropic Gaussian filters conventionally used for smoothing fMRI data are inadequate for denoising the BOLD signal in WM. The fundamental element in the proposed method is a graph-based description of WM that encodes the underlying anisotropy observed across WM, derived from diffusion-weighted MRI data. Based on this representation, and leveraging graph signal processing principles, we design subject-specific spatial filters that adapt to a subject’s unique WM structure at each position in the WM that they are applied at. We use the proposed filters to spatially smooth fMRI data in WM, as an alternative to the conventional practice of using isotropic Gaussian filters. We test the proposed filtering approach on two sets of simulated phantoms, showcasing its greater sensitivity and specificity for the detection of slender anisotropic activations, compared to that achieved with isotropic Gaussian filters. We also present WM activation mapping results on the Human Connectome Project’s 100-unrelated subject dataset, across seven functional tasks, showing that the proposed method enables the detection of streamline-like activations within axonal bundles.

Gliomas are the most common primary brain malignancies, with different degrees of aggressiveness, variable prognosis and various heterogeneous histological sub-regions, i.e., peritumoral edema, necrotic core, enhancing and non-enhancing tumour core. Although brain tumours can easily be detected using multi-modal MRI, accurate tumor segmentation is a challenging task. Hence, using the data provided by the BraTS Challenge 2020, we propose a 3D volume-to-volume Generative Adversarial Network for segmentation of brain tumours. The model, called Vox2Vox, generates realistic segmentation outputs from multi-channel 3D MR images, segmenting the whole, core and enhancing tumor with mean values of 87.20%, 81.14%, and 78.67% as dice scores and 6.44mm, 24.36 mm, and 18.95 mm for Hausdorff distance 95 percentile for the BraTS testing set after ensembling 10 Vox2Vox models obtained with a 10-fold cross-validation. The code is available at https://​github.​com/​mdciri/​Vox2Vox

Training segmentation networks requires large annotated datasets, which in medical imaging can be hard to obtain. Despite this fact, data augmentation has in our opinion not been fully explored for brain tumor segmentation. In this project we apply different types of data augmentation (flipping, rotation, scaling, brightness adjustment, elastic deformation) when training a standard 3D U-Net, and demonstrate that augmentation significantly improves the network’s performance in many cases. Our conclusion is that brightness augmentation and elastic deformation work best, and that combinations of different augmentation techniques do not provide further improvement compared to only using one augmentation technique. Our code is available at https://github.com/mdciri/3D-augmentation-techniques

Existing Bayesian spatial priors for functional magnetic resonance imaging (fMRI) data correspond to stationary isotropic smoothing filters that may oversmooth at anatomical boundaries. We propose two anatomically informed Bayesian spatial models for fMRI data with local smoothing in each voxel based on a tensor field estimated from a T1-weighted anatomical image. We show that our anatomically informed Bayesian spatial models results in posterior probability maps that follow the anatomical structure.