Computational diffusion MRI : MICCAI Workshop最新文献

In this paper, we present a method based on sparse non-negative matrix factorization (NMF) for brain tissue segmentation using diffusion MRI (DMRI) data. Unlike existing NMF-based approaches, in our method NMF is applied to the spherical mean data, computed on a per-shell basis, instead of the original diffusion-weighted images. This is motivated by the fact that the spherical mean is independent of the fiber orientation distribution and is only dependent on tissue microstructure. Applying NMF to the spherical mean data will hence allow tissue signal separation based solely on the microstructural properties, unconfounded by factors such as fiber dispersion and crossing. We show results explaining why applying NMF directly on the diffusion-weighted images fails and why our method is able to yield the expected outcome, producing tissue segmentation with greater accuracy.

Diffusion-weighted magnetic resonance imaging (DW-MRI) allows for non-invasive imaging of the local fiber architecture of the human brain at a millimetric scale. Multiple classical approaches have been proposed to detect both single (e.g., tensors) and multiple (e.g., constrained spherical deconvolution, CSD) fiber population orientations per voxel. However, existing techniques generally exhibit low reproducibility across MRI scanners. Herein, we propose a data-driven technique using a neural network design which exploits two categories of data. First, training data were acquired on three squirrel monkey brains using ex-vivo DW-MRI and histology of the brain. Second, repeated scans of human subjects were acquired on two different scanners to augment the learning of the network proposed. To use these data, we propose a new network architecture, the null space deep network (NSDN), to simultaneously learn on traditional observed/truth pairs (e.g., MRI-histology voxels) along with repeated observations without a known truth (e.g., scan-rescan MRI). The NSDN was tested on twenty percent of the histology voxels that were kept completely blind to the network. NSDN significantly improved absolute performance relative to histology by 3.87% over CSD and 1.42% over a recently proposed deep neural network approach. Moreover, it improved reproducibility on the paired data by 21.19% over CSD and 10.09% over a recently proposed deep approach. Finally, NSDN improved generalizability of the model to a third in vivo human scanner (which was not used in training) by 16.08% over CSD and 10.41% over a recently proposed deep learning approach. This work suggests that data-driven approaches for local fiber reconstruction are more reproducible, informative and precise and offers a novel, practical method for determining these models.

Diffusion MRI affords great value for studying brain development, owing to its capability in assessing brain microstructure in association with myelination. With longitudinally acquired pediatric diffusion MRI data, one can chart the temporal evolution of microstructure and white matter connectivity. However, due to subject dropouts and unsuccessful scans, longitudinal datasets are often incomplete. In this work, we introduce a graph-based deep learning approach to predict diffusion MRI data. The relationships between sampling points in spatial domain (x-space) and diffusion wave-vector domain (q-space) are harnessed jointly (x-q space) in the form of a graph. We then implement a residual learning architecture with graph convolution filtering to learn longitudinal changes of diffusion MRI data along time. We evaluate the effectiveness of the spatial and angular components in data prediction. We also investigate the longitudinal trajectories in terms of diffusion scalars computed based on the predicted datasets.

The human brain develops very rapidly in the first years of life, resulting in significant changes in water diffusion anisotropy. Developmental changes pose significant challenges to longitudinally consistent white matter tractography. In this paper, we will introduce a method to harmonize infant diffusion MRI data longitudinally across time. Specifically, we harmonize diffusion MRI data collected at an earlier time point to data collected at a later time point. This will promote longitudinal consistency and allow sharpening of fiber orientation distribution functions (ODFs) based on information available at the later time point. For this purpose, we will introduce an approach that is based on the method of moments, which allows harmonization to be performed directly on the diffusion-attenuated signal without the need to fit any diffusion models to the data. Given two diffusion MRI datasets, our method harmonizes them voxel-wise using well-behaving mapping functions (i.e., monotonic, diffeomorphic, etc.), parameters of which are determined by matching the spherical moments (i.e., mean, variance, skewness, etc.) of signal measurements on each shell. The mapping functions we use is isotropic and does not introduce new orientations that are not already in the original data. Our analysis indicates that longitudinal harmonization sharpens ODFs and improves tractography in infant diffusion MRI.

Construction of brain atlases is generally carried out using a two-step procedure involving registering a population of images to a common space and then fusing the aligned images to form an atlas. In practice, image registration is not perfect and simple averaging of the images will blur structures and cause artifacts. In diffusion MRI, this is further complicated by the possibility of within-voxel fiber misalignment due to natural inter-subject orientation dispersion. In this paper, we propose a method to improve the construction of diffusion atlases in light of inter-subject fiber dispersion. Our method involves a novel q-space (i.e., wavevector space) patch matching mechanism that is incorporated in a mean shift algorithm to seek the most probable signal at each point in q-space. Our method relies on the fact that the mean shift algorithm is a mode seeking algorithm that converges to the mode of a distribution and is hence robustness to outliers. Our method is therefore in effect seeking the most probable signal profile at each voxel given a distribution of profiles. Experimental results confirm that our method yields cleaner fiber orientation distribution functions with less artifacts caused by dispersion.

Noise in diffusion-weighted (DW) images increases the complexity of quantitative analysis and decreases the reliability of inferences. Hence, to improve analysis, it is often desirable to remove noise and at the same time preserve relevant image features. In this paper, we propose a tight wavelet frame based approach for edge-preserving denoising of DW images. Our approach (i) employs the unitary extension principle (UEP) to generate frames that are discrete analogues to differential operators of various orders; (ii) introduces a very efficient method for solving an ℓ₀ denoising problem that involves only thresholding and solving a trivial inverse problem; and (iii) groups DW images acquired with neighboring gradient directions for collaborative denoising. Experiments using synthetic data with noncentral chi noise and real data with repeated scans confirm that our method yields superior performance compared with denoising using state-of-the-art methods such as non-local means.

Amygdala plays an important role in fear and emotional learning, which are critical for human survival. Despite the functional relevance and unique circuitry of each human amygdaloid subnuclei, there has yet to be an efficient imaging method for identifying these regions in vivo. A data-driven approach without prior knowledge provides advantages of efficient and objective assessments. The present study uses high angular and high spatial resolution diffusion magnetic resonance imaging to generate orientation distribution function, which bears distinctive microstructural features. The features were extracted using spherical harmonic decomposition to assess microstructural similarity within amygdala subfields are identified via similarity matrices using spectral k-mean clustering. The approach was tested on 32 healthy volunteers and three distinct amygdala subfields were identified including medial, posterior-superior lateral, and anterior-inferior lateral.