This dissertation deals with the use of constrained spherical deconvolution (CSD) of diffusion weighted (DW) MRI data for the purpose of improved fiber tractography. The manuscript is divided into two large parts. Part I , provides the necessary background material on diffusion MRI, multi-fiber reconstruction algorithms and fiber tractography. Part II provides an overview of the main contributions of this thesis.;Diffusion-weighted (DW) MRI is a magnetic resonance imaging (MRI) technique that indirectly measures the local mobility of water molecules. It is unique in its ability to measure diffusion non-invasively, making it the method of choice for in vivo diffusion measurements. A key feature of diffusion MRI is that it can provide information about the geometry of the underlying tissue microstructure, at scales much smaller than the imaging resolution. In fibrous tissue, such as in the brain white matter (WM), water molecules tend to diffuse more along the fibers, enabling researchers to obtain information about the orientation and 'integrity' of the underlying tissue. Currently, diffusion tensor imaging (DTI) is the most widely used method for assessing WM orientation and integrity, owing to its modest acquisition requirements. The ability to assess WM orientation and integrity from a single in vivo scan raises huge possibilities for neuroscientific research and there has been a rapid increase in clinical studies using DTI in the last decade. For a detailed review of the principles of diffusion, diffusion MRI and DTI the reader is referred to Chapter 1.;Despite its popularity, DTI has an important limitation in that it can only model a single fiber population per voxel. However, due to partial volume effects between adjacent WM fiber bundles, many voxels contain contributions from several differently oriented fiber populations. In such voxels, DTI orientation and DTI integrity metrics are unreliable. Recently, a number of methods have been proposed that are able to extract multiple fiber orientations from the DW signal, overcoming the limitation of DTI. One particularly promising method is constrained spherical deconvolution (CSD), which recovers the full fiber orientation distribution function (fODF) within each voxel directly from the diffusion data using the concept of spherical deconvolution. By applying a non-negativity constraint on the fODF, CSD allows robust multiple fiber orientation estimation using relatively modest acquisition settings. An in-depth review of the different multi-fiber reconstruction algorithms, CSD in particular, is provided in Chapter 2.;Fiber tractography pieces together the local WM orientations derived with DTI or more advanced multiple fiber reconstruction algorithms in order to infer long-range connectivity patterns between distant brain regions. Diffusion MRI based fiber tractography is unique in its ability to delineate the WM fiber pathways in a non-invasive way, raising possibilities for clinical applications and providing new insights in how the brain is wired up. Fiber tractography algorithms can be classified largely into deterministic and probabilistic algorithms. Deterministic tractography algorithms reconstruct the most likely trajectory emanating from a given point, whereas probabilistic algorithms produce a distribution of trajectories, reflecting the degree of uncertainty of the trajectories. The concepts, limitations, and applications of fiber tractography are introduced in Chapter 3.;Contributions.;As DTI based fiber tractography becomes unreliable in regions of complex fiber configurations, we developed a new deterministic tractography algorithm based on CSD. As CSD is capable of resolving multiple fiber orientations within each voxel, it is expected to improve tractography results in regions of complex fiber architecture. By means of a simple crossing fiber phantom, we showed that the algorithm is able to track through regions containing crossing fibers where DTI tractography fails. In addition, our method was evaluated quantitatively on a more complex fiber phantom, as part of the MICCAI 2009 fiber cup contest. Analysis of the results revealed our solution was characterized by the lowest average error for both the spatial and directional metric and our method was the only one tracing the correct fiber bundles from start to end. In Chapter 4, our algorithm, as well as the quantitative and qualitative evaluation using different MR phantoms is explained in detail. In addition we briefly discuss some applications of the proposed CSD tractography method.;While CSD offers an improved estimate of the fiber orientations in the presence of partial volume effects, diffusion MRS is inherently a noisy technique, resulting in uncertainty associated with each fiber orientation estimate. In Chapter 5, we introduce the use of bootstrapping techniques to quantify the uncertainty of CSD estimated fiber orientations. The performance of bootstrapping was measured in terms of accuracy and precision using Monte Carlo simulations. We looked at both the 'classic repetition bootstrap' approach which estimates the fiber orientation uncertainty by randomly selecting individual measurements from a set of repeated measurements, and the 'residual bootstrap' approach, which estimates the fiber orientation by randomly selecting model residuals, requiring only a single measurement and thus being more clinically feasible. Our simulations showed that the 'classic repetition bootstrap' significantly underestimates the uncertainty when only a few repeated acquisitions are available, which is typically the case. We showed that this large downward bias can be removed by using the bootknife approach, allowing accurate CSD fiber orientation uncertainty estimates with only a limited set of repeated measurements and without making assumptions about the sources of uncertainty in the data. However, in a clinical setting, even a few repeated measurements can render acquisition time unacceptably long. For this reason we also investigated the residual bootstrap, which performs the bootstrapping procedure on the residuals of a model fit, requiring only a single acquisition. Our simulations showed that the combination of the residual bootstrap with the modified spherical harmonics model allows accurate estimates of the CSD fiber orientation uncertainty, bringing it into the clinical realm.;In Chapter 6, we build on the findings of Chapters 4 & 5 to formulate a new probabilistic tractography algorithm based on CSD and the residual bootstrap, overcoming the limitations of DTI tractography and at the same time providing uncertainty measures of the fiber trajectories, using only a single acquisition. Using Monte Carlo simulations, we measured the accuracy and precision of the residual bootstrap method when estimating CSD fiber pathway uncertainty. We also applied our algorithm to clinical DW data and compared our method to state-of-the-art DTI residual bootstrap tractography and to an established probabilistic multi-fiber CSD tractography algorithm which draws samples directly from the fODF. CSD residual bootstrap probabilistic tractography showed advantageous over DTI residual bootstrap probabilistic tractography: in regions of multiple fiber orientations, CSD was much less prone to fiber dispersion, false positives, and false negatives. We also showed the advantages of our method over CSD fODF sampling tractography: in regions of well ordered and sharp peak orientations, our method does not suffer from unrealistically high dispersion and our method has a higher specificity in general.;In Chapter 7, we set out to assess the prevalence of voxels containing multiple fiber orientations, as these are the voxels where multi-fiber reconstruction algorithms would result in improved tractography results. For this purpose, we acquired large, high quality DW data sets and extracted the fiber orientations using both CSD and the bedpostx algorithm. Our results indicated that multiple fiber orientations can be found in a much higher percentage of WM voxels than previously reported, with CSD providing much higher estimates than bedpostx. These findings have obvious and profound implications for both tractography and integrity analyses, and strengthen the growing awareness that fiber tractography and 'WM integrity' metrics derived from DTI need to be interpreted with extreme caution, underlining the importance of the methods developed in the previous chapters.
展开▼
