Fully automated classification of HARDI in vivo data using a support vector machine.

The purpose of this study is the classification of high angular resolution diffusion imaging (HARDI) in vivo data using a model-free approach. This is achieved by using a Support Vector Machine (SVM) algorithm taken from the field of supervised statistical learning. Six classes of image components are determined: grey matter, parallel neuronal fibre bundles in white matter, crossing neuronal fibre bundles in white matter, partial volume between white and grey matter, background noise and cerebrospinal fluid. The SVM requires properties derived from the data as input, the so called feature vector, which should be rotation invariant. For our application we derive such a description from the spherical harmonic decomposition of the HARDI signal. With this information the SVM is trained in order to find the function for separating the classes. The SVM is systematically tested with simulated data and then applied to six in vivo data sets. This new approach is data-driven and enables fully automatic HARDI data segmentation without employing a T1 MPRAGE scan and subjective expert intervention. This was demonstrated on five test in vivo data sets giving robust results. The segmentation results could be used as a priori knowledge for increasing the performance of fibre tracking as well as for other clinical and diagnostic applications of diffusion weighted imaging (DWI).

Gibbs tracking: a novel approach for the reconstruction of neuronal pathways.

Reconstruction of neuronal fibers using diffusion-weighted (DW) MRI is an emerging method in biomedical research. Existing fiber-tracking algorithms are commonly based on the "walker principle." Fibers are reconstructed as trajectories of "walkers," which are guided according to local diffusion properties. In this study, a new method of fiber tracking is proposed that does not engage any "walking" algorithm. It resolves a number of inherent problems of the "walking" approach, in particular the reconstruction of crossing and spreading fibers. In the proposed method, the fibers are built with small line elements. Each line element contributes an anisotropic term to the simulated DW signal, which is adjusted to the measured signal. This method demonstrates good results for simulated fibers. A single in vivo result demonstrates the successful reconstruction of the dominant neuronal pathways. A comparison with the diffusion tensor imaging (DTI)-based fiber assignment with continuous tracking (FACT) method and the probabilistic index of connectivity (PICo) method based on a multitensor model is performed for the callosal fibers. The result shows a strong increase in the number of reconstructed fibers. These almost fill the total white matter (WM) volume and connect a large area of the cortex. The method is very computationally expensive. Possible ways to address this problem are discussed.

Connecting and merging fibres: pathway extraction by combining probability maps.

Probability mapping of connectivity is a powerful tool to determine the fibre structure of white matter in the brain. Probability maps are related to the degree of connectivity to a chosen seed area. In many applications, however, it is necessary to isolate a fibre bundle that connects two areas. A frequently suggested solution is to select curves, which pass only through two or more areas. This is very inefficient, especially for long-distance pathways and small areas. In this paper, a novel probability-based method is presented that is capable of extracting neuronal pathways defined by two seed points. A Monte Carlo simulation based tracking method, similar to the Probabilistic Index of Connectivity (PICo) approach, was extended to preserve the directional information of the main fibre bundles passing a voxel. By combining two of these extended visiting maps arising from different seed points, two independent parameters are determined for each voxel: the first quantifies the uncertainty that a voxel is connected to both seed points; the second represents the directional information and estimates the proportion of fibres running in the direction of the other seed point (connecting fibre) or face a third area (merging fibre). Both parameters are used to calculate the probability that a voxel is part of the bundle connecting both seed points. The performance and limitations of this DTI-based method are demonstrated using simulations as well as in vivo measurements.

Multitensor approach for analysis and tracking of complex fiber configurations.

A multidiffusion-tensor model (MDT) is presented containing two anisotropic and one isotropic diffusion tensors. This approach has the ability to detect areas of fiber crossings and resolve the direction of crossing fibers. The mean diffusivity and the ratio of the tensor compartments were merged to one independent parameter by fitting MDT to the diffusion-weighted intensities of a two-point data acquisition scheme. By an F-test between the errors of the standard single diffusion tensor and the more complex MDT, fiber crossings were detected and the more accurate model was chosen voxel by voxel. The performance of crossing detection was compared with the spherical harmonics approach in simulations as well as in vivo. Similar results were found in both methods. The MDT model, however, did not only detect crossings but also yielded the single fiber directions. The FACT algorithm and a probabilistic connectivity algorithm were extended to support the MDT model. For example, a mean angular error smaller than 10 degrees was found for the MDT model in a simulated fiber crossing with an SNR of 80. By tracking the corticospinal tract the MDT-based tracks reached a significantly greater area of the gyrus precentralis.