A Machine Learning Based Approach to Fiber Tractography Using Classifier Voting
Current tractography pipelines incorporate several modelling assumptions about the nature of the diffusion-weighted signal. We present an approach that tracks fiber pathways based on a random forest classification and voting process, guiding each step of
- PDF / 1,081,900 Bytes
- 8 Pages / 439.363 x 666.131 pts Page_size
- 109 Downloads / 182 Views
Abstract. Current tractography pipelines incorporate several modelling assumptions about the nature of the diffusion-weighted signal. We present an approach that tracks fiber pathways based on a random forest classification and voting process, guiding each step of the streamline progression by directly processing raw signal intensities. We evaluated our approach quantitatively and qualitatively using phantom and in vivo data. The presented machine learning based approach to fiber tractography is the first of its kind and our experiments showed auspicious performance compared to 12 established state of the art tractography pipelines. Due to its distinctly increased sensitivity and specificity regarding tract connectivity and morphology, the presented approach is a valuable addition to the repertoire of currently available tractography methods and promises to be beneficial for all applications that build upon tractography results.
1
Introduction
Fiber tractography on the basis of diffusion-weighted magnetic resonance imaging (DW-MRI) enables the spatial reconstruction of white matter pathways connecting the different regions of the brain. Despite the efforts invested into developing novel diffusion modeling and fiber tractography methods, ranging from local deterministic approaches through probabilistic methods to global tractography, several studies have shown that the task of fiber tractography is far from being solved. Current tractography algorithms still struggle with simultaneously achieving a high sensitivity and specificity regarding inter-regional connectivity as well as tract morphology [1,2]. This directly impacts processing and analysis steps, such as cortical connectivity analysis, that build upon the tractography result. Initial studies have successfully shown the potential of machine learning techniques in the context of DW-MRI analysis, e.g. for the tasks of image quality transfer and tissue micro-structure analysis [3] and to estimate the number of distinct fiber clusters per voxel [4]. In this work we present a purely data-driven and thus fundamentally new approach to reconstruct fiber pathways by directly processing raw signal intensities using machine learning methods. In contrast
Corresponding author.
c Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part I, LNCS 9349, pp. 45–52, 2015. DOI: 10.1007/978-3-319-24553-9_6
46
P.F. Neher et al.
to other current tractography pipelines that incorporate several modelling assumptions about the nature of the underlying diffusion-weighted signal, this model-free approach has several advantages: 1. No simplifying and possibly inaccurate assumptions about the diffusion propagator are made (e.g. Gaussianity). Subsequently, the subtleties of the signal are not blurred by an abstracting modeling approach. 2. Artifacts are directly learned from data. Simplified noise models that are inadequate for modern coil configurations and acquisition methods become obsolete (e.g. Ricianity). 3. Tissue probabilities are learned from the data. M
Data Loading...
