Introduction

Reconstructing and visualizing in vivo White Matter (WM) fibers is a challenging issue in the investigation of brain. The knowledge of these fibers is useful to understand and predict the effects of some neurodegenerative pathologies, like multiple sclerosis. The most accurate method to perform this task is tractography, which is based on the analysis of the main diffusion directions of the water molecules estimated by Diffusion Tensor Imaging (DTI). From an anatomic point of view, particular sets of fibers (called fiber-bundles) represent different WM structures. In order to analyze WM structures, it is crucial to isolate subsets of fibers belonging to the WM regions into consideration. This task is often performed manually by expert neuroanatomists, who define inclusion and exclusion criteria in such a way as to delineate regions of interests and isolate specific WM fiber-bundles. However, this way of proceeding is time consuming and cannot be applied on large cohorts of subjects where an automated or, at least, semi-automated approach is in order. To overcome these limitations, different automated algorithms aimed to isolate and extract WM fiber-bundles have been proposed in the literature. They can be grouped in two categories, namely: (i) atlas-based algorithms, which require an a priori knowledge about the location of certain WM brain regions, and (ii) “unsupervised” algorithms, which do not require this a priori knowledge.

Our Research

In our research, we aim at providing a contribution in the setting of WM fiber-bundles analysis by proposing a new automated approach, which tries to retain the main pros and to overcome the main cons of the two families of automated approaches mentioned in the Introduction. The core “ingredients” (and the main contributions) of our research are:

1. a new string-based formalism allowing an alternative representation of WM fibers,

2. a new string dissimilarity metric,

3. a WM fiber clustering algorithm, and

4. a model-based algorithm to characterize WM fibers.

The proposed string-based techniques for representing and comparing WM fibers take both spatial and orientation information about involved fibers into account. Furthermore, to perform fiber comparison, they do not require fiber registration. This last feature, which is an important “fixed point” of our approaches, leads to a relevant simplification of the fiber clustering task, which (we recall) represents an important step in such analysis.

Our developed approaches [1–5] overcome different limitations of the related ones proposed in the past. For instance, it allows a better integration of a priori information provided by a neuroanatomist. Indeed, the usage of a string-based model, representing the shape of a particular fiber-bundle, allows an easy extraction of just those fibers having the same structure as the provided model.

It is also important to point out that key ideas of our approach can be applied in many other contexts, even very far from the biomedical ones.

Representing fibers as strings

The main purpose of this concept is to represent a three-dimensional fiber in a different format, more compatible with clustering and other different analyses. In the past, several ways for representing a three-dimensional line have been proposed. These different representations depend on both the context and the expected use. In our research, we choose to represent a fiber as a sequence of voxels (volumetric picture elements), representing, in their turn, values on a grid in a three-dimensional space. Then, we employ a minimum variance quantization algorithm in order to assign to each voxel a color w.r.t. its orientation. Finally, each (colored) fiber is mapped to a string according to an automatically derived alphabet.

Examples

Publications

[1] Cauteruccio, F., Stamile, C., Terracina, G., Ursino, D., & Sappey-Marinier, D. (2018). Integrating QuickBundles into a Model-Guided Approach for Extracting “Anatomically-Coherent” and “Symmetry-Aware” White Matter Fiber-Bundles. In Multidisciplinary Approaches to Neural Computing (pp. 39-46). Springer, Cham.
[2] Cauteruccio, F., Stamile, C., Terracina, G., Ursino, D., & Sappey-Marinier, D. (2016). An automated string-based approach to extracting and characterizing White Matter fiber-bundles. Computers in biology and medicine, 77, 64-75.
[3] Cauteruccio, F., Stamile, C., Terracina, G., Ursino, D., & Sappey-Marinier, D. (2016, July). Improving quickbundles to extract anatomically coherent white matter fiber-bundles. In International Conference on Image Analysis and Recognition (pp. 633-641). Springer, Cham.
[4] Stamile, C., Cauteruccio, F., Terracina, G., Ursino, D., Kocevar, G., & Sappey-Marinier, D. (2015, August). A model-guided string-based approach to white matter fiber-bundles extraction. In International Conference on Brain Informatics and Health (pp. 135-144). Springer, Cham.
[5] Cauteruccio, F., Stamile, C., Terracina, G., Ursino, D., & Sappey-Mariniery, D. (2015, July). An automated string-based approach to white matter fiber-bundles clustering. In 2015 International Joint Conference on Neural Networks (IJCNN) (pp. 1-8). IEEE.