White matter fiber classification

The aim of this project was to build a classifier able to classify white matter fibers into anatomically meaningful white matter tracts of the human brain.

Obtaining accurately segmented white matter tracts in the human brain is essential for multiple applications, such as pre-surgical planning and tractometry. Although notable improvements have occurred over the years, the segmentation quality is not yet satisfactory, especially when dealing with datasets with diverse characteristics, such as different tract properties, tracking methods, or data quality. To overcome these limitations, we proposed a novel supervised segmentation method, called Classifyber, which combines in a simple linear model information from the geometry of the fiber paths, from their connectivity patterns, and from anatomy.

Classifyber is a supervised method that performs automatic tract segmentation by learning from example tracts segmented by experts. In particular, Classifyber provides a linear classifier that accurately predicts whether or not a given streamline (i.e. fiber) belongs to the tract of interest. In order to create the linear classifier, first we transform each streamline into a vector that contains both its geometrical and anatomical information. Then, we train a linear classifier, specifically Logistic Regression, with such vectors from multiple participants in which experts segmented the tracts of interest.

We ran multiple experiments over four different datasets, and compared the performances of our method, Classifyber, with other three state-of-the-art- tract segmentation methods through the Dice Similarity Coefficient (DSC) score (the higher the better). Classifyber outperformed the other methods in all cases, and segmented the tracts very accurately. This occured across different kinds of tracts, tractography techniques, expert-made segmentations, and data quality. Classifyber is freely available as an open source web app through the platorm brainlife.io.

Read the full paper here: https://pubmed.ncbi.nlm.nih.gov/32979520/
Read the poster here: classifyber-poster.pdf
Try the brainlife App here: https://doi.org/10.25663/brainlife.app.265
Code: app-classifyber-segmentation

Citation: Bertò G, Bullock D, Astolfi P, Hayashi S, Zigiotto L, Annicchiarico L, Corsini F, De Benedictis A, Sarubbo S, Pestilli F, Avesani P, Olivetti E. Classifyber, a robust streamline-based linear classifier for white matter bundle segmentation. Neuroimage. 2021 Jan 1;224:117402. doi: 10.1016/j.neuroimage.2020.117402. Epub 2020 Sep 23. PMID: 32979520.