This repository contains code that emulates the DeepBundle [1] framework, visualizes its features, and explores the addition of an SVM on top of the network and false-positive mining to improve performance.

A publicly available dataset [2] containing 105 subjects (72 bundles each) from the Human Connectome Project (HCP) [3] is used. Python 3 with Tensorflow 1.15.0 are the primary requirements, the req.txt file contains a listing of other dependencies. To install all the requirements, run the following:

$ pip install -r req.txt

General experiment settings are defined in the params.py file and model parameters in main.py, the code can be run using:

$ python main.py

[1] F. Liu, J. Feng, G. Chen, Y. Wu, Y. Hong, P. T. Yap, and D. Shen, DeepBundle: Fiber Bundle Parcellation with Graph Convolution NeuralNetworks, in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11849 LNCS, 2019, pp. 88–95

[2] J. Wasserthal, P. F. Neher, and K. H. Maier-Hein, Tract orientation mapping for bundle-specific tractography, in Lecture Notes in ComputerScience (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11072 LNCS.Springer Verlag,9 2018, pp. 36–44.

[3] D. C. Van Essen, S. M. Smith, D. M. Barch, T. E. Behrens, E. Yacoub, and K. Ugurbil, The WU-Minn Human Connectome Project: An overview, NeuroImage, vol. 80, pp. 62–79, 10 2013