Department of Computer Science, University of Verona, Verona, Italy; Sherbrooke Connectivity Imaging Laboratory (SCIL), Departement dInformatique, Universite de Sherbrooke, Sherbrooke, Quebec, Canada.
Department of Computer Science, University of Verona, Verona, Italy; Department of Neuroscience, Rehabilitation, Ophthalmology, Genetics, Maternal and Child Health (DINOGMI), University of Genoa, Genova, Italy.
Neuroimage. 2022 Nov;263:119600. doi: 10.1016/j.neuroimage.2022.119600. Epub 2022 Sep 12.
Tractography is a powerful tool for the investigation of the complex organization of the brain in vivo, as it allows inferring the macroscopic pathways of the major fiber bundles of the white matter based on non-invasive diffusion-weighted magnetic resonance imaging acquisitions. Despite this unique and compelling ability, some studies have exposed the poor anatomical accuracy of the reconstructions obtained with this technique and challenged its effectiveness for studying brain connectivity. In this work, we describe a novel method to readdress tractography reconstruction problem in a global manner by combining the strengths of so-called generative and discriminative strategies. Starting from an input tractogram, we parameterize the connections between brain regions following a bundle-based representation that allows to drastically reducing the number of parameters needed to model groups of fascicles. The parameters space is explored following an MCMC generative approach, while a discrimininative method is exploited to globally evaluate the set of connections which is updated according to Bayes' rule. Our results on both synthetic and real brain data show that the proposed solution, called bundle-o-graphy, allows improving the anatomical accuracy of the reconstructions while keeping the computational complexity similar to other state-of-the-art methods.
束描技术是一种强大的工具，可用于研究活体大脑的复杂结构，因为它可以根据非侵入性的扩散加权磁共振成像采集来推断白质中主要纤维束的宏观路径。尽管具有这种独特而引人注目的能力，但一些研究已经暴露了该技术获得的重建的解剖精度较差，并对其研究大脑连接的有效性提出了挑战。在这项工作中，我们描述了一种通过结合所谓生成和判别策略的优势来全面解决束描重建问题的新方法。从输入的束描开始，我们根据基于束的表示来参数化大脑区域之间的连接，这允许大大减少建模束组所需的参数数量。参数空间遵循 MCMC 生成方法进行探索，而判别方法则用于全局评估连接集，该连接集根据贝叶斯规则进行更新。我们在合成和真实脑数据上的结果表明，所提出的解决方案称为束图技术，可提高重建的解剖精度，同时保持与其他最先进方法相似的计算复杂度。
