eduzhai > Applied Sciences > Engineering >

Model-Informed Machine Learning for Multi-component T2 Relaxometry

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 28 pages

Abstract: Recovering the T2 distribution from multi-echo T2 magnetic resonance (MR)signals is challenging but has high potential as it provides biomarkerscharacterizing the tissue micro-structure, such as the myelin water fraction(MWF). In this work, we propose to combine machine learning and aspects ofparametric (fitting from the MRI signal using biophysical models) andnon-parametric (model-free fitting of the T2 distribution from the signal)approaches to T2 relaxometry in brain tissue by using a multi-layer perceptron(MLP) for the distribution reconstruction. For training our network, weconstruct an extensive synthetic dataset derived from biophysical models inorder to constrain the outputs with textit{a priori} knowledge of textit{invivo} distributions. The proposed approach, called Model-Informed MachineLearning (MIML), takes as input the MR signal and directly outputs theassociated T2 distribution. We evaluate MIML in comparison to non-parametricand parametric approaches on synthetic data, an ex vivo scan, andhigh-resolution scans of healthy subjects and a subject with MultipleSclerosis. In synthetic data, MIML provides more accurate and noise-robustdistributions. In real data, MWF maps derived from MIML exhibit the greatestconformity to anatomical scans, have the highest correlation to a histologicalmap of myelin volume, and the best unambiguous lesion visualization andlocalization, with superior contrast between lesions and normal appearingtissue. In whole-brain analysis, MIML is 22 to 4980 times faster thannon-parametric and parametric methods, respectively.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×