eduzhai > Applied Sciences > Engineering >

Appearance Learning for Image-based Motion Estimation in Tomography

  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: In tomographic imaging, anatomical structures are reconstructed by applying apseudo-inverse forward model to acquired signals. Geometric information withinthis process is usually depending on the system setting only, i. e., thescanner position or readout direction. Patient motion therefore corrupts thegeometry alignment in the reconstruction process resulting in motion artifacts.We propose an appearance learning approach recognizing the structures of rigidmotion independently from the scanned object. To this end, we train a siamesetriplet network to predict the reprojection error (RPE) for the completeacquisition as well as an approximate distribution of the RPE along the singleviews from the reconstructed volume in a multi-task learning approach. The RPEmeasures the motioninduced geometric deviations independent of the object basedon virtual marker positions, which are available during training. We train ournetwork using 27 patients and deploy a 21-4-2 split for training, validationand testing. In average, we achieve a residual mean RPE of 0.013mm with aninter-patient standard deviation of 0.022 mm. This is twice the accuracycompared to previously published results. In a motion estimation benchmark theproposed approach achieves superior results in comparison with twostate-of-the-art measures in nine out of twelve experiments. The clinicalapplicability of the proposed method is demonstrated on a motion-affectedclinical dataset.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...