Appearance Learning for Image-Based Motion Estimation in Tomography

Abstract

In tomographic imaging, anatomical structures are reconstructed by applying a pseudo-inverse forward model to acquired signals. Geometric information within this process is usually depending on the system setting only, i. e., the scanner position or readout direction. Patient motion therefore corrupts the geometry alignment in the reconstruction process resulting in motion artifacts. We propose an appearance learning approach recognizing the structures of rigid motion independently from the scanned object. To this end, we train a siamese triplet network to predict the reprojection error (RPE) for the complete acquisition as well as an approximate distribution of the RPE along the single views from the reconstructed volume in a multi-task learning approach. The RPE measures the motioninduced geometric deviations independent of the object based on virtual marker positions, which are available during training. We train our network using 27 patients and deploy a 21-4-2 split for training, validation and testing. In average, we achieve a residual mean RPE of 0.013 mm with an inter-patient standard deviation of 0.022 mm. This is twice the accuracy compared to previously published results. In a motion estimation benchmark the proposed approach achieves superior results in comparison with two state-of-the-art measures in nine out of twelve experiments. The clinical applicability of the proposed method is demonstrated on a motion-affected clinical dataset.

2-D Respiration Navigation Framework for 3-D Continuous Cardiac Magnetic Resonance Imaging

“…3) and ( 3) with our proposed 2-D navigation (2-D Nav ). For the quantitative evaluation, three commonly used image quality metrics in motion compensation [9] were applied (Tab. 1): (1) Histogram entropy H, (2) total variation T V and (3) wavelet-based estimation of the standard deviation of Gaussian noise distribution σ Noise [10].…”

Motion correction for separate mandibular and cranial movements in cone beam CT reconstructions

“…Cases are also observed in which only the mandible is moved against the remaining (resting) skull 24 . Techniques, which are applicable in scenarios more similar to ours, include methods of autofocus, 25–28 consistency conditions 29–31 and learning based approaches 32–34 . However, methods using consistency conditions are currently not able to correct separate cranial and mandibular motions, which is the goal of our work.…”

“…24 Techniques, which are applicable in scenarios more similar to ours, include methods of autofocus, [25][26][27][28] consistency conditions [29][30][31] and learning based approaches. [32][33][34] However, methods using consistency conditions are currently not able to correct separate cranial and mandibular motions, which is the goal of our work. Approaches based on an autofocus metric are typically able to compensate non-rigid transformations, but usually incorporate temporal regularization of some kind to reduce the motion parameter space, hindering their application in cases of inconsistent or sudden motions of the patient.Motion correction methods based on deep learning are another lively field of research and will play an important role in the future.…”

Gradient-based geometry learning for fan-beam CT reconstruction

“…One class of motion compensation approaches defines a quality metric (QM) on the reconstructed image, often referred to as autofocus criterion. This QM is minimized with respect to the geometry parameters to find an acquisition geometry which annihilates the patient motion and maximizes the quality of the reconstructed image (Kingston et al 2011, Sisniega et al 2017, Preuhs et al 2020, Capostagno et al 2021, Huang et al 2022. While the exact formulation of the QM and the parameterization of the motion may vary, all of these approaches commonly rely on gradient-free optimization.…”