Conference Agenda

Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).

Please note that all times are shown in the time zone of the conference. The current conference time is: 17th Sept 2021, 06:25:36am CEST

Session Overview
Oral Session - Motion Management
Tuesday, 20/July/2021:
10:00am - 11:40am

Session Chair: Roger Fulton
Session Chair: Simon Rit
Location: virtual (CEST)

External Resource:
  • Link to the prerecordings
  • Link to the recording of the live session
  • Show help for 'Increase or decrease the abstract text size'
    10:00am - 10:20am

    Collision avoidance trajectories for on-line trajectory optimization in C-arm CBCT

    Sepideh Hatamikia1,2, Ander Biguri3, Gernot Kronreif1, Tom Russ4, Joachim Kettenbach5, Wolfgan Birkfellner2

    1Austrian Center for Medical Innovation and Technology, Wiener Neustadt, Austria; 2Center for Medical Physics and Biomedical Engineering, Medical University of Vienna, Austria; 3Institute of Nuclear Medicine, University College London; 4Computer Assisted Clinical Medicine, Medical Faculty Mannheim, Heidelberg University, Germany; 5Institute of Diagnostic and Interventional Radiology and Nuclear Medicine, Landesklinikum Wiener Neustadt, Austria

    Kinematic constraints due to the additional medical equipment or patient size are common while acquiring C-arm cone beam computed tomography (CBCT). In a previous paper, we proposed a framework to develop patient-specific collision-free trajectories for the scenarios where circular CBCT is not possible. However, the proposed method required kinematic constraints to be known beforehand. As collisions are mainly unpredictable in the operation theater, a real-time trajectory optimization is of great clinical importance. In this study, we introduce a new search strategy which has the potential to optimize trajectories on-the-fly. We propose an optimization procedure which identifies trajectories with the highest information to reconstruct a volume of interest by means of maximizing an objective function; then a local search is consequently performed around the best selected candidates and better trajectory solutions are investigated. The overall time required for the whole optimization process was around three to four minutes using one GPU.

    10:20am - 10:40am

    Reference-Free, Learning-Based Image Similarity : Application to Motion Compensation in CBCT

    Heyuan Huang1, Jeffrey H. Siewerdsen1, Wojciech Zbijewski1, Clifford R. Weiss1, Tina Ehtiati2, Alejandro Sisniega1

    1Johns Hopkins University, United States of America; 2Siemens Healthineers, Germany

    Conventional CBCT autofocus metrics are agnostic to the realistic presentation of anatomical structures in the image and can yield unrealistic solutions. Image structural similarity metrics (e.g., visual information fidelity, VIF) combine measures of image quality and structural similarity to a reference image, offering a potentially ideal basis for autofocus metrics. However, matched motion-free reference images are usually not available. In this work, we propose a reference-free, learning-based image similarity metric obtained using a deep neural network (denoted DL-VIF). A convolutional neural network was trained on simulated motion-corrupted CBCT data to extract features associated with the structural components contributing to VIF. The DL-VIF showed a good correlation with conventional VIF. When DL-VIF was incorporated in an autofocus motion compensation framework, and its performance was compared against a conventional metric (gradient entropy). This work is an important step toward reliable motion compensation in scenarios of complex soft-tissue deformable motion in CBCT.

    10:40am - 11:00am

    DeepSLM: Image Registration Aware of Sliding Interfaces for Motion-Compensated Reconstruction - FULLY3D 2021 AWARD NOMINEE

    Markus Susenburger1, Pascal Paysan2, Ricky Savjani3, Joscha Maier1, Igor Peterlik2, Marc Kachelrie├č1

    1German Cancer Research Center, Germany; 2Varian Medical Systems, Switzerland; 3Varian Medical Systems, CA, USA

    A common deep convolutional neural network architecture for deformable image registration is adapted to fit the needs of anatomical consistency for motion estimation in motion-compensated reconstruction. We introduce a sliding interface motion constraint to decompose the motion into perpendicular and tangential components in the vicinity of organ interfaces. Three separation schemes are evaluated. The results for the proposed approach, referred to as DeepSLM, show comprehensive motion adaption at the lung border for unseen test data. During inference, no additional input is needed to enable sliding lung motion registration. DeepSLM improves the registration quality and is able to learn anatomical features of the sliding lung border. The network is the basis for future investigations in motion-compensated reconstruction with deep learning techniques.

    11:00am - 11:20am

    Prior-aided Volume of Interest CBCT Image Reconstruction - FULLY3D 2021 AWARD WINNER

    Daniel Punzet1, Robert Frysch1, Oliver Speck2, Georg Rose1

    1Institute for Medical Engineering and Research Campus STIMULATE, Otto-von-Guericke-University Magdeburg, Magdeburg, Germany; 2Department of Biomedical Magnetic Resonance and Research Campus STIMULATE, Otto-von-Guericke-University Magdeburg, Magdeburg, Germany

    Typical CBCT acquisition protocols require the user to acquire two orthogonal fluoroscopic images for the purpose of positioning the patient ideally for the following complete acquisition. Since these fluoroscopic positioning images are fairly low-dose, they are typically not truncated even for volume of interest imaging acquisitions, except in the case of a too small detector. Yet, they share the positioning of the CBCT acquisition succeeding them and can therefore provide additional information to the reconstruction from truncated projections.
    We present a prior-aided reconstruction scheme which registers these fluoroscopic images to potentially available priors of the same patient in order to enhance CBCT VOI imaging acquisitions and ultimately reduce patient dose. We make use of a novel registration method which moves the computationally expensive steps of the registration to a timepoint prior to the VOI acquisition and therefore allows for fast registration of priors for usage in interventional VOI imaging settings.

    11:20am - 11:40am

    Data-driven motion compensated SPECT reconstruction for liver radioembolization

    Antoine Robert1,2, Simon Rit2, Julien Jomier1, David Sarrut2

    1Kitware SAS, France; 2Univ.Lyon, INSA-Lyon, Universit\'e Claude Bernard Lyon 1, UJM-Saint Etienne, CNRS, Inserm, CREATIS UMR 5220, U1206, F-69373, Lyon, France.

    Breathing motion is a major issue for quantitation as it leads to misestimation of the tumor activity in the SPECT images. We developed a data-driven motion compensated SPECT reconstruction to account for respiratory motion.

    A respiratory signal was extracted from SPECT list-mode data with the Laplacian Eigenmaps algorithm and used to sort the projections into temporal bins of fixed phase width. A 2D affine motion was then estimated between phases. The transformation parameters were used to re-bin the list-mode data into one set of compensated projections that was then used to reconstruct a 3D motion-compensated SPECT image using all available events of the list-mode data.

    The method was evaluated on simulated and real liver radioembolization SPECT acquisitions, and compared to a respiratory-gated reconstruction. The motion-compensated reconstruction allowed to retrieve larger activity in the tumors compared to a conventional 3D SPECT reconstruction and with a better contrast-to-noise ratio than gated reconstruction.

    Contact and Legal Notice · Contact Address:
    Privacy Statement · Conference: Fully3D 2021
    Conference Software - ConfTool Pro 2.6.141
    © 2001 - 2021 by Dr. H. Weinreich, Hamburg, Germany