Philips J. Res. 51 (1998) 197-201 FOREWORD TO THE SPECIAL ISSUE ON MOTION DETECTION AND COMPENSATION This special issue of Philips Journalof Research includes a number of papers presented at a Philips internal workshop on Motion Detection and Motion Compensation*. The meeting brought together Philips scientists of different disciplines working in the field of motion analysis for image processing applications. The interest in motion detection based on medical and technical image scenes from different image sensors is continuously growing. This is mainly due to the fact that an increasing number of faster imaging techniques allow for real-time applications where motion plays an increasingly more important role. In the medical segment, such applications can be found in fluoroscopic image sequences for digital subtraction angiographies (DSA), portal images from radiotherapy, and magnetic resonance imaging (MRI) as well as computer tomography (CT) image series. In the consumer segment, a large number of algorithms based on motion detection lead to improved television and video applications. In the professional segment, many computer vision and robot vision applications are inconceivable without motion detection. Figure 1 gives an overview of the application fields. The bold-printed number in some of the boxes refers to the respective paper in this issue. An example from the medical segment, which was the main focus of the workshop, is DSA, a standard diagnosis tool for the examination of blood vessels. With this method, X-ray images are taken from the patient, while a radio-opaque contrast agent is injected through a catheter. The first image is usually taken before injection, and is called the mask image. In the case of patient motion between mask and contrasted images, artifacts appear in the subtraction result. These artifacts can be significantly reduced by motion detection and subsequent compensation which then leads to an improved diagnosis. Another medical example is given by radiotherapy. Linear accelerator devices produce images that are called portal images. These portal images can be used for motion tracking. Tracking of patient motion during the radiotherapy treatment is very important for automatic accelerator control because the device must be immediately switched off when the patient leaves the optimal position. * Held at Philips Research Laboratories Hamburg, Germany in June 1997. Philips Journal of Research Vol. SI No.2 1998 197
Fig. I. Overview of the application fields of motion detection. The bold-printed numbers refer to the numbered papers in this issue. If the patient is at rest, motion detection can also be used to track surgical instruments such as catheters, and biopsy needles within minimally invasive procedures which are monitored via CT, M RI or X-ray fluoroscopy devices. Since pre-operatively acquired images are necessary for navigation during such procedures, these images must be related to the patient's coordinate system in the operating room or to an intra-operatively acquired image. In the case of patient movements during treatment (e.g. due to heart beat and breathing), patient movement must be tracked, and the registration must be updated. In real-time applications a compromise is to be found between signal-tonoise ratio (SNR) and temporal resolution. In recent years, for instance, M RI has shown that real-time applications are possible. However, the main drawback of the required fast pulse sequences is the low SNR of the MR-images. In such situations the use of a radial MR-acquisition scheme 198 Philips Journalof Research Vol. SI NO.2 1998
can support the concept ofhierarchical motion estimation in image sequences, which can be utilized for the Correlative Averaging noise-reduction strategy. In addition to the real-time aspect, dose reduction is a general goal in X-ray based medical imaging (fluoroscopy and CT). One example is real-time fluoroscopic visual guidance in diagnostic examinations and therapeutic interventions. The resulting moving images are viewed on-line on a monitor while the clinical procedure is performed. Patient and medical staff are therefore exposed to more radiation than in classical single-shot radiographic imaging. In these situations, low-dose rates are desired, and this becomes possible with the use of motion-compensated filtering. In the technical area, motion detection is also used in many application fields, particularly in television and video systems. Prominent examples of the consumer segment are picture rate conversion, coding and image compression, as well as noise reduction, In the professional computer- and robot vision segment motion tracking finds a huge variety of applications. Flow visualization and analysis (automatic extraction of velocity fields to study turbulence via particle tracking), matching of satellite image sequences, object-tracking in street scenes, and remote velocity measurement to control autonomous vehicles and robots are only a few examples that underline the importance of motion detection. The methods for motion detection are surprisingly similar for all the abovementioned applications. Generally, motion is associated with changes in the images. Hence, corresponding features ofthe object in motion have to be identified in the image sequence. Such features can be external artificial markers attached to the object, or intrinsic object-related image features like edges and corners. However, the most frequently used feature is the gray-value of the object itself, because motion-induced gray-value changes can be quantified in terms of similarity measures. Nevertheless, the similarity measure has to be carefully adapted to image-sensor characteristics to yield acceptable results. The first paper of this special issue, 'Motion Detection and Motion Compensation for Digital Subtraction Angiography Image Enhancement' by Buzug, Weese and Strasters (paper 1), focuses on artifact suppression in subtraction imaging. Special emphasis is given to the analysis of similarity measures that are used for motion detection. This is done because image pairs in digital subtraction angiography are inherently dissimilar and an adequate similarity measure must be adapted to that situation. In the second paper, 'Bayesian Motion Estimation for Temporally Recursive Noise Reduction in X-Ray Fluoroscopy' by Aach and Kunz (paper 2), the authors focus on a robust motion estimator which is able to deal with the high noise levels Philips Journal of Research Vol. SI No.2 1998 199
encountered in low-dose fluoroscopy images. This robustness against noise is achieved by spatial and temporal regularization. In the paper 'Layered Motion Estimation' by Schutten, Pelagotti and de Haan (paper 3), a region based motion-parameter estimation and segmentation algorithm is proposed that permits quasi-simultaneous motion estimation up to a fixed maximum number ofmoving layers within a sequence. The concept yields a very simple approach to the total motion estimation problem, and results in a very low operationsper-pixel-count. A variant of the presented algorithm is scheduled to run in real-time on a Philips TMIOOOprocessor. The paper 'Correlative Averaging for Magnetic Resonance Imaging' by Schäffter, Carlsen and Rasche (paper 4) deals with a hierarchical block matching procedure for motion detection in Magnetic Resonance Imaging (MRI). The determined displacement fields can be used in a Correlative Averaging approach to increase the SNR and in the reconstruction process to reduce motion artifacts. In the paper 'A Projection-Based Method for Motion-Compensated Noise Suppression' by GraB, Rasche, Lüdeke, Pro ksa and Schäffter (paper 5), a projection-based method for the motion-compensated SNR enhancement in radial MR imaging is presented. Since motion of an object is represented in the radial projections used for the reconstruction, a motion-compensated noise suppression method can already be applied to the measured projections. The next two papers deal with motion detection or registration of pretreatment with intra-treatment images. The paper '2D/3D Registration and Motion Tracking for Surgical Interventions' by Weese, Buzug, Penney and Desmedt (paper 6) focuses on the problem of registration of intra-operatively acquired X-ray fluoroscopies with 3D CT images obtained before the intervention. The result can be used to support the placement of pedicle screws in spine surgery as well as to guide the installment of stents in Transfemoral Endovascular Aneurysm Management (TEAM). The different approaches for 2D/3D registration are discussed and a novel voxel-based method is presented. The paper 'A Structure-Based Method for on-line Matching of Port al Images for an Optimal Patient Set-up in Radiotherapy' by Kreuder, Schreiber, Kausch and Dössel (paper 7) describes the registration problem in radiotherapy. Portal images are used to ensure a correct position of the patient relative to the beam during every radiation session. A reliable on-line verification is of clinical interest to interrupt the treatment in time if the patient is not at the desired position. The authors present a structure-based method which is almost insensitive to disturbances. Motion of the patient on the treatment table can then be detected fast and reliably. Another approach of motion artifact reduction is given in the paper 'Real- Time Reduction of Motion Artefacts Using k-space Weighting in Magnetic 200 Philips Journal of Research Vol. SI No.2 1998
Resonance Imaging' by Sinkus and Börnert (paper 8). Since respiratory motion in abdominal MRI gives rise to motion artifacts, the authors merge the benefits of the so-called Diminishing Variance Algorithm (DVA) concept with the ideas of Motion-Adapted Gating (MAG). In contrast to the other methods mentioned above, motion artifacts are reduced within the image acquisition process. I would like to acknowledge all the parties who contributed to the success of the workshop. Especially, I would like to thank Ursula Muuss for secretarial support and the service group of the laboratory. I would also like to thank all the reviewers from Philips Research Hamburg for their valuable help in the reviewing process. Last but not least, I would like to thank all our customers, whose kind support made the results reported in this special issue possible. e-mail: Thorsten M. Buzug Medical Image Processing T.Buzug@pjh.research.philips.com Philips Journalof Research Vol. SI No.2 1998 201