G Sharp

ICP registration using invariant features

Investigates the use of Euclidean invariant features in a generalization of iterative closest point (ICP) registration of range images. Pointwise correspondences are chosen as the closest point with respect to a weighted linear combination of positional and feature distances. It is shown that, under ideal noise-free conditions, correspondences formed using this distance function are correct more often than correspondences formed using the positional distance alone. In addition, monotonic convergence to at least a local minimum is shown to hold for this method. When noise is present, a method that automatically sets the optimal relative contribution of features and positions is described. This method trades off the error in feature values due to noise against the error in positions due to misalignment. Experimental results suggest that using invariant features decreases the probability of being trapped in a local minimum and may be an effective solution for difficult range image registration problems where the scene is very small compared to the model.

Prediction of respiratory tumour motion for real-time image-guided radiotherapy

Image guidance in radiotherapy and extracranial radiosurgery offers the potential for precise radiation dose delivery to a moving tumour. Recent work has demonstrated how to locate and track the position of a tumour in real-time using diagnostic x-ray imaging to find implanted radio-opaque markers. However, the delivery of a treatment plan through gating or beam tracking requires adequate consideration of treatment system latencies, including image acquisition, image processing, communication delays, control system processing, inductance within the motor, mechanical damping, etc. Furthermore, the imaging dose given over long radiosurgery procedures or multiple radiotherapy fractions may not be insignificant, which means that we must reduce the sampling rate of the imaging system. This study evaluates various predictive models for reducing tumour localization errors when a real-time tumour-tracking system targets a moving tumour at a slow imaging rate and with large system latencies. We consider 14 lung tumour cases where the peak-to-peak motion is greater than 8 mm, and compare the localization error using linear prediction, neural network prediction and Kalman filtering, against a system which uses no prediction. To evaluate prediction accuracy for use in beam tracking, we compute the root mean squared error between predicted and actual 3D motion. We found that by using prediction, root mean squared error is improved for all latencies and all imaging rates evaluated. To evaluate prediction accuracy for use in gated treatment, we present a new metric that compares a gating control signal based on predicted motion against the best possible gating control signal. We found that using prediction improves gated treatment accuracy for systems that have latencies of 200 ms or greater, and for systems that have imaging rates of 10 Hz or slower.

Evaluation of Registration Methods on Thoracic CT: The EMPIRE10 Challenge

EMPIRE10 (Evaluation of Methods for Pulmonary Image REgistration 2010) is a public platform for fair and meaningful comparison of registration algorithms which are applied to a database of intra patient thoracic CT image pairs. Evaluation of nonrigid registration techniques is a nontrivial task. This is compounded by the fact that researchers typically test only on their own data, which varies widely. For this reason, reliable assessment and comparison of different registration algorithms has been virtually impossible in the past. In this work we present the results of the launch phase of EMPIRE10, which comprised the comprehensive evaluation and comparison of 20 individual algorithms from leading academic and industrial research groups. All algorithms are applied to the same set of 30 thoracic CT pairs. Algorithm settings and parameters are chosen by researchers expert in the con figuration of their own method and the evaluation is independent, using the same criteria for all participants. All results are published on the EMPIRE10 website (http://empire10.isi.uu.nl). The challenge remains ongoing and open to new participants. Full results from 24 algorithms have been published at the time of writing. This paper details the organization of the challenge, the data and evaluation methods and the outcome of the initial launch with 20 algorithms. The gain in knowledge and future work are discussed.

GPU-based streaming architectures for fast cone-beam CT image reconstruction and demons deformable registration

This paper shows how to significantly accelerate cone-beam CT reconstruction and 3D deformable image registration using the stream-processing model. We describe data-parallel designs for the Feldkamp, Davis and Kress (FDK) reconstruction algorithm, and the demons deformable registration algorithm, suitable for use on a commodity graphics processing unit. The streaming versions of these algorithms are implemented using the Brook programming environment and executed on an NVidia 8800 GPU. Performance results using CT data of a preserved swine lung indicate that the GPU-based implementations of the FDK and demons algorithms achieve a substantial speedup--up to 80 times for FDK and 70 times for demons when compared to an optimized reference implementation on a 2.8 GHz Intel processor. In addition, the accuracy of the GPU-based implementations was found to be excellent. Compared with CPU-based implementations, the RMS differences were less than 0.1 Hounsfield unit for reconstruction and less than 0.1 mm for deformable registration.

Multiview registration of 3D scenes by minimizing error between coordinate frames

This paper addresses the problem of large-scale multiview registration of range images captured from unknown viewing directions. To reduce the computational burden, we separate the local problem of pairwise registration on neighboring views from the global problem of distribution of accumulated errors. We define the global problem as an optimization over the graph of neighboring views, and we show how the graph can be decomposed into a set of cycles such that the optimal transformation parameters for each cycle can be solved in closed form. We then describe an iterative procedure that can be used to integrate the solutions for the set of cycles across the graph of views. This method for error distribution does not require point correspondences between views, and can be used to integrate any method of pairwise registration or robot odometry.

4D-CT lung motion estimation with deformable registration: quantification of motion nonlinearity and hysteresis.

In this article, our goal is twofold. First, we propose and compare two methods which process deformable registration to estimate patient specific lung and tumor displacements and deformation during free breathing from four-dimensional computed tomography (4D-CT) data. Second, we propose techniques to quantify the physiological parameters of motion nonlinearity and hysteresis. A Fréchet distance-based criterion is introduced to measure the motion hysteresis. Experiments were conducted with 4D-CT data of five patients treated in radiotherapy for non-small cell lung cancer. The accuracy of deformation fields assessed against expert-selected landmarks was found to be within image voxel tolerance. The second method gave slightly better results in terms of accuracy and consistency, although the differences were not statistically significant between the two methods. Lung motion nonlinearity and hysteresis are patient specific, and vary across regions within the lung during respiration. For all patients, motion between end-exhale and end-inhale was well approximated with a straight line trajectory for the majority of lung points. Hysteresis was found to be globally correlated with trajectory length. The main limitation to the proposed method is that intensity-based deformable registration is dependent on image quality and image resolution. Another limitation is the absence of gold standard which makes the validation of the computed motion difficult. However, the proposed tools provide patient specific motion information which, in radiotherapy for example, can ease the definition of precise internal margins. In the future, the integration of physiological information from multiple patients could help to create a general lung atlas with different clinical applications.

Evaluation of deformable registration of patient lung 4DCT with subanatomical region segmentations.

Deformable registration is needed for a variety of tasks in establishing the voxel correspondence between respiratory phases. Most registration algorithms assume or imply that the deformation field is smooth and continuous everywhere. However, the lungs are contained within closed invaginated sacs called pleurae and are allowed to slide almost independently along the chest wall. This sliding motion is characterized by a discontinuous vector field, which cannot be generated using standard deformable registration methods. The authors have developed a registration method that can create discontinuous vector fields at the boundaries of anatomical subregions. Registration is performed independently on each subregion, with a boundary-matching penalty used to prevent gaps. This method was implemented and tested using both the B-spline and Demons registration algorithms in the Insight Segmentation and Registration Toolkit. The authors have validated this method on four patient 4DCT data sets for registration of the end-inhalation and end-exhalation volumes. Multiple experts identified homologous points in the lungs and along the ribs in the two respiratory phases. Statistical analyses of the mismatch of the homologous points before and after registration demonstrated improved overall accuracy for both algorithms.

On developing B-spline registration algorithms for multi-core processors.

Spline-based deformable registration methods are quite popular within the medical-imaging community due to their flexibility and robustness. However, they require a large amount of computing time to obtain adequate results. This paper makes two contributions towards accelerating B-spline-based registration. First, we propose a grid-alignment scheme and associated data structures that greatly reduce the complexity of the registration algorithm. Based on this grid-alignment scheme, we then develop highly data parallel designs for B-spline registration within the stream-processing model, suitable for implementation on multi-core processors such as graphics processing units (GPUs). Particular attention is focused on an optimal method for performing analytic gradient computations in a data parallel fashion. CPU and GPU versions are validated for execution time and registration quality. Performance results on large images show that our GPU algorithm achieves a speedup of 15 times over the single-threaded CPU implementation whereas our multi-core CPU algorithm achieves a speedup of 8 times over the single-threaded implementation. The CPU and GPU versions achieve near-identical registration quality in terms of RMS differences between the generated vector fields.

How much margin reduction is possible through gating or breath hold

We determined the relationship between intra-fractional breathing motion and safety margins, using daily real-time tumour tracking data of 40 patients (43 tumour locations), treated with radiosurgery at Hokkaido University. We limited our study to the dose-blurring effect of intra-fractional breathing motion, and did not consider differences in positioning accuracy or systematic errors. The additional shift in the prescribed isodose level (e.g. 95 %) was determined by convolving a one-dimensional dose profile, having a dose gradient representing an 8 MV beam through either lung or water, with the probability density function (PDF) of breathing. This additional shift is a measure for the additional margin that should be applied in order to maintain the same probability of tumour control as without intra-fractional breathing. We show that the required safety margin is a nonlinear function of the peak-to-peak breathing motion. Only a small reduction in the shift of isodose curves was observed for breathing motion up to 10 mm. For larger motion, 20 or 30 mm, control of patient breathing during irradiation, using either gating or breath hold, can allow a substantial reduction in safety margins of about 7 or 12 mm depending on the dose gradient prior to blurring. Clinically relevant random setup uncertainties, which also have a blurring effect on the dose distribution, have only a small effect on the margin needed for intra-fractional breathing motion. Because of the one-dimensional nature of our analysis, the resulting margins are mainly applicable in the superior-inferior direction. Most measured breathing PDFs were not consistent with the PDF of a simple parametric curve such as cos4, either because of irregular breathing or base-line shifts. Instead, our analysis shows that breathing motion can be modelled as Gaussian with a standard deviation of about 0.4 times the peak-to-peak breathing motion.

Multiple template-based fluoroscopic tracking of lung tumor mass without implanted fiducial markers

Precise lung tumor localization in real time is particularly important for some motion management techniques, such as respiratory gating or beam tracking with a dynamic multi-leaf collimator, due to the reduced clinical tumor volume (CTV) to planning target volume (PTV) margin and/or the escalated dose. There might be large uncertainties in deriving tumor position from external respiratory surrogates. While tracking implanted fiducial markers has sufficient accuracy, this procedure may not be widely accepted due to the risk of pneumothorax. Previously, we have developed a technique to generate gating signals from fluoroscopic images without implanted fiducial markers using a template matching method (Berbeco et al 2005 Phys. Med. Biol. 50 4481-90, Cui et al 2007 Phys. Med. Biol. 52 741-55). In this paper, we present an extension of this method to multiple-template matching for directly tracking the lung tumor mass in fluoroscopy video. The basic idea is as follows: (i) during the patient setup session, a pair of orthogonal fluoroscopic image sequences are taken and processed off-line to generate a set of reference templates that correspond to different breathing phases and tumor positions; (ii) during treatment delivery, fluoroscopic images are continuously acquired and processed; (iii) the similarity between each reference template and the processed incoming image is calculated; (iv) the tumor position in the incoming image is then estimated by combining the tumor centroid coordinates in reference templates with proper weights based on the measured similarities. With different handling of image processing and similarity calculation, two such multiple-template tracking techniques have been developed: one based on motion-enhanced templates and Pearsons correlation score while the other based on eigen templates and mean-squared error. The developed techniques have been tested on six sequences of fluoroscopic images from six lung cancer patients against the reference tumor positions manually determined by a radiation oncologist. The tumor centroid coordinates automatically detected using both methods agree well with the manually marked reference locations. The eigenspace tracking method performs slightly better than the motion-enhanced method, with average localization errors less than 2 pixels (1 mm) and the error at a 95% confidence level of about 2-4 pixels (1-2 mm). This work demonstrates the feasibility of direct tracking of a lung tumor mass in fluoroscopic images without implanted fiducial markers using multiple reference templates.