Fast 4D cone-beam CT from 60 s acquisitions.

BACKGROUND & PURPOSE: Four dimensional Cone beam CT (CBCT) has many potential benefits for radiotherapy but suffers from poor image quality, long acquisition times, and/or long reconstruction times. In this work we present a fast iterative reconstruction algorithm for 4D reconstruction of fast acquisition cone beam CT, as well as a new method for temporal regularization and compare to state of the art methods for 4D CBCT. MATERIALS & METHODS: Regularization parameters for the iterative algorithms were found automatically via computer optimization on 60 s acquisitions using the XCAT phantom. Nineteen lung cancer patients were scanned with 60 s arcs using the onboard image on a Varian trilogy linear accelerator. Images were reconstructed using an accelerated ordered subset algorithm. A frequency based temporal regularization algorithm was developed and compared to the McKinnon-Bates algorithm, 4D total variation and prior images compressed sensing (PICCS). RESULTS: All reconstructions were completed in 60 s or less. The proposed method provided a structural similarity of 0.915, compared with 0.786 for the classic McKinnon-bates method. For the patient study, it provided fewer image artefacts than PICCS, and better spatial resolution than 4D TV. CONCLUSION: Four dimensional iterative CBCT reconstruction was done in less than 60 s, demonstrating the clinical feasibility. The frequency based method outperformed 4D total variation and PICCS on the simulated data, and for patients allowed for tumor location based on 60 s acquisitions, even for slowly breathing patients. It should thus be suitable for routine clinical use.

Overcoming foetal motion using interactive real-time magnetic resonance imaging.

OBJECTIVE: Foetal MRI has become an established image modality in the prenatal diagnosis of CNS anomalies, but image quality can be severely affected by foetal movements. The objective was to overcome these inherent motion problems by applying interactive real-time MRI and to evaluate the diagnostic usefulness of the applied real-time MRI sequence in relation to standard protocols. METHODS: Ten healthy foetuses (gestation week 21·3 ± 0·5) were scanned using a system, which allowed visual feedback and interactive slice positioning in real time. The data were compared to a control group of 14 healthy foetuses (gestation week 21·0 ± 0·8) who had previously been scanned using standard MRI. Comparisons were carried out by two radiologists with regard to cerebral anthropometric sizes, presence of important brain structures, degree of movement, clinical image value, image quality and ability to obtain correct slice planes. RESULTS: Two out of eight anthropometric sizes were statistically different between the two groups. Representation of cerebral structures was found in 70-100% in the real-time group. No statistically differences were found in clinical image value and image quality. The mean ability to obtain optimal slice planes was higher in the real-time group, but it was not significant. CONCLUSION: Imaging of the foetal brain using the proposed interactive real-time MRI system is a promising alternative to traditional foetal MRI for anthropometrics or as a supplement for the representation of foetal brain structures in cases in which foetal motion causes challenges in relation to obtaining optimal slice planes using conventional MRI techniques.

A simulation study on proton computed tomography (CT) stopping power accuracy using dual energy CT scans as benchmark.

BACKGROUND: Accurate stopping power estimation is crucial for treatment planning in proton therapy, and the uncertainties in stopping power are currently the largest contributor to the employed dose margins. Dual energy x-ray computed tomography (CT) (clinically available) and proton CT (in development) have both been proposed as methods for obtaining patient stopping power maps. The purpose of this work was to assess the accuracy of proton CT using dual energy CT scans of phantoms to establish reference accuracy levels. MATERIAL AND METHODS: A CT calibration phantom and an abdomen cross section phantom containing inserts were scanned with dual energy and single energy CT with a state-of-the-art dual energy CT scanner. Proton CT scans were simulated using Monte Carlo methods. The simulations followed the setup used in current prototype proton CT scanners and included realistic modeling of detectors and the corresponding noise characteristics. Stopping power maps were calculated for all three scans, and compared with the ground truth stopping power from the phantoms. RESULTS: Proton CT gave slightly better stopping power estimates than the dual energy CT method, with root mean square errors of 0.2% and 0.5% (for each phantom) compared to 0.5% and 0.9%. Single energy CT root mean square errors were 2.7% and 1.6%. Maximal errors for proton, dual energy and single energy CT were 0.51%, 1.7% and 7.4%, respectively. CONCLUSION: Better stopping power estimates could significantly reduce the range errors in proton therapy, but requires a large improvement in current methods which may be achievable with proton CT.

Distributed MRI reconstruction using Gadgetron-based cloud computing.

PURPOSE: To expand the open source Gadgetron reconstruction framework to support distributed computing and to demonstrate that a multinode version of the Gadgetron can be used to provide nonlinear reconstruction with clinically acceptable latency. METHODS: The Gadgetron framework was extended with new software components that enable an arbitrary number of Gadgetron instances to collaborate on a reconstruction task. This cloud-enabled version of the Gadgetron was deployed on three different distributed computing platforms ranging from a heterogeneous collection of commodity computers to the commercial Amazon Elastic Compute Cloud. The Gadgetron cloud was used to provide nonlinear, compressed sensing reconstruction on a clinical scanner with low reconstruction latency (eg, cardiac and neuroimaging applications). RESULTS: The proposed setup was able to handle acquisition and 11 -SPIRiT reconstruction of nine high temporal resolution real-time, cardiac short axis cine acquisitions, covering the ventricles for functional evaluation, in under 1 min. A three-dimensional high-resolution brain acquisition with 1 mm(3) isotropic pixel size was acquired and reconstructed with nonlinear reconstruction in less than 5 min. CONCLUSION: A distributed computing enabled Gadgetron provides a scalable way to improve reconstruction performance using commodity cluster computing. Nonlinear, compressed sensing reconstruction can be deployed clinically with low image reconstruction latency.

The image quality of ion computed tomography at clinical imaging dose levels.

PURPOSE: Accurately predicting the range of radiotherapy ions in vivo is important for the precise delivery of dose in particle therapy. Range uncertainty is currently the single largest contribution to the dose margins used in planning and leads to a higher dose to normal tissue. The use of ion CT has been proposed as a method to improve the range uncertainty and thereby reduce dose to normal tissue of the patient. A wide variety of ions have been proposed and studied for this purpose, but no studies evaluate the image quality obtained with different ions in a consistent manner. However, imaging doses ion CT is a concern which may limit the obtainable image quality. In addition, the imaging doses reported have not been directly comparable with x-ray CT doses due to the different biological impacts of ion radiation. The purpose of this work is to develop a robust methodology for comparing the image quality of ion CT with respect to particle therapy, taking into account different reconstruction methods and ion species. METHODS: A comparison of different ions and energies was made. Ion CT projections were simulated for five different scenarios: Protons at 230 and 330 MeV, helium ions at 230 MeV/u, and carbon ions at 430 MeV/u. Maps of the water equivalent stopping power were reconstructed using a weighted least squares method. The dose was evaluated via a quality factor weighted CT dose index called the CT dose equivalent index (CTDEI). Spatial resolution was measured by the modulation transfer function. This was done by a noise-robust fit to the edge spread function. Second, the image quality as a function of the number of scanning angles was evaluated for protons at 230 MeV. In the resolution study, the CTDEI was fixed to 10 mSv, similar to a typical x-ray CT scan. Finally, scans at a range of CTDEI's were done, to evaluate dose influence on reconstruction error. RESULTS: All ions yielded accurate stopping power estimates, none of which were statistically different from the ground truth image. Resolution (as defined by the modulation transfer function = 10% point) was the best for the helium ions (18.21 line pairs/cm) and worst for the lower energy protons (9.37 line pairs/cm). The weighted quality factor for the different ions ranged from 1.23 for helium to 2.35 for carbon ions. For the angle study, a sharp increase in absolute error was observed below 45 distinct angles, giving the impression of a threshold, rather than smooth, limit to the number of angles. CONCLUSIONS: The method presented for comparing various ion CT modalities is feasible for practical use. While all studied ions would improve upon x-ray CT for particle range estimation, helium appears to give the best results and deserves further study for imaging.

Three-dimensional liver motion tracking using real-time two-dimensional MRI.

PURPOSE: Combined magnetic resonance imaging (MRI) systems and linear accelerators for radiotherapy (MR-Linacs) are currently under development. MRI is noninvasive and nonionizing and can produce images with high soft tissue contrast. However, new tracking methods are required to obtain fast real-time spatial target localization. This study develops and evaluates a method for tracking three-dimensional (3D) respiratory liver motion in two-dimensional (2D) real-time MRI image series with high temporal and spatial resolution. METHODS: The proposed method for 3D tracking in 2D real-time MRI series has three steps: (1) Recording of a 3D MRI scan and selection of a blood vessel (or tumor) structure to be tracked in subsequent 2D MRI series. (2) Generation of a library of 2D image templates oriented parallel to the 2D MRI image series by reslicing and resampling the 3D MRI scan. (3) 3D tracking of the selected structure in each real-time 2D image by finding the template and template position that yield the highest normalized cross correlation coefficient with the image. Since the tracked structure has a known 3D position relative to each template, the selection and 2D localization of a specific template translates into quantification of both the through-plane and in-plane position of the structure. As a proof of principle, 3D tracking of liver blood vessel structures was performed in five healthy volunteers in two 5.4 Hz axial, sagittal, and coronal real-time 2D MRI series of 30 s duration. In each 2D MRI series, the 3D localization was carried out twice, using nonoverlapping template libraries, which resulted in a total of 12 estimated 3D trajectories per volunteer. Validation tests carried out to support the tracking algorithm included quantification of the breathing induced 3D liver motion and liver motion directionality for the volunteers, and comparison of 2D MRI estimated positions of a structure in a watermelon with the actual positions. RESULTS: Axial, sagittal, and coronal 2D MRI series yielded 3D respiratory motion curves for all volunteers. The motion directionality and amplitude were very similar when measured directly as in-plane motion or estimated indirectly as through-plane motion. The mean peak-to-peak breathing amplitude was 1.6 mm (left-right), 11.0 mm (craniocaudal), and 2.5 mm (anterior-posterior). The position of the watermelon structure was estimated in 2D MRI images with a root-mean-square error of 0.52 mm (in-plane) and 0.87 mm (through-plane). CONCLUSIONS: A method for 3D tracking in 2D MRI series was developed and demonstrated for liver tracking in volunteers. The method would allow real-time 3D localization with integrated MR-Linac systems.

Improved proton computed tomography by dual modality image reconstruction.

PURPOSE: Proton computed tomography (CT) is a promising image modality for improving the stopping power estimates and dose calculations for particle therapy. However, the finite range of about 33 cm of water of most commercial proton therapy systems limits the sites that can be scanned from a full 360° rotation. In this paper the authors propose a method to overcome the problem using a dual modality reconstruction (DMR) combining the proton data with a cone-beam x-ray prior. METHODS: A Catphan 600 phantom was scanned using a cone beam x-ray CT scanner. A digital replica of the phantom was created in the Monte Carlo code Geant4 and a 360° proton CT scan was simulated, storing the entrance and exit position and momentum vector of every proton. Proton CT images were reconstructed using a varying number of angles from the scan. The proton CT images were reconstructed using a constrained nonlinear conjugate gradient algorithm, minimizing total variation and the x-ray CT prior while remaining consistent with the proton projection data. The proton histories were reconstructed along curved cubic-spline paths. RESULTS: The spatial resolution of the cone beam CT prior was retained for the fully sampled case and the 90° interval case, with the MTF = 0.5 (modulation transfer function) ranging from 5.22 to 5.65 linepairs/cm. In the 45° interval case, the MTF = 0.5 dropped to 3.91 linepairs/cm For the fully sampled DMR, the maximal root mean square (RMS) error was 0.006 in units of relative stopping power. For the limited angle cases the maximal RMS error was 0.18, an almost five-fold improvement over the cone beam CT estimate. CONCLUSIONS: Dual modality reconstruction yields the high spatial resolution of cone beam x-ray CT while maintaining the improved stopping power estimation of proton CT. In the case of limited angles, the use of prior image proton CT greatly improves the resolution and stopping power estimate, but does not fully achieve the quality of a 360° proton CT scan.

Registration-based reconstruction of four-dimensional cone beam computed tomography.

We present a new method for reconstruction of 4-D cone beam computed tomography from an undersampled set of X-ray projections. The novelty of the proposed method lies in utilizing optical flow based registration to facilitate that each temporal phase is reconstructed from the full set of acquired projections. The reconstruction of each phase thus exhibits limited aliasing despite significant intra-phase undersampling. The method is fully self-contained. Initially an approximate 4-D volume is reconstructed and an inter-phase registration based hereon. A subsequent reconstruction pass integrates the optical flow estimation in a cost function formulation in which the X-ray projections from all temporal phases are considered for the reconstruction of each individual phase. Quantitative and qualitative evaluations were performed through reconstruction of both a numerical phantom and a clinical dataset. The obtained reconstructions are compared to the state-of-the-art alternatives of total variation regularization and prior image constrained compressed sensing. Our studies show that the proposed method is the better overall "compromise" in the depiction of both moving and stationary anatomical structures.

Simple DVH parameter addition as compared to deformable registration for bladder dose accumulation in cervix cancer brachytherapy.

BACKGROUND AND PURPOSE: Variations in organ position, shape, and volume cause uncertainties in dose assessment for brachytherapy (BT) in cervix cancer. The purpose of this study was to evaluate uncertainties associated with bladder dose accumulation based on DVH parameter addition (previously called "the worst case assumption") in fractionated BT. MATERIALS AND METHODS: Forty-seven patients treated for locally advanced cervical cancer were included. All patients received EBRT combined with two individually planned 3D image-guided adaptive BT fractions. D(2cm(3)) and D(0.1cm(3)) were estimated by DVH parameter addition and compared to dose accumulations based on an in-house developed biomechanical deformable image registration (DIR) algorithm. RESULTS: DIR-based DVH analysis was possible in 42/47 patients. DVH parameter addition resulted in mean dose deviations relative to DIR of 0.4±0.3 Gy(αß3) (1.5±1.8%) and 1.9±1.6 Gy(αß3) (5.2±4.2%) for D(2cm(3)) and D(0.1cm(3)), respectively. Dose deviations greater than 5% occurred in 2% and 38% of the patients for D(2cm(3)) and D(0.1cm(3)), respectively. Visual inspection of the dose distributions showed that hotspots were located in the same region of the bladder during both BT fractions for the majority of patients. CONCLUSION: DVH parameter addition provides a good estimate for D(2cm(3)), whereas D(0.1cm(3)) is less robust to this approximation.

Gadgetron: an open source framework for medical image reconstruction.

This work presents a new open source framework for medical image reconstruction called the "Gadgetron." The framework implements a flexible system for creating streaming data processing pipelines where data pass through a series of modules or "Gadgets" from raw data to reconstructed images. The data processing pipeline is configured dynamically at run-time based on an extensible markup language configuration description. The framework promotes reuse and sharing of reconstruction modules and new Gadgets can be added to the Gadgetron framework through a plugin-like architecture without recompiling the basic framework infrastructure. Gadgets are typically implemented in C/C++, but the framework includes wrapper Gadgets that allow the user to implement new modules in the Python scripting language for rapid prototyping. In addition to the streaming framework infrastructure, the Gadgetron comes with a set of dedicated toolboxes in shared libraries for medical image reconstruction. This includes generic toolboxes for data-parallel (e.g., GPU-based) execution of compute-intensive components. The basic framework architecture is independent of medical imaging modality, but this article focuses on its application to Cartesian and non-Cartesian parallel magnetic resonance imaging.

Online real-time reconstruction of adaptive TSENSE with commodity CPU/GPU hardware.

Adaptive temporal sensitivity encoding (TSENSE) has been suggested as a robust parallel imaging method suitable for MR guidance of interventional procedures. However, in practice, the reconstruction of adaptive TSENSE images obtained with large coil arrays leads to long reconstruction times and latencies and thus hampers its use for applications such as MR-guided thermotherapy or cardiovascular catheterization. Here, we demonstrate a real-time reconstruction pipeline for adaptive TSENSE with low image latencies and high frame rates on affordable commodity personal computer hardware. For typical image sizes used in interventional imaging (128 x 96, 16 channels, sensitivity encoding (SENSE) factor 2-4), the pipeline is able to reconstruct adaptive TSENSE images with image latencies below 90 ms at frame rates of up to 40 images/s, rendering the MR performance in practice limited by the constraints of the MR acquisition. Its performance is demonstrated by the online reconstruction of in vivo MR images for rapid temperature mapping of the kidney and for cardiac catheterization.

Four-dimensional (4D) flow of the whole heart and great vessels using real-time respiratory self-gating.

Four-dimensional (4D) flow imaging has been used to study flow patterns and pathophysiology, usually focused on specific thoracic vessels and cardiac chambers. Whole-heart 4D flow at high measurement accuracy covering the entire thoracic cardiovascular system would be desirable to simplify and improve hemodynamic assessment. This has been a challenge because compensation of respiratory motion is difficult to achieve, but it is paramount to limit artifacts and improve accuracy. In this work we propose a self-gating technique for respiratory motion-compensation integrated into a whole-heart 4D flow acquisition that overcomes these challenges. Flow components are measured in all three directions for each pixel over the complete cardiac cycle, and 1D volume projections are obtained at certain time intervals for respiratory gating in real time during the acquisition. The technique was tested in 15 volunteers, in which stroke volumes (SVs) in the great arteries showed excellent agreement with standard 2D phase-contrast (PC) scans. In contrast, nongated 4D flow with two averages had substantial disagreement with 2D flow. Applied to patients with congenital cardiac left-to-right shunting, the precision of flux data was highly beneficial. The methodology presented here has the potential to allow a complete study of flow pathophysiology of the thoracic cardiovascular system from a single free-breathing scan.

Real-time reconstruction of sensitivity encoded radial magnetic resonance imaging using a graphics processing unit.

A barrier to the adoption of non-Cartesian parallel magnetic resonance imaging for real-time applications has been the times required for the image reconstructions. These times have exceeded the underlying acquisition time thus preventing real-time display of the acquired images. We present a reconstruction algorithm for commodity graphics hardware (GPUs) to enable real time reconstruction of sensitivity encoded radial imaging (radial SENSE). We demonstrate that a radial profile order based on the golden ratio facilitates reconstruction from an arbitrary number of profiles. This allows the temporal resolution to be adjusted on the fly. A user adaptable regularization term is also included and, particularly for highly undersampled data, used to interactively improve the reconstruction quality. Each reconstruction is fully self-contained from the profile stream, i.e., the required coil sensitivity profiles, sampling density compensation weights, regularization terms, and noise estimates are computed in real-time from the acquisition data itself. The reconstruction implementation is verified using a steady state free precession (SSFP) pulse sequence and quantitatively evaluated. Three applications are demonstrated; real-time imaging with real-time SENSE 1) or k- t SENSE 2) reconstructions, and 3) offline reconstruction with interactive adjustment of reconstruction settings.

Developing and evaluating virtual cardiotomy for preoperative planning in congenital heart disease.

Careful preoperative planning is of outmost importance - in particular when considering complex corrective surgery on congenitally malformed hearts. As an aid to such decision-making we describe a system for virtual reconstruction of patient-specific morphology from 3D-capable imaging modalities such as MRI and CT. We introduce and illustrate the concept of virtual cardiotomy as a new tool to preoperatively evaluate the feasibility of different surgical strategies by investigating the anatomical spatial relations through any number of virtual incisions. We review the technical and clinical implementation of the various components of the system, namely 3D imaging, segmentation and reconstruction, visualization, and simulation of tissue elasticity. Finally we summarize the main findings from a recent evaluation study on 42 infants and children.

Three dimensional three component whole heart cardiovascular magnetic resonance velocity mapping: comparison of flow measurements from 3D and 2D acquisitions.

BACKGROUND: Two-dimensional, unidirectionally encoded, cardiovascular magnetic resonance (CMR) velocity mapping is an established technique for the quantification of blood flow in large vessels. However, it requires an operator to correctly align the planes of acquisition. If all three directional components of velocity are measured for each voxel of a 3D volume through the phases of the cardiac cycle, blood flow through any chosen plane can potentially be calculated retrospectively. The initial acquisition is then more time consuming but relatively operator independent. AIMS: To compare the curves and volumes of flow derived from conventional 2D and comprehensive 3D flow acquisitions in a steady state flow model, and in vivo through planes transecting the ascending aorta and pulmonary trunk in 10 healthy volunteers. METHODS: Using a 1.5 T Phillips Intera CMR system, 3D acquisitions used an anisotropic 3D segmented k-space phase contrast gradient echo sequence with a short EPI readout, with prospective ECG and diaphragm navigator gating. The 2D acquisitions used segmented k-space phase contrast with prospective ECG and diaphragm navigator gating. Quantitative flow analyses were performed retrospectively with dedicated software for both the in vivo and in vitro acquisitions. RESULTS: Analysis of in vitro data found the 3D technique to have overestimated the continuous flow rate by approximately 5% across the entire applied flow range. In vivo, the 2D and the 3D techniques yielded similar volumetric flow curves and measurements. Aortic flow: (mean +/- SD), 2D = 89.5 +/- 13.5 ml & 3D = 92.7 +/- 17.5 ml. Pulmonary flow: 2D = 98.8 +/- 18.4 ml & 3D = 94.9 +/- 19.0 ml). Each in vivo 3D acquisition took about 8 minutes or more. CONCLUSION: Flow measurements derived from the 3D and 2D acquisitions were comparable. Although time consuming, comprehensive 3D velocity acquisition could be relatively operator independent, and could potentially yield information on flow through several retrospectively chosen planes, for example in patients with congenital or valvular heart disease.

Virtual cardiotomy based on 3-D MRI for preoperative planning in congenital heart disease.

BACKGROUND: Patient-specific preoperative planning in complex congenital heart disease may be greatly facilitated by virtual cardiotomy. Surgeons can perform an unlimited number of surgical incisions on a virtual 3-D reconstruction to evaluate the feasibility of different surgical strategies. OBJECTIVE: To quantitatively evaluate the quality of the underlying imaging data and the accuracy of the corresponding segmentation, and to qualitatively evaluate the feasibility of virtual cardiotomy. MATERIALS AND METHODS: A whole-heart MRI sequence was applied in 42 children with congenital heart disease (age 3 +/- 3 years, weight 13 +/- 9 kg, heart rate 96 +/- 21 bpm). Image quality was graded 1-4 (diagnostic image quality > or =2) by two independent blinded observers. In patients with diagnostic image quality the segmentation quality was also graded 1-4 (4 no discrepancies, 1 misleading error). RESULTS: The average image quality score was 2.7 - sufficient for virtual reconstruction in 35 of 38 patients (92%) older than 1 month. Segmentation time was 59 +/- 10 min (average quality score 3.5). Virtual cardiotomy was performed in 19 patients. CONCLUSION: Accurate virtual reconstructions of patient-specific cardiac anatomy can be produced in less than 1 h from 3-D MRI. The presented work thus introduces a new, clinically feasible noninvasive technique for improved preoperative planning in complex cases of congenital heart disease.

Acceleration and validation of optical flow based deformable registration for image-guided radiotherapy.

MATERIALS AND METHODS: Two registration methods based on optical flow estimation have been programmed to run on a graphics programming unit (GPU). One of these methods by Horn & Schunck is tested on a 4DCT thorax data set with 10 phases and 41 landmarks identified per phase. The other method by Cornelius & Kanade is tested on a series of six 3D cone beam CT (CBCT) data sets and a conventional planning CT data set from a head and neck cancer patient. In each of these data sets 6 landmark points have been identified on the cervical vertebrae and the base of skull. Both CBCT to CBCT and CBCT to CT registration is performed. RESULTS: For the 4DCT registration average landmark error was reduced by deformable registration from 3.5+/-2.0 mm to 1.1+/-0.6 mm. For CBCT to CBCT registration the average bone landmark error was 1.8+/-1.0 mm after rigid registration and 1.6+/-0.8 mm after deformable registration. For CBCT to CT registration errors were 2.2+/-0.6 mm and 1.8+/-0.6 mm for rigid and deformable registration respectively. Using GPU hardware the Horn & Schunck method was accelerated by a factor of 48. The 4DCT registration can be performed in 37 seconds. The head and neck cancer patient registration takes 64 seconds. DISCUSSION: Compared to image slice thickness, which limits accuracy of landmark point determination, we consider the landmark point accuracy of the registration acceptable. The points identified in the CBCT images do not give a full impression of the result of doing deformable registration as opposed to rigid registration. A larger validation study is being planned in which soft tissue landmarks will facilitate tracking the deformable registration. The acceleration obtained using GPU hardware means that registration can be done online for CBCT.

GPU accelerated viscous-fluid deformable registration for radiotherapy.

Repeated acquisitions of computed tomography and magnetic resonance imaging are increasingly used during radiotherapy treatment to accurately deliver radiation while limiting side effects. This is only feasible however after all acquisitions have been correlated to a single reference scan using a deformable registration method. This paper presents a parallel implementation of one such method, the viscous-fluid registration method, on modern graphics hardware. A significant speedup close to two orders of magnitudes was observed when comparing to a CPU based implementation. As a consequence of the reduced registration time it is now feasible to perform larger scale clinical evaluation of the method. An example of registration results obtained during a treatment course is included in the paper.

A framework for shape matching in deformable image registration.

Many existing image registration methods have difficulties in accurately describing significant rotation and bending of entities (e.g. organs) between two datasets. A common problem in this case is to ensure that the resulting registration is physically plausible, i.e. that the registration describes the actual bending/rotation occurring rather than just introducing expansion in some areas and shrinkage in others. In this work we developed a general framework for deformable image registration of two 3D datasets that alleviates this problem. To ensure that only physically feasible and plausible solutions to the registration problem are found, a soft tissue deformable model is used to constrain the search space for the desired correspondence map while minimizing a similarity metric between the source and reference datasets. Results from a deformable phantom experiment were used to verify and evaluate the framework.