Search

Lung stereotactic-body radiotherapy (SBRT) places additional requirements on targeting accuracy over standard approaches. In treatment planning, a tumour volume is geometrically expanded and the resulting planning target volume (PTV) is covered with the prescribed dose. This ensures full dose delivery despite various uncertainties encountered during treatment. We developed a retrospective technique for optimizing the PTV expansion for a patient population. The method relies on deformable image registration (DIR) of the planning CT to a treatment cone-beam CT (CBCT). The resulting transformation is used to map the planned target onto the treatment geometry, allowing the computation of the achieved target/PTV overlap. Basic validation of the method was performed using an anthropomorphic respiratory motion phantom. A self-validation technique was also implemented to allow estimation of the DIR error for the data being analyzed. Our workflow was used to retrospectively optimize PTV margin for 25 patients treated over 93 fractions. Targets for these patients were contoured on 4D CT images. SBRT delivery followed CBCT acquisition and a couch correction. A post-treatment CBCT was also acquired in some cases. Our basic validation demonstrated that the DIR-based technique is capable of transforming target volumes from planning CTs to treatment CBCTs with sub-mm accuracy. Our clinical analysis showed that the minimum percentages of target volumes covered for 3, 4, and 5 mm PTV margins were 92.1, 97.6, and 99.2, respectively. Analyzing data acquired before and just after treatment demonstrated that margins exceeding 5 mm did not significantly improve coverage. Finally, a 5 mm PTV margin achieved ⩾95% target volume coverage with ⩾95% probability. Our technique is accurate, automated, self-validating, and incorporates complex ITV shapes/deformations to allow PTV margin optimization. The analysis of clinical data indicates a 5 mm PTV margin is optimal for our process. This approach is generalizable to other disease sites and treatment strategies.

Intensity modulated radiation therapy (IMRT) is an advanced technique used to concentrate the prescribed dose in the tumour while minimizing exposure to healthy tissues. Success in IMRT is greatly dependent upon the localization of the target volume and normal tissue, thus accurate contouring is crucial. In this paper, we describe an automated atlas-based image contouring system and our approach for improving the system by performing a full-scale optimization of registration parameters using high-performance computing. To achieve this, we use manually pre-contoured CT images of ten head and neck patients. For any parameter set, each patient data is registered with the remaining patients. Accuracy of the resulting contours is determined automatically by comparing their overlap with manually defined targets using Dice's similarity coefficient (DSC). This allows us to compare all permutations of the image registration parameter sets and input data to investigate their impact on final contour accuracy. Investigating the parameter space required 27,000 image registrations and 216,000 DSC computations. To perform these registrations we introduced a large cluster of high-performance computers and developed a parallel testing harness. The metrics collected from the tests show a wide range of performance, indicating that parameter selection is crucial in our contouring system. By selecting an optimized parameter set, we increased the mean overlap of the automatically contoured regions of interest by 50% and reduced registration time by 50% compared to the original parameters. Our findings illustrate that full-scale optimization is an effective method for improving the performance of the automated image contouring system.

Stereotactic body radiation therapy (SBRT) requires precise delivery of radiation to the target; intra- and inter-fraction lung tumour motion may adversely impact local tumour control. The purpose of this study was to retrospectively evaluate the impact of planning target volume (PTV) margin size on the coverage of the internal target volume (ITV) as localized in pre- and post-treatment cone-beam computed tomography (CBCT) images. Data from two patients undergoing SBRT were evaluated. For planning, free-breathing and 4DCT scans were performed, and used to contour the ITV. A 5mm margin was added to create the PTV. During treatment, 14 CBCTs were collected pre- and post-beam delivery. A data set comprising the average 4DCT intensities where available and treatment planning CT intensities for voxels that were beyond the field of view of the 4DCT was constructed. Registration of the combined planning image to each CBCT was performed using a deformable image registration algorithm. The transformations aligning the combined planning image with the CBCTs were applied to the planning ITV to obtain the treatment ITVs. For each CBCT, the fraction of treatment ITV within the PTV was determined using Boolean logic. This was repeated for various PTV margins ranging from 0 to 10 mm at 1mm intervals. The 3 and 5 mm PTV margins covered 95.1 ± 5.9% and 99.0 ± 2.0% of the ITV, respectively. Analysis of additional patients will be performed to confirm these preliminary results, which reinforce the use of a 5mm PTV margin for lung SBRT.

Purpose: Varian’s On-Board Imager is a linac-integrated cone-beam CT(CBCT) system used at the authors’ institution to acquire images prior to delivering each fraction of prostate intensity modulated radiotherapy. The images are used to determine a couch shift that realigns the tumor with the position obtained in the planning CT. However, this manual image-guided radiotherapy (IGRT) technique is operator dependent, time consuming, offers limited degrees of freedom, and requires significant imagingdose over the course of treatment. To overcome these problems, the authors propose two fully automatic IGRT techniques that require significantly less imagingdose. Methods: Dose is reduced by lowering the x-ray tube mA s during CBCT acquisition at the cost of increasing imagenoise. In “forward” IGRT, the CBCTimage is automatically registered to the planning CT to obtain the necessary couch shift. The “reverse” technique offers additional degrees of freedom as it involves nonrigid registration of the planning CT to the CBCT. Both techniques were evaluated using images of an anthropomorphic phantom with simulated motion and by retrospectively analyzing data from ten prostate cancer patients. Results: IGRT error for the phantom data at 100% relative imagingdose was 8.2 ± 3.7 , 3.5 ± 1.2 , and 2.1 ± 0.6 mm for setup only, forward, and reverse techniques, respectively. For patient images acquired at 100% relative imagingdose, the errors were 5.4 ± 1.7 , 5.0 ± 1.6 , 5.0 ± 2.0 , and 4.0 ± 1.6 mm for setup only, manual forward (performed clinically), automatic forward, and reverse IGRT, respectively. Furthermore, imagingdose could be reduced to 20% without a significant loss in image guidance accuracy. Conclusions: The presented image guidance methods are accurate while requiring only 20% of the standard imagingdose. The combination of low dose, automation, and accuracy enables frequent corrections during treatment, possibly leading to reduced margins and improved treatment outcomes.

This work presents the initial development and implementation of a novel 3D biomechanics-based approach to measure the mechanical activity of myocardial tissue, as a potential non-invasive tool to assess myocardial function. This technique quantifies the myocardial contraction forces developed within the ventricular myofibers in response to electro-physiological stimuli. We provide a 3D finite element formulation of a contraction force reconstruction algorithm, along with its implementation using magnetic resonance (MR) data. Our algorithm is based on an inverse problem solution governed by the fundamental continuum mechanics principle of conservation of linear momentum, under a first-order approximation of elastic and isotropic material conditions. We implemented our technique using a subject-specific ventricle model obtained by extracting the left ventricular anatomical features from a set of high-resolution cardiac MR images acquired throughout the cardiac cycle using prospective electrocardiographic gating. Cardiac motion information was extracted by non-rigid registration of the mid-diastole reference image to the remaining images of a 4D dataset. Using our technique, we reconstructed dynamic maps that show the contraction force distribution superimposed onto the deformed ventricle model at each acquired frame in the cardiac cycle. Our next objective will consist of validating this technique by showing the correlation between the presence of low contraction force patterns and poor myocardial functionality.

Two reasons for the recent rise in radiation exposure from CT are increases in its clinical applicability and the desire to maintain high SNR while acquiring smaller voxels. To address this emerging dose problem, several strategies for reducing patient exposure have already been proposed. One method employed in cardiacimaging is ECG-driven modulation of the tube current between 100% at one time point in the cardiac cycle and a reduced fraction at the remaining phases. In this paper, we describe how images obtained during such acquisition can be used to reconstruct 4D data of consistent high quality throughout the cardiac cycle. In our approach, we assume that the mid-diastole (MD) phase is imaged with full dose. The MD image is then independently registered to lower dose images (lower SNR) at other frames, resulting in a set of transformations. Finally, the transformations are used to warp the MD frame through the cardiac cycle to generate the full 4D image. In addition, the transformations may be interpolated to increase the temporal sampling or to generate images at arbitrary time points. Our approach was validated using various data obtained with simulated and scanner-implemented dose modulation. We determined that as little as 10% of the total dose was required to reproduce full quality images with a 1 mm spatial error and an error in intensity values on the order of the imagenoise. Thus, our technique offers considerable dose reductions compared to standard imaging protocols, with minimal effects on the quality of the final data.

Atlas based-segmentation has recently been evaluated for use in prostate radiotherapy. In a typical approach, the essential step is the selection of an atlas from a database that the best matches of the target image. This work proposes an atlas selection strategy and evaluate it impacts on final segmentation accuracy. Several anatomical parameters were measured to indicate the overall prostate and body shape, all of these measurements obtained on CT images. A brute force procedure was first performed for a training dataset of 20 patients using image registration to pair subject with similar contours; each subject was served as a target image to which all reaming 19 images were affinity registered. The overlap between the prostate and femoral heads was quantified for each pair using the Dice Similarity Coefficient (DSC). Finally, an atlas selection procedure was designed; relying on the computation of a similarity score defined as a weighted sum of differences between the target and atlas subject anatomical measurement. The algorithm ability to predict the most similar atlas was excellent, achieving mean DSCs of 0.78 ± 0.07 and 0.90 ± 0.02 for the CTV and either femoral head. The proposed atlas selection yielded 0.72 ± 0.11 and 0.87 ± 0.03 for CTV and either femoral head. The DSC obtained with the proposed selection method were slightly lower than the maximum established using brute force, but this does not include potential improvements expected with deformable registration. The proposed atlas selection method provides reasonable segmentation accuracy.

This work presents the first steps towards the development and implementation of a novel 3D biomechanical- based method for assessing the viability of myocardial tissue, with particular interest for its application in myocardial infarction (MI) diagnosis. This assessment technique quantifies the myocardial contraction forces developed within the ventricular myofibrils in response to the electrophysiological stimulus. In this manuscript we provide a 3D finite element (FE) formulation of a contraction force reconstruction algorithm based on an inverse problem solution of linear elasticity, along with its implementation using clinical data. This algorithm has been applied to patient-specific models obtained by extracting anatomical features from high- resolution, high-contrast magnetic resonance (MR) cardiac images. The input consists of motion information extracted by nonrigid registration of the mid-diastole reference image to the remaining images of the 4D data set, acquired using ECG- gating throughout the cardiac cycle. The result consists of a display-map of the contraction force distribution superimposed on the anatomical ventricle model, which allows the clinician to identify regions of low contractility in the myocardium.

High quality images have many uses, such as the testing of image processing algorithms and serving as templates for registration in image guided surgery. This work describes the creation of four high quality images or "appearance models", for 1. myocardium, 2. right atrium and ventricle, 3. left atrium and aorta, and 4. epicardial surface. The appearance models are created by registering and averaging together a series of high-resolution MR images of the same volunteer. These are then corrected to represent the average shape as determined from MR data of ten other volunteers. Thus, we show how single volunteer imaging can be combined with image registration to generate average-shape appearance models with high resolution and signal-to-noise ratio. Our final data consist of four 3D average shape appearance models (each depicting one of the sub-heart components) with 1.5 times 1.5 times 1.5 mm voxels and a signal-to-noise ratio increase of 6.6 versus raw data

While most currently available minimally invasive robotically assisted cardiac surgical systems do not employ 3D image guidance, such support can be generated using pre operative images such as CT. Previously we demonstrated a virtual model of the thorax with simulated surgical instruments, and a pulsating virtual model of the coronary arteries. In this paper we report the overlay of optical endoscopic images of a beating heart phantom with CT-based dynamic volumetric images of the phantom. Spatial matching is obtained through optical tracking of the endoscope and of the phantom, while time synchronization of the display of the model utilizes ECG gating. The spatial accuracy between the optical and virtual images varies from about 0.8 mm to -2.6 mm, while the time discrepancy depends on the frame-rate at which the virtual model is refreshed, and is typically 50-100 ms. Although the CT-based dynamic images are sufficient for animation of the model, artefacts associated with the image registration prevent seamless animation. Instead, to reconstruct the various phases of heart pulsation, we used a high-quality semi-static image of the diastolic phase of the phantom, and warped it to match the CT-based images corresponding to other phases of the heart pulsation.

Currently, minimally invasive cardiac surgery (MICS) faces several limitations, including inadequate training methods using non-realistic models, insufficient surgery planning using 2D images, and the lack of global, 3D guidance during the procedure. To address these issues we are developing the Virtual Cardiac Surgery Platform (VCSP) – a virtual reality model of the patient specific thorax, derived from pre-procedural images. Here we present an image registration-based method for customizing a geometrical template model of the heart to any given patient, and validate it using manual segmentation as the gold standard. On average, the process is accurate to within 3.3 ± 0.3 mm in MR images, and 2.4 ± 0.3 mm in CT images. These results include inaccuracies in the gold standard, which are on average 1.6 ± 0.2 and 0.9 ± 0.2 mm for MR and CT images respectively. We believe this method adequately prepares templates for use within VCSP, prior to and during MICS.

A four dimensional computed tomography (4DCT) image is acquired for all radically treated, lung cancer patients to define the internal target volume (ITV), which encompasses tumour motion due to breathing and subclinical disease. Patient set-up error and anatomical motion that is not due to breathing is addressed through an additional 1 cm margin around the ITV to obtain the planning target volume (PTV). The objective of this retrospective study is to find the minimum PTV margin that provides an acceptable probability of delivering the prescribed dose to the ITV. Acquisition of a kV cone beam computed tomography (CBCT) image at each fraction was used to shift the treatment couch to accurately align the spinal cord and carina. Our method utilized deformable image registration to automatically position the planning ITV on each CBCT. We evaluated the percentage of the ITV surface that fell within various PTVs for 79 fractions across 18 patients. Treatment success was defined as a situation where at least 99% of the ITV is covered by the PTV. Overall, this is to be achieved in at least 90% of the treatment fractions. The current approach with a 1cm PTV margin was successful ∼96% of the time. This analysis revealed that the current margin can be reduced to 0.8cm isotropic or 0.6×0.6×1 cm3 non-isotropic, which were successful 92 and 91 percent of the time respectively. Moreover, we have shown that these margins maintain accuracy, despite intrafractional variation, and maximize CBCT image guidance capabilities.

Automated atlas-based segmentation has recently been evaluated for use in planning prostate cancer radiotherapy. In the typical approach, the essential step is the selection of an atlas from a database that best matches the target image. This work proposes an atlas selection strategy and evaluates its impact on the final segmentation accuracy. Prostate length (PL), right femoral head diameter (RFHD), and left femoral head diameter (LFHD) were measured in CT images of 20 patients. Each subject was then taken as the target image to which all remaining 19 images were affinely registered. For each pair of registered images, the overlap between prostate and femoral head contours was quantified using the Dice Similarity Coefficient (DSC). Finally, we designed an atlas selection strategy that computed the ratio of PL (prostate segmentation), RFHD (right femur segmentation), and LFHD (left femur segmentation) between the target subject and each subject in the atlas database. Five atlas subjects yielding ratios nearest to one were then selected for further analysis. RFHD and LFHD were excellent parameters for atlas selection, achieving a mean femoral head DSC of 0.82 ± 0.06. PL had a moderate ability to select the most similar prostate, with a mean DSC of 0.63 ± 0.18. The DSC obtained with the proposed selection method were slightly lower than the maximums established using brute force, but this does not include potential improvements expected with deformable registration. Atlas selection based on PL for prostate and femoral diameter for femoral heads provides reasonable segmentation accuracy.

Purpose: A 2D/3D registration algorithm is proposed for registering orthogonal x-ray images with a diagnostic CT volume for high dose rate (HDR) lung brachytherapy. Methods: The algorithm utilizes a rigid registration model based on a pixel/voxel intensity matching approach. To achieve accurate registration, a robust similarity measure combining normalized mutual information, image gradient, and intensity difference was developed. The algorithm was validated using a simple body and anthropomorphic phantoms. Transfer catheters were placed inside the phantoms to simulate the unique image features observed during treatment. The algorithm sensitivity to various degrees of initial misregistration and to the presence of foreign objects, such as ECG leads, was evaluated. Results: The mean registration error was 2.2 and 1.9 mm for the simple body and anthropomorphic phantoms, respectively. The error was comparable to the interoperator catheter digitization error of 1.6 mm. Preliminary analysis of data acquired from four patients indicated a mean registration error of 4.2 mm. Conclusions: Results obtained using the proposed algorithm are clinically acceptable especially considering the complications normally encountered when imaging during lung HDR brachytherapy.

Image‐guidedradiotherapy(IGRT) is becoming the standard treatment for prostate cancer. One approach employs a linear accelerator‐mounted, cone‐beam computed tomography(CBCT) system to acquire 3D images at treatment. Typically, radiation therapists manually register such images to the planning CT to determine couch shifts, thereby ignoring anatomical rotations and deformations. Furthermore, CBCT systems deliver significant imagingdoses over long fractionation schedules. Therefore, we continue developing image guidance (IG) techniques relying on automatic registration and low doseCBCTimages. Our “global” method computes couch corrections while the “local” variant provides a deformable transformation that can be used to adapt the original plan. Previous validation with Varian's On‐board Imager (OBI v1.3) showed that IG error is maintained despite reducing the mAs to 15% of the standard 1300. Recent improvements in OBI v1.4 result in similar image quality at 680 mAs (“pelvis” mode). Additionally, “pelvis spotlight” mode was introduced with additional lateral collimation and 720 mAs employed over 200 degrees. Retesting showed that global IG error is 3.5 ±1.0 mm irrespective of the OBI version down to 10% of the standard dose. Local IG required 20% of the standard dose to achieve 1.8 ± 0.6 mm accuracy with OBI v1.4 pelvis, while the 2.0 ± 0.5 mm error was maintained down to 15% with OBI v1.3 and v1.4 spotlight. In absolute terms, the dose savings achieved by our IG methods are cumulative with those offered by the upgrade. Our local IG technique has great potential to significantly improve the precision of radiation therapy.

Currently, we are using Varian's On Board Imager to acquire cone beam computed tomography(CBCT)images prior to prostate intensity modulated radiotherapy(IMRT). The images are used to determine the post‐setup prostate position, and thus, to shift the couch towards the situation assumed during treatment planning. This “manual forward” image guided radiotherapy(IGRT) technique is time consuming and requires significant imagingdose over the course of treatment. Therefore, we present two low‐dose IGRT methods that are entirely automatic. The dose is reduced by lowering the mAs during CBCT acquisition at the cost of increased noise in the final images. In the “auto forward” IGRT technique, the CBCTimage is registered to the planning CT and the result is used to shift the patient on the couch. The “auto reverse” technique involves non‐rigid registration of the planning CT to the CBCT. Both techniques were evaluated using a retrospective analysis of data from ten patients, including one planning CT and five CBCTimages acquired at evenly spaced fractions per patient. At the standard imagingdose,IGRT error was 4.3 ± 1.6, 3.9 ± 1.2, 3.8 ± 1.8, and 3.1 ± 1.1 mm for no correction, manual forward, auto forward, and auto reverse techniques, respectively. Errors exceeded seven mm in 4% of the fractions when employing the auto reverse technique, highlighting a possible need for manual surveillance. The imagingdose could be reduced to 20% without an appreciable loss in accuracy of the automatic methods.

Minimally invasive cardiac surgery (MICS) has already been shown to reduce hospital stays, but the full potential is not yet realized, largely due to limitations in pre‐ and intra‐operative visualization inside the closed chest. To address these issues, we are developing the Virtual Cardiac Surgery Platform (VCSP) — a 4D (3D + time), virtual reality model of the patient specific thorax, derived from pre‐procedural images. In this abstract, we discuss the accuracy of our image registration‐based method for deforming geometrical template models of the heart sub‐anatomy (myocardium, right atrium + ventricle, left atrium + aorta, epicardium) to 10 different volunteers (“patients”). The template models are built by manually segmenting a high quality magnetic resonance(MR)image (this template image is an average of 20 acquisitions of the same volunteer, 1.53 mm3 voxels). The template image is mapped to a much lower quality patient image (1.5×1.5×6.0 mm3 voxels) obtained in a clinically feasible manner, by maximizing the normalized mutual information (NMI) between the two images. The resulting global (affine) and local (free form deformation) transformation is applied to one of the four template models to transform it into patient space. The registration accuracy is assessed by comparing the mapped template to the manual segmentation of the patient. On average, the customization process is accurate to within 2.4 ± 0.2 mm, whereas, the difference between two manual segmentations (gold standards) was 1.3 ± 0.2 mm. We believe our method adequately prepares templates for use within VCSP, prior to and during MICS.