D Markel

Novel multimodality segmentation using level sets and Jensen-Renyi divergence

PURPOSE Positron emission tomography (PET) is playing an increasing role in radiotherapy treatment planning. However, despite progress, robust algorithms for PET and multimodal image segmentation are still lacking, especially if the algorithm were extended to image-guided and adaptive radiotherapy (IGART). This work presents a novel multimodality segmentation algorithm using the Jensen-Rényi divergence (JRD) to evolve the geometric level set contour. The algorithm offers improved noise tolerance which is particularly applicable to segmentation of regions found in PET and cone-beam computed tomography. METHODS A steepest gradient ascent optimization method is used in conjunction with the JRD and a level set active contour to iteratively evolve a contour to partition an image based on statistical divergence of the intensity histograms. The algorithm is evaluated using PET scans of pharyngolaryngeal squamous cell carcinoma with the corresponding histological reference. The multimodality extension of the algorithm is evaluated using 22 PET/CT scans of patients with lung carcinoma and a physical phantom scanned under varying image quality conditions. RESULTS The average concordance index (CI) of the JRD segmentation of the PET images was 0.56 with an average classification error of 65%. The segmentation of the lung carcinoma images had a maximum diameter relative error of 63%, 19.5%, and 14.8% when using CT, PET, and combined PET/CT images, respectively. The estimated maximal diameters of the gross tumor volume (GTV) showed a high correlation with the macroscopically determined maximal diameters, with a R(2) value of 0.85 and 0.88 using the PET and PET/CT images, respectively. Results from the physical phantom show that the JRD is more robust to image noise compared to mutual information and region growing. CONCLUSIONS The JRD has shown improved noise tolerance compared to mutual information for the purpose of PET image segmentation. Presented is a flexible framework for multimodal image segmentation that can incorporate a large number of inputs efficiently for IGART.

Automatic Segmentation of Lung Carcinoma Using 3D Texture Features in 18-FDG PET/CT

Target definition is the largest source of geometric uncertainty in radiation therapy. This is partly due to a lack of contrast between tumor and healthy soft tissue for computed tomography (CT) and due to blurriness, lower spatial resolution, and lack of a truly quantitative unit for positron emission tomography (PET). First-, second-, and higher-order statistics, Tamura, and structural features were characterized for PET and CT images of lung carcinoma and organs of the thorax. A combined decision tree (DT) with K-nearest neighbours (KNN) classifiers as nodes containing combinations of 3 features were trained and used for segmentation of the gross tumor volume. This approach was validated for 31 patients from two separate institutions and scanners. The results were compared with thresholding approaches, the fuzzy clustering method, the 3-level fuzzy locally adaptive Bayesian algorithm, the multivalued level set algorithm, and a single KNN using Hounsfield units and standard uptake value. The results showed the DTKNN classifier had the highest sensitivity of 73.9%, second highest average Dice coefficient of 0.607, and a specificity of 99.2% for classifying voxels when using a probabilistic ground truth provided by simultaneous truth and performance level estimation using contours drawn by 3 trained physicians.

Technical Note: Dose-volume histogram analysis in radiotherapy using the Gaussian error function

A mathematical model based on the Gaussian error and complementary error functions was proposed to describe the cumulative dose-volume histogram (cDVH) for a region of interest in a radiotherapy plan. Parameters in the model (a, b, c) are related to different characteristics of the shape of a cDVH curve such as the maximum relative volume, slope and position of a curve drop off, respectively. A prostate phantom model containing a prostate, the seminal vesicle, bladder and rectum with cylindrical organ geometries was used to demonstrate the effect of interfraction prostate motion on the cDVH based on this error function model. The prostate phantom model was planned using a five-beam intensity modulated radiotherapy (IMRT), and a four-field box (4FB), technique with the clinical target volume (CTV) shifted in different directions from the center. In the case of the CTV moving out of the planning target volume (PTV), that is, the margin between the CTV and PTV is underestimated, parameter c (related to position of curve drop off) in the 4FB plan and parameters b (related to the slope of curve) and c in the IMRT plan vary significantly with CTV displacement. This shows that variation of the cDVH is present in the 4FB plan and such variation is more serious in the IMRT plan. These variations of cDVHs for 4FB and IMRT are due to the different dose gradients at the CTV edges in the anterior and posterior directions for the 4FB and IMRT plan. It is believed that a mathematical representation of the dose-volume relationship provides another viewpoint from which to illustrate problems with radiotherapy delivery such as internal organ motion that affect the dose distribution in a treatment plan.

Technical note: calculation of normal tissue complication probability using Gaussian error function model.

PURPOSE The Gaussian error function was first used and verified in normal tissue complication probability (NTCP) calculation to reduce the dose-volume histogram (DVH) database by replacing the dose-volume bin set with the error function parameters for the differential DVH (dDVH). METHODS Seven-beam intensity modulated radiation therapy (IMRT) treatment planning was performed in three patients with small(40cm3), medium (53cm3), and large (87cm3) prostate volume, selected from a group of 20 patients. Rectal dDVH varying with the interfraction prostate motion along the anterior-posterior direction was determined by the treatment planning system (TPS) and modeled by the Gaussian error function model for the three patients. Rectal NTCP was then calculated based on the routine dose-volume bin set of the rectum by the TPS and the error function model. The variations in the rectal NTCP with the prostate motion and volume were studied. RESULTS For the ranges of prostate motion of 8-2, 4-8, and 4-3 mm along the anterior-posterior direction for the small, medium, and large prostate patient, the rectal NTCP was determined varying in the ranges of 4.6%-4.8%, 4.5%-4.7%, and 4.6%-4.7%, respectively. The deviation of the rectal NTCP calculated by the TPS and the Gaussian error function model was within ±0.1%. CONCLUSIONS The Gaussian error function was successfully applied in the NTCP calculation by replacing the dose-volume bin set with the model parameters. This provides an option in the NTCP calculation using a reduced size of dose-volume database. Moreover, the rectal NTCP was found varying in about ±0.2% with the interfraction prostate motion along the anterior-posterior direction in the radiation treatment. The dependence of the variation in the rectal NTCP with the interfraction prostate motion on the prostate volume was found to be more significant in the patient with larger prostate.

A 4D biomechanical lung phantom for joint segmentation/registration evaluation.

At present, there exists few openly available methods for evaluation of simultaneous segmentation and registration algorithms. These methods allow for a combination of both techniques to track the tumor in complex settings such as adaptive radiotherapy. We have produced a quality assurance platform for evaluating this specific subset of algorithms using a preserved porcine lung in such that it is multi-modality compatible: positron emission tomography (PET), computer tomography (CT) and magnetic resonance imaging (MRI). A computer controlled respirator was constructed to pneumatically manipulate the lungs in order to replicate human breathing traces. A registration ground truth was provided using an in-house bifurcation tracking pipeline. Segmentation ground truth was provided by synthetic multi-compartment lesions to simulate biologically active tumor, background tissue and a necrotic core. The bifurcation tracking pipeline results were compared to digital deformations and used to evaluate three registration algorithms, Diffeomorphic demons, fast-symmetric forces demons and MiMVistas deformable registration tool. Three segmentation algorithms the Chan Vese level sets method, a Hybrid technique and the multi-valued level sets algorithm. The respirator was able to replicate three seperate breathing traces with a mean accuracy of 2-2.2%. Bifurcation tracking error was found to be sub-voxel when using human CT data for displacements up to 6.5 cm and approximately 1.5 voxel widths for displacements up to 3.5 cm for the porcine lungs. For the fast-symmetric, diffeomorphic and MiMvista registration algorithms, mean geometric errors were found to be [Formula: see text], [Formula: see text] and [Formula: see text] voxels widths respectively using the vector field differences and [Formula: see text], [Formula: see text] and [Formula: see text] voxel widths using the bifurcation tracking pipeline. The proposed phantom was found sufficient for accurate evaluation of registration and segmentation algorithms. The use of automatically generated anatomical landmarks proposed can eliminate the time and potential innacuracy of manual landmark selection using expert observers.

Dosimetric comparison between the prostate intensity-modulated radiotherapy (IMRT) and volumetric-modulated arc therapy (VMAT) plans using the planning target volume (PTV) dose–volume factor

Background We demonstrated that our proposed planning target volume (PTV) dose–volume factor (PDVF) can be used to evaluate the PTV dose coverage between the intensity-modulated radiotherapy (IMRT) and volumetric-modulated arc therapy (VMAT) plans based on 90 prostate patients. Purpose PDVF were determined from the prostate IMRT and VMAT plans to compare their variation of PTV dose coverage. Comparisons of the PDVF with other plan evaluation parameters such as D 5% , D 95% , D 99% , D mean , conformity index (CI), homogeneity index (HI), gradient index (GI) and prostate tumour control probability (TCP) were carried out. Methods and materials Prostate IMRT and VMAT plans using the 6 MV photon beams were created from 40 and 50 patients, respectively. Dosimetric indices (CI, HI and GI), dose–volume points ( D 5% , D 95% , D 99% and D mean ) and prostate TCP were calculated according to the PTV dose–volume histograms (DVHs) of the plans. All PTV DVH curves were fitted using the Gaussian error function (GEF) model. The PDVF were calculated based on the GEF parameters. Results From the PTV DVHs of the prostate IMRT and VMAT plans, the average D 99% of the PTV for IMRT and VMAT were 74·1 and 74·5 Gy, respectively. The average prostate TCP were 0·956 and 0·958 for the IMRT and VMAT plans, respectively. The average PDVF of the IMRT and VMAT plans were 0·970 and 0·983, respectively. Although both the IMRT and VMAT plans showed very similar prostate TCP, the dosimetric and radiobiological results of the VMAT technique were slightly better than IMRT. Conclusion The calculated PDVF for the prostate IMRT and VMAT plans agreed well with other dosimetric and radiobiological parameters in this study. PDVF was verified as an alternative of evaluation parameter in the quality assurance of prostate treatment planning.

MO-AB-BRA-09: Temporally Realistic Manipulation a 4D Biomechanical Lung Phantom for Evaluation of Simultaneous Registration and Segmentation.

Purpose: To produce multi-modality compatible, realistic datasets for the joint evaluation of segmentation and registration with a reliable ground truth using a 4D biomechanical lung phantom. The further development of a computer controlled air flow system for recreation of real patient breathing patterns is incorporated for additional evaluation of motion prediction algorithms. Methods: A pair of preserved porcine lungs was pneumatically manipulated using an in-house computer controlled respirator. The respirator consisted of a set of bellows actuated by a 186 W computer controlled industrial motor. Patient breathing traces were recorded using a respiratory bellows belt during CT simulation and input into a control program incorporating a proportional-integral-derivative (PID) feedback controller in LabVIEW. Mock tumors were created using dual compartment vacuum sealed sea sponges. 65% iohexol,a gadolinium-based contrast agent and 18F-FDG were used to produce contrast and thus determine a segmentation ground truth. The intensity distributions of the compartments were then digitally matched for the final dataset. A bifurcation tracking pipeline provided a registration ground truth using the bronchi of the lung. The lungs were scanned using a GE Discovery-ST PET/CT scanner and a Phillips Panorama 0.23T MRI using a T1 weighted 3D fast field echo (FFE) protocol. Results: The standard deviation of the error between the patient breathing trace and the encoder feedback from the respirator was found to be ±4.2%. Bifurcation tracking error using CT (0.97×0.97×3.27 mm3 resolution) was found to be sub-voxel up to 7.8 cm displacement for human lungs and less than 1.32 voxel widths in any axis up to 2.3 cm for the porcine lungs. Conclusion: An MRI/PET/CT compatible anatomically and temporally realistic swine lung phantom was developed for the evaluation of simultaneous registration and segmentation algorithms. With the addition of custom software and mock tumors, the entire package offers ground truths for benchmarking performance with high fidelity.

SU-F-R-40: Robustness Test of Computed Tomography Textures of Lung Tissues to Varying Scanning Protocols Using a Realistic Phantom Environment

PURPOSE The reliability of computed tomography (CT) textures is an important element of radiomics analysis. This study investigates the dependency of lung CT textures on different breathing phases and changes in CT image acquisition protocols in a realistic phantom setting. METHODS We investigated 11 CT texture features for radiation-induced lung disease from 3 categories (first-order, grey level co-ocurrence matrix (GLCM), and Laws filter). A biomechanical swine lung phantom was scanned at two breathing phases (inhale/exhale) and two scanning protocols set for PET/CT and diagnostic CT scanning. Lung volumes acquired from the CT images were divided into 2-dimensional sub-regions with a grid spacing of 31 mm. The distribution of the evaluated texture features from these sub-regions were compared between the two scanning protocols and two breathing phases. The significance of each factor on feature values were tested at 95% significance level using analysis of covariance (ANCOVA) model with interaction terms included. Robustness of a feature to a scanning factor was defined as non-significant dependence on the factor. RESULTS Three GLCM textures (variance, sum entropy, difference entropy) were robust to breathing changes. Two GLCM (variance, sum entropy) and 3 Laws filter textures (S5L5, E5L5, W5L5) were robust to scanner changes. Moreover, the two GLCM textures (variance, sum entropy) were consistent across all 4 scanning conditions. First-order features, especially Hounsfield unit intensity features, presented the most drastic variation up to 39%. CONCLUSION Amongst the studied features, GLCM and Laws filter texture features were more robust than first-order features. However, the majority of the features were modified by either breathing phase or scanner changes, suggesting a need for calibration when retrospectively comparing scans obtained at different conditions. Further investigation is necessary to identify the sensitivity of individual image acquisition parameters.

Development of a multi-modality 4D biomechanical phantom for evaluation of simultaneous registration/segmentation algorithms

A package for evaluating joint registration/ segmentation (“regmentation” algorithms) and motion prediction systems was developed using a pair of preserved swine lungs pneumatically controlled with a custom-built respirator. The phantom is MRI, CT and PET compatible and moves in a realistic 4D non-rigid fashion. The segmentation and registration ground truths are provided by a dual compartment mock tumor and a bifurcation tracking pipeline. The mock tumor consists of two vacuum-sealed sea sponges in separate compartments, allowing for the injection of radiotracer to approximate an active tumor and surrounding healthy tissue. Injection is possible through catheters connected to each compartment. The boundary of the inner compartment, segmented post-extraction, provides a ground truth for the boundary of the tumor region. Bifurcations of the bronchi of the lungs were used as anatomical landmarks, providing a registration ground truth between two sets of images of the lungs using in-house bifurcation detection and matching software. The accuracy of the bifurcation tracking pipeline was found to be on the order of a voxel width for human and swine lung datasets, with tracking capabilities up to deformations of 7.8 and 3.4 cm respectively. Accuracy was evaluated using a known virtual deformation. The computer-controlled respirator was found capable of mimicking human breathing traces using the swine lungs within a maximum error of ±2.2% and an average error of ±0.5%. PET/CT and MRI scans of the lungs were acquired for various levels of image noise.

Poster — Thur Eve — 71: A 4D Multimodal Lung Phantom for Regmentation Evaluation

Segmentation and registration of medical imaging data are two processes that can be integrated (a process termed regmentation) to iteratively reinforce each other, potentially improving efficiency and overall accuracy. A significant challenge is presented when attempting to validate the joint process particularly with regards to minimizing geometric uncertainties associated with the ground truth while maintaining anatomical realism. This work demonstrates a 4D MRI, PET, and CT compatible tissue phantom with a known ground truth for evaluating registration and segmentation accuracy. The phantom consists of a preserved swine lung connected to an air pump via a PVC tube for inflation. Mock tumors were constructed from sea sponges contained within two vacuum-sealed compartments with catheters running into each one for injection of radiotracer solution. The phantom was scanned using a GE Discovery-ST PET/CT scanner and a 0.23T Phillips MRI, and resulted in anatomically realistic images. A bifurcation tracking algorithm was implemented to provide a ground truth for evaluating registration accuracy. This algorithm was validated using known deformations of up to 7.8 cm using a separate CT scan of a human thorax. Using the known deformation vectors to compare against, 76 bifurcation points were selected. The tracking accuracy was found to have maximum mean errors of −0.94, 0.79 and −0.57 voxels in the left-right, anterior-posterior and inferior-superior directions, respectively. A pneumatic control system is under development to match the respiratory profile of the lungs to a breathing trace from an individual patient.