Gregg Tracton

Deformable M-Reps for 3D Medical Image Segmentation

M-reps (formerly called DSLs) are a multiscale medial means for modeling and rendering 3D solid geometry. They are particularly well suited to model anatomic objects and in particular to capture prior geometric information effectively in deformable models segmentation approaches. The representation is based on figural models, which define objects at coarse scale by a hierarchy of figures—each figure generally a slab representing a solid region and its boundary simultaneously. This paper focuses on the use of single figure models to segment objects of relatively simple structure.A single figure is a sheet of medial atoms, which is interpolated from the model formed by a net, i.e., a mesh or chain, of medial atoms (hence the name m-reps), each atom modeling a solid region via not only a position and a width but also a local figural frame giving figural directions and an object angle between opposing, corresponding positions on the boundary implied by the m-rep. The special capability of an m-rep is to provide spatial and orientational correspondence between an object in two different states of deformation. This ability is central to effective measurement of both geometric typicality and geometry to image match, the two terms of the objective function optimized in segmentation by deformable models. The other ability of m-reps central to effective segmentation is their ability to support segmentation at multiple levels of scale, with successively finer precision. Objects modeled by single figures are segmented first by a similarity transform augmented by object elongation, then by adjustment of each medial atom, and finally by displacing a dense sampling of the m-rep implied boundary. While these models and approaches also exist in 2D, we focus on 3D objects.The segmentation of the kidney from CT and the hippocampus from MRI serve as the major examples in this paper. The accuracy of segmentation as compared to manual, slice-by-slice segmentation is reported.

Image Registration: An Essential Part of Radiation Therapy Treatment Planning

PURPOSE We believe that a three-dimensional (3D) registration of nonplanning (diagnostic) imaging data with the planning computed tomography (CT) offers a substantial improvement in tumor target identification for many radiation therapy patients. The purpose of this article is to review and discuss our experience to date. METHODS AND MATERIALS We reviewed the charts and treatment planning records of all patients that underwent 3D radiation treatment planning in our department from June 1994 to December 1995, to learn which patients had image registration performed and why it was thought they would benefit from this approach. We also measured how much error would have been introduced into the target definition if the nonplanning imaging data had not been available and only the planning CT had been used. RESULTS Between June 1994 and December 1995, 106 of 246 (43%) of patients undergoing 3D treatment planning had image registration. Four reasons for performing registration were identified. First, some tumor volumes have better definition on magnetic resonance imaging (MRI) than on CT. Second, a properly contrasted diagnostic CT sometimes can show the tumor target better than can the planning CT. Third, the diagnostic CT or MR may have been preoperative, with the postoperative planning CT no longer showing the tumor. Fourth, the patient may have undergone cytoreductive chemotherapy so that the postchemotherapy planning CT no longer showed the original tumor volume. In patients in whom the planning CT did not show the tumor volume well an analysis was done to determine how the treatment plan was changed with the addition of a better tumor-defining nonplanning CT or MR. We have found that the use of this additional imaging modality changed the tumor location in the treatment plan at least 1.5 cm for half of the patients, and up to 3.0 cm for 1/4 of the patients. CONCLUSIONS Multimodality and/or sequential imaging can substantially aid in better tumor definition in many patients undergoing 3D treatment planning. In some patients the appropriate nonplanning imaging source can change the perceived tumor location by several centimeters and is thus essential for proper treatment planning.

A method and software for segmentation of anatomic object ensembles by deformable m‐reps

Deformable shape models (DSMs) comprise a general approach that shows great promise for automatic image segmentation. Published studies by others and our own research results strongly suggest that segmentation of a normal or near-normal object from 3D medical images will be most successful when the DSM approach uses (1) knowledge of the geometry of not only the target anatomic object but also the ensemble of objects providing context for the target object and (2) knowledge of the image intensities to be expected relative to the geometry of the target and contextual objects. The segmentation will be most efficient when the deformation operates at multiple object-related scales and uses deformations that include not just local translations but the biologically important transformations of bending and twisting, i.e., local rotation, and local magnification. In computer vision an important class of DSM methods uses explicit geometric models in a Bayesian statistical framework to provide a priori information used in posterior optimization to match the DSM against a target image. In this approach a DSM of the object to be segmented is placed in the target image data and undergoes a series of rigid and nonrigid transformations that deform the model to closely match the target object. The deformation process is driven by optimizing an objective function that has terms for the geometric typicality and model-to-image match for each instance of the deformed model. The success of this approach depends strongly on the object representation, i.e., the structural details and parameter set for the DSM, which in turn determines the analytic form of the objective function. This paper describes a form of DSM called m-reps that has or allows these properties, and a method of segmentation consisting of large to small scale posterior optimization of m-reps. Segmentation by deformable m-reps, together with the appropriate data representations, visualizations, and user interface, has been implemented in software that accomplishes 3D segmentations in a few minutes. Software for building and training models has also been developed. The methods underlying this software and its abilities are the subject of this paper.

Beam orientation selection for intensity-modulated radiation therapy based on target equivalent uniform dose maximization

PURPOSE To develop an automated beam-orientation selection procedure for intensity-modulated radiotherapy (IMRT), and to determine if a small number of beams picked by this automated procedure can yield results comparable to a large number of manually placed orientations. METHODS AND MATERIALS The automated beam selection procedure maximizes an unconstrained objective function composed of target equivalent uniform dose (EUD) and critical structure dose-volume histogram (DVH) constraints. Beam orientations are selected from a large feasible set of directions through a series of alternating fluence optimization and orientation alteration steps, until convergence to a stable orientation set. The fluence optimization step adjusts fluences to maximize the objective function. The orientation alteration step substitutes beams in the orientation set currently under consideration with beams of the parent set in the immediate angular vicinity; the altered orientation set is deemed current if it produces a higher objective function value in the fluence optimization step. RESULTS AND CONCLUSIONS It is demonstrated, for prostate IMRT planning, that a modest number of appropriately selected beam orientations (3 or 5) can provide dose distributions as satisfactory as those produced by a large number of unselected equispaced orientations. Such selected beam orientations can reduce overall treatment time, thus making IMRT more clinically practical.

Training models of anatomic shape variability

Learning probability distributions of the shape of anatomic structures requires fitting shape representations to human expert segmentations from training sets of medical images. The quality of statistical segmentation and registration methods is directly related to the quality of this initial shape fitting, yet the subject is largely overlooked or described in an ad hoc way. This article presents a set of general principles to guide such training. Our novel method is to jointly estimate both the best geometric model for any given image and the shape distribution for the entire population of training images by iteratively relaxing purely geometric constraints in favor of the converging shape probabilities as the fitted objects converge to their target segmentations. The geometric constraints are carefully crafted both to obtain legal, nonself-interpenetrating shapes and to impose the model-to-model correspondences required for useful statistical analysis. The paper closes with example applications of the method to synthetic and real patient CT image sets, including same patient male pelvis and head and neck images, and cross patient kidney and brain images. Finally, we outline how this shape training serves as the basis for our approach to IGRT/ART.

Clinical experience with 3-dimensional surface matching-based deep inspiration breath hold for left-sided breast cancer radiation therapy

PURPOSE Three-dimensional (3D) surface matching is a novel method to administer deep inspiration breath-hold (DIBH) radiation therapy for left-sided breast cancer to reduce cardiac exposure. We analyzed port (x-ray) films to assess patient setup accuracy and treatment times to assess the practical workflow of this system. METHODS AND MATERIALS The data from 50 left-sided breast cancer patients treated with DIBH were studied. AlignRT (London, UK) was used. The distance between the field edge and the anterior pericardial shadow as seen on the routine port films (dPORT), and the corresponding distance seen on the digitally reconstructed radiographs (DRR) from the planning (dDRR) were compared as a quantitative measure of setup accuracy. Variations of dPORT - dDRR over the treatment course were assessed. In a subset of 21 patients treated with tangential beams alone, the daily treatment durations were analyzed to assess the practical workflow of this system. RESULTS Considering all 50 patients, the mean absolute systematic uncertainty between dPORT and dDRR was 0.20 cm (range, 0 to 1.22 cm), the mean systematic uncertainty was -0.07 cm (range, -1.22 to 0.67 cm), and their mean random uncertainty was 0.19 cm (range, 0 to 0.84 cm). There was no significant change in dPORT - dDRR during the course of treatment. The mean patient treatment duration for the 21 patients studied was 11 minutes 48 seconds. On intrapatient assessments, 15/21 had nonsignificant trends toward reduced treatment durations during their course of therapy. On interpatient comparisons, the mean treatment times declined as we gained more experience with this technique. CONCLUSIONS The DIBH patient setup appears to provide a fairly reproducible degree of cardiac sparing with random uncertainties of ≈ 0.2 cm. The treatment durations are clinically acceptable and appear not to change significantly over time on an intrapatient basis, and to improve over time on an interpatient basis.

Multi-scale 3-D Deformable Model Segmentation Based on Medial Description

This paper presents a Bayesian multi-scale three dimensional deformable template approach based on a medial representation for the segmentation and shape characterization of anatomical objects in medical imagery. Prior information about the geometry and shape of the anatomical objects under study is incorporated via the construction of exemplary templates. The anatomical variability is accommodated in the Bayesian framework by defining probabilistic transformations on these templates. The modeling approach taken in this paper for building exemplary templates and associated transformations is based on a multi-scale medial representation. The transformations defined in this framework are parameterized directly in terms of natural shape operations, such as thickening and bending, and their location. Quantitative validation results are presented on the automatic segmentation procedure developed for the extraction of the kidney parenchyma-including the renal pelvis-in subjects undergoing radiation treatment for cancer. We show that the segmentation procedure developed in this paper is efficient and accurate to within the voxel resolution of the imaging modality.

A new three-dimensional dose distribution reduction scheme for tubular organs

In tubular structures, spatial aspects of the dose distribution may be important in determining the normal tissue response. Conventional dose-volume-histograms (DVHs) and dose-surface-histograms (DSHs) lack spatial information and may not be adequate to represent the three-dimensional (3D) dose data. A new 3D dose distribution data reduction scheme which preserves its longitudinal and circumferential character is presented. Dose distributions were generated at each axial level for esophagus or rectum in 123 patients with lung cancer or prostate cancer. Dose distribution histograms at each axial level were independently analyzed along the esophageal or rectal circumference to generate dose-circumference-histogram (DCH) sheets. Two types of plots were then generated from the DCH sheet. The first considered the percentage of the circumference at each axial level receiving various doses. The second considered the minimum dose delivered to any percentage of the circumference at each axial level. The DCH as a treatment planning tool can be easily implemented in a 3D planing system and is potentially useful for the study of the relationship between the complication risk and the longitudinal and circumferential dose distributions.

Prostate Shape Modeling Based on Principal Geodesic Analysis Bootstrapping

The use of statistical shape models in medical image analysis is growing due to the ability to incorporate prior organ shape knowledge for tasks such as segmentation, registration, and classification.

Segmentation of Single-Figure Objects by Deformable M-reps

This paper describes the basis and behavior of segmentation of single figures in 3D by deformable m-reps models. Results are given for the segmentation of kidneys from CT and of hippocampi from MR images. Special focus is made on multi-scale-level stages of segmentation, on intrinsic correspondences under deformation that are provided by m-reps, and on the match against model-relative templates provided by both theoretical edge strength templates and templates derived from training images.