Automatic learning-based beam angle selection for thoracic IMRT.

PURPOSE: The treatment of thoracic cancer using external beam radiation requires an optimal selection of the radiation beam directions to ensure effective coverage of the target volume and to avoid unnecessary treatment of normal healthy tissues. Intensity modulated radiation therapy (IMRT) planning is a lengthy process, which requires the planner to iterate between choosing beam angles, specifying dose-volume objectives and executing IMRT optimization. In thorax treatment planning, where there are no class solutions for beam placement, beam angle selection is performed manually, based on the planner's clinical experience. The purpose of this work is to propose and study a computationally efficient framework that utilizes machine learning to automatically select treatment beam angles. Such a framework may be helpful for reducing the overall planning workload. METHODS: The authors introduce an automated beam selection method, based on learning the relationships between beam angles and anatomical features. Using a large set of clinically approved IMRT plans, a random forest regression algorithm is trained to map a multitude of anatomical features into an individual beam score. An optimization scheme is then built to select and adjust the beam angles, considering the learned interbeam dependencies. The validity and quality of the automatically selected beams evaluated using the manually selected beams from the corresponding clinical plans as the ground truth. RESULTS: The analysis included 149 clinically approved thoracic IMRT plans. For a randomly selected test subset of 27 plans, IMRT plans were generated using automatically selected beams and compared to the clinical plans. The comparison of the predicted and the clinical beam angles demonstrated a good average correspondence between the two (angular distance 16.8° ± 10°, correlation 0.75 ± 0.2). The dose distributions of the semiautomatic and clinical plans were equivalent in terms of primary target volume coverage and organ at risk sparing and were superior over plans produced with fixed sets of common beam angles. The great majority of the automatic plans (93%) were approved as clinically acceptable by three radiation therapy specialists. CONCLUSIONS: The results demonstrated the feasibility of utilizing a learning-based approach for automatic selection of beam angles in thoracic IMRT planning. The proposed method may assist in reducing the manual planning workload, while sustaining plan quality.

Vision 20/20: perspectives on automated image segmentation for radiotherapy.

Due to rapid advances in radiation therapy (RT), especially image guidance and treatment adaptation, a fast and accurate segmentation of medical images is a very important part of the treatment. Manual delineation of target volumes and organs at risk is still the standard routine for most clinics, even though it is time consuming and prone to intra- and interobserver variations. Automated segmentation methods seek to reduce delineation workload and unify the organ boundary definition. In this paper, the authors review the current autosegmentation methods particularly relevant for applications in RT. The authors outline the methods' strengths and limitations and propose strategies that could lead to wider acceptance of autosegmentation in routine clinical practice. The authors conclude that currently, autosegmentation technology in RT planning is an efficient tool for the clinicians to provide them with a good starting point for review and adjustment. Modern hardware platforms including GPUs allow most of the autosegmentation tasks to be done in a range of a few minutes. In the nearest future, improvements in CT-based autosegmentation tools will be achieved through standardization of imaging and contouring protocols. In the longer term, the authors expect a wider use of multimodality approaches and better understanding of correlation of imaging with biology and pathology.

Auto-segmentation of normal and target structures in head and neck CT images: a feature-driven model-based approach.

PURPOSE: Intensity modulated radiation therapy (IMRT) allows greater control over dose distribution, which leads to a decrease in radiation related toxicity. IMRT, however, requires precise and accurate delineation of the organs at risk and target volumes. Manual delineation is tedious and suffers from both interobserver and intraobserver variability. State of the art auto-segmentation methods are either atlas-based, model-based or hybrid however, robust fully automated segmentation is often difficult due to the insufficient discriminative information provided by standard medical imaging modalities for certain tissue types. In this paper, the authors present a fully automated hybrid approach which combines deformable registration with the model-based approach to accurately segment normal and target tissues from head and neck CT images. METHODS: The segmentation process starts by using an average atlas to reliably identify salient landmarks in the patient image. The relationship between these landmarks and the reference dataset serves to guide a deformable registration algorithm, which allows for a close initialization of a set of organ-specific deformable models in the patient image, ensuring their robust adaptation to the boundaries of the structures. Finally, the models are automatically fine adjusted by our boundary refinement approach which attempts to model the uncertainty in model adaptation using a probabilistic mask. This uncertainty is subsequently resolved by voxel classification based on local low-level organ-specific features. RESULTS: To quantitatively evaluate the method, they auto-segment several organs at risk and target tissues from 10 head and neck CT images. They compare the segmentations to the manual delineations outlined by the expert. The evaluation is carried out by estimating two common quantitative measures on 10 datasets: volume overlap fraction or the Dice similarity coefficient (DSC), and a geometrical metric, the median symmetric Hausdorff distance (HD), which is evaluated slice-wise. They achieve an average overlap of 93% for the mandible, 91% for the brainstem, 83% for the parotids, 83% for the submandibular glands, and 74% for the lymph node levels. CONCLUSIONS: Our automated segmentation framework is able to segment anatomy in the head and neck region with high accuracy within a clinically-acceptable segmentation time.

A feature-based approach for refinement of model-based segmentation of low contrast structures.

Accuracy and robustness are fundamental requirements of any automated method used for segmentation of medical images. Model-based segmentation (MBS) is a well established technique, where uncertainties in image content can be to a certain extent compensated by the use of prior shape information. This approach is, however, often problematic in cases where image information does not allow for generating a strong feature response, one example being soft tissue organs in CT data, which typically appear in low contrast. In this paper, we enhance our recently proposed framework for voxel classification-based refinement of MBS using a level-set segmentation technique with shape priors. We also introduce a novel feature weighting methodology that improves the performance of the classifier, demonstrating results superior to the previous feature selection method. Results of fully automated segmentation of low contrast organs in head and neck CT are presented. Compared to our previous approach, we have achieved an increase of up to 22% in segmentation accuracy.

Towards automatic patient positioning and scan planning using continuously moving table MR imaging.

A concept is proposed to simplify patient positioning and scan planning to improve ease of use and workflow in MR. After patient preparation in front of the scanner the operator selects the anatomy of interest by a single push-button action. Subsequently, the patient table is moved automatically into the scanner, while real-time 3D isotropic low-resolution continuously moving table scout scanning is performed using patient-independent MR system settings. With a real-time organ identification process running in parallel and steering the scanner, the target anatomy can be positioned fully automatically in the scanner's sensitive volume. The desired diagnostic examination of the anatomy of interest can be planned and continued immediately using the geometric information derived from the acquired 3D data. The concept was implemented and successfully tested in vivo in 12 healthy volunteers, focusing on the liver as the target anatomy. The positioning accuracy achieved was on the order of several millimeters, which turned out to be sufficient for initial planning purposes. Furthermore, the impact of nonoptimal system settings on the positioning performance, the signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) was investigated. The present work proved the basic concept of the proposed approach as an element of future scan automation.

Automated planning of scan geometries in spine MRI scans.

Consistency of MR scan planning is very important for diagnosis, especially in multi-site trials and follow-up studies, where disease progress or response to treatment is evaluated. Accurate manual scan planning is tedious and requires skillful operators. On the other hand, automated scan planning is difficult due to relatively low quality of survey images ("scouts") and strict processing time constraints. This paper presents a novel method for automated planning of MRI scans of the spine. Lumbar and cervical examinations are considered, although the proposed method is extendible to other types of spine examinations, such as thoracic or total spine imaging. The automated scan planning (ASP) system consists of an anatomy recognition part, which is able to automatically detect and label the spine anatomy in the scout scan, and a planning part, which performs scan geometry planning based on recognized anatomical landmarks. A validation study demonstrates the robustness of the proposed method and its feasibility for clinical use.

Spine detection and labeling using a parts-based graphical model.

The detection and extraction of complex anatomical structures usually involves a trade-off between the complexity of local feature extraction and classification, and the complexity and performance of the subsequent structural inference from the viewpoint of combinatorial optimization. Concerning the latter, computationally efficient methods are of particular interest that return the globally-optimal structure. We present an efficient method for part-based localization of anatomical structures which embeds contextual shape knowledge in a probabilistic graphical model. It allows for robust detection even when some of the part detections are missing. The application scenario for our statistical evaluation is spine detection and labeling in magnetic resonance images.

Assessment of a model-based deformable image registration approach for radiation therapy planning.

PURPOSE: The aim of this study is to develop a surface-based deformable image registration strategy and to assess the accuracy of the system for the integration of multimodality imaging, image-guided radiation therapy, and assessment of geometrical change during and after therapy. METHODS AND MATERIALS: A surface-model-based deformable image registration system has been developed that enables quantitative description of geometrical change in multimodal images with high computational efficiency. Based on the deformation of organ surfaces, a volumetric deformation field is derived using different volumetric elasticity models as alternatives to finite-element modeling. RESULTS: The accuracy of the system was assessed both visually and quantitatively by tracking naturally occurring landmarks (bronchial bifurcations in the lung, vessel bifurcations in the liver, implanted gold markers in the prostate). The maximum displacements for lung, liver and prostate were 5.3 cm, 3.2 cm, and 0.6 cm respectively. The largest registration error (direction, mean +/- SD) for lung, liver and prostate were (inferior-superior, -0.21 +/- 0.38 cm), (anterior-posterior, -0.09 +/- 0.34 cm), and (left-right, 0.04 +/- 0.38 cm) respectively, which was within the image resolution regardless of the deformation model. The computation time (2.7 GHz Intel Xeon) was on the order of seconds (e.g., 10 s for 2 prostate datasets), and deformed axial images could be viewed at interactive speed (less than 1 s for 512 x 512 voxels). CONCLUSIONS: Surface-based deformable image registration enables the quantification of geometrical change in normal tissue and tumor with acceptable accuracy and speed.

An adaptive irregular grid approach for 3D deformable image registration.

Deformable registration is an important application in medical image analysis and processing. We propose a physics-based parametric approach for deformable image registration, where non-rigid transformations are computed using an irregular grid of control points distributed within the image domain. The image is modelled as a three-dimensional (3D) homogeneous infinite elastic medium. It is assumed that a Gaussian-shaped force is applied at every control point, where the strengths, directions and influence areas of the forces as well as the positions of the control points are considered as free parameters whose optimization leads to maximization of the similarity measure between the images to be registered. For optimization, a computationally efficient Levenberg-Marquardt method is used. The proposed approach has certain advantages over traditional landmark-based methods or the registration methods based on regular grids, for example B-splines, since comparable results can be achieved by using less control points. Experimental results with 3D clinical images demonstrate that our method is capable of successfully coping with complex registration tasks.

Automated model-based organ delineation for radiotherapy planning in prostatic region.

PURPOSE: Organ delineation is one of the most tedious and time-consuming parts of radiotherapy planning. It is usually performed by manual contouring in two-dimensional slices using simple drawing tools, and it may take several hours to delineate all structures of interest in a three-dimensional (3D) data set used for planning. In this paper, a 3D model-based approach to automated organ delineation is introduced that allows for a significant reduction of the time required for contouring. METHODS AND MATERIALS: The presented method is based on an adaptation of 3D deformable surface models to the boundaries of the anatomic structures of interest. The adaptation is based on a tradeoff between deformations of the model induced by its attraction to certain image features and the shape integrity of the model. To make the concept clinically feasible, interactive tools are introduced that allow quick correction in problematic areas in which the automated model adaptation may fail. A feasibility study with 40 clinical data sets was done for the male pelvic area, in which the risk organs (bladder, rectum, and femoral heads) were segmented by automatically adapting the corresponding organ models. RESULTS: In several cases of the validation study, minor user interaction was required. Nevertheless, a statistically significant reduction in the time required compared with manual organ contouring was achieved. The results of the validation study showed that the presented model-based approach is accurate (1.0-1.7 mm mean error) for the tested anatomic structures. CONCLUSION: A framework for organ delineation in radiotherapy planning is presented, including automated 3D model-based segmentation, as well as tools for interactive corrections. We demonstrated that the proposed approach is significantly more efficient than manual contouring in two-dimensional slices.

Automated segmentation of the left ventricle in cardiac MRI.

We present a fully automated deformable model technique for myocardium segmentation in 3D MRI. Loss of signal due to blood flow, partial volume effects and significant variation of surface grey value appearance make this a difficult problem. We integrate various sources of prior knowledge learned from annotated image data into a deformable model. Inter-individual shape variation is represented by a statistical point distribution model, and the spatial relationship of the epi- and endocardium is modeled by adapting two coupled triangular surface meshes. To robustly accommodate variation of grey value appearance around the myocardiac surface, a prior parametric spatially varying feature model is established by classification of grey value surface profiles. Quantitative validation of 121 3D MRI datasets in end-diastolic (end-systolic) phase demonstrates accuracy and robustness, with 2.45 mm (2.84 mm) mean deviation from manual segmentation.

Automated 3-D PDM construction from segmented images using deformable models.

In recent years, several methods have been proposed for constructing statistical shape models to aid image analysis tasks by providing a priori knowledge. Examples include principal component analysis of manually or semiautomatically placed corresponding landmarks on the learning shapes [point distribution models (PDMs)], which is time consuming and subjective. However, automatically establishing surface correspondences continues to be a difficult problem. This paper presents a novel method for the automated construction of three-dimensional PDM from segmented images. Corresponding surface landmarks are established by adapting a triangulated learning shape to segmented volumetric images of the remaining shapes. The adaptation is based on a novel deformable model technique. We illustrate our approach using computed tomography data of the vertebra and the femur. We demonstrate that our method accurately represents and predicts shapes.