Stochastic formulation of patient positioning using linac‐mounted cone beam imaging with prior knowledge

Abstract

Purpose:In this work, a novel stochastic framework for patient positioning based on linac-mounted CB projections is introduced. Based on this formulation, the most probable shifts and rotations of the patient are estimated, incorporating interfractional deformations of patient anatomy and other uncertainties associated with patient setup. Methods: The target position is assumed to be defined by and is stochastically determined from positions of various features such as anatomical landmarks or markers in CB projections, i.e., radiographs acquired with a CB-CT system. The patient positioning problem of finding the target location from CB projections is posed as an inverse problem with prior knowledge and is solved using a Bayesian maximum a posteriori ͑MAP͒ approach. The prior knowledge is three-fold and includes the accuracy of an initial patient setup ͑such as in-room laser and skin marks͒, the plasticity of the body ͑relative shifts between target and features͒, and the feature detection error in CB projections ͑which may vary depending on specific detection algorithm and feature type͒. For this purpose, MAP estimators are derived and a procedure of using them in clinical practice is outlined. Furthermore, a rule of thumb is theoretically derived, relating basic parameters of the prior knowledge ͑initial setup accuracy, plasticity of the body, and number of features͒ and the parameters of CB data acquisition ͑number of projections and accuracy of feature detection͒ to the expected estimation accuracy. Results: MAP estimation can be applied to arbitrary features and detection algorithms. However, to experimentally demonstrate its applicability and to perform the validation of the algorithm, a water-equivalent, deformable phantom with features represented by six 1 mm chrome balls were utilized. These features were detected in the cone beam projections ͑XVI, Elekta Synergy ® ͒ by a local threshold method for demonstration purposes only. The accuracy of estimation ͑strongly varying for different plasticity parameters of the body͒ agreed with the rule of thumb formula. Moreover, based on this rule of thumb formula, about 20 projections for 6 detectable features seem to be sufficient for a target estimation accuracy of 0.2 cm, even for relatively large feature detection errors with standard deviation of 0.5 cm and spatial displacements of the features with standard deviation of 0.5 cm. Conclusions:The authors have introduced a general MAP-based patient setup algorithm accounting for different sources of uncertainties, which are utilized as the prior knowledge in a transparent way. This new framework can be further utilized for different clinical sites, as well as theoretical developments in the field of patient positioning for radiotherapy.

Bayesian Estimation Applied to Stochastic Localization with Constraints due to Interfaces and Boundaries

“…Constraints due to interfaces or surfaces, for example, might occur in radiotherapy (or interventional radiology, image-guided surgical procedure) where the location of an implanted fiducial marker or another feature has to be determined in real-time based on X-ray projections (radiographs or tomosynthesis projections) [3,4]. Let us focus on a specific application of breast cancer patient setup in radiotherapy.…”

“…Due to their inherent ability to use prior knowledge, Bayesian approaches have gained interest in many fields in the recent years, such as in computer vision [1,2] and medical applications [3][4][5][6], as well as robotics (Kalman-Filter, Particle-Filter) [7,8] and metrology [9]. In this work, we present a Bayesian framework for the localization of objects with constraints derived from interfaces (surfaces) or boundaries specific to a given localization problem.…”

Learning Perspective Deformation in X-Ray Transmission Imaging

“…As a consequence, acquired images suffer from geometric distortions, which is called perspective deformation. Perspective deformation causes difficulty in direct, accurate geometric assessments of structures of interest (SOI) in many practical applications, e.g., anatomic landmark detection [6], [7], fluoroscopic image stitching [8], fiducial marker registration [9]- [11], and dual-modality image fusion [12]. Therefore, orthogonal projections of SOI are preferred over perspective projections in many applications.…”

Separating components of variation in measurement series using maximum likelihood estimation. Application to patient position data in radiotherapy

“…Several studies have used MLE based techniques for the estimation of object position and trajectory for target tracking in IGRT (Poulsen et al 2008, Yue et al 2011. A few studies have used MLE to calculate the most probable patient position from image registrations (Munbodh et al 2009, Hoegele et al 2011. All of these demonstrate the application of MLE to statistical estimation problems, especially in the presence of unknown parameters.…”