An automated dose tracking system for adaptive radiation therapy
Introduction
In adaptive radiation therapy (ART), improvements in treatment plan (re-planning) are made during the course of a multi-fraction schedule based on the delivered dose during initial fractions [1]. About 40 percent of the head&neck plans were adapted through re-planning in our clinic. Critical steps in the implementation of ART involve the estimation of the radiation dose received by the patient for each treatment fraction and the assessment of the estimated dose. Daily cone-beam computed tomography (CBCT) imaging provides volumetric information of the patient position, tumor and normal organs immediately prior to each treatment fraction. This affords a practical means to assess the patient position and movement of the tumor and critical, normal structures including dose limiting normal tissues defined in the initial treatment plan [2].
The re-planning process begins by estimating variations in dose distributions typically from dose volume histograms (DVH's). Three steps are involved. First, the plan is simulated from the daily CBCT image dataset to calculate the estimated actual delivered daily dose for the given treatment fraction. Second, structures of interest are delineated to obtain daily DVHs to provide dose metrics for the tumor and organs-at-risk (OARs) from which oncologists can evaluate treatment plan effectiveness. Third, the doses to the therapeutic target and OARs are modified, if necessary, to meet the dose constraints in the original treatment plan, or to afford further improvement in plan quality, if possible. Practical execution of the various tasks for ART in clinical practice is not trivial for several reasons. For instance, the process for a physician to manually delineate the target and surrounding normal tissue structures is cumbersome, typically taking hours even for a skilled clinician. This becomes a procedural challenge given the number of individual fractions (i.e. 35–40) comprising a complete treatment, and given that each fraction requires the full set of structures to be delineated. The limited image quality of daily CBCT places practical constraints on the ability to manually delineate structures as well as estimate actual daily doses. Furthermore, even though the daily organ dose is available through the above steps, in the absence of accurate point-to-point correspondence between the CBCT datasets at each fraction, it is not possible to accurately estimate the cumulative dose at each voxel within the image dataset, an obvious limitation.
One solution is to use DIR to establish point correspondence. DIR can also be used to transfer structure contours from one image to another through a deformation vector field (DVF) so that structures only need to be delineated once. There is a clinical need for such a tool in fractionated radiation therapy. Practically speaking, however, DIR in general is challenging due to the unknown individual biomechanics and potentially large organ deformation between treatments.
There remains debate as to the appropriateness of deforming dose along with DIR, especially under conditions where the tumor and/or surrounding normal organs undergo mass changes [3], [4], [5]. We believe that: (1) when inter-fraction motion is small, the DIR can be accurate in local regions; (2) when automated DIR produces visually large errors; user interaction can greatly improve the accuracy [6]. Thus, the DIR quality needs to be sensibly and frequently monitored while appropriate tools need to be implemented for uncertainty detection and correction. Currently there are several commercial software and open source projects dedicated to DIR. Although commercial software solutions provide sophisticated evaluation tools and fast calculation, they currently lack sufficient robustness to handle difficult cases such as prostate registration as well as patient specific optimization [7], [8]. Alternatively, open source projects with similar functions are being developed. Despite their flexibility and accuracy, these research endeavors are not tailored to the clinical environment; it takes great effort and resources to upgrade them based on rigorous quality assurance, based on extensive, independent tests for accuracy/robustness and their ability to fit into a complex and heavily regulated hospital IT structure [9]. In addition, many existing solutions lack automation and scripting capabilities. One cost effective solution explored in this report is a custom system to fulfill clinical needs that address efficiency and accuracy at the same time. Lei et al. [10] outlined the critical computational components required for implementation of ART. Each component is investigated separately so as to identify the steps that can be safely automated to achieve maximum efficiency. The work presented herein represents the first step toward this goal, providing the following:
- (a)
Demonstration of the clinical use of DIR for daily dose calculation and accumulation, which has been validated and applied to clinical data sets.
- (b)
Presentation of a highly integrated and automated software system for medical physicists and physicians to perform ART with accurate dose tracking based on daily CBCT image datasets. Our goal is for readers to be able to implement a similar custom system at their respective institutions using available resources, and the ideas presented.
It is important to note that re-planning, another major component of ART, is outside of the scope of the current paper. This article is organized as follows: Section 2 describes our system in detail; Section 3 illustrates an example case and evaluates utility; Section 4 discusses the limitations of the current system, and Section 5 presents conclusions and future work.
Section snippets
Methods and materials
Daily dose calculation and dose accumulation are the 2 major steps necessary to track radiation dose during adaptive radiation therapy [10]. We use a modularized design to ensure the flexibility of different dose calculation and accumulation modules catered for different clinics, and we store all data in a separate database/workspace to minimize the occupation of the clinical database. To automate the modules, a major software component enables free flow of data between modules, including the
Clinical implementation
The system described is designed for routine clinical use of dose tracking for ART. Fig. 3 illustrates the clinical workflow. The process is initiated by a physician's request (Step1). After the treatment simulation and planning, the plan information is exported to the ART database (Step 2), automated using the Eclipse scripting application programming interface (ESAPI, Varian Medical System, Palo Alto, CA). On each day of the treatment, the ESAPI module detects a new CBCT image acquisition,
Discussions and limitations
Although different aspects of dose tracking have been studied over the past decade, to the best of our knowledge, very few implementations have been reported [26], [27], [28]. Yang et al. [27] reported on a DIR algorithm for dose-tracking, developed in Matlab, which limited for use as a research-only system. Pinter et al. [26] generated a ‘’plug-in’’ for Slicer3D [29] for dose tracking; however, it lacks the integration tools necessary for clinical use, as opposed to our comprehensive system.
Conclusions and future work
In this paper, we present an automated dose tracking system for adaptive radiation therapy, which includes 3 components: ART engine, user interactive tools, and integration tools. Each component has several modules that can run independently in a distributed Windows environment. It seamlessly integrates open source projects such as Elastix, and commercial software such as Eclipse. In addition we have developed versatile GUIs to validate and improve DIRs. Another noteworthy feature of the system
Conflict of interest
None
Disclosure
Henry Ford Health System holds research agreements with Varian Medical Systems (Palo Alto, CA) and Philips Healthcare (Best, Netherlands).
References (34)
- et al.
Semi-automated CT segmentation using optic flow and Fourier interpolation techniques
Comput. Methods Programs Biomed.
(2006) - et al.
Increasing the impact of medical image computing using community-based open-access hackathons: the NA-MIC and 3D slicer experience
Med. Image Anal.
(2016) - et al.
Computational challenges for image-guided radiation therapy: framework and current research
Semina. Radiat. Oncol.
(2007) - et al.
3D Slicer as an image computing platform for the quantitative imaging network
Magnetic Resonance Imaging
(2012) - et al.
Reducing metal artifacts in cone-beam CT images by preprocessing projection data
Int. J. Radiation Oncol. Biol. Phys.
(2007) - et al.
Adaptive radiation therapy
Phys. Med. Biol.
(1997) - et al.
Evaluation of on-board kV cone beam CT (CBCT)-based dose calculation
Phys. Med. Biol.
(2007) - et al.
Point/counterpoint: it is not appropriate to "deform" dose along with deformable image registration in adaptive radiotherapy
Med. Phys.
(2012) - et al.
Dose mapping sensitivity to deformable registration uncertainties in fractionated radiotherapy - applied to prostate proton treatments
BMC Med. Phys.
(2013) - et al.
Caution must be exercised when performing deformable dose accumulation for tumors undergoing mass changes during fractionated radiotherapy
Int. J. Radiation Oncol. Biol. Phys.
(2016)