4. Capítulo 4: Propuesta final de diseño
4.2 Funcionamiento de la propuesta
External beam radiotherapy (EBRT) involves a detailed planning phase and has a relatively complex workflow diagramed in Fig. 131. Patients are imaged, and this planning image is segmented to identify target regions and nearby “at-risk” structures. A dosimetrist then proposes a therapy plan by arranging several virtual “beams” around the planning image. The expected dose with respect to each of the important structures is computed, and the plan is refined accordingly. The goal of the planning is to deliver adequate radiation “dose” to the target region while sparing as much as possible the nearby normal tissue.
Given a plan, the therapy itself is divided up into a course of treatment “fractions” that will be delivered over several sessions. This makes the task of reproducibly aligning the patient with the plan extremely important. Patient setup is typically limited to aligning external markers, such as tattoos, with intersecting lasers that indicate a planning space reference position. However, the patient’s internal anatomy may also move or change shape from day to day or even from moment to moment. For example, the target prostate tissue can be shifted and deformed between treatment fractions as the bladder empties and fills. Respiration can cause similar kinds of anatomic changes in the abdomen within a single treatment fraction.
Modified treatment plan Segmentation & treatment plan Treatment Planning or reference image
Patient scanned Segmentation and planning Segmentation & treatment plan Expected dose on target and objects at risk Patient setup (alignment with plan)
Treatment image
Simulation Compute relationship between I0 and I1 Registration & modified treatment plan Planning Treatment Repeat for several fractions of treatment
I0 S0
D
Ii
H0i
Fig. 130 Linear accelerator used for external beam radiotherapy (EBRT).
Fig. 131. Workflow for adaptive radiotherapy. Main components are planning and treatment. The MGR applications described here
could be used in the
segmentation, planning, and treatment setup phases.
Adjusting an EBRT plan to account for treatment-time anatomic shape change is called “adaptive radiotherapy” (ART) (Yan D 2000)(Wong JR 2005). ART is based on mappings between planning images and images collected at treatment-time. Evaluating the effects of such mappings on the expected dose delivered to the various important anatomic structures in the scene is a complex spatial task; yet it is typically limited to 2D views and a few quantitative summary measurements such as the volume overlaps of important regions before and after a registration. Planning in the presence of such anatomic change or potential patient setup error can account for considerable clinical time.
The next subsections demonstrate how MGR methods can be used to improve the particular tasks of segmentation, planning, and patient setup. From a high level, the MGR improvements allow the clinician to do 3D work in a regionally appropriate image modality, in the presence of potential error, and in a setting that ties the planning image to the patient in the world. The three clinical application views described are for 3D cross-modal segmentation, interactive planning under error, and patient setup verification. Each section includes a short overview of the problem being addressed, the view itself, and a discussion of what MGR methods were used to improve comprehension for the task.
Appendix: Implementing the Planning Under Error View Using mgrView
on page 152, walks the reader through the implementation of the planning project using the mgrView library, detailing how to preprocess input data for mgrView’s file loaders and how to create a new shader program and integrate it with the library.
The examples shown here are interactive mock-ups or vignettes. At present, mgrView does not provide a complete solution for allowing a clinician to interactively segment, plan, or update patient positioning: it is limited to providing views on data. Integrating mgrView functions into a framework such as 3D Slicer could provide many of the editing capabilities required for a complete solution.
5.2.2 3D Cross-Modal Segmentation View
Identifying the boundaries of target and at-risk regions is a crucial task in preparing a patient image for planning. A clinician usually delineates these boundaries slice by slice. Many structures are difficult to discern in a CT image; for example, prostate tissue is relatively homogeneous
and in CT images has no clear boundary with some neighboring structures. The combination of difficult to identify structures and views limited to axial slices leads to ad hoc segmentation heuristics like “find a particular bony landmark, then count down three slices and assume the prostate starts there”.
The goal of this vignette is to provide a clinician with both local detail and 3D context near a particular anatomic region to support a segmentation task.
The MGR vignette shown in Fig. 132 addresses each of these issues. First, while the user can still draw on a 2D slice, the slice is presented in its volume rendered 3D context and can take a continuous range of orientations. If a neighboring structure such as the bladder has already been identified, the slice can be oriented and positioned automatically according to local model coordinates and directions. An example might be positioning and orienting the slice to naturally emphasize the most proximal points between the bladder and the likely position of the prostate.
Second, while CT is the most important imaging modality for radiotherapy planning because it directly measures the tissue properties required for computing dose distribution, MRI can also be a useful imaging modality despite its potential geometric inaccuracies. In this case, the CT values of homogeneous regions such as the prostate have been swapped out to allow the clinician to pick landmarks or delineate boundaries with stronger tissue type differentiation from MRI (see Fig.
Fig. 132. 3D segmentation in mixed modes. Volume rendered structures from the CT image provide global context while the clinician can segment on a slice drawn from a corresponding MRI. Fig. 134 shows how the CT values near the prostate had been corrupted by artifacts from the metal fiducial marker visible in the center of the prostate region.
MGR Methods Required Model-coordinate driven clipping
Regional color mapping from an alternate modality
133) in the region of interest but with the familiar context of the CT data elsewhere.
One particular example of the utility of this is when metal fiducial markers embedded in the prostate for tracking create CT reconstruction artifacts that obscure the target region, as in Fig. 134. Here, MGR’s special capabilities for integrating multiple image sources has been used to replace the CT values on the entire working slice with data from a fused MRI. More complex potential solutions could be imagined using MGR’s color mapping methods. One particularly interesting application might be using the marker positions identified in the MRI to identify untrustworthy regions in the CT so that they can be replaced with an appropriate reference solid texture. Such solid texture “in-painting” for gas bubbles in the rectum based on the methods of (Cheung, Frey and Jojic 2005) was originally discussed by Joshua Levy in an unpublished report from 2005.
Fig. 134 The prostate region in the CT-only volume rendering on the left is obscured by the artifacts from the fiducial markers. The hybrid rendering on the right preserves the clear tissue distinction in the target region.
Fig. 133 Standard slice-by-slice view used during segmentation; the colored contours are the region boundaries drawn on this slice. The CT image on the left
shows very little tissue
differentiation between the
circled prostate region and its neighbors compared to the MR slice on the right.