Electronic Portal Imaging Verification of Radiation Therapy

One of the first application of the chamfer matching algorithm in the medical profession was in the field of treatment verification for radiotherapy. The aim of external beam radiotherapy is to eradicate a tumor by irradiating it repeatedly by high-energy photons with multiple beams (from multiple directions). Since the irradiation is typically repeated 20 or more times over a period of several weeks, the patient has to be repeatedly aligned on the treatment machine with high accuracy. To verify the patient position, a planar radiograph can be made during irradiation by placing a film cassette in the treatment beam at the opposite side of the patient (portal imaging). To facilitate more extensive quality assurance, several groups have developed electronic portal imaging devices to replace the film cassette [8]. Such systems can be based on fluorescent screen/video camera combinations (e.g., [25]) on a matrix of ionization chambers [32], or on large-area amorphous silicon detectors [1]. An example of a electron portal imaging device mounted on a linear accelerator for radiotherapy is shown in Fig. 2. The quality of the portal images is inherently poor because of the high photon energy that is used for therapeutic irradiation (4 to 50 MV acceleration potential). The images have a poor spatial resolution, and the contrast of the interesting anatomical structures (bone) is extremely low (Fig. 2B). After some enhancement, the bones become visible (Fig. 3B).

To verify the patient position, the portal image is compared with a reference image that is either a diagnostic energy X-ray

FIGURE 2 (A) Typical radiotherapy equipment. A patient is accurately positioned on the treatment table using side lasers. The head of the linear accelerator is above the treatment table, while below the treatment table one sees the electronic portal imaging device (mounted on a robot arm) that is used to verify the position of the patient relative to the treatment beam. The accelerator and table can rotate to allow targeting of a tumor from any direction. Photo courtesy of Varian Medical Systems, Palo Alto, CA. (B) The images made with the portal imager inherently have a poor image quality because of the high energy of the treatment beam. This is a portal image of a treatment in the pelvic area, where the black square is the treatment field.

FIGURE 3 (A) Reference image for portal image analysis. This is a simulated radiograph. The green dots were automatically placed on highlights in the image to describe the bony anatomy and form the input drawing for the chamfer matching procedure on anatomy. (B) Enhanced portal image with automatically detected field edge. (C) After matching the field edge, the portal image and the reference image have the same scale and alignment. Note the misalignment of the bony structures due to a small misalignment of the patient. (D) By matching the anatomy, the displacement of the patient with respect to the treatment beam is measured. See also Plate 72.

FIGURE 3 (A) Reference image for portal image analysis. This is a simulated radiograph. The green dots were automatically placed on highlights in the image to describe the bony anatomy and form the input drawing for the chamfer matching procedure on anatomy. (B) Enhanced portal image with automatically detected field edge. (C) After matching the field edge, the portal image and the reference image have the same scale and alignment. Note the misalignment of the bony structures due to a small misalignment of the patient. (D) By matching the anatomy, the displacement of the patient with respect to the treatment beam is measured. See also Plate 72.

image made on a treatment simulator or a digitally reconstructed radiograph (DRR), which is a simulated X-ray image that is constructed by ray tracing a CT scan of the patient along the beam direction (Fig. 3A). The analysis of the portal images needs to answer the following question: is the patient in the correct orientation with respect to the treatment beam? To answer that question, the reference image is matched twice with the portal image. First the borders of the treatment field are matched, so that scale and position of the portal image and the reference image are brought into correspondence (Fig. 3C). Next, the anatomy is matched, so that the displacement of the patient with respect to the treatment beam is quantified (Fig. 3D). For both registration purposes chamfer matching has been applied successfully. Note that the image analysis is done in two dimensions, for each beam direction separately. Fully 3D portal image analysis methods have been developed [16]. However, these are not based on chamfer matching and therefore fall outside the scope of this chapter.

Field Edge Matching

The field edge in the reference image can be delineated manually, or it can be retrieved from the planning system (if the reference image is a DRR) or detected automatically from the simulator image. The field edge in the portal image is easily detectable because it has a very high contrast. Several authors have presented algorithms for this purpose, e.g., based on histogram driven segmentation and/or gradient tracing [4]. Early attempts to register these shapes applied moment-based shape descriptors [5].

Because of the high quality of the segmented features, field edge registration by means of chamfer matching works extremely well, and this method has gained wide acceptance (Fig. 4). However, in some cases there are small errors in the shape of the field edge, which complicates the analysis [15]. With (simulated) perfect shapes, both the MEAN and RMS

cost functions work perfectly, independent of the distance transform used. However, when there is a small aspect ratio difference, e.g., because of a small misalignment of the shielding blocks used to shape the beam, the MAX and the RMS cost function perform much better. In contrast, when there is an outlier, such as a missing shielding block, the MEAN cost function is the only cost function that works well. For this reason, these authors implemented a two step procedure in which both the MEAN and then the MAX cost functions are used for matching and the results are compared to detect common shape errors. The residuals using both cost functions are an indication of the quality of the field shape and warnings are given if these exceed threshold levels. Variations on this algorithm are given by Leszczynski et al. [22,23]; Cai et al. [9] compared its performance with that of a contour-based matching algorithm.

0 0

Responses

  • Bacco
    How to draw portal imaging anatomy?
    5 years ago

Post a comment