Automated quality assurance for image-guided radiation therapy

Abstract

The use of image-guided patient positioning requires fast and reliable Quality Assurance (QA) methods to ensure the megavoltage (MV) treatment beam coincides with the integrated kilovoltage (kV) or volumetric cone-beam CT (CBCT) imaging and guidance systems. Current QA protocol is based on visually observing deviations of certain features in acquired kV in-room treatment images such as markers, distances, or HU values from phantom specifications. This is a time-consuming and subjective task because these features are identified by human operators. The method implemented in this study automated an IGRT QA protocol by using specific image processing algorithms that rigorously detected phantom features and performed all measurements involved in a classical QA protocol. The algorithm was tested on four different IGRT QA phantoms. Image analysis algorithms were able to detect QA features with the same accuracy as the manual approach but significantly faster. All described tests were performed in a single procedure, with acquisition of the images taking approximately 5 minutes, and the automated software analysis taking less than 1 minute. The study showed that the automated image analysis based procedure may be used as a daily QA procedure because it is completely automated and uses a single phantom setup.

Figure 1

Difference between the automated and manual QA procedure. In the standard procedure, certain distances and locations have to be measured manually, as illustrated in (a) where a marker's center shift from isocenter is assesed using the distance tool. The proposed QA procedure automatically detects marker shape, represented as a yellow line (b). Based on the contour, its center (yellow dot) and its shift from isocenter (red dot) can be automatically assessed.

Figure 2

Example of our automated feature extraction tool applied on four different phantoms. (a) The Varian cube phantom consists of a central 2 mm marker embedded in a cube. (b) The Exact Track phantom consist of five radioopaque markers arranged in a star pattern. (c) The Modus phantom consists of five spheres of different sizes embedded in a acrylic cube.

Figure 3

Effect of different image analysis filters used to automate the QA process. The first picture (a) shows a volumetric representation of the CT scan of a Modus phantom. Phantom features (b) are detected using a contouring algorithm with a HU threshold of –500. To select individual features, a labeling algorithm is used. The output of the algorithm is color‐coded in (c), where each individual phantom feature is represented using a different color.

Figure 4

Central marker extraction for the Modus phantom in the kV(left) and MV(right) OBI images. The auto‐extracted contour, represented in red, closely matches marker borders.

Figure 5

Comparison of the expected (white) and actual (green) marker location for the Varian cube (left) and Modus (right) phantoms. The axes and the corresponding labels are used to measure deviation of the actual marker position from the expected location.

Figure 6

Winston‐Lutz tests are performed directly on the OBI‐acquired projections of a CBCT scan. The gantry sag with rotation on the X and Y directions is plotted in these graphs.