Purpose: To use computer vision to automate monthly mechanical QA tests on a linac, including table travel, ODI, field size detection, table and collimator angle, and table and collimator walkout.
Methods: A 5MP digital camera with an 8.5mm focal length lens was connected to the interface mount of a Varian TrueBeam with a custom holder. A custom board was created with ARUCO markers, which give their location and a unique identifier for each marker, allowing orientation of the calibration pattern to be unambiguously determined. A user-friendly GUI has a button for each test that changes color to indicate passing or failure according to MPPG8a/9a criteria. The GUI also shows the image analysis regions for each test. Reproducibility measurements were taken across four days. Comparisons to ground truth were performed by graph paper or digital level. Twelve physicists were used to set the field size and to read the ODI as no ground truth could be established for these tests.
Results: The system takes five minutes to setup and seven minutes to complete all tests. All tests were reproducible within 0.5mm and 0.5°, except for the 40cm field size Y jaws as these have a large light penumbra, and ODI measurements that were reproducible within 0.1cm. Table travel, ODI, and walkout measurements agreed with ground truth within 0.5mm and 0.4°, except vertical table motions that agreed within 0.9mm due to lens focus optimized for the 100cm SSD plane. The average computer vision reading of the field size set by the physicists matched the desired field size within one standard deviation in most cases. Some readings exceeded this value due to the difficulty of precisely locating the broad Y jaw penumbra with the human eye.
Conclusion: The computer vision mechanical QA system can reproducibly perform mechanical QA tests with high accuracy.