Computer vision mechanical QA: Development, characterization, and five years of clinical performance.

Area of Science:

• Medical Physics
• Radiation Oncology
• Computer Vision Applications

Background:

• Radiation therapy quality assurance (QA) is vital for accurate patient treatments.
• Current manual QA tasks are time-consuming and limited by human accuracy.

Purpose of the Study:

• To automate monthly mechanical QA tests using a computer vision-based quality assurance (CVQA) system.
• To enhance measurement accuracy and precision while reducing inter-user variability.

Main Methods:

• Developed a CVQA system using ArUco markers and OpenCV for automated tests (couch translation, angles, walkout, ODI, field size).
• Created a GUI for user guidance, pass/fail determination, and troubleshooting.
• Validated system reproducibility and compared CVQA measurements against manual methods and human variability.

Main Results:

• CVQA system setup in 5 min, test execution in 7 min.
• High reproducibility for all tested parameters (within 0.5 mm/0.5° for most, 1 mm for ODI).
• CVQA measurements closely agreed with manual methods, outperforming human consistency in physicist comparisons.

Conclusions:

• CVQA effectively automates mechanical QA, offering an efficient, precise, and reproducible solution.
• The open-source CVQA system is freely available, promoting adoption in academic institutions.
• Implementation of CVQA can significantly improve workflow efficiency and treatment consistency in clinical settings.