An error detection method for real-time EPID-based treatment delivery quality assurance

Area of Science:

• Medical Physics
• Radiation Oncology
• Image Analysis

Background:

• Accurate radiation therapy delivery is crucial for patient safety and treatment efficacy.
• Existing quality assurance (QA) methods may not detect all subtle delivery errors.
• Real-time monitoring of treatment delivery can enhance safety and accuracy.

Purpose of the Study:

• To quantify the error detection power of a novel method using real-time electronic portal imaging device (EPID) images to validate monitor unit (MU) resolved beam apertures.
• To assess the method's ability to detect various simulated delivery errors and distinguish between similar treatment plans.

Main Methods:

• Utilized cine-EPID images (~10 Hz) from clinical and nonclinical VMAT/SBRT plans.
• Interpolated planned multileaf collimator (MLC) positions for real-time aperture validation.
• Simulated delivery errors by perturbing MLC positions, control point (CP) weights, and collimator angles.
• Trained a logistic regression model to identify error detection thresholds and evaluated accuracy using tenfold cross-validation.
• Assessed error detection using Matthews correlation coefficient (MCC) and false positive rate (FPR).

Main Results:

• The aperture check method achieved high detection sensitivity (MCC=1.00) for various errors, including MLC shifts, CP weight perturbations, collimator rotations, and stuck MLC leaves, with a low per-frame false positive rate (0.02%).
• Optimal error detection occurred at a threshold of 0.3% of expected aperture area, with errors identified within 1% of beam delivery for several error modes.
• The method demonstrated poor detection for EPID imager shifts (minimum MCC of 0.14) but reliably distinguished highly similar treatment plans, unlike gamma analysis.

Conclusions:

• The aperture check method provides sensitive and specific detection of radiation therapy delivery errors, including MLC deviations, CP MU shifts, and stuck MLC leaves.
• Its high performance and low false positive rate make it suitable for pretreatment and during-treatment quality assurance (QA).
• The method's ability to detect errors rapidly and distinguish similar plans positions it as a valuable tool for advanced radiotherapy techniques like on-line adaptive radiotherapy.