SU-E-T-445: Gross Error Detection Based On EPID-Image-Extracted MLC-Leaf-Position Monitoring

Authors


Abstract

Purpose:

To develop a robust real-time, log-file independent MLC position validation metric capable of preventing gross treatment delivery errors.

Methods:

MLC leaf positions are identified on individual during-treatment EPID images acquired in real time at 10 Hz during VMAT/IMRT patient treatments. MLC positions are identified as follows: (1) the image is rotated counterclockwise by 90°-collimatorAngle so that MLC motion direction is vertical; (2) pixel-columns corresponding to individual leaves are assigned based upon the imager SDD and known leaf positions on the isocenter plane; (3) inflection points are identified and averaged across pixel-column intensity-profiles for all central (50 % of total) pixel-columns of each leaf; (4) an empirical position-based offset is added to each leaf's average inflection point to arrive at its final estimated position. These leaf positions are appropriately interpolated between EPID images and compared to their planned positions at every control point and statistically analyzed for in-field active (moving) leaves on image. Machine log and planned positions were compared for the same deliveries.

Results:

The mean EPID-detected leaf offset from 51 fractions was 0.03±0.25 cm. 90.3 % of EPID-detected leaves were within 3 mm. Machine-log-based analysis found an average of 0.0002±0.02 cm and 100% within.7 mm. Large (>3mm) EPID-detected deviations resulted from algorithm failure in cases where multiple inflection points persisted in the profiles.

Conclusions:

Real-time EPID-based quality control can provide an independent validation of MLC leaf positions with sufficient resolution for gross error detection. Although further algorithmic improvements are required to match log-file-based accuracy, the current EPID-based leaf position algorithm is suitable for clinical gross error detection.

Ancillary