MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Towards Instantaneous, Patient Specific, Measurement Based Pre-Treatment QA by Utilizing Machine Learning of Trajectory Files and Routine MLC QA

L Lay1*, J Adamson2, W Giles3, K Chuang4, (1) Duke Medical Physics Graduate Program, Durham, NC, (2) Duke University Medical Center, Durham, NC, (3) Duke University Medical Center, Durham, NC, (4) Duke Kunshan University, Zhubei City, HSQ, TW

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: Online adaptive radiotherapy will require instantaneous pre-treatment QA. We recently developed an algorithm to predict MLC positional discrepancies that occur at treatment delivery for a new treatment plan based on machine learning of the linear accelerator’s prior trajectory log files, which can be applied for DVH based pre-treatment QA. However, one weakness of this approach is that trajectory files do not always detect MLC discrepancies. This work entails the first steps towards incorporating MLC uncertainties measured during routine MLC QA.

Methods: In addition to verifying static MLC from picket fence or MPC, we developed a novel technique to measure discrepancy during dynamic motion. Throughout an integrated EPID acquisition, each leaf travels across a 3cm gap at constant speed until the leaves abut, with leaf openings interspersed throughout the field to minimize effect of scatter. The acquired image is normalized to an “open” field in which MLC openings are static at the 3cm maximum. The relationship between image intensity and systematic MLC discrepancy is defined empirically using a series of calibration images with intentional lags (0.1mm to 2.0mm). A proof of principle plan verifying a single leaf was delivered repeatedly (n=3) at two leaf speeds (0.1cm/s 300 MU, 2.5cm/s 12 MU).

Results: The linear fit of the intensity profile had an R² of 1.000 for both leaf speeds; the intercept of the linear fit quantified systematic lag of the moving leaf, with a linear relationship between leaf lag and y-intercept (R² = 1.000). The standard deviation of measured systematic MLC errors was 0.73mm for both 0.1cm/s and 2.5cm/s.

Conclusion: Positional discrepancies during dynamic MLC motion can be measured using a simple QA plan with an integrated EPID acquisition. Future work will include improving the measurement precision, applying to all MLCs, and incorporating the measured uncertainties into the pre-treatment QA workflow.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email