Program Information
EPID-Based Real-Time Patient Misalignment Detection Algorithm
M Ahmed*, H Nourzadeh , B Neal , W Watkins , J Siebers , University of Virginia Health System, Charlottesville, VA
Presentations
TH-AB-FS1-6 (Thursday, August 3, 2017) 7:30 AM - 9:30 AM Room: Four Seasons 1
Purpose: To quantify day-to-day and during treatment patient variations using exit fluence measurements on EPID.
Methods: Radiation fluence exiting the patient recorded by the Electronic Portal Imager Device (EPID) is streamed to “intrEPID”, an in-house real-time verification system, at ~8 frames/sec. Each frame undergoes normalized iso-intensity image segmentation into N≥3 levels. The same segmentation is performed for a corresponding prediction based on the patient’s CT images. Measured and predicted segmentations differences are quantified, currently, using the Hausdorff distance (HD). Algorithm performance is demonstrated via comparing predicted treatment frames with respect to frames predicted through the patient CT offset in the x, y, and z directions.
Results: For a VMAT treatment (500+ frames distributed over a full gantry rotation), the HD and the HD spread increases as a function of the rigid offset applied. The HD dispersion is due to variations in the frames beams-eye-view of the applied offset as a function of gantry angle. Box-and-Whisker plots illustrate the quasi-linear increase of the maximum and the upper quartile values of the HD data set with the increase of the shifts. The Pearson correlation coefficients for the maximum spread of the HD data and the shifts applied in the x, y, and z directions are 0.981, 0.964, and 0.967 respectively. For the median, Pearson’s r values read 0.984, 0.922, and 0.92, respectively. On current hardware, the full algorithm requires <80 ms, within the time constraints of frame-by-frame EPID acquisition.
Conclusion: A real-time patient misalignment/movement detection algorithm utilizing EPID was developed. Statistical variations (i.e., the spread of the Hausdorff data set) is linearly correlated to patient positioning variations.
Funding Support, Disclosures, and Conflict of Interest: The work is partially supported by Varian Medical systems
Contact Email: