Room: Exhibit Hall | Forum 5
Purpose: Breath hold monitoring for DIBH treatment of breast cancer is being performed by various means involving external surrogates. The use of continuous MV imaging with an electronic portal imaging device (EPID) holds the potential to more accurately ensure positioning of the target without additional imaging dose or extra imaging hardware needed. Using real time image analysis, the position of the anatomy can be verified. The work discusses a new image analysis algorithm to reliably identify the irradiation geometry.
Methods: Images are obtained parallel to the linear accelerator with frame grabber hard hardware and provided for analysis by in-house software using MATLAB/SIMULINK. Deviation from the anatomy in the reference image (motion) is assessed for each incoming MV image: First the image is rotated to 0 degree using a Hough transform algorithm. Next, a region of interest (ROI) is defined in the middle band of the image. Each line of intensity values inside the ROI is extracted and compared to the corresponding line from the reference image. Normalized Cross Correlation (NCC) is used to identify motion distance based on best match. To assess robustness of the proposed method, image sets of 10 patients were retrospectively re-analyzed.
Results: The method worked robustly on all images. For the 10 patients, the average of absolute mean motion (Â±1SD) was 0.45 (Â±0.13) mm with the average standard deviation (Â±1SD) of 0.72 (Â±0.09) mm. The average of the maximum motion (Â±1SD) was 2.82 (Â±1.70) mm. This matched results from earlier algorithms. Execution of the algorithm took approximately 0.007 (Â±0.005) seconds per image on a standard PC.
Conclusion: The new algorithm is reliable and fast enough for real time image analysis in continuous EPID based monitoring of DIBH treatments. The high image analysis speed allows for additional image evaluations, like comparisons to performance at previous treatment fractions.
Funding Support, Disclosures, and Conflict of Interest: This work has been supported by National Health and Medical Research Council (NHMRC) grant 1147533 of the Australian Government.