Room: ePoster Forums
Purpose: Periodic cone-beam computed tomography (CBCT) image evaluation is important to ensure high quality images for accurate patient setup verification and position adjustment. Although we expect image quality evaluation by two different operators using the same evaluation method, or by one operator repeating the measurement at two different times to produce essentially the same result, this is not always the case. Automated CBCT image analysis should reduce inter-user, and intra-user measurement variability, and thereby increase reproducibility. Here, we compare manual CBCT image analysis with semi-automated analysis using Radiological Imaging Technologyâ€™s (RIT) software module, Radia, using The Phantom Laboratoryâ€™s Catphan 504 phantom CBCT scans.
Methods: We evaluated monthly CBCT QA of a Varian TrueBeam linear accelerator over 34 months for the Catphan 504â€™s four test module sections: CTP515 (sub-slice and supra-slice low contrast disks), CTP528 (21-line pair high resolution), CTP486 (image uniformity and noise), and CTP404 (sensitometry, or Hounsfield number constancy). Using Bland Altman analysis, regression analysis, and Paired t-test, we compared RITâ€™s semi-automatic analysis results with those obtained manually.
Results: For the CTP486 test module, we found that RITâ€™s semi-automated analysis provided statistically different results (t=4.692 and p=2.6E-05) when compared to the manual analysis completed by different users for monthly QA. For both the CTP528 and CTP515 modules, our data shows that RITâ€™s semi-automated and manual analysis are comparable. For the CTP404 module, linear regression analysis shows high correlation (R= 0.999) between the CT numbers obtained both manually and via RIT, for the various density inserts.
Conclusion: Semi-automatic image analysis software provides a viable and reproducible method for monthly CBCT QA analysis that can produce reliable data for analysis of system performance. The degree of added reproducibility depends on the module analyzed. Future work includes analysis considering both inter- and intra-user variability.