ESTRO 2024 - Abstract Book

S4747

Physics - Quality assurance and auditing

ESTRO 2024

Secondly, the analysis results obtained from the in-house tool were accessed by correlating them with the gamma analysis results obtained from ArcCheck measurement, specifically for the MLC positional error introduced plans. Those plans were selected from clinical treatment plans in different treatment sites, namely Abdomen, Cranium, Head & Neck, Gynaecological Pelvic, and Prostatic Pelvic. These introduced errors (σ) for the open MLC were normally distributed with standard deviation ranging from 0.0 to 3.0mm with increments of 0.5mm. Finally, a receiver operating characteristic (ROC) analysis was performed to determine the capability of the in-house tools in detecting different magnitudes of MLC positional errors. The area under curve (AUC) was also calculated to evaluate the performance of the in-house tool. The optimal threshold of mean MLC positional error was determined by identifying the point on the ROC that was nearest to the coordinate (0, 1), where the maximum values for sensitivity and specificity were achieved.

Results:

The PF results in general agrees with our in-house QA tool results. A strong association has been discovered between the deviations in the junction position of the PF test and the occurrence of an MLC machine interlock event. It is worth noting that the deviation tends to escalate as it gets close to the MLC machine interlock event. As depicted in Figure 1, the average MLC deviation reaches its minimum value after a successful maintenance procedure triggered by an

MLC machine interlock event.

For all the 6 unmodified plans and 36 σ-introduced plans, their corresponding gamma analysis results were plotted against the mean MLC positional error obtained from the in-house tool. All the σ-introduced modified plans exhibit a linear relationship, indicating a close correlation between the mean MLC positional error and the gamma analysis result. Figure 2a pictures the ROC obtained by varying the passing threshold of mean MLC positional error for each introduced σ. Subsequently, the AUC is plotted against σ in Figure 2b, revealing that an excellent sensitivity and specificity in detection of all σ of interest. Figure 2c plotted the determined ideal mean MLC positional error thresholds when implementing the tool for MLC positional error detection.