Purpose: To investigate the use of Statistical Process Control (SPC) to monitor use of a web-based auto-contouring and planning tool.
Methods: The Radiation Planning Assistant (RPA) is such a tool that provides plans calculated on standard linacs, which must then be recalculated by the user in their own treatment planning system for their own linac. Thirty-three 4-field box cervix plans (8 structures) and 34 VMAT head and neck plans (30 structures) were generated by the RPA. These plans were downloaded and imported into Raystation v10B where they were recalculated using fixed MUs. The final plans were uploaded back to the RPA where the differences in mean dose for each structure was calculated. Utilizing SPC, control charts were generated with mean and control limits at the 3 standard deviation. Mean percent dose differences that fell outside of our control limits were removed from the calculation of the mean and standard deviation.
Results: Head and neck mean percent dose difference and control limits varied per structure. For example, PTV1 (2.28%, [1.00%, 3.56%]), PTV2 (2.19%, [5.59%, -1.21%]), PTV3 (1.67%, [-0.67%, 4.01%), and spinal cord (0.59%, [1.09, 0.09%]), and similarly, for cervix plans, L5 contour (-2.59%, [-3.43%, -1.74%]) and right femoral head (-1.79%, [-0.77%, -2.81%]). This indicates that SPC will give more individualized results per contour than can be achieved with a fixed threshold (e.g., 5%). Variability in SPC thresholds may be partially attributed to differences in dose calculations near the surface, particularly for targets near the surface for head and neck VMAT cases, and to differing body contours between Eclipse and Raystation.
Conclusion: Specific control limits for individual contours based on percent mean dose difference can be determined and differ across anatomical sites. More study is needed to understand the underpinning of these differences and if they are effective in indicating plan errors.
Statistical Analysis, Quality Assurance