Purpose: The purpose of this work is to generate an atlas for the automatic segmentation of contours for post-prostate seed implant evaluation and to investigate the dosimetric impact of using these contours for post-plan analysis.
Methods: A prostate atlas was generated in the MIM software package using expert (radiation oncologist) defined contours for the bladder, rectum, and prostate for 40 patients. A workflow was developed that implemented several pre-processing steps to improve the contouring accuracy. Processing steps included the automated detection of metal (seeds) and air and masking these image values in order to allow the atlas registration framework to focus on the patient anatomy and not on the metal/air that could bias the result. Post-plans from 10 additional patients were exported from the clinically implemented VariSeed TPS to MIM for dosimetric analysis. The RTplan from VariSeed was used to calculate a new dose distribution in MIM using the same seed locations and source strengths. Relevant dosimetric parameters were calculated using both the expert and atlas defined contours.
Results: An atlas and workflow was successfully generated in MIM for the automatic segmentation of the prostate, rectum, and bladder. Minimal dosimetric differences were observed between VariSeed and MIM when using expert defined contours, consistent with previously published results. When atlas generated contours were used, there were larger differences in dosimetric parameters for the prostate and rectum. The average difference in prostate volume, V₁₀₀, and D₉₀ was -0.3+/-4.3 cm³, -0.7+/-1.4%, and -1.3+/-6.6%, respectively. The average difference in V₅₀ for the rectum was 0.6+/-2.6 cm³.
Conclusion: Despite large differences in dosimetric parameters observed when using atlas defined contours for post-plan analysis, the magnitude of these differences are generally small with respect to clinical tolerances. In addition, the improved efficiency of using automatically generated contours has the potential to significantly improve clinical workflow.