Dosimetric Quality of An Automatic IMRT Planning Method for Head and Neck Cancer Cases
L Yuan1*, T Pang2, Y Ge3, T Li4, F Yin5, Q Jackie Wu6, (1) Duke University Medical Center, Durham, NC, (2) Peking Union Medical College Hospital, Beijing, Beijing, (3) University of North Carolina at Charlotte, Charlotte, North Carolina, (4) Duke University Medical Ctr., Durham, NC, (5) Duke University Medical Center, Durham, NC, (6) Duke University Medical Center, Durham, NCTH-C-137-8 Thursday 10:30AM - 12:30PM Room: 137
Purpose: To evaluate the dosimetric quality of an automatic IMRT planning method for Head and Neck (HN) cancer cases.
Methods: Eighteen HN cancer patients are planned with nine equally spaced IMRT beams. The OAR dose sparing objectives for each plan (primary and boost) are generated automatically by the OAR DVH prediction models based on the patient anatomy. The DVH models are trained by prior IMRT plans. The predicted DVHs are sampled and transformed into a template based dose-volume constraints and optimization priorities which are utilized by the plan optimization algorithm. Only one round of optimization was performed for each case without any human intervention. The dosimetric parameters of the OARs and PTV in the automatically generated plans (auto-plans) are evaluated against those from the clinical plans by a paired comparison method.
Results: The average PTV dose homogeneity (D2%-D99%) in the auto-plans is 5.5 Gy, which is comparable with that in the clinical plans (5.4 Gy). For the OARs, the means (standard deviation in parenthesis) of the differences of the dosimetric parameters in the auto-plans compared with the clinical plans are (in Gy): parotids D50%: -2.5 (9.4), oral cavity D50%: -0.9 (3.5), larynx D50%: -1.3 (9.7), pharynx D2%: -0.3 (3.6), spinal cord D2%: -2.2 (4.0), brainstem D2%: -0.9 (3.9), mandible D2%: -0.4 (1.3). The negative mean values indicate the OARs can be further spared in average. However, the differences of the dosimetric parameters are not statistically significant for all the other OARs except spinal cord (p=0.002), which is due to large spinal cord dose reduction in several auto-plans even when the spinal cord dose in the corresponding clinical plans have met the physician's dose sparing constraint.
Conclusion: The automatic IMRT planning method utilizing the DVH prediction models can generate plans with dosimetric quality comparable to clinical IMRT plans generated by experts.
Funding Support, Disclosures, and Conflict of Interest: Partly supported by NIH/NCI and a Varian master research grant.