Yunlong Liu1, Ryan T Flynn2, Yusung Kim2, Xiaodong Wu3. 1. Department of Electrical and Computer Engineering, University of Iowa, 4016 Seamans Center, Iowa City, Iowa 52242. 2. Department of Radiation Oncology, University of Iowa, 200 Hawkins Drive, Iowa City, Iowa 52242. 3. Department of Electrical and Computer Engineering, University of Iowa, 4016 Seamans Center, Iowa City, Iowa 52242 and Department of Radiation Oncology, University of Iowa, 200 Hawkins Drive, Iowa City, Iowa 52242.
Abstract
PURPOSE: It is important to reduce fluence map complexity in rotating-shield brachytherapy (RSBT) inverse planning to improve delivery efficiency while maintaining plan quality. This study proposes an efficient and effective RSBT dose optimization method which enables to produce smooth fluence maps. METHODS: Five cervical cancer patients each with a high-risk clinical-target-volume (HR-CTV) larger than 40 cm(3) were considered as the test cases. The RSBT source was a partially shielded electronic brachytherapy source (Xoft Axxent™). The anchor RSBT plans generated by the asymmetric dose-volume optimization with smoothness control (ADOS) method were compared against those produced by the dose-surface optimization (DSO) method and inverse-planning with simulated annealing (IPSA). Either L1-norm or L2-norm was used to measure the smoothness of a fluence map in the proposed ADOS method as one weighted term of the objective function. Uniform dwell-time scaling was applied to all plans such that HR-CTV D90 was maximized without violating the D2cc tolerances of the rectum, bladder, and sigmoid colon. The quality of the anchor plans was measured with HR-CTV D90 of the anchor plans. Single-shielded RSBT [(S-RSBT), RSBT with single, fix sized delivery window] and dynamic-sheilded RSBT [(D-RSBT), RSBT with dynamically varying sized delivery window] delivery plans generated based on the anchor plans were also measured, with delivery time constraints of 10, 20, and 30 min/fraction (fx). RESULTS: The average HR-CTV D90 values of the anchor plans achieved by the ADOS, DSO, and IPSA methods were 111.5, 94.2, and 107.4 Gy, respectively, where the weighting parameter β used in ADOS with L2-norm was set to be 100. By using S-RSBT sequencing and 20 min/fx delivery time, the corresponding D90 values were 88.8, 81.9, and 83.4 Gy; while using D-RSBT sequencing with 20 min/fx delivery time, the corresponding D90 values were 91.4, 88.3, and 78.9 Gy, respectively. The average optimization times for ADOS, DSO, and IPSA were, respectively, 77, 4, and 1800 s. By using L1-norm instead of L2-norm in the ADOS method, the optimization time was increased by 20 s, while the D90 was reduced by 6.8 Gy on average. ADOS-L1 was found to be more sensitive to the weighting parameter than ADOS-L2. If β was increased to 10 000, the D90 drop with ADOS-L1 was 38 Gy, while the drop with ADOS-L2 is 13 Gy. CONCLUSIONS: The ADOS method had a reasonable optimization time cost, while achieving comparable RSBT dose plans as the IPSA method, which is of much higher time complexity. Compared to the DSO and IPSA methods, ADOS is able to generate anchor plans which are more suitable for RSBT delivery while preserving the high quality of the original plans. Compared to ADOS-L1, ADOS-L2 is able to achieve better quality of anchor plans more efficiently.
PURPOSE: It is important to reduce fluence map complexity in rotating-shield brachytherapy (RSBT) inverse planning to improve delivery efficiency while maintaining plan quality. This study proposes an efficient and effective RSBT dose optimization method which enables to produce smooth fluence maps. METHODS: Five cervical cancerpatients each with a high-risk clinical-target-volume (HR-CTV) larger than 40 cm(3) were considered as the test cases. The RSBT source was a partially shielded electronic brachytherapy source (Xoft Axxent™). The anchor RSBT plans generated by the asymmetric dose-volume optimization with smoothness control (ADOS) method were compared against those produced by the dose-surface optimization (DSO) method and inverse-planning with simulated annealing (IPSA). Either L1-norm or L2-norm was used to measure the smoothness of a fluence map in the proposed ADOS method as one weighted term of the objective function. Uniform dwell-time scaling was applied to all plans such that HR-CTV D90 was maximized without violating the D2cc tolerances of the rectum, bladder, and sigmoid colon. The quality of the anchor plans was measured with HR-CTV D90 of the anchor plans. Single-shielded RSBT [(S-RSBT), RSBT with single, fix sized delivery window] and dynamic-sheilded RSBT [(D-RSBT), RSBT with dynamically varying sized delivery window] delivery plans generated based on the anchor plans were also measured, with delivery time constraints of 10, 20, and 30 min/fraction (fx). RESULTS: The average HR-CTV D90 values of the anchor plans achieved by the ADOS, DSO, and IPSA methods were 111.5, 94.2, and 107.4 Gy, respectively, where the weighting parameter β used in ADOS with L2-norm was set to be 100. By using S-RSBT sequencing and 20 min/fx delivery time, the corresponding D90 values were 88.8, 81.9, and 83.4 Gy; while using D-RSBT sequencing with 20 min/fx delivery time, the corresponding D90 values were 91.4, 88.3, and 78.9 Gy, respectively. The average optimization times for ADOS, DSO, and IPSA were, respectively, 77, 4, and 1800 s. By using L1-norm instead of L2-norm in the ADOS method, the optimization time was increased by 20 s, while the D90 was reduced by 6.8 Gy on average. ADOS-L1 was found to be more sensitive to the weighting parameter than ADOS-L2. If β was increased to 10 000, the D90 drop with ADOS-L1 was 38 Gy, while the drop with ADOS-L2 is 13 Gy. CONCLUSIONS: The ADOS method had a reasonable optimization time cost, while achieving comparable RSBT dose plans as the IPSA method, which is of much higher time complexity. Compared to the DSO and IPSA methods, ADOS is able to generate anchor plans which are more suitable for RSBT delivery while preserving the high quality of the original plans. Compared to ADOS-L1, ADOS-L2 is able to achieve better quality of anchor plans more efficiently.
Authors: Yunlong Liu; Ryan T Flynn; Yusung Kim; Hossein Dadkhah; Sudershan K Bhatia; John M Buatti; Weiyu Xu; Xiaodong Wu Journal: Med Phys Date: 2015-10 Impact factor: 4.071