Treatment plan optimization for rotating-shield brachytherapy

In this thesis, we aim to develop fundamentally new techniques and algorithms for efficiently computing rotating-shield brachytherapy (RSBT) treatment plans.

We propose that these algorithms will pave the way for making RSBT available in clinical practices. RSBT is an intensity modulated high-dose-rate brachytherapy (HDR-BT) technique. Theoretically, RSBT offers advantages over the conventional HDR-BT. Although this technique is promising in theory, its application in practice is still at an early stage. The RSBT technique entails rotating a radiation-attenuating shield about a brachytherapy source to directionally modulate the radiation in an optimized fashion. The unshielded brachytherapy source used in conventional HDR-BT delivers radially symmetric dose distributions, thus the intensity modulation capability of the conventional HDR-BT is limited. With the capability of making anisotropic radiation, RSBT will revolutionize the brachytherapy technique through superior dose conformity, increased flexibility and inherent accuracy. Due to the enhanced power of intensity-modulation, RSBT will also enable dose escalation without increasing toxicity to the organs-at-risk, thus improving quality of life for millions of cancer patients. Although the first conceptual RSBT method was proposed more than ten years ago, there are still tremendous challenges for applying it in clinical practices. Creating efficient and automated treatment planning system is one of the major technical obstacles for making RSBT deliverable in the clinic. The time-critical nature of the application significantly increases the difficulty of RSBT treatment planning, demanding innovative techniques for information integration. Therefore, we propose that fundamentally novel technology and algorithms for RSBT treatment planning can make RSBT clinically accessible. The fundamental concept used for this thesis is to decompose the dose optimization step for RSBT treatment planning into two steps, namely anchor plan optimization and optimal sequencing. The degree of freedom in anchor plan optimization is controlled at a low level compared to single-step dose optimization, and the optimal sequencing algorithms can efficiently calculate treatment plans by reusing the solutions from anchor plan optimization. Thus, by decomposing the dose optimization, the computational complexity in the two-step method is greatly reduced compared to the single-step method. In the anchor plan optimization, an abstract RSBT delivery model is assumed. The abstract RSBT delivery model assumes that only beams with fixed small azimuthal emission angle, which are called beamlets, will be used during the delivery. An anchor plan is created based on this assumption that only these beamlets will be used. Generally, an anchor plan will be of high quality in the sense of dose distribution, but of low quality in the sense that it has prohibitory long delivery time. In the optimal sequencing step, beamlets will be superposed into beams to reduce the delivery time. By limiting the delivery time to a clinically acceptable level, the anchor plans turn into deliverable plans. Unlike anchor plan optimization, where an abstract RSBT delivery model is assumed, the optimal sequencing step depends on more concrete RSBT delivery models. Specifically, we will study three methods of RSBT, namely the single rotating-shield brachytherapy (S-RSBT), the dynamic rotating-shield brachytherapy (D-RSBT) and the paddle rotating-shield brachytherapy (P-RSBT). We proposed a novel anchor plan dose optimization method as well as novel optimal sequencing methods for each of the RSBT delivery methods studied in this work. We have implemented all the proposed algorithms and experimented with them using real medical data. With the methods proposed in this thesis, the optimization time for creating delivery plans can be controlled within 15 minutes based on the data from our experiments. Compared to the conventional brachytherapy techniques, the three methods studied in this work can produce more conformal dose distributions at an acceptable level of delivery time increase. With 15 min/fx delivery time, S-RSBT, D-RSBT and P-RSBT averagely increased the D90 (the minimum dose received by the hottest 90% of the tumor) by 17, 9 and 5 Gy compared to conventional interstitial plus intracavitary brachytherapy, whose D90 is 79 Gy. The best choice depends on the specified delivery time or quality requirement, as well as the complexity of building the equipment. Roughly speaking, among the three RSBT methods studied in this thesis, P-RSBT has the most complex applicators as well as the highest plan qualities. S-RSBT has the simplest applicators, and its plan qualities is generally better than D-RSBT with limited delivery time (/fx). With sufficient delivery time (~30 min/fx), D-RSBT may be considered as the best solution in the sense of balancing the complexity of applicators and the dose qualities.