Fully automated treatment planning of spinal metastases – A comparison to manual planning of Volumetric Modulated Arc Therapy for conventionally fractionated irradiation

Fractionated or single-dose radiotherapy to spinal metastases of solid tumors is one of the most frequently performed radiation treatments [1]. Highly conformal radiation techniques such as Stereotactic Body Radiation Therapy (SBRT), Intensity Modulated Radiation Therapy (IMRT) or VMAT are increasingly being used as their application becomes easier and dose distributions confer theoretical advantages [2], albeit so far without proof of superior clinical outcome. On the other hand, generation of an optimal IMRT or VMAT plan is often time consuming and it has been shown that staff limitations are correlated with restricted use of new techniques such as IMRT even in the developed world [3, 4]. Automated planning of IMRT and VMAT may reduce the workload which is associated with manual “trial-and-error” approaches by around 50% [5].

Erasmus-iCycle, developed at the Erasmus MC-Cancer Institute, is an optimizer for multicriterial beam profile optimization and beam angle selection for coplanar and non-coplanar IMRT [5‐10]. Other solutions for multicriterial beam angle optimization have been proposed by Schreibmann et al. [11], Craft and Monz presented a full multibeam space Pareto navigation tool [12]. However, both are a posteriori methods, i.e., the algorithm generates different sets of beam angles and intensity profiles from which the user selects the plan afterward. The Erasmus-iCycle solution is an a priori approach which enables the user to define a site specific set of criteria (wish-list) which may not be violated (constraints) or have to be met as well as possible, or better (objectives). Objectives have assigned priorities to steer the multicriterial planning towards favourable trade-offs between the various treatment goals. If the priority of an objective is higher, the probability that the corresponding objective is met increases. Hard constraints are always respected in Erasmus-iCycle plans.

Beam directions are selected from candidate directions which can be restricted, e.g., in case of coplanar treatments [6, 7, 13]. For fully automated generation of plans that are clinically delivered, Erasmus-iCycle auto plans are automatically reconstructed and segmented in the clinical treatment planning system (TPS) [13]. This study intends to validate VMAT plans for treatment of spinal metastases, generated with this approach, and to compare the quality of automatically generated plans with plans that were manually created by experienced treatment planners. Different spine regions pose different optimization problems as a consequence of different OARs being relevant for the treatment. Therefore cervical, thoracic, and lumbar targets were included in the study design.

All automatically created plans were clinically deliverable and acceptable. Figures 1 and 2 show dose distributions, and dose-volume histogram comparisons for a kidney-level target, and a lung-level target, respectively.

Though highly conformal radiotherapy is not mandatory in the treatment of painful bone metastases [16], intensity modulation with static beams or VMAT allows to deliver a highly focused dose distribution in every clinical situation and may be beneficial for patients with oligometastatic disease [17] who require dose escalation or patients who had received prior courses of radiotherapy limiting doses to OARs in case of reirradiation [18, 19]. Highly conformal radiotherapy is, however, not always applied when appropriate [3] for its perceived resource intensity, both in personnel and planning time. Unexpectedly, the reason for not performing advanced radiotherapy techniques despite a clear indication seems to be a shortage of sufficiently trained personnel rather than machine shortage [3, 4, 20]. AlDuhaiby et al. described in a Canadian survey in 2012 that limitations in CT simulator or linear accelerator configuration impaired IMRT implementation in only 10% of cases [20]. The most relevant factors hindering IMRT implementation was the need to train existent treatment planners (50%) as well as the need to hire more planners (30%) [20]. Personal shortage is even more relevant in VMAT settings as these are more time consuming as compared to fixed field IMRT [21]; “increased planning time required for generating the VMAT plans” has been described by Rao et al. as the main disadvantage of VMAT [22]. Therefore, there is a need for tools that reduce planning time and ease the process of developing a radiotherapy plan without compromising plan quality.

Various auto-planning strategies are currently being developed [8], these include approaches for beam angle selection (e.g., [23]), and integrated beam weight optimization algorithms. The latter can be based on global optimization [24], or on sequential beam selection [25]. As discussed above, multicriterial optimization systems have been proposed [11, 12], that enable the user to select the plan out of different sets of beam angles and intensity profiles after optimization (a posteriori setting). Template based auto-planning solutions are already commercially available and have been applied to various target paradigms such as head and neck [26], breast [27] and esophageal targets [28].

Erasmus-iCycle plans, which provide a template-free approach have been successfully applied to head and neck [29], prostate [5], and cervical cancer [13] target volumes to create clinically deliverable plans.

Our results show that fully automated VMAT treatment planning with the Erasmus-iCycle/Monaco system for spinal metastases was non-inferior to conventional treatment planning by expert dosimetrists or medical physicists. Automatically generated plans with this system outperformed manual plans in terms of sparing of OARs in 3 of 4 predefined organs and also PTV coverage was favorable when compared to manual plans. We observed one hotspot in an iCycle plan in the spinal cord (33.2 Gy, D_50%: 30 Gy). This problem can be addressed by modifying the wish list to include a maximum dose cost function on the spinal cord. In our analysis, however, we applied a quadratic overdose cost function which lead to maximum doses that were comparable to maximum doses observed in manual plans in all but this patient and allowed for more degrees of freedom regarding dose reduction to other OAR. If the maximum dose has to be strictly controlled, e.g., in case of reirradiation, a maximum dose cost function should be applied. Otherwise, the situation would be evaluated during the manual approval process which might prompt occasional replanning with a maximum dose cost function. Monitor units and estimated treatment time were higher in auto plans; however, this difference was not considered clinically relevant in the treated patient population and resulted in a negligible prolongation of treatment times. The difference in treatment times was expected as the aim of this study was to generate high-quality plans with clinically acceptable delivery times. Monitor units and delivery times could be reduced in the auto-plans if required. In contrast to a study in a more complex setting such as head and neck cancer [29], the observed dosimetric advantages of Erasmus-iCycle for treatment of spinal metastases were on average of low clinical relevance although advantages across the population were also observed.

External audits have shown that experienced treatment centers may yield superior IMRT plans [30]. Therefore, as postulated most recently by Fogliata et al. [28], quality improvements of automated planning vs. manual plan generation may even be more pronounced in target paradigms that are not frequently treated in a particular center.

Our data add to the growing evidence [5, 13, 29] that automated treatment planning might be an alternative to manual planning, reducing the workload of medical phyicists and dosimetrists while maintaining or improving plan quality.