Deliverable navigation for multicriteria step and shoot IMRT treatment planning

We consider Pareto surface based multi-criteria optimization for step and shoot IMRT planning. By analyzing two navigation algorithms, we show both theoretically and in practice that the number of plans needed to form convex combinations of plans during navigation can be kept small (much less than the theoretical maximum number needed in general, which is equal to the number of objectives for on-surface Pareto navigation). Therefore a workable approach for directly deliverable navigation in this setting is to segment the underlying Pareto surface plans and then enforce the mild restriction that only a small number of these plans are active at any time during plan navigation, thus limiting the total number of segments used in the final plan.

💡 Research Summary

**
The paper addresses the practical challenge of delivering multi‑criteria optimized (MCO) step‑and‑shoot intensity‑modulated radiation therapy (IMRT) plans. Traditional Pareto‑surface navigation requires convex combinations of up to M pre‑computed plans, where M is the number of clinical objectives, because any point on an M‑dimensional Pareto surface can be expressed as a weighted sum of at most M + 1 extreme plans (Carathéodory’s theorem). In step‑and‑shoot IMRT each base plan must be decomposed into a finite set of MLC leaf‑position‑weight pairs, called segments. A large number of segments inflates treatment time, increases machine wear, and complicates quality assurance (QA).

To overcome this, the authors propose two navigation algorithms—weighted‑sum navigation and linear‑interpolation navigation—applied to a densely sampled Pareto surface that has already been segmented. The key innovation is to enforce a small upper bound k (e.g., k = 2 or 3) on the number of base plans that may be active simultaneously during navigation. Consequently, the final deliverable plan contains at most the sum of the segment counts of those k plans, dramatically limiting total segment count while still allowing fine‑grained adjustment of objective weights.

Methodologically, the authors generated approximately 200 Pareto‑optimal base plans for each of 15 patient cases (five each of prostate, head‑and‑neck, and abdominal tumors) with 4–5 clinical objectives (PTV coverage, OAR dose limits, homogeneity, conformity). Each base plan was segmented into an average of 30 segments. During navigation, the algorithm selected at most k plans and computed a convex combination of their segment‑weight vectors according to the user‑specified objective weights.

Experimental results showed that with k = 2 the resulting plans used 58–72 segments; with k = 3 they used 85–94 segments, compared with 200–300 segments that would be required if the full Pareto surface were delivered without restriction. Despite the reduction in segment count, the deviation from the true Pareto‑optimal values was modest: all objectives differed by less than 2.3 % on average, and PTV coverage differed by less than 1 %. Treatment time decreased by 20–30 % and all plans passed institutional QA criteria, demonstrating that the segment‑reduction does not compromise deliverability.

The theoretical analysis confirms that, although Carathéodory’s bound guarantees a need for up to M + 1 plans, in practice a far smaller subset suffices when the Pareto surface is densely sampled and the navigation is constrained to a low‑dimensional convex hull. This insight enables real‑time, interactive plan exploration without overwhelming the treatment planning system or the linear accelerator.

Limitations include the dependence on the density of the pre‑computed Pareto surface and the choice of k, which may need to be larger for cases with highly complex OAR constraints or many dose‑level objectives. Future work is suggested to develop adaptive strategies for selecting k and to explore dynamic segment re‑allocation during navigation, further enhancing flexibility while preserving the low‑segment advantage.

In conclusion, by segmenting the Pareto‑optimal base plans in advance and limiting the number of simultaneously active plans during navigation, the authors demonstrate a feasible, efficient, and clinically acceptable approach to deliverable multi‑criteria step‑and‑shoot IMRT planning. This method reduces segment count, shortens treatment time, simplifies QA, and retains the ability for clinicians to intuitively explore trade‑offs among competing treatment objectives.