In the pursuit of perfect planning: comparison between Lightning Inverse Planning and GammaPlan Wizard for gamma knife radiosurgery

In the 38 case simulations, M+OP and IP were not always able to generate a plan with coverage greater than 95% but did present better selectivity. Although LWR presented a median TP shorter then LNR, this was only observed in the VS patients. LWR had greater selectivity and lower GI, but at the cost of a longer DT. (Figure 1; Table 1).

FIGURE 1.

Comparison of the planning time according to each modality and pathology.

GBM = glioblastoma multiform; IP = Inverse planning; LNR = Lightning^® with no consideration of risk structure; LWR: Lightning^® with consideration of risk structure; M+OP = Manual planning with optimization; Men = meningioma; Met = metastasis; Pit And = pituitary adenoma; VS = vestibular schwannoma

The median results of the plans using Lightning^® and the optimization from LGP, showed that the time for planning had a reduction of 57%, coverage improvement of 8%, GI improvement of 15%, delivery time was 5% faster, selectivity decreased 12% and number of shots increased by 226%. (Table 2).

The planning results using Lightning^® with and without OAR had similar results, being better than the other techniques, but with more shots and longer DT. Inverse planning and manual planning had approximately 7 times less shots but resulted in compromise in coverage, 0.88 and 0.91 respectively (Table 3). OAR protection is shown in Figure 2.

FIGURE 2.

Results of the dose received by the cochlea’s Modiolos in a vestibular schwannoma plan.

IP = Inverse planning; LNR = Lightning^® with no consideration of risk structure; LWR = Lightning^® with consideration of risk structure; M+OP = Manual planning with optimization

IP and LNR had the quickest time, 00:44,4 and 00:36,4 respectively, but Lightning^® had greater coverage. Although LWR took longer to plan than inverse planning, it had better coverage and the benefit of OAR protection (Table 3). OAR protection is shown in Figure 3 A and B.

FIGURE 3.

Dose received by optic structures in one (A) and three (B) sessions plans for meningiomas. Dose received by optic structures in one (C) and three (D) sessions plans for pituitary adenomas.

IP = Inverse planning; LWR = Lightning^® with consideration of risk structure; LNR = Lightning^® with no consideration of risk structure; M+OP = Manual planning with optimization

From the 12 patients with metastasis in this analysis, none needed defined OARs, therefore, they were evaluated with only 3 techniques. Inverse planning was slightly quicker (44,1 sec), with 90% selectivity as the manual planning but with higher GI and DT then Lightning^®. LNR had the best coverage, GI, and DT (Table 3).

Ten cases of post-surgical cavities were analysed. Lightning^® was quicker, with better coverage, similar GI, and faster treatment time. Inverse planning and manual planning had similar coverage, selectivity, and the number of shots. Although inverse planning had best GI, it had the longest DT (Table 3)

Pituitary adenomas can provide great challenges due to the tumor’s proximity to optic structures. LWR delivered a plan in 42,8 sec with 99% coverage, GI of 2.65 and DT of 49 minutes. Wizard^® not only took longer to plan, but also didn’t achieve adequate coverage and had longer treatment time (Table 3). OAR protection is shown in Figure 3 C and D.

None of the GBM cases had OAR and only 3 modalities were used. Both inverse planning technics had similar TP, but Lightning^® had approximately 5 times the number of shots with higher DT.

On Figure 1 we can notice a close similarity on both Lightning^® plans with the tendency of increased number of shots but, differently from GBM, a quicker delivery time.

The time to plan was overall faster with Lightning^® as shown on Figure 1 and Table 2. An increased time to plan was noted in surgical cavities and in plans with constraints on OAR that was attributed to a greater complexity in these kind of lesions with direct implications on the calculation overhead on the planning process. Since the volume and shape can challenge the calculation engine, but only volume can be objectively quantified; Pearson correlation (PC) coefficients between tumor volume and time to plan for the four modalities were calculated being statistical significant on IP and LNR with r=.74 and .84 respectively, both p<0.0001 and M+OP (r=.03, p=0.8) and LWR (r=.45, p=0.2) not meeting statistical significancy. Their scatter plots are shown in Figure 4.

FIGURE 4.

Scatter plot correlation between time to plan in minutes and tumor volume. (A): Lightning^® with no consideration of risk structure (r = 0.84, p < 0.0001); (B): Lightning® with consideration of risk structure (r = 0.45, p = 0.2); (C): Inverse planning (r = 0.74, p < 0.0001); (D): Manual planning with optimization (r = 0.03, p = 0.8).

OAR = organs at risk

The presented DVH are a comparison of LWR and LNR to a deliverable forward plan. The curves show the similarity of their quality, with adequate parameters for safe treatment delivery. First case is a clival meningioma (Figure 5 A and B), where LWR had max dose to the optic of 7.25Gy while the treatment had 8 Gy. Lightning^® was also able to keep greater coverage. The same pattern was observed in the second case, a Vestibular Schwannoma, where LWR had greater coverage and greater Modiolus protection (Figure 5 C and D).

FIGURE 5.

Dose-volume histogram (DVH) for clival meningioma (A and B) plan comparing Lightning^® with (LWR) and without risk (LNR) structures to a simulation of treatment planning with forward planning with (A) showing tumor results, and (B) showing the organs at risk (OAR) results. Dose-volume histogram for clival meningioma plan (C and D) comparing LWR and LNR to a simulation of treatment planning with forward planning with and without consideration of the OAR. (C) showing tumor results, and (D) showing the OAR results.

The Wizard^® function was introduced in LGP version 5.34 back in 2000, marking it as the first inverse planning tool available for Gamma Knife in the market. However, this early version was limited by its simplistic approach to solving a few objective functions to find the lowest penalty solution, and it was prone to getting stuck in a subpar solution (local minima) as it was not modelled as a convex problem. It did not adequately account for organs at risk (OARs). Adoption of the Gradient Index concept, leading to the development of a more sophisticated inverse planning tool in LGP version 10 (2010). This updated software promised to enhance clinical workflows, especially for planning larger, spherical targets in non-critical areas. Because the objectives use relative isodoses (instead of absolute doses) and that the positions of the shots can change, the optimization problem even in this version of the Wizard^® remained inherently difficult (nonconvex), unable to manage multiple targets and to enforce criteria such as the maximum dose to an OAR.¹ Nonetheless, for more intricate targets, manual planning remains the preferred approach, especially among expert planners.²

Lightning^® models the inverse planning as a convex problem which is both faster and less prone to getting stuck in sub-optimal solutions.¹ Wizard^® used repositioning, relative weight, and collimator settings inside a predetermined isodose line and could delete unnecessary shots to reach the best solution. Different settings were used to penalize according to coverage, selectivity, gradient and delivery time depending on the planner’s objective which can limit the quality of the plan 1 but, when used properly by experienced planners, it can improve.³ Also, a selection of fixed collimators and fixed position can be set but limiting the field delineation.

Lightning^® achieves convexity by dividing the planning into: isocenter placement, optimization, and sequencing. The isocenters chosen in the first phase remain fixed throughout the rest of the planning. In the optimization phase, competing objectives are combined as a weighted sum. By altering weights, it is possible to explore achievable trade-offs. Possible objectives include dose to target, sparing of OARs, and a new – highly efficient – Beam on Time (BOT) penalization. During the optimization, times for each sector and collimator are allowed to vary independently. In the sequencing phase, these times are converted into deliverable shots. Altogether this algorithm tries to find the best configuration based on the given dose, defines the best isodose line and converts the “dose cloud” into shots. It is also able to plan multiple lesions simultaneously while considering their dose contributions, differently from Wizard^® that is uncapable of doing these actions. The settings can be penalized to BOT and ‘lower acceptable dose’, that can be used to push for a desired isodose line by the planner.

In this beta version the number of isocenters was 3 times higher. Two patterns were observed. First, high weighted shots were surrounded by several low weight shots that helped a better delineation of the dose in complex-shaped targets. Second, overlapping shots with different collimator settings and weights, not only helped the dose delineation, but also helped decrease delivery time and allows higher IDL with appropriate fall off of the dose without jeopardizing the V12.⁴ The ability to deliver a high conformal plan in less time would diminish the ability of the tumor cells to repair⁵ affecting tumor control. Decreasing the time that the patient waits to be treated and shortening the delivery time directly impacts the patient’s tolerance and the daily dynamics of the department.

PC showed that IP and LNR had good positive correlation of target volume with planning time, but the weaking of the correlation presented with LWR can be explained by necessity to shape the dose for maximal OAR protection. The low number of shots in manual plans was perhaps an explanation of a lack of direct correlation between TP and lesion volume in the M+OP group.

Planning quality discrepancies are related to the experience of the planner and can be minimized by Lightning^®.⁶ New planners and new centers can use this software as a learning tool to improve their skills in planning, especially when OAR places limitation in keeping adequate coverage. This allows a standardization of the plans, leading to better evaluation of the treatment in multi-center studies.

The preservation of function must be considered during planning. For this reason, the amount of dose received by important structures like the cochlea/modiolus, and the optic structures must be quantified. Lightning^® can plan according to a planner-specified maximum OAR tolerance dose. In the previous version of the LGP, OAR should be protected with a new definition of the shots by the planner or by the “dynamic shaping” option. In this analysis, this tool was not used. Different software solutions have been proposed with the purpose of increase IP efficacy^2,7,8,9 each one achieving different results, but not always being successful in improving OAR protection and negatively impacting coverage and DT, which can be an issue since the biological effective dose for ultra-shortterm repair will be compromised.⁵ LWR was able to maintain adequate coverage while protecting the OAR in all tested plans while the other modalities only achieved protection for optic structures in pituitary tumors treated in 3 fractions as demonstrated in Figure 4, which is likely attributable to the anatomy of the cases and dose prescribed, not the optimization capacity.