Patient Selection and Structure Definition
Ten patients with PTV lengths between 49.8 and 85.0 cm, with media length of 66.5 cm, who received CSI were retrospectively included in this study. All the patients were positioned supine, with arms resting at both sides and immobilized by thermoplastic masks in our regular procedure. Computed tomography (CT) images were acquired using a 16-slice GE Revolution CT scanner (General Electric Medical Systems, USA) of 3-mm thickness and transferred to the RayStation TPS, version 9A. All plans were designed for an Elekta Versa HDTM Linac (Elekta Medical System, UK) equipped with a high-resolution multileaf collimator with a leaf width of 5 mm at the isocenter.
The cranial clinical target volume (CTV) included the whole brain within the inner table of the skull, while the spinal CTV included the entire subarachnoid space that encompasses the extensions along with the nerve roots laterally. The cranial PTV (PTV_C) was created by expanding the cranial CTV by 5 mm, while the spinal PTV (PTV_S) was created by expanding the spinal CTV by 7 mm. The OARs to be delineated included the eyes, lenses, cochlea, parotid, salivary glands, larynx, esophagus, thyroid gland, breasts in females, lungs, heart, kidneys, intestines, and gonads. The actual prescription doses were 23.4 - 36 Gy to PTV with 13 - 20 fractions for the ten patients. However, the same dose of 23.4 Gy was prescribed to each patient with 13 fractions, and every plan was normalized to 95% of PTV in this retrospective study.
Automatic CSI Planning Workflow
An automatic CSI planning script based on Python language was created in RayStation TPS. The workflow was described below and showed in Figure 1.
Isocenter setting and beamsets arrangement
The number of isocenters depends on the length of PTV_S, which is denoted as “L.” If L was < 38 cm, two isocenters were needed, wherein the first one was placed in PTV_C and the second in the center of PTV_S. If L was between 38 cm and 42 cm, two isocenters were still needed; however, the second one was moved 10 cm toward the patient’s back. If L was between 42 and 72 cm, three isocenters were needed, wherein the first one was still placed in PTV_C, and the second and third isocenters were placed in the midline of PTV_S. If L was between 72 and 80 cm, three isocenters were needed, but the second and third were moved 10 cm toward the patient’s back. Since L is generally smaller than 80 cm for a majority of patients, cases in which L is larger than 80 cm were not considered. A case with L between 42 and 72 cm was shown in Figure 2 to illustrate the determination of isocenters and overlapping of adjacent fields. The coordinates of the center of PTV_C and PTV_S were extracted and denoted as (xc, yc, zc) and (xs, ys, zs), respectively, with the positive x, y, and z coordinates toward the patient’s left arm, abdomen, and head, respectively. All the isocenters have the same x and y coordinates; therefore, only a longitudinal shift of couch was performed during the treatment. The coordinates of the three isocenters can be expressed as follows:
Iso1 (x, y, z) = (xs, ys, zc),
Iso2 (x, y, z) = (xs, ys, zs + (L-F)/2),
Iso3 (x, y, z) = (xs, ys, zs - (L-F)/2),
where “L” is the length of PTV_S and can be derived by the GetBoundingBox () built-in function in RayStation. “F” is the achievable largest field length assuming that the maximum field length of standard Linac is 40 cm. However, considering beam divergence and penumbra, F was set to 38 cm.
The cranial beamset with the first isocenter (Iso1) consisted of 6 fields with gantry angles of 180, 260, 300, 60, and 100 degrees and Y2 jaw limit not exceeding the fifth cervical vertebra level to avoid shoulders irradiation. Both the upper and lower spinal beamsets with the second isocenter (Iso2) and the third isocenter (Iso3), respectively, consisted of three posterior or near posterior fields with gantry angles of 140, 180, and 220 degrees. The cranial and upper spinal beamsets overlapped at the neck from the first to the fifth cervical vertebra. The upper and lower spinal beamsets overlap at the middle spinal with a length equal to 2×F-L. If the overlapping length was set to be at least 4 cm and F was known to be 38 cm, the maximum value of L was 72 cm. However, when the second and third isocenters were moved 10 cm toward the patient’s back, F was magnified by 1.1 times, and L could reach almost 80 cm.
Plan Optimization
The initial optimization functions are listed in Table 1. PTV dose constraints were set to robustness with independent isocenters position uncertainty of 5 mm in superior and inferior directions. The total scenarios to compute were 27. Two auxiliary structures, Ring_PTV_C and Ring_PTV_S, were involved in the optimization to improve the target conformity. The Ring_PTV_C was derived by firstly expanding PTV_C with large margins of 2 cm in anterior, posterior, left and right directions, and small margins of 0.5 cm in all six directions, then subtracting the smaller expanded PTV_C from the larger expanded PTV_C. The Ring_PTV_S was derived in the same way but with large margins of 4 cm in anterior, posterior, left and right directions, and small margins of 1 cm in anterior, left, right, superior, and inferior, and 2 cm in posterior directions. The larger posterior margin in Ring_PTV_S is due to posterior or near posterior fields for PTV_S. Function values of each optimization objective were computed after one round of optimization. If the function value is out of range of 0.002-0.006 and the organ type is not target, the script will tighten or loosen dose constraints until all the values are within the range. Then optimization and adjustment were repeated until the function values were within the range, or the iteration number reached 10. At the end of the optimization, a very highly weighted max dose constraint was added to the external body, then the last round of optimization was performed. The non-robust plans had the same initial optimization functions except that PTV dose constraints were set to non-robustness.
Table 1
The initial optimization functions of robust plans
Structures
|
Optimization functions
|
Weight
|
Robust
|
Cranial PTV (PTV_C)
|
Min Dose: 2340 cGy
Min DVH: 2370 cGy to 95% volume
Max Dose: 2450 cGy
|
300
300
300
|
Yes
Yes
Yes
|
Spinal PTV (PTV_S)
|
Min Dose: 2340 cGy
Min DVH: 2370 cGy to 95% volume
Max Dose: 2450 cGy
|
300
300
300
|
Yes
Yes
Yes
|
Ring_PTV_C
|
Max dose: 2200 cGy
Max EUD: 500 cGy, Parameter A 1
|
20
1
|
No
No
|
Ring_PTV_S
|
Max dose: 2200 cGy
Max EUD: 500 cGy, Parameter A 1
|
20
1
|
No
No
|
Lens_L
|
Max dose: 800 cGy
Max EUD: 400 cGy, Parameter A 1
|
20
1
|
No
No
|
Lens_R
|
Max dose: 800 cGy
Max EUD: 400 cGy, Parameter A 1
|
20
1
|
No
No
|
Lungs
|
Max DVH: 500 cGy to 38% volume
Max EUD: 450 cGy, Parameter A 1
|
20
1
|
No
No
|
Heart
|
Max DVH: 1000 cGy to 10% volume
Max EUD: 390 cGy, Parameter A 1
|
20
1
|
No
No
|
Kidneys
|
Max DVH: 1000 cGy to 10% volume
Max EUD: 390 cGy, Parameter A 1
|
20
1
|
No
No
|
Setup Error Sensitivity
The sensitivity of the automatic CSI plans to setup error was evaluated by deliberately shifting Iso2 5 mm superiorly and then inferiorly. Doses were recalculated with applied shifts but without altering any planning parameters. A dose comparison was performed for the shifted and un-shifted plans.
Plan evaluation
The dose homogeneity and conformity referred to the PTV were determined using the homogeneity index (HI) [19] and conformity index (CI) [20] through the following equations: HI = (D2-D98) /D50, CI = (VT,P × VT,P) / (VT × VP), where D50 is the median dose to the PTV, and D2 and D98 are the near maximum and minimum doses that cover 2% and 98% volume of the PTV, respectively. VT,P is the volume of the target enclosed by the prescription dose. VP is the volume of tissues including the target covered by the prescription dose, and VT is the volume of the target. D1, D2, D98, and D99 of PTV_C and PTV_S, D1 of Lens, and mean dose of lungs, heart, and kidneys were also evaluated, where Di is the dose that covers i% volume of PTV. The total planning time was recorded to evaluate the efficiency of the automatic CSI planning.
Statistical methods
Wilcoxon matched-pairs signed-rank test was performed with p < 0.05 to identify differences between robust and non-robust optimization plans. The computations were performed with the use of the SPSS statistics program (Version 22, IBM SPSS Inc., Chicago, IL).