Data collection
Ninety-eight lung cancer patients treated in Shanghai chest hospital from May 2019 to April 2021 were retrospectively enrolled in this study. All cases had early-stage inoperable NSCLC or oligometastatic lung cancer who were consulted with at least two radiation oncologists before receiving SBRT. The prescription dose was defined depending on clinical stage, tumor size, location, and patient's physical condition. When the study began, all the patients signed informed consent and completed their radiotherapy. The study was approved by the Institutional Ethics Committee (the committee's reference Number: KS1863).
Structure delineation
Patients were scanned with a Siemens Definition AS computed tomography (CT) Scanner System (Siemens Healthcare, Erlangen, Germany) to obtain free-breathing CT and four-dimensional CT (4DCT). All targets were delineated on a MIM Maestro Station (MIM Vista Corp, Cleveland, US-OH) by experienced radiation oncologists. The GTV was defined in the 10 phases of 4DCT. 10 GTVs were merged to generate internal target volume (ITV). PTV was obtained by expanding 0.5 cm of ITV in three dimensions. All structures were reviewed and approved by an independent radiation oncologist before being used for planning design.
Treatment Planning
Treatment plans were planned on the averaged 4DCT using the Pinnacle treatment planning system (TPS) (V16.2, Philips Radiation Oncology Systems, Fitchburg, WI, USA) for an Edge™ linear accelerator (Varian Medical Systems, Palo Alto, CA) equipped with a high-definition HD 120 multileaf collimator (MLC)™. HD120 MLC™ has 120 leaves with a leaf width projected at the isocenter of 2.5 mm for the central 8.0 cm region and 5.0 mm for the two 7.0 cm peripheral regions. The planning method was similar to our previous study. [8, 20, 21] All plans employed ten or more 6MV fields, and the angular intervals of the fields were either 15 or 20 degrees. Collimator and couch angles were adjusted according to the individual situation. The direct machine parameter optimization (DMPO) and the collapsed cone convolution (CCC) algorithms were used for plan optimization and dose calculation with a calculation resolution of 1.0 mm, respectively. The treatment plan goals and constraints recommended in RTOG 0813 [18] and 0915 [19] were followed, where the prescription dose covers 95% of the PTV volume.
In order to achieve a sharp dose gradient, six dose-limiting shells were generated for PTV before plan optimization. [20] The generation of those shells listed in Table 1 is briefly described as follows. First, the PTV was expanded to a specific boundary (3mm, 5mm, 8mm, 15mm, 25mm) to generate an intermediate structure. The intermediate structure was then subtracted from the outline to generate a shell. This step was completed by pre-written scripts.
Table 1
Planning constraints used for the process of optimization
Structure
|
Plan-Auto
|
Plan-Manual
|
Objective
|
Constraint
|
Priority
|
Compromise
|
Objective
|
Constraint
|
Weight
|
Constrain
|
PTV
|
Dp
|
-
|
-
|
-
|
VDp>95%
|
|
|
Selected
|
|
|
|
|
Dmin>95%Dp
|
|
|
Selected
|
Shell3mm (% of Dp)
|
-
|
Dmax< 56
|
High
|
Not selected
|
-
|
Dmax< 56
|
100
|
-
|
Shell5mm (% of Dp)
|
-
|
Dmax< 50
|
High
|
Not selected
|
-
|
Dmax< 50
|
90
|
-
|
Shell8mm (% of Dp)
|
-
|
Dmax<40
|
High
|
Not selected
|
-
|
Dmax<40
|
80
|
-
|
Shell11mm (% of Dp)
|
-
|
Dmax<34
|
High
|
Not selected
|
-
|
Dmax<34
|
70
|
-
|
Shell15mm (% of Dp)
|
-
|
Dmax<28
|
High
|
Not selected
|
-
|
Dmax<28
|
60
|
-
|
Shell25mm (% of Dp)
|
-
|
Dmax<18
|
High
|
Not selected
|
-
|
Dmax<18
|
50
|
-
|
Total Lung (% of Volume)
|
-
|
V5Gy < 25
|
High
|
Selected
|
-
|
V5Gy < 25
|
50
|
-
|
V10Gy < 15
|
High
|
Selected
|
-
|
V10Gy < 15
|
50
|
-
|
V20Gy < 6
|
High
|
Selected
|
-
|
V20Gy < 6
|
50
|
-
|
Dp: total prescription dose. V5Gy, V10Gy, V20Gy, and VDp : The percentage of volume of total lung excluding ITV receiving 5Gy, 10Gy, 20Gy, and the prescription dose. Dmin: Minimum dose. Dmax: maximum dose. |
The optimization objectives and constraints of manual and automatic planning are optimized according to standard clinical practices. The goals and constraints for initial optimization were set by a pre-set script. The details were shown in Table 1. Some other constraints of normal tissue were individually set according to the positional relationship between each patient's target and normal tissues. In the optimization process of manual planning, objective, constraint, weight, constrain were adjusted to achieve a dose distribution as good as possible. Similarly, after optimizing automatic planning, fine-tuning was allowed by adding, reducing, or changing constraints, priority, and compromise options to achieve desirable results.
Plan evaluation
The metric of plan quality is a comprehensive evaluation index (CoI) that integrates conformity index (CI), gradient index (GI), heterogeneity Index (HI), and monitor units (MU). CI is a parameter to evaluate the similarity between the prescription isodose line and the outer contour of PTV, which characterizes the precise coverage of prescription dose to PTV. CI [22] was computed as CI = VT,Rx2/(VT*VRx),where VT,Rx is the PTV volume covered by prescription dose, VT is the target volume, and VRx is the volume covered by prescription dose. CI ranges from 0 to 1, and CI = 1 indicates the best conformability. GI is a metric of dose fall-off velocity around the target area, which represents the radiation dose of OARs. GI [23] is calculated as GI = V50%Rx/VRx, where V50%Rx is the volume receiving half the prescription dose. A lower GI represents a faster dose fall-off in normal tissue from the target. HI is an index to measure the dose uniformity of the PTV, indicating the hot spots in the target. HI is defined as HI = Dmax/DRx༌where Dmax corresponds to the maximum dose of the PTV volume. DRx is the prescription dose. Lower HI means a more uniform radiation distribution. For SBRT, a higher dose inside the tumor may translate to enhanced clinical efficacy in treating hypoxic tumors. [21, 24] Therefore, a higher HI within the clinically acceptable range of the SBRT plan represents a better plan quality. MU is an index of the complexity of a plan, which is closely related to the treatment time and delivery accuracy. [20, 21, 25] The smaller MU represents a shorter treatment time and higher delivery accuracy.
CoI is defined as
CoI=(CI*HI)/[GI*(MU/MUref)]
In order to avoid the excessive proportion of MU in the formula, we normalized MU with reference mu (MUref), which is defined as twice the fractional prescription dose (unit cGy). Due to the higher CI and HI, [21, 24] lower GI and MU represent better plan quality. Therefore, a larger CoI indicates a better plan.
Data analysis
Statistical analyses were performed with SPSS 20.0(IBM Corp., Armonk, NY, USA). The PTV volumes of all patients were counted in this study.
Firstly, CoIs obtained from the manual plan and automatic plan were categorized according to whether the manual plan was superior to the automatic plan or not. Then, receiver operating characteristics (ROC) analysis with PTV volume as a predictor of the classification results was performed to investigate the ability of PTV volume to predict the suitability of manual and automatic plans. The predictive ability of PTV volume is reflected by the AUC value in ROC analysis. When p < 0.05, the greater the AUC value, the better the predictive ability of PTV volume.
Secondly, if PTV volume can be used to predict the suitability of manual planning and automatic planning for lung SBRT, then the cut-off point of PTV volume for predicting the suitability of automatic planning and manual planning was tried to find. The coordinates of the ROC curve were analyzed, and the Youden index was used to determine the predictable PTV volume cut-off point. The Youden Index is a frequently used summary measure of the ROC curve. It both, measures the effectiveness of a diagnostic marker and enables the selection of an optimal cut-off point for the marker.[26] It is defined as Youden Index = sensitivity + specificity − 1, which could be practical in daily routine with a combination of high sensitivity and specificity. The sensitivity, specificity, and positive and negative predictive values were presented when a significant cut-off value was observed. If the patient's PTV volume is larger than this cut-off point, one plan is better. Otherwise, the other plan is better.
Finally, all patients were divided into two groups below and above the cut-off point, and the Wilcoxon signed-rank test was used for the dosimetric comparison and analysis of those two groups. A p-value of less than 0.05 was considered statistically significant.