A complete 4DCT‐ventilation functional avoidance virtual trial: Developing strategies for prospective clinical trials

Abstract Introduction 4DCT‐ventilation is an exciting new imaging modality that uses 4DCT data to calculate lung‐function maps. Because 4DCTs are acquired as standard of care for lung cancer patients undergoing radiotherapy, 4DCT‐ventiltation provides functional information at no extra dosimetric or monetary cost to the patient. The development of clinical trials is underway to use 4DCT‐ventilation imaging to spare functional lung in patients undergoing radiotherapy. The purpose of this work was to perform a virtual trial using retrospective data to develop the practical aspects of a 4DCT‐ventilation functional avoidance clinical trial. Methods The study included 96 stage III lung cancer patients. A 4DCT‐ventilation map was calculated using the patient's 4DCT‐imaging, deformable registration, and a density‐change‐based algorithm. Clinical trial inclusion assessment used quantitative and qualitative metrics based on the patient's spatial ventilation profile. Clinical and functional plans were generated for 25 patients. The functional plan aimed to reduce dose to functional lung while meeting standard target and critical structure constraints. Standard and dose‐function metrics were compared between the clinical and functional plans. Results Our data showed that 69% and 59% of stage III patients have regional variability in function based on qualitative and quantitative metrics, respectively. Functional planning demonstrated an average reduction of 2.8 Gy (maximum 8.2 Gy) in the mean dose to functional lung. Conclusions Our work demonstrated that 60–70% of stage III patients would be eligible for functional planning and that a typical functional lung mean dose reduction of 2.8 Gy can be expected relative to standard clinical plans. These findings provide salient data for the development of functional clinical trials.


| INTRODUCTION
Despite considerable recent technologic advances in thoracic radiotherapy, symptomatic radiation pneumonitis and fibrosis remain serious complications occurring in an estimated 5-50% of patients. 1 One proposed method for reducing pulmonary toxicity for lung cancer patients is functional avoidance radiotherapy which utilizes functional imaging to selectively spare functional portions of lung in favor of irradiating nonfunctional regions. [2][3][4][5][6][7][8][9][10][11][12][13] The idea is that sparing functional portions of the lung can reduce the incidence of pulmonary complications. 12,14 Studies have demonstrated functional avoidance using single-photon emission computed tomography (SPECT), 2,6-8,10 positron emission tomography (PET), 9 and magnetic resonance imaging (MRI). 5 A new lung function imaging technique has been proposed for purposes of functional avoidance 3,4,13 that calculates ventilation using the patient's 4-dimensional computed tomography (4DCT) data. 15,16 Compared with SPECT, PET, and MRI, 4DCT-based ventilation (4DCT-ventilation) offers perhaps the most attractive means toward achieving functional avoidance in radiation oncology because the functional information is obtained using data that is already acquired as standard of care; no additional imaging procedure is necessary. Furthermore, 4DCT-ventilation does not require a radioactive contrast agent and offers a faster imaging procedure, improved spatial resolution, and an imaging modality that by definition provides anatomical (4DCT) and functional information (4DCTventilation).
Retrospective work has been done to validate 4DCT-ventilation against nuclear medicine ventilation-perfusion scans (VQ), [17][18][19] pulmonary function tests, 17,20 xenon-based CT, 16 MRI, 21 and PET. 22 Proof of principle studies have demonstrated the use of 4DCT-ventilation for functional avoidance, 3,4,13 and Vinogradskiy et al. 12 retrospectively showed that doses to 4DCT-based functional lung better predict for clinical lung toxicity compared with dose alone, suggesting that prospectively incorporating 4DCT-ventialtion-based functional information can decrease toxicity.
The retrospective work on 4DCT-ventilation has paved the way for the development of clinical trials to use 4DCT-ventilation imaging for functional avoidance radiotherapy at other institutions 23

2.B | 4DCT-ventilation imaging
Each patient's simulation 4DCT data were used to calculate 4DCTventilation maps. 15,25 The lungs were segmented on the end-inhale and end-exhale phases. Lung voxel elements were then mapped from inhale to exhale phase using a deformable image registration (DIR) algorithm with spatial accuracy on the order of 1.25 mm. 26 The registration was used to apply the HU density-change equation 15 to calculate ventilation: where V in and V ex are the inhale and exhale volumes and HU in and HU ex are the inhale and exhale Hounsfield units of the individual lung voxels. Equation 1 calculates the local change in air content for each voxel and produces a 3D map of ventilation function (example shown in Fig. 1). For each 4DCT-ventilation image, the 4DCT was reviewed for motion artifacts and the DIR was reviewed for discontinuities and errors.

2.C | Clinical trial inclusion criteria
One critical consideration in determining functional avoidance trial eligibility is a patient's spatial lung function profile. If a patient has homogenous lung function, there is no basis to preferentially spare any regions. Conversely, if a patient's ventilation image is heterogeneous and displays a major ventilation defect, functional avoidance can preferentially deposit dose in the defect area as opposed to the functional region. We previously developed both quantitative and qualitative metrics to assess spatial lung function. 27 We used the previously developed metrics to aid in the evaluation of trial eligibility of our stage III cohort. The qualitative assessment included a binary metric of whether a ventilation defect was present using consensus from three reviewers. For quantitative analysis, we derived quantitative metrics intended to reflect the degree of heterogeneity of the ventilation image. We computed two metrics using the percent ventilation in each lung third. The percent ventilation in each lung third is a standard metric used in VQ imaging 18 and is intended to geometrically approximate the ventilation in each lobe (schematically shown in In the absence of quantitative guidelines for assessing ventilation heterogeneity, we sought to evaluate how well the quantitative metrics predicted for observer-identified defects. The assessment was done using logistic regression and receiver operator characteristic (ROC) analysis using the area under the curve (AUC) metric.

2.D | Functional planning techniques
Based on observer-identified defects and the %VTA metrics, we determined which patients would be eligible for functional planning (Appendix A). Of the eligible patients, we reduced the subset to patients that were originally treated with IMRT and randomly 20 Gy or more (V20 Gy) less than 37%, esophagus mean less than 34 Gy, and less than 2/3 of the heart volume receiving 45 Gy.
The aim was to have 95% cover the planning target volume (PTV) with the prescription dose with a hotspot that could not exceed 120%. In situations where the RTOG 0617 OAR or PTV constraints could not be met, the clinical plan aimed to meet the dosimetry of the original plan used to treat the patient. The reasoning was that the original plan was deemed clinically acceptable at the time of treatment by the clinician and by matching dosimetry parameters, the generated clinical plan could also be considered clinically acceptable.
The functional plan aimed to maximize functional lung sparing while trying to meet RTOG 0617 criteria. A "functional-avoid" structure was created with auto-segmentation of functional portions of lung tissue using a threshold of 15%. The 15% threshold was determined using AUC analysis to determine the %VTA value which produced an optimal operative point (largest AUC). The optimal operative point was calculated to be a 15% reduction in ventilation in a given lung third. The 15% reduction was subsequently used to derive the lower limit needed for the auto-segmentation for the functional-avoid structure. In other words, the functional-avoid structure was derived using auto-segmentation of any lung with no less than a 15% reduction in ventilation. Once auto-segmentation was applied, the gross or internal target volumes (GTV or ITV) were subsequently subtracted from the functional-avoid structure.
Both the clinical and functional plans were generated in Eclipse

2.E | Dose-function assessment
In the case of functional avoidance radiotherapy, evaluating dosevolume parameters alone will be insufficient, rather an assessment will be needed that combines both dose and function. We compared the clinical and functional plans using standard dose-volume and dosefunction metrics. Standard dose metrics included PTV coverage, homogeneity index (HI) (defined as D90%/D5%), 8 conformity index (CI), 9 max cord, mean esophagus, mean heart, and mean lung doses (MLD). based on the dose-function histogram. 28 Dose-function metrics based on the "functional-avoid" structure are preceded with "S" and metrics based on the entire ventilation image are proceeded with "Im". Dose-function metrics were compared between the functional plan and the clinical plans using t-tests.

3.B | Functional planning techniques
To generate the functional plans, a median number of 3 arcs (range 2-4) was utilized. All plans were designed using coplanar arcs except for one case which used a sagittal arc. Twelve (48%) of the plans were generated using directional (ipsilateral) arcs, and 52% used full, 360°arcs.

4.B | Functional planning techniques
The functional planning approach we used was to generate a functional-avoid structure from the 4DCT-ventilation image. Some authors proposed a similar structure-based approach, [2][3][4][5]7,9,13 while others directly incorporated the 4DCT-ventilation image into the optimization. 8 All the functional plans were created using rotational IMRT which is the current standard in our clinic and is gaining momentum for thoracic treatments due to its relative simplicity. Standard practice in rotational therapy employs one-sided partial arcs for lateral disease and full 360°arcs for disease with mostly mediastinal involvement. This treatment paradigm held true for functional planning with the exception of patients that had mediastinal involvement but large ipsilateral ventilation defects; in those instances, one-sided arcs were dosimetrically beneficial.   The logic behind the %VTA metric is that a ventilation defect may appear downstream of the tumor.

4.C | Dose and dose-function assessment
In the absence of quantitative guidelines for assessing ventilation heterogeneity, we sought to evaluate how well the quantitative metrics predicted for observer-identified defects. The assessment was done using logistic regression, receiver operator characteristic (ROC) analysis, and area under the curve (AUC) evaluation.
We found that the quantitative metric that best predicted for observer-based defects was the %VTA metrics (percent ventilation in the third containing the tumor or any adjacent third) with an AUC of 0.83 | 151 (Table 1). We performed two further analyses using the %VTA metric.
The first analysis compared the %VTA metric and user-identified defects as means to evaluating a patient's heterogeneity profiles. The second analysis used the %VTA metric to help derive a threshold for generating the functional-avoid structure (presented in the manuscript).
To analyze the relationship between the %VTA metric and useridentified defects, we calculated standard ROC metrics including the sensitivity, specificity, and overall accuracy (Table A1). It is instructive to compare instances where the %VTA metric and useridentified defects disagreed. The false positive rate (FPR) in the analysis alludes to instances where the %VTA metric identified a defect but users did not. This scenario can be attributed to situations where the quantitative differences in percent ventilation in each third were more sensitive than what the human observers could decipher. The false negative rate (FNR) refers to situations where the %VTA metric predicted no defect while the users identified a defect. The FNR scenario arose when defects were present but not in an adjacent third relative to the tumor (i.e., further away from the tumor). The distant defects could be due to prior irradiation or chronic obstructive pulmonary disease. In this regard, we believe the %VTA metric provides the more pertinent result for functional avoidance because it is not practical to design treatment plans that place dose through nonventilated lung far away from the tumor. For example, if the patient has a left lower lobe (LLL) mass and a defect in the right upper lobe (RUL), it is not feasible to design a plan to treat the LLL mass by putting dose through the ventilation defect in the RUL.