Clinical investigation: lung
Measurement of lung tumor volumes using three-dimensional computer planning software

https://doi.org/10.1016/S0360-3016(02)02783-9Get rights and content

Abstract

Purpose: To examine the interclinician variation in the definition of gross tumor volume (GTV) in patients undergoing radiotherapy for non-small-cell lung cancer (NSCLC), develop methods to minimize this variation, and test these methods.

Methods and Materials: The radiotherapy planning computed tomography (CT) scans of 6 consecutive patients with NSCLC in which the radiologist was able to define and outline the GTV were used. Six oncologists independently contoured the tumors with the radiologist’s markings as a guide using a three-dimensional treatment planning system. Separate contours were prepared using only mediastinal window settings and using both mediastinal and lung window settings. The volumes were calculated using the planning system software (series 1). Factors that resulted in interclinician variation were determined, and, after a 3-year interval, 5 of the 6 clinicians redefined the GTVs using a revised protocol aimed at minimizing variation (series 2).

Results: For series 1, the interclinician variation in the measurement of volumes ranged from 5%, in the most tightly measured tumor, to 42%, in the most variable, but was, on average, 20%. Statistically significant differences were noted among the clinicians (p = 0.002), that is, some clinicians tended to record relatively small and some relatively large volumes. The reasons for the variation among the oncologists included a tendency to include regions with a low probability of containing tumor, as if the oncologist were contouring a target volume; inclusion of adjacent atelectasis (ignoring the radiologist’s outline); and variable treatment of spicules. When the exercise was repeated using the revised protocol (series 2), the degree of interclinician variation was reduced, with a range of 7–22% (average 13%). In series 2, the differences among the clinicians were not statistically significant (p = 0.25).

Conclusion: Despite major radiologic input, significant variation occurred in the delineation of the three-dimensional GTVs of NSCLC among oncologists. Standardization of the approach with guidelines resulted in a reduction in this variation.

Introduction

Report No. 50 of the International Commission on Radiation Units and Measurements defines the gross tumor volume (GTV) as “the gross palpable or visible/demonstrable extent and location of the malignant growth” (1). The report acknowledges that the GTV “may seemingly be different in size and shape … depending on what examination technique is used for evaluation.” A precise definition of the GTV in patients with lung cancer is best achieved with a CT scan of the thorax, but its appearance can be altered by varying the window width and level of the Hounsfield unit (HU) settings. Accurate and reproducible measurement of the GTV on the computed tomography (CT) image therefore requires standardization of both the conditions under which the volume is delineated and the method used by the radiation oncologist or radiologist to contour the tumor. Very little guidance is available in the literature on how to define the GTV, from which the clinical target volume (CTV) is derived, and whose extent is of fundamental importance in the creation of three-dimensional conformal treatment plans. For example, the recommendations based on a recent CT contouring exercise omitted any reference to the optimum CT window settings to display the GTV (2). Similarly, in an evaluation of a target contouring protocol for three-dimensional conformal radiotherapy (3D-CRT) of non-small-cell lung cancer (NSCLC), the CT window settings were not specified, although that was a concluding recommendation resulting from the study (3).

In addition to its importance in treatment planning, the GTV may also have prognostic significance. In a review of the influence of tumor volume on the probability of tumor control by radiotherapy (RT), the clinical data demonstrated a consistent adverse effect of increasing tumor volume on local control in patients with cancer of the head and neck, cervix, and breast and melanoma (4). The prognostic importance of tumor size is reflected to varying degrees in the staging systems for cancers of the head and neck, breast, and cervix and melanoma, but in NSCLC, T3 and T4 tumors can be any size (5). The implication is that these less resectable lung tumors cannot be effectively treated by other means. In practice, many of these patients are referred for RT, in which case, the tumor size may ironically be a major determinant of outcome.

To develop our own set of guidelines for lung cancer patients, we undertook an exercise in which six oncologists contoured the GTV in 6 cases of NSCLC. The main aims of this study were to measure the variability among clinicians in the measurement of tumor volume, to investigate whether any systematic differences in GTV measurement occurred among the clinicians, and finally, to identify factors explaining the differences among the clinicians in GTV measurement to establish guidelines for the measurement of the GTV with minimal error. After a 3-year interval, the exercise was repeated using the new guidelines to determine whether they resulted in a reduction in the interclinician variation. A secondary aim was to investigate the relationship between GTV and maximal tumor diameter, as a possible surrogate for the GTV.

Section snippets

Methods and materials

Six consecutive patients with NSCLC who had undergone planning for radical RT, and in whom the radiologist (A.McK.) was able to define the GTV on a planning CT scan of the thorax, were selected for study. All patients had undergone a diagnostic CT scan with contrast, as well as a planning CT scan that was performed with the patient in the treatment position. The planning CT scan was performed without contrast and with cuts at 1-cm intervals. The CT data were transferred to the treatment

Series 1

The volumes measured by the six clinicians for each of the 6 tumors are given, with summary statistics, in Table 1, Table 2. The tumors and clinicians appear in order of increasing mean volume.

Figure 1 illustrates the measurements of the clinicians relative to one another and the amount of variation in the measurement of each tumor volume. This interclinician variability (as %CV) ranged from 5% for Tumor 5 to 39% for Tumor 3. Averaged over the tumors, the interclinician variability of

Discussion

In this study, we demonstrated significant variations in the volumes contoured by oncologists when independently delineating GTVs in patients with lung tumors. We also showed that these variations can be reduced using a standardized approach, following the protocol included in the Appendix. These variations were dependent on the contouring oncologist, CT window setting, and tumor morphology. This occurred even though each oncologist was provided with the same set of hard copy CT images on which

Cited by (81)

  • A method to combine target volume data from 3D and 4D planned thoracic radiotherapy patient cohorts for machine learning applications

    2018, Radiotherapy and Oncology
    Citation Excerpt :

    The maximum volume ratio over all cases is 1.69 for the linear fit. Typical Vmax/Vmin values from inter-observer delineation studies quoted in the literature, have been observed to be in the range of 1.1–10.1 [26–29]. Interestingly, contrary to existing knowledge of lung tumour motion, [18], the derived kernels were smaller in the cranial–caudal direction than in the left–right and anterior–posterior directions.

  • Decision support systems for personalized and participative radiation oncology

    2017, Advanced Drug Delivery Reviews
    Citation Excerpt :

    Technological progress in imaging – including enhanced temporal and spatial resolution, faster scanners and protocol standardization – has rapidly advanced the area towards the identification of quantitative noninvasive imaging biomarkers, also known as “Radiomics”. [21,130,131](Fig. 6). Metrics built on tumor volume and location are the most frequently employed image derived features of tumor response to therapy and survival [132–139], and utilize CT or magnetic resonance imaging (MRI) technology for 3D measurement [140–142]. Although used in clinical practice, tumor volume and location are subject to inter-observer variability that can be ascribed to variances in tumor delineations [143,144].

  • Uncertainties in volume delineation in radiation oncology: A systematic review and recommendations for future studies

    2016, Radiotherapy and Oncology
    Citation Excerpt :

    These studies established if there was greater IOV between one dataset and another, independent of the magnitude, shape or other descriptors of the volumes being delineated. These studies were generally undertaken to determine if there was a factor which resulted in greater concordance between clinicians such as a different imaging modality [20,21,24–26,32,37,41,45,57,58,68,69,73,76–79,84,87,88,96,101,132], fiducials [16], training [36,56,62,75,122], guidelines or atlas [46,55,74,83,97,107,108,114,126,133] or provision of autosegmented contours to edit [15,51,73,80,82,94]. For large datasets comparing only two situations, the paired T-test was commonly used for assessing statistical significance [15,18,45,78].

  • Non-Small Cell Lung Cancer

    2015, Clinical Radiation Oncology
View all citing articles on Scopus
View full text