Comparison of primary target volumes delineated on four-dimensional CT and ¹⁸ F-FDG PET/CT of non-small-cell lung cancer

Worldwide, lung cancer is the most common cause of cancer-related mortality [1]. About 80% of the cases of lung cancer are non-small-cell lung cancer (NSCLC) [2]. Radiotherapy plays a major role in the management of patients with NSCLC who cannot tolerate or refuse surgery. Unfortunately, the prognosis of patients with NSCLC remains poor because of high rates of local failure and distant metastases [3, 4], with local control rates of approximately 50% after radical radiotherapy [5]. A geometric target miss induced by tumor motion during radiotherapy is considered as one of the main reasons for local failure [6]. In order to account for geometric uncertainties due to internal variations in tumor position, size, and shape, the International Commission on Radiation Units and Measurements (ICRU) report 62 introduced the concept of an internal target volume (ITV) [7]. For lung tumors, respiratory motion is the major consideration for the ITV.

Currently, four-dimensional CT (4DCT) is widely used for the simulation of lung cancer. It is a reliable and effective tool for assessing tumor and organ motion [8, 9] and can provide patient-specific information about tumor position, shape, and size at different phases of the respiratory cycle. The internal gross target volume (IGTV) can make the determination of the ITV more efficient [6]. IGTV₁₀ is generated by combining all 10 individual gross tumor volumes (GTVs) contoured in each phase of the 4DCT dataset, which is thought to encompasses the motion information for the tumor in the whole respiratory cycle [10]. Therefore, it is expected to provide the most accurate IGTV based on a given 4DCT dataset [11].

As a functional imaging modality, ¹⁸ F-fluorodeoxyglucose (¹⁸ F-FDG) positron emission tomography/computer tomography (PET/CT) images have been shown to have greater specificity and sensitivity than CT alone for the diagnosis and staging of NSCLC patients [12]. Furthermore, the interobserver variability, as well as the intraobserver variability, could be significantly reduced when the ¹⁸ F-FDG PET image was used for tumor volume delineation [13, 14]. In addition, since three-dimensional PET (3D-PET) images are acquired over several minutes and represent the accumulated traces of multiple respiratory cycles, they may be capable of accounting for movement by indicating the average location of a tumor over time [15]. A phantom study by Caldwell et al. [16] concluded that PET imaging could more accurately depict the 3D volume of a moving phantom compared with spiral CT. Therefore, 3D-PET/CT might represent the ITV of a tumor. However, the optimal threshold values for patients with NSCLC have never been reported.

Hence, We perform this study to investigate the appropriateness of the threshold method to determine the best volumetric match to 4DCT-based IGTV₁₀ when contouring the primary tumor volume of NSCLC. Additionally, the feasibility of 3D-PET/CT images was evaluated with respect to incorporating tumor motion into the radiation target for NSCLC.

For lesion 15 and 16, the SUV_max in the tumor was 6.09 and 6.07, we could not obtain GTV_PET15% target volumes as the volume obtained from the SUV15% contours were indistinguishable from background lung activity.

Accurate definition of the target volume, ideally incorporating metabolic information, becomes paramount importance in the current trend in NSCLC treatment planning [21]. A number of studies compared 3DCT volumes with ¹⁸ F-FDG PET/CT volumes of NSCLC [22‐24]. However, the best methodology for applying ¹⁸ F-FDG PET/CT to IGTV definition is not currently well established. To the best of our knowledge, there is few study that has compared tumor sizes and CI values between GTV_PET and IGTV₁₀ in contouring NSCLC.

In this study, we determined the IGTV₁₀ from 10 phases of the 4DCT dataset and used them as the reference to find the optimal threshold that yield the best match between the GTV_PET and IGTV₁₀ in both the target size and the spatial conformity. Our study revealed that GTV_PET using a threshold setting of SUV15% approximated most closely to the IGTV₁₀ with the lowest median percentage volume changes. When using the threshold level of ≥ SUV25% and/or ≥ SUV2.5, the PET-based tumor sizes were estimated to be smaller than the IGTV₁₀. Therefore, on the basis of the results of our study, the SUV threshold setting of ≥25% and/or ≥2.5 is not suitable for IGTV contouring in NSCLC.

Analogously, Hanna et al. [25] compared volumes from a manual method and five automated PET segmentation techniques to 4DCT-derived ITV and found that none of the PET target volumes approximated closely to the 4DCT target volumes. However, in their study, the patient’s PET/CT and 4DCT scans were not acquired on the same day or in identical position. In this circumstance, it was possible that changes in tumor geometry or size occurred and potentially increased the likelihood of a mismatch between the PET-based contours and the 4DCT-based contours [25]. Caldwell et al. [16] reported that using a threshold as low as 15% of the maximum value could account for respiratory motion and more accurately depict the true extension of the moving target. Another phantom study by Okubo et al. [26] concluded that when a threshold value of 35% of the measured maximum FDG activity was adopted, the sizes of PET delineation were almost the same for static and moving phantom spheres of 22 mm or more in the axial plane. Our study was similar to the result of Caldwell et al., but smaller than the result of Okubo et al. This is possible because patients enrolled in our study had a range of tumor sizes and positions. Moreover, the 15% threshold method was not suitable for contouring some lung tumors that have low SUV, because it might fail to distinguish tumor from background lung activity. Nevertheless, Okubo et al. suggested that the threshold of 35% of measured maximum FDG activity was only a provisional criterion for tumors of 2–4 cm given that appropriate threshold values could be changed on the basis of the tumor size [26]. In addition, it should be acknowledged that any phantom studies versus clinical comparison is limited. For example, unlike in real tumors, the FDG distribution in the spheres of the phantoms and in the background was homogenous.

Similarity in absolute volume does not mean identity in the space location. Our results indicated that GTV_PET15%, GTV_PET20% and GTV_PET2.0 showed no significant difference with IGTV₁₀ in target volume. However, the CIs of them were significantly lower than 1.0. The best CI was between IGTV₁₀ and GTV_PET15%, which was only 0.57. It is not surprising as GTV_PET15% is the biggest volume and hence has the greatest degree of potential overlap. Based on this consideration, this may not make it the most accurate. The poor CIs suggested great unconformity between what was indicated abnormal on PET image and on CT image. One of the reasons is that shape and/or positional alterations between IGTV₁₀ and GTV_PET had occurred. Our study showed that the minimum variation in the centroid coordinate position in the 3D direction of GTV_PET and IGTV₁₀ was 0.38 cm (median). Moreover, the CIs were inversely correlated with the centroid shifts in 3D directions. Although patients in our study were immobilized in the same position for both 4DCT and PET/CT, millimetric precision in set-up using immobilization devices may not be feasible. Furthermore, a rigid registration might not be sufficient for lung tumors. Hence, registration error may inevitable affect the spatial position between GTV_PET and IGTV₁₀. In addition, it is possible that some of this difference may be related to differences in the patient’s breathing pattern between acquiring the PET/CT and 4DCT. Different breathing pattern can influence tumor size, shape and distribution of activity on the free-breathing PET images [27].

Hanna et al. [25] used the Dice similarity coefficient (DSC) to assess volumetric, shape and positional similarity in the PET-generated target volumes and 4DCT target volumes. Their study revealed that the highest DSCs (mean) were 0.64. Grills et al. [21] investigated the impact of PET/CT for GTV definition in NSCLC using the matching index (similar to CI in our study) to compare the GTV, as defined by CT, with the GTV defined by PET co-registered with CT. In their study, the mean matching index was 0.65. Gondi et al. [23] demonstrated that CI values of NSCLC with the incorporation of FDG-PET and CT were 0.44. The results of our study were similar to data published in prior studies [21, 23, 25]. In addition, although CI can provide the most information on both volume change and positional change [28], it cannot quantify the percentage of one volume included by another volume. Further analyzing the inclusion relation between GTV_PET15% and IGTV₁₀, there was 20% of (median) GTV_PET15% not included in IGTV₁₀ and 40% of (median) IGTV₁₀ not included in GTV_PET15%. It suggested that ITGV₁₀ did not encompass GTV_PET15% completely or vice versa. Therefore, we concurred with Gondi [21] who concluded that although the quantitative absolute target volume could sometimes be similar between CT and PET, the qualitative target locations can be significantly different.

One limitation of our study is the small number of patients studied, so that subgroup analyses were limited. Therefore, a larger cohort of patients with many different tumor sizes and locations should be conducted to further investigate the relationship of using 3D-PET/CT and 4DCT for contouring IGTV of NSCLC. We are continuing our work to enroll more patients for further clinical investigations.

None of the PET based contours had both close spatial and volumetric approximation to the 4DCT IGTV₁₀. At present 3D-PET/CT should not be used for IGTV generation.