Interreader Agreement and Variability of FDG PET Volumetric Parameters in Human Solid Tumors
Abstract
OBJECTIVE. The purpose of this article is to evaluate the interreader agreement and variability of two 18F-FDG PET parameters, metabolic tumor volume and total lesion glycolysis, in human solid tumors.
MATERIALS AND METHODS. One hundred eleven patients (mean [± SD] age, 61.9 ± 12.5 years) with baseline staging FDG PET/CT scans were included. Two readers independently read the scans and segmented metabolic tumor volume and total lesion glycolysis using two fixed thresholds, 40% and 50% of the lesion's maximum standardized uptake value (SUVmax). The impact of the lesion's FDG avidity and location on reader agreement and variability was established. Intraclass correlation coefficient (ICC), precision, and Bland-Altman analysis were used to evaluate agreement and variability.
RESULTS. The ICCs for 40% and 50% SUVmax segmentations of metabolic tumor volume between the readers were 0.987 and 0.995, and the corresponding values for 40% and 50% SUVmax segmentations of total lesion glycolysis were 0.987 and 0.986, respectively (p = 0.0001). The corresponding precisions were 0.5%, 0.2%, 0.5%, and 0.5%, respectively. The mean biases between the readers for 40% and 50% SUVmax segmentations of metabolic tumor volume were −1.78 ± 8.42 mL and −0.46 ± 2.1 mL and for 40% and 50% SUVmax segmentations of total lesion glycolysis were −7.3 ± 31.6 g and −2.97 ± 12.86 g, respectively. Subgroup analysis showed better precision and lesser variability for 50% SUVmax segmentations of metabolic tumor volume and total lesion glycolysis in patients with the highest and lowest FDG-avid primary tumors. The precision was highest and variability was lowest for lung tumors.
CONCLUSION. There is excellent interreader agreement for measurement of metabolic tumor volume and total lesion glycolysis with 40% and 50% SUVmax threshold segmentations in human solid tumors.
Fluorine-18 FDG PET metabolic activity has been increasingly used as a quantitative biomarker in evaluating treatment response [1–3]. The FDG PET maximum standardized uptake value (SUVmax) is a commonly used FDG imaging parameter in oncology [4–9] because of its simplicity and excellent reader reliability. However, SUVmax, representing maximum single-pixel metabolic information of the tumor, may not accurately represent tumor biology. Imaging parameters that measure FDG, including metabolic tumor volume and total lesion glycolysis of the tumor, are emerging as imaging parameters for the prediction of outcome and therapy response in patients with solid tumors [10–14].
For metabolic tumor volume and total lesion glycolysis to be useful in clinical practice as reliable prognostic and predictive parameters, the reader agreement and variability of these parameters must be established. Semiautomatic segmentation methods have been shown to have less variability than manual methods [15] in measuring tumor parameters. Semiautomatic isocontour region of interest based on a fixed percentage of SUVmax is a common and widely used method because of its availability across imaging workstations and ease of use [14]. We previously studied the intrareader agreement of FDG tumor volumetric parameters in human solid tumors [14]. However, the interreader reliability of metabolic tumor volume and total lesion glycolysis has not been well studied in a large patient population and needs to be established for clinical translation and reliability across readers.
The objective of this study is to establish the interreader reliability and variability of segmenting metabolic tumor volume and total lesion glycolysis of primary tumor using two fixed thresholds, 40% and 50% of lesion SUVmax, in patients with head and neck, lung, and breast cancers and to investigate the impact of the lesion FDG avidity and lesion location on interreader reliability and variability.
Materials and Methods
Patients and Study Design
This study is a retrospective evaluation of PET/CT images. Approval from our institutional review board was granted with a waiver of informed consent. Patients with lung, head and neck, and breast cancers who underwent a baseline PET/CT at our institution in 2009 were included in the study. Patients who underwent prior local or systemic therapy were excluded. The study population included 111 patients (mean [± SD] age, 61.9 ± 12.5 years). We analyzed 121 patients in our previous study to establish the intrareader agreement [16], using only one reader. We used the same patient cohort as described in the previous study, except for 10 of these patients, because of technical failure of transfer of images from our archival system into the dedicated workstation. Hence, our study cohort was a total of 111 patients with head and neck, breast, and lung cancers.
PET/CT
All PET/CT studies were performed on a PET/CT scanner (Discovery STE 16, GE Healthcare) according to the institutional standard clinical protocol. A dedicated head and neck protocol was instituted for all patients with head and neck cancer. These dedicated head and neck scans were done from skull base to aortic arch with the arms down and were followed by a second scan from clavicle to mid thigh with the arms up to complete imaging from the skull base to the mid thighs. For patients with breast and lung cancers, the scans were done from skull base to mid thigh with the arms up. The average patient blood glucose level was 101.5 ± 24.8 mg/dL. Patients were injected with a mean dose of 13.1 ± 2.9 mCi (484.7 ± 107.3 MBq) of FDG, and the mean uptake time was 91.7 ± 25.2 minutes.
Image Analysis
All PET/CT studies were retrieved from the electronic archival system and were reviewed on a workstation (Advantage, software version 4.2, GE Healthcare). To establish the interreader reliability, two readers, a junior and senior faculty member, performed all reads independently. The junior faculty is board certified in radiology, is nuclear radiology fellowship trained, and had 4 years of experience reading PET/CT (reader 1). The senior faculty is dual board certified in radiology and nuclear medicine and had 10 years of experience reading PET/CT (reader 2). Reader 1 was the original reader in the prior intrareader study. The bias of his measurements for this study is minimized because this study was conducted with different analytic software, all measurements are independent of intrareader study, PET/CT scans were reviewed in random order, and there was a 6-month interval between the two studies.
For patients who had more than one lesion, the index lesion was a priori identified so that both readers would segment the same lesion for metabolic tumor volume and total lesion glycolysis. Thus, only one lesion was segmented in each patient. PET, CT, and fused PET/CT images were reviewed in axial, coronal, and sagittal planes. For the purposes of this study, the relevant imaging parameter measurements were the primary tumor SUVmax, metabolic tumor volume, and total lesion glycolysis segmented from PET. Metabolic tumor volume was defined as the tumor volume with FDG uptake segmented by two fixed thresholds, 40% and 50% of SUVmax. The total lesion glycolysis was defined as metabolic tumor volume multiplied by the mean SUV. We did not segment the FDG tumor volumetric parameters using a gradient segmentation method, because this method was not available in the workstation we used.
The fixed SUVmax threshold contouring method relies on including all voxels that are greater than a defined percentage of the maximal voxel within an operator-defined sphere. In this study, we have chosen two fixed thresholds, 40% and 50% of lesion SUVmax, because threshold values of 40–50% have been commonly used in previous studies [17–19]. Cross-sectional circles are displayed in all three projections (axial, sagittal, and coronal) to ensure 3D coverage of the primary tumor (Figs. 1A and 1B). The edges of the primary tumor are semiautomatically calculated and outlined by the software. Because the readers define the cross-sectional circles, the reader variability may be introduced, if the readers are not careful to include all parts of the FDG-avid lesion in three dimensions or if the readers included adjacent FDG-avid structures. Either of these would lead to errors in calculated semiautomatic metabolic tumor volume and total lesion glycolysis. The threshold segmentation methods of volume measurement available in the Advantage Workstation software have been previously used [14].
Fig. 1A —Primary tumor segmentations in two different patients.
A, 59-year-old woman with right upper lobe non–small cell lung cancer and distant atelectasis with maximum standardized uptake value (SUVmax) of 27.8. Images show 50% SUVmax segmentation in axial (left), sagittal (middle), and coronal (right) planes generated by PET (top row) and fused PET/CT (bottom row). Metabolic tumor volume was 19.85 mL, and total lesion glycolysis was 367.1 g.
Fig. 1B —Primary tumor segmentations in two different patients.
B, 59-year-old man with right floor of mouth squamous cell carcinoma with SUVmax of 18.3. Images show 50% SUVmax segmentation in axial (left), sagittal (middle), and coronal (right) planes generated by PET (top row) and fused PET/CT (bottom row). Metabolic tumor volume was 2.3 mL, and total lesion glycolysis was 25.9 g.
Statistical Methods
We present our summary statistics as the mean ± SD for continuous variables or frequency and percentage for categoric variables. The reliability of metabolic tumor volume and total lesion glycolysis was measured using the intraclass correlation coefficient (ICC), as generated by two-way random effects model and an absolute agreement definition, and is reported as a point estimate with 95% CI. The independent variables are the patients, and the dependent variables are the readers’ measurements. Readers and patients are considered as randomly selected from a large pool of readers and patients. The absolute agreement definition considers that systematic differences between the two readers are relevant in estimating the ICC. The ICC ranges between 0 and 1.0, with values closer to 1.0 representing better reproducibility [20]. The reproducibility of ICCs based on their precision (half width of 95% CI × 100%) was also established. Variability of measurements between the readers was measured with Bland-Altman analysis with bias, SD, and 95% CI. Analysis was performed for absolute difference in metabolic tumor volume and total lesion glycolysis between the two readers.
To investigate the impact of FDG avidity of the primary tumors on interreader reliability and variability of volumetric segmentation, we performed ICCs and Bland-Altman subgroup analysis of the 30 lesions (patients) with lowest SUVmax and the 30 lesions (patients) with highest SUVmax of the study population. Furthermore, we investigated the impact of location of the tumors (head and neck or lung or breast) on the interreader reliability and variability of volumetric segmentation. We used the Prism 5 (GraphPad Software) and SPSS 20 (SPSS) statistical packages for all analyses, and all hypothesis tests are two-sided with a significance level of 0.05.
Results
There were 111 patients (mean age, 61.9 ± 12.5 years), including 54 men (48.6%) (mean age, 57.8 ± 13.4 years) and 57 women (51.4%) (mean age, 65.1 ± 10.9 years). Among these, 39 patients had lung cancer (35.1%), 37 patients had head and neck cancer (33%), and 35 patients had breast cancers (31.5%). American Joint Committee on Cancer stages included stage I (n = 14), stage II (n = 24), stage III (n = 35), and stage IV (n = 38).
SUVmax, Metabolic Tumor Volume, and Total Lesion Glycolysis
The mean SUVmax of primary tumors was 12.78 (range, 2.5–32.4) for reader 1 and 12.8 (range, 2.5–32.4) for reader 2. The mean metabolic tumor volume and total lesion glycolysis calculated using 40% and 50% SUVmax thresholds for reader 1 and reader 2 are summarized for the entire study population; for patients with lung cancer, head and neck cancer, and breast cancer; for the 30 patients with primary tumors of the highest SUVmax; and for the 30 patients with primary tumors of the lowest SUVmax (Table 1).
| Patient Group | SUVmax | 40% SUVmax Segmentations of Metabolic Tumor Volume (mL) | 50% SUVmax Segmentations of Metabolic Tumor Volume (mL) | 40% SUVmax Segmentations of Total Lesion Glycolysis (g) | 50% SUVmax Segmentations of Total Lesion Glycolysis (g) | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| Reader 1 | Reader 2 | Reader 1 | Reader 2 | Reader 1 | Reader 2 | Reader 1 | Reader 2 | Reader 1 | Reader 2 | |
| All patients (n = 111) | 12.78 ± 6.4 | 12.8 ± 6.4 | 21.27 ± 34.4 | 21.97 ± 34.0 | 14.77 ± 24.3 | 14.99 ± 24.4 | 194.1 ± 350.5 | 197.1 ± 353.5 | 150.3 ± 272.2 | 151.9 ± 274.3 |
| Lung tumors (n = 39) | 12.99 ± 5.9 | 13.0 ± 5.9 | 32.79 ± 49.2 | 33.36 ± 49.9 | 22.69 ± 35.2 | 22.93 ± 35.4 | 299.3 ± 467.7 | 303.4 ± 474.2 | 231 ± 369.1 | 233.3 ± 373.3 |
| Head and neck tumors (n = 37) | 15.3 ± 6.2 | 15.3 ± 6.2 | 8.99 ± 7.6 | 9.86 ± 7.8 | 6.44 ± 6.0 | 6.73 ± 6.0 | 97.3 ± 99.6 | 102.1 ± 99.6 | 78.0 ± 83.2 | 79.9 ± 83.6 |
| Breast tumors (n = 35) | 9.87 ± 6.4 | 9.93 ± 6.3 | 27.5 ± 44.8 | 31.6 ± 55.6 | 19.13 ± 33.1 | 20.0 ± 34.4 | 183.7 ± 242.7 | 197.2 ± 357.1 | 140.6 ± 256.7 | 145.4 ± 261.4 |
| High-FDG-avid tumors (n = 30) | 21.15 ± 4.08 | 21.16 ± 4.08 | 26.25 ± 29.1 | 26.69 ± 29.4 | 18.58 ± 20.81 | 18.81 ± 21.0 | 343.9 ± 404.2 | 348.8 ± 408.7 | 152.5 ± 382.2 | 154.3 ± 386.5 |
| Low-FDG-avid tumors (n = 30) | 5.32 ± 1.3 | 5.3 ± 1.3 | 11.0 ± 9.2 | 12.0 ± 10.5 | 7.02 ± 6.0 | 7.33 ± 6.3 | 35.81 ± 34.7 | 38.64 ± 38.4 | 26.42 ± 25.9 | 27.39 ± 27.3 |
Note—Data are mean ± SD.
Metabolic Tumor Volume and Total Lesion Glycolysis: Interreader Agreement and Precision
The interreader agreement for metabolic tumor volume and total lesion glycolysis was established using ICCs for all 111 primary tumors and also for the breast, lung, and head and neck tumors separately. In all primary tumors, the ICCs for 40% and 50% SUVmax segmentations of metabolic tumor volume between the readers were 0.987 (95% CI, 0.981–0.991; precision, 0.5%) and 0.995 (95% CI, 0.993–0.997; precision, 0.2%), respectively (Table 2). The ICCs for 40% and 50% SUVmax segmentations of total lesion glycolysis segmentations 50% SUV max Segmentations of between the readers were 0.987 (95% CI, 0.98–0.991; precision, 0.5%) and 0.986 (95% CI, 0.98–0.991; precision, 0.5%), respectively (p = 0.0001 for all point estimates) (Table 3).
| Patient Group | 40% SUVmax Segmentations | 50% SUVmax Segmentations | ||
|---|---|---|---|---|
| ICC (95% CI) | Precision (%) | ICC (95% CI) | Precision (%) | |
| All patients (n = 111) | 0.987 (0.981–0.991) | 0.5 | 0.995 (0.993–0.997) | 0.2 |
| Lung tumors (n = 39) | 1.0 (0.999–1.0) | 0.05 | 1.0 (1.0–0.999) | 0.05 |
| Head and neck tumors (n = 37) | 0.989 (0.966–0.996) | 1.5 | 0.997 (0.993–0.999) | 0.3 |
| Breast tumors (n = 35) | 0.977 (0.955–0.989) | 1.7 | 0.997 (0.994–0.999) | 0.25 |
| High-FDG-avid tumors (n = 30) | 0.999 (0.999–0.999) | 0 | 1.0 (0.999–1.0) | 0 |
| Low-FDG-avid tumors (n = 30) | 0.974 (0.944–0.988) | 2 | 0.997 (0.993–0.998) | 0.2 |
Note—Precision is defined as (95% CI × 100%) / 2. SUVmax = maximum standardized uptake value, ICC = intraclass correlation coefficient.
| Patient Group | 40% SUVmax Segmentations | 50% SUVmax Segmentations | ||
|---|---|---|---|---|
| ICC (95% CI) | Precision (%) | ICC (95% CI) | Precision (%) | |
| All patients (n = 111) | 0.987 (0.98–0.991) | 0.5 | 0.986 (0.98–0.991) | 0.5 |
| Lung tumors (n = 39) | 1.0 (1.0–0.999) | 0.05 | 1.0 (1.0–1.0) | 0 |
| Head and neck tumors (n = 37) | 0.998 (0.994–0.999) | 0.25 | 0.999 (0.998–1.0) | 0.1 |
| Breast tumors (n = 35) | 0.994 (0.988–0.997) | 0.45 | 0.998 (0.997–0.99) | 0.1 |
| High-FDG-avid tumors (n = 30) | 1.0 (0.999–1.0) | 0 | 1.0 (0.999–1.0) | 0 |
| Low-FDG-avid tumors (n = 30) | 0.962 (0.92–0.982) | 3 | 0.99 (0.978–0.995) | 0.8 |
Note—Precision is defined as (95% CI × 100%) / 2. SUVmax = maximum standardized uptake value, ICC = intraclass correlation coefficient.
The ICCs for 40% and 50% SUVmax segmentations of metabolic tumor volume and total lesion glycolysis with corresponding precision in the subgroups of lung, head and neck, and breast tumors are summarized in Tables 2 and 3. In addition, subgroup analysis among the 30 patients with highest SUVmax of the primary tumor and the lowest SUVmax of the primary tumor are also summarized in Tables 2 and 3. The precision was narrower for 50% SUVmax segmentations of metabolic tumor volume and total lesion glycolysis than for 40% SUVmax segmentations of metabolic tumor volume and total lesion glycolysis, respectively, for all subgroups. The narrowest precision for metabolic tumor volume was observed for high-FDG-avid tumors (0%) followed by lung tumors (0.05%) using 50% SUVmax segmentation. The widest precision for metabolic tumor volume was observed for low-FDG-avid tumors (2%) followed by breast tumors (1.7%) using 40% SUVmax segmentation. A similar pattern was evident for total lesion glycolysis segmentation.
Metabolic Tumor Volume and Total Lesion Glycolysis: Interreader Variability
The mean biases between the readers for metabolic tumor volume 40% and 50% SUVmax segmentations for all study patients were −1.78 ± 8.42 mL (95% CI, −18.29 to 14.72 mL) and −0.46 ± 2.1 mL (95% CI, −4.52 to 3.58 mL), respectively. The corresponding absolute biases for total lesion glycolysis were −7.3 ± 31.6 g (95% CI, −69.3 to 54.7 g) and −2.97 ± 12.86 g (95% CI, −28.19 to 22.24 g), respectively (Tables 4 and 5 and Fig. 2).
| Patient Group | 40% SUVmax Segmentations | 50% SUVmax Segmentations | ||
|---|---|---|---|---|
| Mean Bias (95% CI) | SD | Mean Bias (95% CI) | SD | |
| All patients (n = 111) | −1.78 (−18.29 to 14.72) | 8.42 | −0.46 (−4.52 to 3.58) | 2.1 |
| Lung tumors (n = 39) | −0.56 (−4.19 to 3.05) | 1.85 | −0.24 (−2.02 to 1.52) | 0.9 |
| Head and neck tumors (n = 37) | −0.86 (−3.56 to 1.8) | 1.37 | −0.29 (−1.39 to 0.81) | 0.56 |
| Breast tumors (n = 35) | −4.11 (−32.88 to 24.65) | 14.67 | 0.89 (−7.77 to 5.98) | 3.5 |
| High-FDG-avid tumors (n = 30) | −0.44 (−3.22 to 2.33) | 1.41 | −0.23 (−1.68 to 1.22) | 0.24 |
| Low-FDG-avid tumors (n = 30) | −1.05 (−6.2 to 4.11) | 2.83 | −0.31 (−1.99 to 1.38) | 0.86 |
Note—Data are metabolic tumor volumes measured in milliliters.
| Patient Group | 40% SUVmax Segmentations | 50% SUVmax Segmentations | ||
|---|---|---|---|---|
| Mean Bias (95% CI) | SD | Mean Bias (95% CI) | SD | |
| All patients (n = 111) | −7.3 (−69.3 to 54.7) | 31.6 | −2.97 (−28.19 to 22.24) | 12.86 |
| Lung tumors (n = 39) | −4.09 (−36.9 to 28.7) | 16.7 | −2.31 (−21.93 to 17.3) | 10.0 |
| Head and neck tumors (n = 37) | −4.77 (−20.4 to 10.84) | 7.97 | −1.94 (9.61 to 5.72) | 3.91 |
| Breast tumors (n = 35) | −13.5 (−11.72 to 90.19) | 52.9 | −4.79 (−44.10 to 34.51) | 20.05 |
| High-FDG-avid tumors (n = 30) | −4.89 (−38.55 to 28.76) | 17.17 | −2.98 (−23.47 to 17.49) | 10.45 |
| Low-FDG-avid tumors (n = 30) | −2.82 (−18.9 to 13.2) | 8.21 | −0.96 (−6.95 to 5.01) | 3.05 |
Note—Data are total lesion glycolysis measured in grams.
The Bland-Altman biases, SD, and 95% CI for 40% and 50% SUVmax metabolic tumor volume and total lesion glycolysis segmentations in the subgroups of lung, head and neck, and breast tumors are summarized in Tables 2 and 3. In addition, subgroup analysis among the 30 patients with highest SUVmax of the primary tumor and the lowest SUVmax of the primary tumor are also summarized in Tables 4 and 5 and in Figures 3 and 4.
Discussion
The goal of our study was to establish the interreader reliability and variability of metabolic tumor volume and total lesion glycolysis using two fixed threshold isocontouring methods in various human solid tumors for quantitative imaging. We also investigated the impact of index lesion SUVmax on the interreader reliability and variability of metabolic tumor volume and total lesion glycolysis. Our results show that interreader reliability is excellent for both metabolic tumor volume and total lesion glycolysis segmented using 40% and 50% SUVmax. The precision is narrower for 50% SUVmax segmentation than for 40% SUVmax segmentation for the entire study population and for the breast, head and neck, and lung cancer subgroups. For low-FDG-avid lesions, the ICC precision width was also narrower for 50% SUVmax segmentation for metabolic tumor volume and total lesion glycolysis than 40% SUVmax segmentations. For high-FDG-avid lesions, the ICC precision is similar for both metabolic tumor volume and total lesion glycolysis segmented by both threshold methods. The Bland-Altman plots show that the variability between the readers (95% CI width of bias) is narrower for 50% SUVmax segmentations for both metabolic tumor volume and total lesion glycolysis for the entire study population as well as for breast, head and neck, and lung cancer subgroups and low- and high-FDG-avid lesions. The variability was also narrower for high-FDG-avid lesions for both metabolic tumor volume and total lesion glycolysis than for low-FDG-avid lesions.
The implications of these findings are significant for quantitative imaging with FDG PET volumetric imaging parameters. The FDG avidity and location of the tumors affect the precision and interreader variability of the measurements of FDG volumetric parameters. Fifty-percent SUVmax segmentations may be appropriate to use because they show narrower precision and variability with all subgroups. Because FDG avidity of the lesions influences the precision and variability, it may be necessary to establish a level of FDG avidity as a cutoff point, above which the volumetric parameters can be applied for therapy assessment in oncology. The lower interreader variability for 50% SUVmax segmentation can be explained by the fact that it includes central higher metabolically active volume than does 40% SUVmax segmentation. This is similar to the results from prior phantom studies showing lesser variability with 50% SUVmax segmentation than with 42% SUVmax segmentation in repeatability studies [21]. In our study, the ICCs for 50% SUVmax segmentation approach 1.0, similar to the ICCs observed for SUVmax in prior studies. Hence, the 50% SUVmax segmentation for metabolic tumor volume and total lesion glycolysis can be used for therapy response assessment and as prognostic FDG parameters with high interreader reliability and minimal variability.
There are advantages and disadvantages for threshold-based segmentations. The threshold segmentations based on a fixed percentage of SUVmax are very easy to implement and use [22]. The repeatability of fixed threshold segmentations is better than adaptive threshold segmentation, because the latter is dependent on a manually drawn region of interest leading to variable delineation [21]. However, the accuracy of threshold-based approaches can be limited for lesions with a nonhomogeneous distribution [23] of FDG uptake.
The results of this study provide preliminary data for more challenging investigation, in the future, for establishing pre- and post-therapy metabolic tumor volume and total lesion glycolysis assessment and interreader reliability and variability of these measurements in human solid tumors. More challenging variables, such as radiation-induced inflammation, would affect the interreader reliability and variability of these measurements in the post-therapy settings. It is important to establish the interreader agreement and variability so that treatment-related changes can be estimated more accurately for therapy assessment. It appears that FDG volumetric tumor parameters are superior to SUVmax and peak SUV in predicting the outcome of oncologic patients with multiple solid tumors [24–27] and are very likely to be used in routine clinical practice in the future. This requires careful investigations to establish the biologic and reader variability of these volumetric parameters, before deployment in clinical practice. The 50% SUVmax segmentation has a precision width of 0.2% and Bland-Altman bias 95% CI of 8.1 mL for metabolic tumor volume for the entire study population. Hence, the measurements may be more reliable for tumors with more than 8 mL of metabolic tumor volume.
Our results are limited because we included only two isocontour fixed threshold segmentation methods in this study. Further studies are needed to assess the other segmentation methods, such as the gradient and adaptive segmentation methods, to establish the interreader agreement of metabolic tumor volume and total lesion glycolysis in human solid tumors, compared with the 50% SUVmax threshold. We used a single vendor's software to segment the tumors, and it would be necessary to establish the interreader agreement of volumetric FDG parameters across different viewing platforms and software programs from different commercial vendors for clinical translation. In addition, the study was limited to only two readers from the same institution, albeit with different experience levels in reading PET scans and a priori–identified lesions. It would be ideal to perform the study with four or five readers from different institutions and with varied professional experience to simulate clinical practice. We included only untreated tumors and only three different tumor types. However, measuring metabolic tumor volume and total lesion glycolysis after treatment may be challenging because treatment-related inflammation may affect adjacent normal tissue. In addition, other FDG-avid tumors, especially those with low FDG avidity that is similar to that of background, such as a subgroup of pancreatic cancers, may be difficult to segment, and measuring the metabolic tumor volume and total lesion glycolysis can be challenging. We also identified the index lesion, a priori, if there was more than one lesion. Hence, the interreader agreement and variability estimates reported in this study are likely high-end estimates. We would need to perform the study with readers from different institutions who have varied experience and with both treated and untreated and unidentified lesions in future.
In conclusion, there is excellent interreader agreement, with ICCs approaching 1.0, for measurement of metabolic tumor volume and total lesion glycolysis with 40% and 50% SUVmax segmentations in various human solid tumors. The precision width is narrower for 50% SUVmax than for 40% SUVmax segmentation. FDG avidity of the tumors and location of the tumors influence the interreader reliability, precision, and variability of FDG volumetric parameter measurements. The 50% SUVmax segmentation has a precision width of 0.2% and a Bland-Altman bias 95% CI of 8.1 mL for metabolic tumor volume for the entire study population. Hence, the measurements may be more reliable for tumors with greater than 8 mL of metabolic tumor volume.
Footnote
R. M. Subramaniam was supported by a GE-AUR research fellowship at the Department of Radiology, Boston University School of Medicine.
References
1.
Larson SM, Schwartz LH. 18F-FDG PET as a candidate for “qualified biomarker”: functional assessment of treatment response in oncology. J Nucl Med 2006; 47:901–903
2.
Weber WA. Positron emission tomography as an imaging biomarker. J Clin Oncol 2006; 24:3282–3292
3.
Wahl RL, Jacene H, Kasamon Y, Lodge MA. From RECIST to PERCIST: evolving considerations for PET response criteria in solid tumors. J Nucl Med 2009; 50:122S–150S
4.
Davison JM, Ozonoff A, Imsande HM, Grillone GA, Subramaniam RM. Squamous cell carcinoma of the palatine tonsils: FDG standardized uptake value ratio as a biomarker to differentiate tonsillar carcinoma from physiologic uptake. Radiology 2010; 255:578–585
5.
Subramaniam RM, Wilcox B, Aubry MC, Jett J, Peller PJ. 18F-fluoro-2-deoxy-D-glucose positron emission tomography and positron emission tomography/computed tomography imaging of malignant pleural mesothelioma. J Med Imaging Radiat Oncol 2009; 53:160–169; quiz, 170
6.
Subramaniam RM, Clayton AC, Karantanis D, Collins DA. Hibernoma: 18F FDG PET/CT imaging. J Thorac Oncol 2007; 2:569–570
7.
Karantanis D, O'Neill BP, Subramaniam RM, et al. Contribution of F-18 FDG PET-CT in the detection of systemic spread of primary central nervous system lymphoma. Clin Nucl Med 2007; 32:271–274
8.
Jackson T, Chung MK, Mercier G, Ozonoff A, Subramaniam RM. FDG PET/CT interobserver agreement in head and neck cancer: FDG and CT measurements of the primary tumor site. Nucl Med Commun 2012; 33:305–312
9.
Sacks A, Peller PJ, Surasi DS, Chatburn L, Mercier G, Subramaniam RM. Value of PET/CT in the management of primary hepatobiliary tumors. Part 2. AJR 2011; 197:[web]W260–W265
10.
Shum WY, Ding HJ, Liang JA, Yen KY, Chen SW, Kao CH. Use of pretreatment metabolic tumor volumes on PET-CT to predict the survival of patients with squamous cell carcinoma of esophagus treated by curative surgery. Anticancer Res 2012; 32:4163–4168
11.
Park GC, Kim JS, Roh JL, Choi SH, Nam SY, Kim SY. Prognostic value of metabolic tumor volume measured by 18F-FDG PET/CT in advanced-stage squamous cell carcinoma of the larynx and hypopharynx. Ann Oncol 2013; 24:208–214
12.
Abelson JA, Murphy JD, Trakul N, et al. Metabolic imaging metrics correlate with survival in early stage lung cancer treated with stereotactic ablative radiotherapy. Lung Cancer 2012; 78:219–224
13.
Dibble EH, Lara Alvarez AC, Truong MT, Mercier G, Cook EF, Subramaniam RM. 18F-FDG metabolic tumor volume and total glycolytic activity of oral cavity and oropharyngeal squamous cell cancer: adding value to clinical staging. J Nucl Med 2012; 53:709–715
14.
Lim R, Eaton A, Lee NY, et al. 18F-FDG PET/CT metabolic tumor volume and total lesion glycolysis predict outcome in oropharyngeal squamous cell carcinoma. J Nucl Med 2012; 53:1506–1513
15.
Hatt M, Visvikis D, Le Rest CC. Autocontouring versus manual contouring. J Nucl Med 2011; 52:658; author reply, 658–659
16.
Shah B, Srivastava N, Hirsch AE, Mercier G, Subramaniam RM. Intra-reader reliability of FDG PET volumetric tumor parameters: effects of primary tumor size and segmentation methods. Ann Nucl Med 2012; 26:707–714
17.
Nestle U, Schaefer-Schuler A, Kremp S, et al. Target volume definition for 18F-FDG PET-positive lymph nodes in radiotherapy of patients with non-small cell lung cancer. Eur J Nucl Med Mol Imaging 2007; 34:453–462
18.
Daisne JF, Sibomana M, Bol A, Doumont T, Lonneux M, Gregoire V. Tri-dimensional automatic segmentation of PET volumes based on measured source-to-background ratios: influence of reconstruction algorithms. Radiother Oncol 2003; 69:247–250
19.
Erdi YE, Mawlawi O, Larson SM, et al. Segmentation of lung lesion volume by adaptive positron emission tomography image thresholding. Cancer 1997; 80:2505–2509
20.
Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977; 33:159–174
21.
Hatt M, Cheze Le Rest C, Albarghach N, Pradier O, Visvikis D. PET functional volume delineation: a robustness and repeatability study. Eur J Nucl Med Mol Imaging 2011; 38:663–672
22.
Dewalle-Vignion AS, Yeni N, Petyt G, et al. Evaluation of PET volume segmentation methods: comparisons with expert manual delineations. Nucl Med Commun 2012; 33:34–42
23.
Hatt M, Cheze-le Rest C, van Baardwijk A, Lambin P, Pradier O, Visvikis D. Impact of tumor size and tracer uptake heterogeneity in 18F-FDG PET and CT non-small cell lung cancer tumor delineation. J Nucl Med 2011; 52:1690–1697
24.
Paidpally V, Chirindel A, Lam S, Agrawal N, Quon H, Subramaniam RM. FDG-PET/CT imaging biomarkers in head and neck squamous cell carcinoma. Imaging Med 2012; 4:633–647
25.
Romesser PB, Qureshi MM, Shah BA, et al. Superior prognostic utility of gross and metabolic tumor volume compared to standardized uptake value using PET/CT in head and neck squamous cell carcinoma patients treated with intensity-modulated radiotherapy. Ann Nucl Med 2012; 26:527–534
26.
Davison J, Mercier G, Russo G, Subramaniam RM. PET-based primary tumor volumetric parameters and survival of patients with non–small cell lung carcinoma. AJR 2013; 200:635–640
27.
Bazan JG, Koong AC, Kapp DS, et al. Metabolic tumor volume predicts disease progression and survival in patients with squamous cell carcinoma of the anal canal. J Nucl Med 2013; 54:27–32
Information & Authors
Information
Published In
Copyright
© American Roentgen Ray Society.
History
Submitted: February 22, 2013
Accepted: June 18, 2013
Keywords
Authors
Metrics & Citations
Metrics
Citations
Export Citations
To download the citation to this article, select your reference manager software.