Diagnostic Accuracy of 18F-Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography in Preoperative Mediastinal/Extramediastinal Nodal Staging of Non–Small-Cell Lung Carcinoma

Hong Kong J Radiol 2023 Mar;26(1):6-13

 

 

Nuclear Medicine Unit and Clinical PET Centre, Queen Elizabeth Hospital, Hong Kong

 

  Full Article in PDF

In many countries, lung cancer has the highest incidence and mortality among all malignancies.[1] Mediastinal/extramediastinal nodal status is the most important factor determining the management of early non–small-cell lung carcinoma (NSCLC). For N0 or N1 disease, according to the 7th and 8th editions of American Joint Committee on Cancer staging,[2] [3] curative surgical resection can be offered.[4] For N2 (subcarinal or ipsilateral mediastinal metastasis) disease, the usual treatment is chemoradiotherapy (CRT).[4] Accurate mediastinal nodal staging is critical but challenging. Whereas mediastinoscopy is the gold standard, this invasive procedure carries 0.5% life-threatening risks of major complications including arrhythmia, respiratory failure, and infection.[5] Although staging with computed tomography based on nodal diameter is non-invasive, it has limited accuracy.[6] [7] ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) positron emission tomography/computed tomography (PET/CT) is useful in NSCLC staging.[8] [9] [10] [11] Both semiquantitative assessment and qualitative visual interpretation are recommended for distinguishing benign from malignant lymph nodes (LNs).[10] [11] [12] [13] [14] In a semiquantitative approach, the maximum standardised uptake value (SUV_max) is used.[10] [11] [12] Yet, SUV_max depends on many physiological as well as technical factors, such as injection time, uptake period, and blood glucose level.[15] [16] Thus, absolute SUV_max is difficult to compare between different PET/CT systems or subjects. Studies have compared the accuracies of simple SUV_max and visual score generated by different comparisons with the activity in the aorta and other locations, as well as with a simple unified windowing technique.[13] [14] We aimed to evaluate the performance of ¹⁸F-FDG PET/CT on mediastinal/extramediastinal nodal staging of NSCLC based on SUV_max, visual score, and nodal diameter. The disparity between radiological and pathological staging was also reviewed.

 

Cases of newly diagnosed lung cancer patients who underwent whole-body ¹⁸F-FDG PET/CT for staging in our centre at Queen Elizabeth Hospital from 1 January 2016 to 31 December 2016 were retrospectively analysed. Cases were included only if they had (1) histological evidence of primary NSCLC and (2) histological mediastinal/extramediastinal LN staging. Cases were excluded if (1) the time interval between PET/CT and histological LN examination was >2 months, or (2) treatment was started before PET/CT.

 

All PET/CT examinations were acquired with a Discovery 710 (General Electric, Milwaukee [WI], US) according to the 2010 procedural protocol of the European Association of Nuclear Medicine for oncological PET imaging.[17] The mean ¹⁸F-FDG activity administrated was 370 ± 44.4 MBq. After a mean uptake time of 60 minutes (standard deviation = 6.24), image data were acquired from the skull vertex to the mid-thighs in 7 to 8 bed positions (3 minutes per bed position) with mean axial bed coverage of 15.2 cm per bed and 9 sections bed overlap in the 2-dimensional acquisition mode. Reconstruction using optimisation of ordered subset expectation maximisation was performed with 4.2 mm section thickness in a 128 × 128 matrix and processed through a standard filter. Non-contrast CT was acquired for anatomical correlation and attenuation correction with the following parameters: 120 mA tube current, 140 kV tube voltage, 0.8 s gantry rotation speed, 0.75 pitch, 0.5 mm section thickness and 512 × 512 matrix. The mean blood glucose level was 5.5 mmol/L (standard deviation = 0.91).

 

This study evaluated the accuracy of staging PET/CT in the diagnosis of mediastinal/extramediastinal LNs. A nuclear medicine physician, blinded to the patients’ clinical background and histological results, assessed the PET/CT images according to the Mountain and Dresler nodal station scheme[18] with an Advantage Workstation Volume Viewer 4.7 (General Electric, Milwaukee [WI], US). If the histologically evaluated LNs were identifiable on PET/CT, three nodal features were analysed: SUV_max, visual score (as illustrated below) and diameter (short axis in axial plane). SUV is defined as the activity measured in a volume of interest divided by the injected ¹⁸F-FDG dose, based on body weight:

 

 

A five-point visual score of LNs was used, with a standardised lower threshold set at 0 and the upper threshold at 2 times the liver’s mean standardised uptake value (SUV_mean).[14] The LNs were rated based on the maximum intensity projection image with grey scale images (Figure 1):

Score 0: No LN uptake

Score 1: LN uptake < mediastinal blood pool

Score 2: mediastinal blood pool ≤ LN uptake < SUV_mean of liver

Score 3: SUV_mean of liver ≤ LN uptake <2 × SUV_mean of liver

Score 4: LN uptake ≥2 × SUV_mean of liver

 

Figure 1. Representative maximum intensity projection for visual scores of 1 to 4, with bloodpool, liver, and primary tumour labelled (arrows).

 

For primary tumour with uptake similar to that of liver, the respective LNs were scored 4 if their uptakes were similar to the uptake of the primary.[14] If a LN was evaluated histologically but unidentifiable on PET/CT, a score of 0 was assigned and a 2-cm sphere was marked as a volume of interest in the corresponding anatomical location for SUV_max.

 

Mediastinal/extramediastinal LNs were sampled by cardiothoracic surgeons or respiratory physicians via lobectomy (59%), mediastinoscopy (14%), video-assisted thoracoscopic surgery (11%), endobronchial ultrasound (8%), pneumonectomy (5%) or wedge resection (3%). Anatomical staging was evaluated by pathologists according to Mountain and Dresler scheme.[18]

 

Statistical analysis was performed using SPSS (Window version 20.0; IBM Corp, Armonk [NY], US). Receiver operating characteristics (ROC) curves as well as corresponding areas under the curve (AUCs) based on SUV_max, visual score, and nodal diameter were compared. The optimal cut-off to distinguish benign from malignant LNs in PET/CT is the minimum of √ (1−specificity)² + (1−sensitivity)² . The respective sensitivities, specificities, accuracies, false negative and positive values were determined. Subgroup analyses were also examined according to primary tumour T stage (T1-2 vs. T3-4), histology (adenocarcinoma vs. squamous cell carcinoma), epidermal growth factor receptor (EGFR) status (wild type vs. mutated), primary tumour SUV_max (<10 vs. ≥10), and LN location (hilar vs. other mediastinal).

 

A total of 467 mediastinal/extramediastinal LNs from 97 patients (62 male, 35 female) of mean age 66 ± 8.4 years were included. The characteristics of the study population are summarised in Table 1. Adenocarcinoma was the most commonly reported histological finding. The properties of LNs are outlined in Table 2.

 

Table 1. Characteristics of the study population (n = 97).

 

Table 2. Properties of lymph nodes (n = 467).

 

Figure 2 shows the ROC curves based on visual score, SUV_max, and nodal diameter. Their corresponding AUCs are listed in Table 3. The visual score achieved the highest AUC of 0.901 (95% confidence interval [CI] = 0.870-0.926), compared with 0.897 (95% CI = 0.866-0.923) for SUV_max and 0.804 (95% CI = 0.737-0.872) for nodal diameter, respectively. Whereas the difference in AUCs for visual score and SUV_max were not statistically significant (p = 0.796), their AUCs were significantly greater than those for nodal diameter (p = 0.0003). Figure 3 shows √ (1−specificity)² + (1−sensitivity)²  against visual score (lower X-axis) and SUV_max (upper X-axis). As mentioned above, the optimal diagnostic cut-off corresponds to the minimum of √ (1−specificity)² + (1−sensitivity)²  . The cut-off for visual score was 3, implying that LNs with uptakes greater than or equal to that of liver should be suspicious for metastases. The optimal cut-off for SUV_max was 2.5 as demonstrated in Figure 3. Similar analysis for nodal diameter (not shown) gave a cut-off of 10 mm.

 

Figure 2. Receiver operating characteristics curves based on visual score, maximum standardised uptake value, and nodal diameter.

 

Table 3. Diagnostic performance based on visual score, maximum standardised uptake value, and nodal diameter.

 

Figure 3. √ (1−specificity)² + (1−sensitivity)²  versus visual score (lower X-axis) and maximum standardised uptake value (upper X-axis).

 

Based on the cut-offs, the corresponding sensitivities, specificities, accuracies, positive predictive values (PPVs) and negative predictive values (NPVs) are evaluated in Table 3. Among SUV_max, visual score and nodal diameter, the visual score yielded the highest specificity (0.932, 95% CI = 0.901-0.953), accuracy (0.916, 95% CI = 0.880-0.953), PPV (0.623, 95% CI = 0.507-0.733), and NPV (0.972, 95% CI = 0.949-0.985). They were higher than those based on nodal diameter (p = 0.0003), although they had no statistically significant difference from the respective values of SUV_max (p > 0.05). On the other hand, the PET/CT achieved limited sensitivity based on nodal diameter (0.660, 95% CI = 0.517-0.785).

 

Subject-based staging rates are shown in Table 4. Based on visual interpretation, 77 subjects (79%) had correct nodal staging by visual score, whereas 12 subjects (12%) were false positive and eight (8%) were false negative. Based on SUV_max, 68 subjects (70%) had correct staging with 23 (24%) wrong upstaging and six (6%) wrong downstaging. Based on nodal diameter, 72 subjects (74%) had correct staging with 11 (11%) wrong upstaging and 14 (14%) wrong downstaging. The overall nodal staging performance achieved no significant difference between visual interpretation and nodal diameter in McNemar’s test (p = 0.190), while that of SUV_max and nodal diameter had statistical significance (p = 0.001).

 

Table 4. Overall nodal staging performance with reference to histology

 

Table 5. Subject-based staging rates according to visual score, maximum standardised uptake value, and nodal diameter.

 

Figure 4 shows the ROC curves of visual score according to T stage (T1-2 vs. T3-4), histology (adenocarcinoma vs. squamous cell carcinoma), EGFR status (wild type vs. mutated), primary tumour SUV_max (<10 vs. ≥10), and LN location (hilar vs. mediastinal). Their corresponding AUC values ranged from 0.838 to 0.961 (Table 6). For all subgroup analyses, the optimal cut-off of visual score was 3. To assess the significance of the difference between the AUCs, two-tailed tests were evaluated within the subgroups. The p values were > 0.05 for all subgroup analyses, except for histology, which had a p value of 0.0458.

 

Figure 4. Subgroup receiver operating characteristics curves according to (a) T stage, (b) histology, (c) epidermal growth factor receptor status, (d) maximum standardised uptake value of primary tumour, and (e) lymph node location based on visual score.

 

Table 6. Subgroup areas under the curve according to T stage, histology, epidermal growth factor receptor status, primary maximum standardised uptake value, and nodal location.

 

Different radiological criteria have been proposed for nodal staging in NSCLC, including semiquantitative SUV_max,[10] [11] qualitative visual interpretation,[13] [14] and nodal diameter.[6] [7] For visual interpretation, the current study focused on a five-point system[14] because of its convenient applicability (with reference to mediastinal blood pool and liver). We demonstrated good performance of the visual score. Subgroup analyses showed satisfactory AUCs in different T stages, histology, EGFR status, LN locations or primary SUV_max. There were no significant differences among the subgroups analyses, except for the histology, which had a borderline p value of 0.0458.

 

There were limited studies comparing the performance of visual score with SUV_max or nodal diameter.[13] Our study suggested that the visual interpretation achieved satisfactory AUC, specificity, accuracy, and NPVs. Such performance was statistically significant (p = 0.0003) compared with that of nodal diameter. While the cut-off for nodal diameter was 10 mm (axial plane), the smallest true positive LN identifiable by visual interpretation had a diameter of 7 mm with a SUV_max of 2.5 (corresponding primary tumour SUV_max = 11 and liver SUV_mean = 1.49). This example demonstrates the higher sensitivity of PET/CT to detect small malignant LNs compared with nodal diameter in CT alone. Although the visual score resulted in higher specificity, accuracy, PPV and NPV compared with SUV_max of 2.5, the differences were not statistically significant. A possible explanation is the similar cut-off magnitudes. A visual score with a cut-off of 3 corresponds to the liver’s SUV_mean (range, 1.22-3.16). This is close to the SUV_max cut-off of 2.5 and therefore yields similar performance.

 

The limited sensitivity of 0.810 in visual interpretation was reviewed by examining the false negative cases. Among the eight false negative LNs, three (25%) may be explained by the limited PET/CT spatial resolution, as their pathology findings showed only a few clusters of tumour cells. Two cases (16.7%) may have been attributable to the close proximity of the primary tumour and LN. As illustrated in Figure 5, a large primary tumour limited the visualisation of the nearby intrapulmonary LN, which was proven to be malignant histologically. Whereas low ¹⁸F-FDG uptake by the primary tumour may limit the identification of nodal metastases,[19] the 38 cases with nodal metastases all had ¹⁸F-FDG–avid primaries. The least avid primary tumour with nodal metastasis had an SUV_max of 3.07, which was still distinguishable from the liver SUV_mean of 2.2. All nine (9.3%) cases with non–¹⁸F-FDG avid primaries had no nodal metastases. We do not know what was behind the remaining seven false negative cases (58.3%).

 

Figure 5. A large primary tumour can hinder the identification of an adjacent intrapulmonary lymph node (arrow), which was malignant in histology.

 

Nodal staging is crucial for treatment of early-stage NSCLC.[4] For N1 and N2 disease treated with surgery, the adjuvant treatment is chemotherapy (N1) and sequential CRT (N2). For inoperable N2 disease, the primary treatment is CRT. Of the SUV_max, visual score and nodal diameter, the visual score achieved the most reliable overall staging, with an accuracy of 79% (Table 4). If a patient is mistakenly upstaged to N2 (10% subjects in the current report), his/her opportunity for curative surgery may be missed as they may receive CRT. If a patient is mistakenly downstaged from N2 disease (2% currently), his/her option for CRT instead of surgery may also be missed.

 

The cut-off of visual score in the current study was 3, implying that the LNs were suspicious for malignancy if their uptakes were greater than or equal to that of the liver (Figure 3). The identical cut-off of 3 was consistently observed in subgroup analyses (Figure 4). On the other hand, the cut-off in a previous study was >3,[14] implying malignancy if the uptake was >2 times the liver uptake or as elevated as that of the primary. The discrepancy of the two cut-offs may be due to different scanners, acquisition protocols, or reconstruction methods. Indeed, the same study suggested malignancy for a score of 3 as it was 3.3 times as likely to be malignant as a LN with a score of 2.[14]

 

In our study, the PPV based on visual interpretation was 0.623, which is less than the approximate value of 0.75 in the previous report.[14] This discrepancy may be explained by the fact that the lower the ratio of histologically malignant/total LNs, the lower the PPV. This ratio was 0.126 in the current study, which is lower than 0.194 in the previous report.[14]

 

The current study was limited by several factors. This was a retrospective study, thus it had no randomisation, and it was difficult to control confounders. The cases were all fit for histological workup and had no evidence of distant metastases. The study evaluated only 97 subjects and all were Chinese. The SUV depended on many factors, including scanner mode, acquisition protocol, and reconstruction method of the PET/CT.

 

PET/CT offers noninvasive preoperative nodal staging of NSCLC. A convenient visual interpretation is demonstrated to have diagnostic performance better than and similar to that of nodal diameter and SUV_max, respectively. The limited sensitivity can be attributed to the spatial resolution of PET/CT.