2. Materials and Methods
2.1. Patients
The Institutional Review Board (IRB) of King Hussein Cancer Center (KHCC) approved this study (ID: 22 KHCC 123). Informed consent was waived because of the retrospective nature of the study. The study was conducted following the Good Clinical Practice and the 1964 Declaration of Helsinki and its later amendments.
A total of 66 patients diagnosed with LSCC from December 2014 to January 2021 were retrospectively enrolled. The patient was included if he/she fulfilled the following inclusion criteria: (1) diagnosed with biopsy-proven LSCC; (2) had baseline 18F-FDG PET/CT and neck MRI performed within 4 weeks before surgery; and (3) underwent surgical total laryngectomy with bilateral or unilateral neck dissection. Patients were excluded if they had synchronous or metachronous neoplasms or baseline PET scans performed >4 weeks before surgery.
Clinicopathologic factors, including patients’ gender, age at diagnosis, tumor stage (based on American Joint Committee on Cancer, 8th edition (AJCC 8th) [
9]), and surgical intervention data, were collected. The data of the primary tumor, including size, location, grade, and laterality, were registered.
2.2. Neck MR Imaging and Interpretation
Magnetic resonance imaging (MRI) of the neck was performed using 1.5-Tesla (1.5 T) and 3-Tesla (3 T) units. A head coil with a width of 30 cm was employed to encompass the region spanning from the frontal sinuses to the C5-C6 level. All patients underwent imaging using axial, sagittal, and coronal T1-weighted turbo spin echo (TSE) sequences, as well as axial and coronal T2-weighted m-DIXON sequences. Additionally, post-contrast material m-DIXON sequences were acquired in axial, coronal, and sagittal planes using T1 weighting. The acquisition of axial T1-weighted fast spin echo (FSE) images involved using a repetition time (TR) of 670 ms and an echo time (TE) of 18 ms. The echo train length was set to 5, and two signals were utilized. The field of view (FOV) was 240 mm, and the matrix size was 225 × 240. Each image section had a thickness of 4 mm, with a 0.4 mm gap between sections. Axial T2-weighted m-DIXON images were obtained using a repetition time (TR) of 2500 ms and an echo time (TE) of 80 ms. The echo train length was set to 19, with a field of view (FOV) of 240 mm. The matrix size was 250 × 380, and the acquired sections had a thickness of 4 mm with a 0.4 mm gap between them. Axial post-contrast material m-DIXON images were obtained using a T1-weighted sequence. The acquisition parameters included a repetition time of 500 ms and an echo time of 7 ms. The echo train length was set to 5, with a field of view of 240 mm and a matrix size of 225 × 225. Each image section had a thickness of 4 mm, with a 0.4 mm gap between sections. In the contrast-enhanced series, a bolus injection of gadolinium-based contrast agent was administered intravenously at a rate of 2 mL/s, with a dosage of 0.2 mmol/kg of body weight. The series of neck MRI studies were evaluated by a proficient radiologist. The acquisition of these images occurred within a maximum interval of two weeks or simultaneously with the acquisition of the
18F-FDG PET/CT images. The metastatic status of cervical LNs was determined based on the presence of central necrosis or a heterogeneous enhancement pattern. The determination of metastatic nodal disease relies on the application of diameter cutoff values exceeding 1.5 cm in the jugulodigastric and submandibular regions, and 1 cm in all other levels of cervical LNs [
10]. The process is deemed significant when the ratio of the maximum longitudinal nodal length to the maximum axial nodal length is less than 2, or when a cluster of three or more LNs nears the threshold for metastasis. The classification of regional LN groups is based on the locations and boundaries of cervical LN groups, resulting in six distinct levels [
11].
2.3. 18F-FDG PET/CT Imaging Protocol
All patients were asked to fast for at least 4–6 h, and their serum blood sugar was below 11.1 mmol/L. PET/CT images were acquired 60 min after injection of 3–5 MBq/kg of
18F-FDG [
12]. A comprehensive imaging procedure was conducted, wherein a whole-body CT scan was performed from the vertex to the mid-thigh. This scan utilized a free breathing signal. The
18F-FDG PET/CT protocol was implemented utilizing a Biograph mCT 64 PET/CT system (Siemens, Erlanger, Germany). PET reconstructions were conducted with and without attenuation correction. CT image acquisition was performed using a Biograph mCT flow CT scanner (64 slices), and PET images were acquired using Flow Motion technology (Erlangen, Germany). The image reconstruction was performed using the order-subset expectation maximization (OSEM) algorithm. Attenuation correction and anatomical localization were achieved through the utilization of low-dose CT without the administration of intravenous contrast. The thickness of each slice was 5 mm. The acquisition process employed a table speed of 1 mm/s, corresponding to a duration of 3 min per bed position.
2.4. Image Analysis and PET Parameter Quantification
Two experienced nuclear medicine specialists blind to the final pathology results analyzed the
18F-FDG PET/CT images using Syngo.via software (Version VB40) (Siemens, Erlanger, Germany) (access date 13 August). The region of interest (ROI) was manually drawn around LNs using the fixed percentile (SUV > 40%) contouring method as recommended by the EANM guideline [
13]. The PET parameters, including SUVmax, MTV, and TLG (MTV × SUVmean), were automatically generated. Using Syngo.via software, the size of LNs can be measured using the software’s measurement tools. The software provides various measurement options, such as diameter, volume, and SUVmax. Each and every visualized LN was detailed in terms of its morphologic and metabolic features, in addition to site and laterality, to be correlated with histopathologic references.
2.5. Statistical Analyses
Conventional statistical analyses were performed. Continuous variables are reported as the median and interquartile range (Q1–Q3), and categorical data are reported as frequencies and percentages. Mann–Whitney U tests were performed for continuous variables to assess differences between metastatic and non-metastatic LN groups. Sensitivity, specificity, and overall accuracy were computed for each diagnostic test. The McNemar test was employed to assess statistical differences between both modalities in nodal staging.
The area under the curve (AUC) and Youden index analyses using the receiver operating characteristic (ROC) curve were performed to determine the optimal cutoff values for continuous variables significantly associated with the LN metastases. Univariate logistic regression analysis was performed to assess the association between the PET parameters and LN metastatic status. Before conducting multivariate logistic regression analysis, the Spearman correlation coefficient (Spearman rho) was used to exclude strong relationships between the obtained variables. Spearman rho exceeding 0.8 indicated strong correlations, while values below 0.5 indicated weak correlations, and other values indicated moderate correlations. Factors with strong correlation were excluded from the multivariable analysis to avoid the collinearity effect.
Finally, to assess the predictive value of PET-derived factors for the LN false-positivity rate, a multivariable logistic regression analysis was performed by incorporating the LN false-positivity rate as a dependent variable. Following this, a decision tree was formulated to construct a user-friendly clinical algorithm for assessing LN false positivity. The analysis employed the exhaustive Chi-squared Automatic Interaction Detector (CHAID) estimation procedure with a 10-fold cross-validation approach, and the results were corroborated using the Chi-squared Residual Tree (CRT) method. A p-value lower than 0.05 was employed to attain results of statistical significance. All statistical analyses were performed using SPSS version 27 software (IBM Corporation, Armonk, NY, USA).
2.6. Reference Standard
The findings from imaging modalities were compared to histopathologic results, which were considered as the reference standard. A skilled surgeon reviewed the surgical report and identified the chosen surgical method for neck dissection. Likewise, an experienced histopathologist examined the histopathology findings for each surgical procedure. The LN site, size, laterality, and anatomical levels were annotated.
For patient-based analysis, a thorough methodology for evaluating cervical LNs in LSCC was carried out. This approach was primarily intended to calculate the accuracy for both 18F-FDG PET/CT and neck MRI modalities. For neck MRI and 18F-FDG PET/CT, morphological factors were examined to identify essential characteristics like size, shape, location, and any morphological anomalies within the LNs. 18F-FDG metabolic activity was assessed to determine SUVmax and detect hypermetabolic LNs. The findings were then compared with biopsy results to determine the presence or absence of metastatic processes. Patients were classified into four groups: true positives, false positives, true negatives, and false negatives. True outcomes occur when the imaging modality aligns with biopsy findings, while false outcomes occur when conflicting results are documented.
For lesion-based analysis, all LN morphologic and metabolic 18F-FDG PET parameters were computed for every visualized LN, irrespective of its metabolic activity. Each visualized LN was compared to the histopathology results of LNs of similar size and anatomical location in the same patient. Importantly, we excluded any visualized LNs that lacked available matching histopathology results and were located in unexplored anatomical levels or sides. Subsequently, these LNs were included in ROC analysis, where the histopathologic findings served as the basis for establishing the threshold cutoff values. These identified threshold values were then utilized in both univariate and multivariate logistic regression analyses. Lastly, the predictive capabilities of these variables were evaluated in the context of LN false positivity, with the false-positivity rate of LNs serving as the dependent variable in the multivariate analysis.
4. Discussion
This study demonstrates the efficacy of 18F-FDG PET/CT in diagnosing and distinguishing between benign and metastatic LNs in LSCC. This was made feasible by taking into account a range of metabolic and morphological factors. 18F-FDG PET/CT provides high sensitivity for detecting LNs in LSCC patients. Despite this, 18F-FDG PET/CT may be associated with false-positive LNs, as reflected by its limited specificity. Therefore, an assessment of the parameters associated with LN false positivity is vital. The primary significance of such evidence lies in its capacity to provide nuclear medicine physicians with a comprehensive understanding of the advantages inherent in adopting a collective approach that transcends the sole reliance on SUVmax and size criteria for a given LN. In clinical practice, the inclusion of all PET-derived parameters in our study served to mitigate the occurrence of false-positive LNs. This reduction in false positives can contribute to the enhancement of the accuracy of this modality and strengthens its potential to influence therapeutic decision-making in clinical settings.
The present study utilized multivariate analysis to examine potential indicators of LN metastasis in patients with LSCC. This analysis revealed that LN MTV and LN size exhibit strong predictive capabilities for the occurrence of LN metastasis. Additionally, the use of LN MTV was found to improve FDG PET/CT specificity in our studied sample. To the best of our current understanding, this study represents an initial exploration of the discriminatory capacities of 18F-FDG PET/CT in the identification of metastatic LNs in LSCC. Furthermore, this study provides the most comprehensive analysis of LNs conducted thus far, with the objective of assessing the effectiveness of 18F-FDG PET/CT in this context.
Accurate preoperative LN assessment is essential for achieving optimal planning in patients with LSCC [
14,
15]. MRI is frequently employed to assess nodal disease in terms of morphological factors including size, the existence of central necrosis, and/or the presence of unclear nodal margins [
16]. In various studies, the sensitivity and specificity of MRI in detecting neck LN metastases in HNC vary, with reported values ranging from approximately 40% to 80% and from 50% to 99%, respectively [
17,
18,
19,
20,
21,
22]. Several studies have been conducted to assess the diagnostic accuracy of
18F-FDG PET/CT in the identification of neck LN metastases. A recent meta-analysis revealed that the combined sensitivity and specificity of
18F-FDG PET/CT for detecting nodal disease were 91% and 87%, respectively, on a per-patient basis [
23]. In a previous retrospective study, it was found that the accuracy of
18F-FDG PET/CT in diagnosing nodal disease was very high, surpassing 95% [
24]. Our study also found that
18F-FDG PET/CT was highly sensitive in detecting nodal disease when analyzing patients. Furthermore, when compared to neck MRI,
18F-FDG PET/CT performed better and showed a significant difference in results. These findings support what was previously found in a recent meta-analysis [
25]. Additionally, in our study,
18F-FDG PET/CT led to a change in treatment plan for 11% of our patients. However, it is important to note that the specificity of
18F-FDG PET/CT was limited due to a high rate of false positives [
26]. In routine practice, the analysis of
18F-FDG PET/CT images of the head and neck region is challenging [
27,
28]. The intricate anatomical composition of the head and neck region, its proximity to vital structures, and the potential overlap of physiological and pathological radiotracer uptake patterns make it difficult to accurately differentiate benign from metastatic LNs [
29]. Moreover, several LN factors, including size and necrosis, that could lead to low metabolic activity contribute to the high incidence of false-negative findings on PET/CT [
30]. Consequently, the misinterpretation caused by these limitations results in inappropriate management decisions. Therefore, finding predictive factors that help evaluate the nodal disease extent accurately in LSCC is clinically meaningful and an unmet need.
The findings of our study indicate that the utilization of LN MTV resulted in a sensitivity of 97% and a specificity of 76.8% for the detection of LN metastasis. This implies that the LN MTV parameter yields more precise outcomes in detecting nodal disease, consequently mitigating the occurrence of false-negative diagnoses as a result of its heightened sensitivity. In alternative terms, LN MTV has the potential to serve as a valuable tool for the purpose of excluding LN metastasis. Furthermore, the LN size and LN TLG parameters exhibited a specificity exceeding 80%. This approach may assist in addressing the problem of inaccurate test outcomes caused by the presence of small benign LN lesions exhibiting slightly increased metabolic activity, ruling out false positives. However, it is crucial to note that nuclear medicine physicians cannot depend solely on a single criterion or factor to consistently differentiate between benign and metastatic LNs. Therefore, in order to effectively implement the most advantageous discrimination strategy, it is recommended to adopt a collective approach rather than a selective one. Due to this rationale, the utilization of 18F-FDG PET/CT holds significant promise in discerning nodal disease through the integration of various metabolic and morphologic characteristics.
Previous studies have employed different approaches to discriminate or identify metastatic LNs in cases of HNC, relying on a predetermined SUV threshold. For example, Nakagawa and colleagues conducted a study in which they examined 31 metastatic LNs that were pathologically confirmed [
6]. These LNs were visualized using PET/CT in a cohort of 11 patients [
6]. The researchers made the observation that utilizing an SUVmax cutoff value of 3.5 yielded a sensitivity of 75% and specificity of 94% in the identification of metastasis in enlarged cervical LNs [
6]. In a study conducted by Murakami et al., ROC curve analysis was employed to assess the utility of size-based SUVmax cutoff values in 23 patients with HNC [
7]. The authors proposed specific LN SUVmax cutoff values based on the LN maximum diameter [
7]. The first LN SUVmax cutoff was 1.9 for LNs with a maximum diameter of less than 1 cm [
7]. A larger LN SUVmax threshold of 2.5 was offered for LNs with a diameter between 1 and 1.5 cm [
7]. Finally, a third cutoff of 3.0 for LNs measuring greater than 1.5 cm was proposed [
7]. The aforementioned cutoff values resulted in a sensitivity of 79% and a specificity of 99% for the detection of cervical LN metastasis [
7]. Recently, a retrospective ROC analysis of cervical LN metastases in HNC patients revealed a relatively high SUVmax threshold [
8]. An LN SUVmax threshold of 5.8 was found to be statistically significant, with an observed sensitivity of 71.4% and specificity of 72.7% [
8]. Moreover, the literature has documented various size criteria for evaluating the enlargement of cervical LNs [
16]. A study by Curtin et al. investigated patients with metastatic HNC, finding that metastatic LNs with largest axial diameter greater than 1 cm had a sensitivity of 88% and a specificity of 39% [
17]. Furthermore, using a cutoff of 1.5 cm, the sensitivity for detecting metastatic LNs was 56%, and the specificity was 84% [
17]. In contrast, the Response Evaluation Criteria In Solid Tumors (RECIST) rely on evaluating LNs by measuring their short axis on axial images [
31]. According to these criteria, LNs with a measurement of ≥1.5 cm are considered pathologically enlarged [
31]. Additionally, smaller cervical LNs measuring between 1 and 1.4 cm in the short axis are considered to be pathologic non-targets [
16]. In an attempt to assess the LN diagnostic efficacy of
18F-fluorothymidine (FLT) PET/CT in relation to
18F-FDG PET/CT for HNC patients, Schaefferkoetter and colleagues performed a comparative study [
32]. They observed that the
18F-FLT PET/CT modality exhibited superior performance in the detection of greater numbers of LNs [
32]. Nevertheless,
18F-FDG PET/CT demonstrated greater accuracy in distinguishing between benign and metastatic LNs through SUVmax cutoffs [
32]. Pietrzak et al. employed an alternative methodology to evaluate the discriminative effectiveness of
18F-FDG PET/CT by analyzing sequential imaging parameters. The study revealed a significant statistical distinction in delayed SUVmax values between physiological and pathological LNs [
33]. All the aforementioned studies did not incorporate a thorough examination of all morphologic and metabolic factors. Additionally, they acknowledged limitations arising from small sample sizes, biases in parameter selection, and the presence of tumor heterogeneity. Notably, these limitations can complicate the observed results and introduce inconsistencies.
The present study has some limitations, including its retrospective nature and single-centric experience. Moreover, the assessment of morphological aspects is hindered by the dependence on low-dose CT scanners integrated with PET/CT consoles. Nonetheless, it remains the first and only study to examine the discriminative power of 18F-FDG PET/CT specifically for LNs in LSCC patients.