Diagnostic Value of Seven Different Imaging Modalities for Patients with Neuroblastic Tumors: A Network Meta-Analysis

Objective We performed a systematic review and network meta-analysis (NMA) to compare the diagnostic value of seven different imaging modalities for the detection of neuroblastic tumors in diverse clinical settings. Methods PubMed, Embase, Medline, and the Cochrane Library were searched to identify eligible studies from inception to Sep 29, 2020. Quality assessment of included studies was appraised with Quality Assessment of Diagnostic Accuracy Studies. Firstly, direct pairwise meta-analysis was conducted to calculate the pooled estimates of odds ratio (OR) and 95% confidence interval (CI) of the sensitivity, specificity, NPV, PPV, and DR. Next, NMA using Bayesian methods was performed. The superiority index was assessed to quantify the rank probability of a diagnostic test. The studies performed SPECT/CT or SPECT were analyzed separately from the ones only performed planar imaging. Results A total of 1135 patients from 32 studies, including 7 different imaging modalities, were eligible for this NMA. In the pairwise meta-analysis, 18F-FDOPA PET/CT had a relatively high value of all the outcomes (sensitivity: 10.195 [5.332–19.493]; specificity: 17.906 [5.950–53.884]; NPV: 16.819 [7.033–40.218]; PPV: 11.154 [4.216–29.512]; and DR 5.616 [3.609–8.739]). In the NMA, 18F-FDOPA PET/CT exhibited relatively high sensitivity in all subgroups (all data: 0.94 [0.87–0.98]; primary tumor: 0.89 [0.53–1]; bone/bone marrow metastases: 0.96 [0.83–1]; and primary tumor and metastases (P + M): 0.92 [0.80–0.97]), the highest specificity in the subgroup of P + M (0.85 [0.61–0.97]), and achieved the highest superiority index in the subgroups of all data (8.57 [1–15]) and P + M (7.25 [1–13]). Conclusion 18F-FDOPA PET/CT exhibited the best diagnostic performance in the comprehensive detection of primary tumor and metastases for neuroblastic tumors, followed by 68Ga-somatostatin analogs, 123I-meta-iodobenzylguanidine (MIBG), 18F-FDG, and 131I-MIBG tomographic imaging.


Introduction
Neuroblastic tumors (NTs) are the most common extracranial solid tumors of children, which are derived from the primitive neural crest. NTs include neuroblastoma, ganglioneuroblastoma, and ganglioneuroma. Nearly 48% of neuroblastomas present with metastasis at the time of diagnosis [1,2]. erefore, accurate identification of all lesions is of importance for staging and establishing therapy protocol [3].
Imaging, especially nuclear medicine functional imaging, plays an indispensable role in the diagnosis, staging, surgical planning, response assessment, and follow-up of NTs. Since neuroblastic tumor cells specifically express the noradrenaline transporter, iodine radioisotope-labelled meta-iodobenzylguanidine (MIBG), a noradrenaline analogue, becomes an ideal tracer for imaging of the tumor lesions. MIBG was labelled with 131 I at the beginning. Nowadays, 123 I-labelled MIBG ( 123 I-MIBG) is the mainstay of radiopharmaceutical in the diagnosis and management of NTs. Considering of the limitation in small lesions and prolonged acquisition time (24-48 hours) of the MIBG scan, positron emission tomography (PET) imaging is increasingly being applied in current clinical practice. In particular, when the tumor uptake of MIBG is weak or negative, 18 Ffluorodeoxyglucose (FDG) PET imaging is recommended as a second-line imaging by the International Neuroblastoma Risk Group (INRG) guidelines [3] and the European Association of Nuclear Medicine (EANM) 2018 guidelines [2]. Various PET tracers have been utilized for imaging in neuroblastoma patients, including the metabolic compounds such as 18 F-FDG and L-3,4-dihydroxy-6-18 F-fluorophenylalanine (FDOPA), as well as the receptormediated compounds such as 68 Ga-DOTA peptides and somatostatin analogues (SSAs) [4][5][6]. Traditional imaging, computed tomography (CT) and magnetic resonance imaging (MRI), also has an essential role in the staging and evaluation of surgical risks for the disease [7].
In the EANM 2018 guidelines for neuroblastoma, discussion is still ongoing on the effectiveness of various imaging modalities and the applicability of different tracers in diverse clinical settings. An increasing number of studies reported the utility of different imaging in the diagnosis of NTs. However, considering the radiation burden and imaging acquisition inconvenience to pediatric patients, headto-head studies are few and most of them are of small sample sizes. Although there have been a few previous meta-analyses [8][9][10] of the diagnostic value of imaging modalities for NTs, there were some limitations in their data grouping. Moreover, all of them were conventional meta-analysis that evaluated one single imaging technique or simply compared two imaging modalities. Network meta-analysis (NMA) extends conventional meta-analysis, which is a novel synthesis of evidence. In contrast to the conventional pairwise meta-analysis, NMA draws together evidence from both direct and indirect comparison of multiple tests simultaneously [11]. NMA can calculate the effect size and quantify the rank probability of each diagnostic test between groups through indirect study comparison, even if there is no direct head-to-head study. Moreover, NMA with an arm-based model provides more natural variance-covariance matrix structures which make it more appropriate than the traditional meta-analysis. erefore, we conducted an NMA and comprehensive systematic review to directly and indirectly compare the diagnostic value of all enrolled imaging modalities in NTs.

Materials and Methods
is systematic literature review and meta-analysis was performed according to the "Preferred Reporting Items for Systematic Reviews and Meta-Analyses" (PRISMA) guidelines and was registered in PROSPERO (CRD42020206862).

Search Strategy.
A systematic literature search was conducted based on the Population, Interventions, Comparator, Outcome, and Study design (PICOS) principle. PubMed, the Cochrane Library, Embase, and Medline were searched from inception to Sep 29, 2020.
Exclusion criteria were as follows: (a) case reports, reviews, meeting abstracts, or comments; (b) non-human studies or non-English articles; (c) sample size ＜ 10; (d) lack of essential data, including true positive (TP), true negative (TN), false positive (FP), and false negative (FN) values; (e) study enrolling recurrent or refractory patients, and (f ) overlapping patients reported. Once articles enrolling overlapping patients were identified, the recently published article with more patients was included.

Data Extraction.
Two researchers independently performed the literature searching, screening, and data extracting. e differences were discussed until reaching consensus. e following information was collected: basic information of studies (first author, publication year, study period, original country, study design, and follow-up time), patient characteristics (sample size, age, and gender), type of lesions (P, primary tumor; BM, bone and bone marrow metastases; and P + M, primary tumor and metastases), type of imaging modality, standard reference, and raw diagnostic data (TP, FP, TN, and FN).

Quality Assessment.
e quality of enrolled studies was appraised with Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) [12] by two authors. Discrepancies between the authors were resolved by discussion. e QUADAS-2 includes four domains: patient selection, reference standard, index text, and flow and timing. Each domain is assessed in terms of risk of bias, and the first 3 domains are also appraised in terms of concerns regarding applicability. RevMan (Version 5.3.5, the Cochrane Collaboration, Oxford, UK) was used to conduct the assessment.

Statistical Analysis and Data Synthesis.
Traditional pairwise meta-analysis was conducted to calculate the pooled estimates of odds ratio (OR) and 95% confidence interval (CI) of sensitivity, specificity, NPV, PPV, and detection rate (DR) of various imaging modalities. Heterogeneity was assessed by the χ 2 test and I 2 statistics. A fixedeffect model would be applied if P > 0.1 and/or I 2 < 50%. Otherwise, a random-effect model would be conducted. Subgroup analyses were conducted based on diverse clinical settings. e publication bias was assessed by Deeks' funnel plot asymmetry test. Traditional meta-analyses were performed using STATA (version 15.0, StataCorp, College Station, TX).
Next, the evidence network structure was performed with package gemtc (v 0.  in R software (Version 4.0.3, Comprehensive R Archive Network). Each node stands for a different diagnostic test, and the thickness of lines between nodes represents the number of studies that directly compared the two tests. en, Bayesian NMA was performed by an arm-based model, which was developed by Nyaga et al. [13]. We run three chains in parallel until there was convergence. Trace plots were applied to visually check whether the distributions of the three simulated chains were mixed properly and were stationary. e superiority index [13] was estimated to quantify the rank probability of a diagnostic test. e larger the superiority index was, the more accurately a test was expected to predict the targeted condition compared to other screening tests. A two-sided P value < 0.05 was considered statistically significant in all statistical tests. All NMA were performed using R software, package rstan (v 2.21.2), package loo (v 2.3.1), and package plyr (v 1.8.6).

Literature Search Results.
A total of 1,094 studies were initially retrieved. After excluding irrelevant articles (n � 826) and duplicated records (n � 38), the remaining 230 studies were further assessed. A total of 51 studies were evaluated for eligibility by full-text review, after excluding non-English articles (n � 24), non-human studies (n � 24), irrelevant studies (n � 73), reviews (n � 31), cases (n � 14), meeting abstracts or comments (n � 4), and the studies with incomplete data (n � 9). After full-text review, irrelevant studies (n � 2), studies with insufficient data (n � 8) or ineligible reference standard (n � 3), studies focusing on recurrent or refractory patients (n � 2), and inaccessible full text (n � 4) were ruled out. Finally, thirtytwo diagnostic studies  met the inclusion criteria ( Figure 1).

Outcome of Pairwise Meta-Analysis.
A direct paired comparison of the 7 different imaging modalities of the diagnostic value for NTs was performed. e estimated OR and 95% CI of the sensitivity, specificity, NPV, PPV, and DR are summarized in Table 2. In both sensitivity and specificity, 18

Evidence Network.
e evidence network included seven imaging modalities. e result revealed the number of studies investigating 123 I-MIBG was the highest. Studies comparing 123 I-MIBG with 18 F-FDG were the most, followed by 131 I-MIBG, CT or MRI and 18 F-FDOPA imaging ( Figure 3).
After removing the studies only conducted with planar imaging and traditional imaging (  ese guidelines presented general indications, advantages, and limitations along with recommendations on imaging protocols, interpretation of findings, and reporting results for nuclear medicine imaging in neuroblastoma. However, discussion regarding the clinical settings that may benefit most from the use of one tracer over the others is still ongoing. It is well known that the accumulation and decarboxylation of L-DOPA in neuroendocrine tumors (NETs) make it an excellent tracer for catecholamine metabolism in NETs, including NTs. 18 F-FDOPA has already been applied in the diagnosis of pheochromocytoma (PCCs) [46,47] and recommended as a first-line PET/CT tracer for the detection of medullary thyroid carcinoma [48].
e EANM 2018 guidelines also suggested 18 F-FDOPA may currently be the best PET tracer alternative to 123 I-MIBG for the assessment of NTs [2]. It showed a remarkable performance in the diagnosis of neuroblastoma in the current meta-analysis as well as other existing research [15,22,49]. In the present NMA, 18 F-FDOPA PET/CT exhibited relatively higher sensitivity in all clinical settings, the highest sensitivity, and specificity in the subgroup of P + M, which ranked the first according to superiority index. erefore, 18 F-FDOPA PET/ CT may become a promising diagnostic tool for neuroblastoma in the future. 68 Ga-SSAs can bind to specific somatostatin receptors on the cell surface of NETs, which is also an imaging tracer of choice in NETs [50]. 68 Ga-SSAs PET or PET/CT ranked the second in the diagnostic value of NTs. Nevertheless, there are only 2 studies focused on this radiopharmaceutical.
e study from Pezhman [14] suggested 68 Ga-SSAs PET/CT was superior to 131 I-MIBG SPECT/CT in providing valuable information for both primary staging and follow-up in patients with neural crest tumors, including NTs and PCCs. Another study [41] did not enroll TN patients, and the specificity cannot be evaluated. So, the results should be interpreted cautiously. Interestingly, both 18 F-FDOPA and 68 Ga-SSAs are relatively new tracers for NETs, and a number of studies which did comparison between them in the detection of neuroendocrine tumors and other diseases have been reported [51,52]. However, the study that head to head compares 18 F-FDOPA and 68 Ga-SSAs in patients with neuroblastic tumors has not been found. us, further study is expected. 123 I-MIBG imaging, the most commonly utilized molecular imaging modality for the identification and     18 F-FDG displayed a higher superiority index compared with MIBG when it was employed to comprehensively evaluate the primary tumor and metastases in the whole body. at may be related with the higher resolution of PET than SPECT. e study of Henriette [24] reported the false negative results of 123 I-MIBG were due to small lesion size (mean lesion diameter 1.7 cm) and low uptake. Combined with our results, it suggested that 18 F-FDG could play a complementary role of MIBG when the lesion is small or non-MIBG avid. Additionally, beyond lesion recognition, 18 F-FDG may be helpful in tumor staging, treatment evaluation, and prognostic assessment of neuroblastoma [40,42,53]. Higher FDG uptake was observed in patients with higher-stage MYCN amplification [54] or advanced stage [55]. Maximum standardized uptake value (SUVmax) was all significantly higher in patients with worse overall survival [54]. 131 I-MIBG imaging exhibited moderate diagnostic characteristics based on the superiority index. It may be limited by the unfavorable imaging characteristics of the isotope 131 I. Nowadays, considering higher radiation dose of 131 I-MIBG compared with 123 I-MIBG, many studies [4,56] recommend that diagnostic 131 I-MIBG was indicated only when 123 I-MIBG is unavailable or 131 I-MIBG therapy is contemplated. Preliminary studies of 124 I-MIBG [57,58] as well as 18 F-meta-fluorobenzylguanidine ( 18 F-MFBG) [59,60] and MIBG variants [61] ( 18 F-fluoropropylbenzylguanidine, 18 F-FPBG) are ongoing. ose imaging agents are proved to improve image quality and demonstrate promising performance in the diagnosis of NTs.
CT and MR are widely available and routinely used in clinical practices, whereas they only showed moderate sensitivity and low specificity in the detection of BM in the NMA. e high incidence of false positive findings was probably related with the fact that the traditional imaging modality cannot distinguish posttherapy bone marrow changes from residual tumor. In recent years, increasing studies suggested that whole-body "diffusion-weighted imaging with background body signal suppression" is feasible for assessment of the primary lesions [62,63] and lymph node metastases [18] of NTs. However, according to our results, it should be carefully used in NT patients due to its low specificity in the identification of skeleton lesions. Regarding the diagnostic value of 111 In-pentetreotide, only one study [31] was included, and the data of this study are incomplete.
erefore, more high-quality studies are expected.
ree previous meta-analyses of the diagnostic value of different imaging modalities for NTs were identified. All of them evaluated one single imaging technique [8,9] or simply compared two imaging modalities [10]. Moreover, the studies conducted with SPECT/CT and planar imaging were not analyzed separately in the meta-analysis from the work of Jia X et al. [10], which was irrational. Because there is a  Shahrokhi, 2020 Sharp, 2009 Syed, 2004 Taggart [9] attempted to assess the diagnostic accuracy of PET(CT) in patients with neuroblastoma in their metaanalysis, but they did not calculate the data of FDG-PET(CT) and FDOPA PET(CT) separately. erefore, the reference value of this study for clinical practice is fairly limited. is NMA has several limitations. Firstly, subgroup analyses were not conducted based on lesion-based analysis versus patient-/scan-based analysis because in the subgroup of primary tumor, all included studies were patient-/scanbased analysis. In the other two subgroups, most of the enrolled studies were lesion-based analysis, and patient-/ scan-based analyses were performed in only 9 studies for the evaluation of 6 imaging modalities. Secondly, due to the small number of included studies for each imaging modality, subgroup analyses were not performed according to other variables such as study design, patients' baseline characteristics, interval between injection and acquisition, and other imaging protocols.

Contrast Media & Molecular Imaging
irdly, CT and MR were not analyzed separately. In the included studies, half (3/6) of the included studies mixed CT and MR together to compare with nuclear medicine imaging. Only two studies investigated the performance of MR, and just one study reported the data of CT. Finally, the estimated specificity of several groups displayed a wide range of 95% credible intervals (0-1), such as 18 F-FDOPA in the BM and P subgroups, 111 Inpentetreotide in the P subgroup, and 68 Ga-SSAs and 131 I-MIBG in the P + M subgroup. is demonstrates that the specificity was unavailable because some studies did not       Contrast Media & Molecular Imaging enroll TN cases. Further direct comparative studies with standardized data would be necessary.

Conclusions
In conclusion, 18 F-FDOPA PET or PET/CT exhibited the best diagnostic performance in the comprehensive detection of primary tumor and metastases in NTs, followed by 68 Ga-SSAs, 123 I-MIBG, 18  Data Availability e data used in this study are available on reasonable request to the corresponding author.

Conflicts of Interest
e authors declare no conflicts of interest.