Diagnostic Accuracy of 18F-FDG-PET/CT and MRI in Predicting the Tumor Response in Locally Advanced Cervical Carcinoma Treated by Chemoradiotherapy: A Meta-Analysis

Objective The aim of this meta-analysis was to compare the diagnostic accuracy of 18F-FDG-PET/CT and MRI in predicting the tumor response in locally advanced cervical carcinoma (LACC) treated by chemoradiotherapy (CRT). Method This meta-analysis has been performed according to PRISMA guidelines. Systematic searches were conducted using PubMed and Embase databases for articles published from January 1, 2010, to January 1, 2020. By using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool, the reviewers assessed the methodological quality scores of the selected studies. We analyzed the sensitivity, specificity, and accuracy of two diagnostic methods using Meta-DiSc 1.4 and Stata 15. Results An overall of 15 studies including 1132 patients were included. Sensitivities of PET/CT and MRI were 83.5% and 82.7%, while the corresponding rates for specificities were 77.8% and 68.4%, respectively. The DOR, PLR, and NLR for MRI were 15.140, 2.92, and 22.6. PET/CT had a DOR of 25.21. The PLR and NLR for PET/CT were 4.13 and 0.215, respectively. The diagnostic sensitivity and specificity of PET/CT for the detection of residual tumor were 86% and 95%, respectively. The corresponding rates for MRI were 73% and 96%, respectively. The diagnostic sensitivity and specificity of PET/CT for the detection of tumor metastases were 97% and 99%, while the corresponding rates for MRI were 31% and 98%, respectively. Conclusion 18F-FDG PET/CT seemed to have a better overall diagnostic accuracy in the evaluation of treatment response to chemoradiotherapy in LACC patients. MRI showed a really poor sensitivity in the detection of metastases, and PET/CT performed significantly better. However, the difference between these two methods in the detection of residual disease was not significant. More studies are needed to be conducted in order to approve that 18F-FDG PET/CT can be a standard option to assess the treatment response.


Introduction
Cervical cancer is one of the most prevalent cancers among women [1]. About 40% of these patients show locally advanced cervical carcinoma (LACC) at initial diagnosis [2]. e standard treatment for LACC is specific cisplatin-based radiotherapy chemotherapy (CRT) [3]. However, usually, in 33% of cases, the tumor recurs about 2 years after CRT, and the overall 5-year survival is approximately 70% [3]. Lymph node status, response to treatment, and clinical stage are the main predictors of recurrence. Among the new strategies, neoadjuvant CRT followed by radical surgery is performed with the purpose of removing residual tumors which are potentially radio-and chemo-resistant and also improving local control and survival [4,5]. In addition, radical surgery provides useful prognostic information, which is the pathological response to treatment. Women who obtain a complete pathological response to neoadjuvant CRT showed significantly longer overall survival and disease-free survival than women with partial response [6,7]. However, applying this approach has some intraoperative and postoperative complications [8]. erefore, there is a need for imaging techniques that can evaluate the tumor response accurately during and after treatment, so personalized treatment can be made possible. Imaging techniques such as magnetic resonance imaging (MRI), computed tomography (CT), and positron emission tomography (PET) with 18F-fluorodeoxyglucose (FDG) are more useful tools for evaluating how extensive the tumor is [9]. In the evaluation of tumor size and invasion and the local extent of the disease, MRI is superior to CT. Unlike CT and MRI, PET is used to evaluate tumor metabolic function. Pathological tumor size in cervical cancer and tumor size measured by PET have been shown to have strong correlation [10]. PET has been used to evaluate pretreatment situation and daily surveillance of patients with cervical cancer after treatment [11]. 3-month posttreatment PET can be applied to predict the therapeutic response [12]. However, patients who have a poor response to CRT can be detected as soon as possible, and their treatment plan can be changed as well, provided that the tumor response can be estimated during or just after concomitant chemoradiotherapy. 18F-FDG-PET/CT is important in LACC staging due to its capability to detect distant metastases and involved lymph nodes; it, therefore, improves treatment planning [13]. Furthermore, 18F-FDG-PET/CT can predict better prognosis in patients who developed complete metabolic response when performed after exclusive CRT [13]. In contrast, the role of 18F-FDG-PET/CT performed during treatment is not defined apparently [4]. e aim of this metaanalysis is to compare the diagnostic accuracy of 18F-FDG-PET/CT and MRI in predicting the tumor response in locally advanced cervical carcinoma treated by chemoradiotherapy.

Literature Search.
is meta-analysis has been performed according to Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [14]. Systematic searches were conducted using PubMed and Embase databases for articles published from January 1, 2010, to January 1, 2020. e search query was done using the key terms "cervical cancer," "magnetic resonance imaging or MRI," "18F-FDG PET/CT," "accuracy," "specificity," "sensitivity," and "prognosis," and related terms are as follows: (cervix or cervical) and (PET or positron emission tomography) or (FDG or fluorodeoxyglucose). ere was no limitation regarding the language of the studies. e bibliography of retrieved early studies was cross-examined in order to find other related papers and articles. e inclusion criteria consisted of the following: studies evaluating PET/CT and MRI diagnostic accuracy in the treatment of LACC. Pathological results were considered as the "gold standard." e diagnostic criteria were as follows: there are 2 × 2 contingency tables (Table 1); direct or indirect access to true positive, false positive, true negative, false negative, specificity, and sensitivity. We also included some additional studies as they provided data which helped to complete our manuscript and make it more understandable. ese data were analyzed separately from the diagnostic accuracy data. e exclusion criteria were as follows: publication type other than the authentic research papers (i.e., review articles and conference abstracts), not in the field of the researchers' interest. e searching process and selection of the articles were done by two independent reviewers with 3 years of meta-analysis experience. Disagreements were resolved through discussion.

Data Extraction and Quality Assessment.
e two abovementioned authors separately assessed each study and then extracted the data by applying standardized dataabstraction forms. e following characteristics were extracted: study characteristics included institution, publication year, first author, country, patient enrollment period, design (prospective or retrospective), number of patients, and reference standard, and clinical and radiologic characteristics included patient FIGO stages, histology of the tumor, age, lymph node metastasis, tumor size, treatment, 3-year overall survival (OS), disease-free interval (DFI), true positive, false positive, true negative, false negative, specificity, and sensitivity. 2 × 2 contingency tables were constructed, and we calculated the specificity, sensitivity, and likelihood ratios (LR). By using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool, the reviewers assessed the methodological quality scores of the selected studies including 11 standard items, applying review manager software program (RevMan, version 5.0.2, Nordic Cochrane Centre, Copenhagen, Denmark). e answers "yes," "no," or "unclear" to the 11 standard questions represented that the risk of bias can be judged to be low, the potential for bias exists, or inadequate data are reported to permit a judgment, respectively.

Statistical Analysis.
We analyzed the sensitivity, specificity, and accuracy of two diagnostic methods by Meta-DiSc 1.4 and Stata 15 with their 95% confidence intervals (CI). Also, we have drawn the hierarchical summary receiver operating characteristic (HSROC) curves. e random effect model or fixed effect model was used to evaluate the effect values based on the results of the heterogenicity test.

Results
After a comprehensive computerized search, searching for articles and selecting them were done, and reference lists were cross-checked, as well. We recorded 1,435 files, of  [20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37]. e features of the included studies are presented in Table 2. e detailed method of study selection in the current meta-analysis is shown in Figure 1. studies, focused on the correlation between the results of PET/CT and pathologic complete response. ey suggested following the changes through the treatment process using delta SUV and delta TLG in order to achieve the most accurate diagnosis [14,15,20]. Choi et al. also suggested SUV max of 4.0 as an optimal cutoff on the posttreatment PET/CT [15]. Scarsbrook et al. defined a five-point qualitative response assessment scoring system using which they reached a high sensitivity for PET/CT (28).

MRI.
Gui et al. reported a relatively low sensitivity and specificity for MRI. eir results indicated that MRI performance is not sufficient in distinguishing post-CRT inflammation from the residual tumor which can lead to a high number of false positives. However, the negative predictive value of MRI was high with a low risk of false negative [25]. e results of Atstupeṅaitėwere inline with this study and indicated high specificity and low sensitivity for MRI in post-CRP evaluation of patients [26].

PET/CT and MRI. Perron et al. evaluated PET/CT and MRI accuracies in the same population.
ey reported a significantly higher Cohen Kappa coefficient between follow-up findings and PET/CT results compared to the findings of MRI. However, Vandecasteele et al., in their study on 27 cervical cancer patients, indicated a very low sensitivity for PET/CT and considered MRI as the preferred modality for recurrence assessment as it provided a specificity of 100% associated with a 74% NPV in their study [20,21]. However, this difference could be caused by the small sample size in Vandecasteele et al.'s study. Su et al., on the other hand, confirmed the superiority of PET/CT in post-CRP patients (P � 0.025) [24] (Table 1).

Meta-Analysis of SUV mean in Patients with Complete and Partial Response.
e overall SUV mean for predicting treatment response in patients with locally advanced cervical cancer was reported to be 1.9 for patients who developed complete response and 5.10 for patients who did not, based on 4 articles including 372 patients [17][18][19]22] (Table 3).

Fagan's Nomogram for the Calculation of Posttest Probabilities.
A pretest probability of 50% for all three diagnostic tools was fixed, which was estimated by the number

Discussion
Among the most common cancers in women, cervical cancer is the third malignant tumor worldwide after breast and colorectal cancer. Cervical cancer has a significant impact on women's health because of its younger onset age, high prevalence, and posttreatment recurrence. e prevalence of residual disease and lymph node metastasis after chemotherapy were 12.7% and 49%, respectively. Predictors of recurrence of cervical cancer include the posttreatment   [29][30][31]. Patients with no significant tumors after treatment are reported to have a 5-year survival of 76%. However, it has been reported that this rate is lower in patients who have findings which are suggestive of a tumor or who are diagnosed with persistent tumor (42% and 8%, respectively) [32]. According to the results of the present study, the overall 3-year survival for patients undergoing CRT was 92.4%. Due to the lack of data in the studies, we could not perform a subgroup analysis to divide patients into two groups with a complete response and partial response. However, we evaluated the overall DFI and PFS of these patients, which showed that the overall DFI was 74% and the mean PFS was evaluated to be 15 months. In cervical cancer, the surgeon's goal is to rule out the progression of the disease rather than macroscopically    invasive because this treatment is not selective on parameters, but is provided by surgery. In cervical cancer, the role of pretreatment 18F-FDG-PET/CT and MRI is well defined. In case of suspected involvement in bladder, cervix, vagina, or rectum, MRI performs better [33,34]. However, in the diagnosis of lymph node metastases or mesenteric, peritoneal, gastrointestinal, mediastinal, and pleural involvement, 18F-FDG-PET/CT is a more sensitive diagnostic tool [35]. erefore, both of these diagnostic tools seem to be useful in choosing the treatment and in radiation therapy planning [36,37]. However, studies assessing the response to chemotherapy are few, although this evaluation plays an important role in deciding on subsequent treatments [18,39,40]. Clinical and radiological features are required to diagnose residual disease. Unfortunately, it is difficult to perform gynecological examinations after radiation therapy. In fact, the accurate visualization of the cervix is interfered by vaginal adhesions and fibrosis after radiation [41]. MRI after radiotherapy may not assess the response, accurately due to heterogeneous gad-contrast enhancement and inflammation in areas which are hyperintense in T2W [42].
e evaluation of 18F-FDG-PET/CT can also be interfered by the inflammation and necrosis caused by radiotherapy [43]. erefore, it is required to define the roles of these imaging modalities after chemoradiotherapy. In this metaanalysis, we assessed the role of 18F-FDG-PET/CT and MRI in predicting tumor response in LACC after CRT. is metaanalysis includes a total of 15 studies. Due to the fact that most individual studies have a limited number of cases, more data can be used in meta-analysis and also more reliable results are provided. Our study indicated better diagnostic sensitivity for MRI follow-up data (0.86) compared to PET/ CT (0.83) in assessing the response to chemoradiotherapy for primary tumor and distant metastasis. However, the sensitivity of these two methods is quite similar, and there is no statistically significant difference between them. We also performed a subgroup analysis to decrease heterogeneity assessing the accuracy of these two methods among patients with residual disease compared to patients with distant metastases. MRI showed a real poor sensitivity in the detection of metastases (31% vs. 97%), and PET/CT performed significantly better. However, the difference among these two methods in the detection of residual disease was not significant (73% vs. 86%). Woo et al., in their meta-analysis, reported 73% sensitivity and 93% specificity for MRI in the diagnosis of parametric lesions in patients with cervical cancer [44], and Sakurai et al. declared that metabolic activity and standardized uptake value (SUV) depend on the tumor lesion (>1 cm); SUV was reported to have an average of 3.90 and 2.31 in tumoral and nontumoral lesions, respectively (P > 0.05) [45]. In our study, the mean SUV was 1.9 for patients who had a complete response and 10.10 for those who did not. Rufini Figure 5: DOR of 18F-FDG PET/CT (b) and MRI (a) for predicting treatment response in LACC patients after CRT based on countries. 8 Contrast Media & Molecular Imaging parameters. However, they showed that the final assessment was not accurate enough to predict the pathological CR of the primary tumor [16]. Another study examined 25 patients with LACC (stages IB2-IIIB), evaluated them using both MRI and 18F-FDG PET/CT, before and after CRT, and indicated that 18F-FDG PET/CT provided important information which led to treatment planning changes in half of the patients. However, MRI detected the pelvic tumors in 2 patients which were missed by 18F-FDG-PET/CT [42]. In contrast, a meta-analysis of 15 studies represented by Maedes et al. assessed the additional diagnostic accuracy of 18F-FDG PET/CT in the whole body compared to conventional imaging in women with suspected recurrent/ persistent cervical cancer. He reported that using 18F-FDG PET/CT in these patients was not supported by previous studies. However, the included studies in his meta-analysis had not compared PET/CT with MRI or CT [46]. is metaanalysis had several limitations. Firstly, not all studies in this meta-analysis reported the specific techniques applied. Some scanning parameters may affect the accuracy of PET/CT and MRI. Secondly, due to the small number of published articles in this field, there is a possibility of bias in the present study. irdly, as there were not enough articles comparing these modalities in the same sample size, we had to include articles evaluating one of the modalities and then compare their results in a meta-analysis. is can be our major source of bias. Finally, the number of patients in the included studies was relatively small, which may lead to bias in the final results.

Conclusion
18F-FDG PET/CT seems to have a better overall diagnostic accuracy in the evaluation of the treatment response to chemotherapy in LACC patients. MRI showed a really poor sensitivity in the detection of metastases, and PET/CT performed significantly better. However, the difference between these two methods in the detection of residual disease was not significant. More studies need to be conducted in order to approve that 18F-FDG PET/CT can be a standard option to assess the treatment response.  Contrast Media & Molecular Imaging 9 MRI:

Abbreviations
Magnetic resonance imaging LACC: Locally advanced cervical cancer CRT: Chemoradio therapy QUADAS-2: Quality Assessment of Diagnostic Accuracy Studies 2 DOR: Diagnostic odds ratio PLR: Positive likelihood ratio NLR: Negative likelihood ratio CT: Computed tomography PET: Positron emission tomography DFI: Disease-free interval PFS: Progression-free survival OS: Overall survival CI: Confidence interval SUV: Standardized uptake value.

Data Availability
e data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest
e authors declare no conflicts of interest.

Authors' Contributions
SS and FP participated in conception and design of the study, library searches and assembling relevant literature, critical review of the paper, supervising writing of the paper, and database management. MS participated in data collection, library searches and assembling relevant literature, writing the paper, and critical review of the paper. All authors have read and approved the final version of manuscript.