Diagnostic Accuracy of MRI for Detecting Cervical Invasion in Patients with Endometrial Carcinoma: A Meta-Analysis

Objectives: To evaluate the diagnostic accuracy of magnetic resonance imaging (MRI) in the preoperative assessment of cervical invasion and to analyse the influence of different imaging protocols in patients with endometrial carcinoma. Methods: An extensive search of articles about MRI for assessing cervical invasion in patients with endometrial carcinoma was performed on PubMed, Embase, Web of Science, Cochrane Library, and Clinical Trials from January 2000 to July 2020. Two reviewers independently evaluated the methodological quality of each study by using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2). Diagnostic accuracy results and additional useful information were extracted. The pooled estimation data was obtained by statistical analysis. Results: A total of 42 eligible studies were included in the meta-analysis. Significant evidence of heterogeneity was found for detecting cervical invasion (I2 = 74.1%, P = 0.00 for sensitivity and I2 = 56.2%, P = 0.00 for specificity). The pooled sensitivity and specificity of MRI were 0.58 and 0.95 respectively. The use of higher field strength (3.0 T) demonstrated higher pooled sensitivity (0.74). Using diffusion weighted imaging (DWI) alone presented higher pooled sensitivity (0.86) than using other sequences. The studies that used dynamic contrast-enhanced MRI (DCE-MRI) alone showed higher sensitivity (0.80) and specificity (0.96) than those that used T2-weighted imaging (T2WI) alone. Conclusions: MRI shows high specificity for detecting cervical infiltration in endometrial carcinoma. Using DWI or a 3.0-T device may improve the pooled sensitivity. DCE-MRI demonstrates higher pooled sensitivity and specificity than T2WI.


Introduction
Endometrial carcinoma is one of the most common gynaecological malignancies [1]. Cervical invasion is an important prognostic factor for endometrial carcinoma and is associated with a higher risk of lymph node metastases [2,3]. Hysterectomy and bilateral salpingo-oophorectomy are the primary treatments for endometrial carcinoma [4]. However, in patients with cervical infiltration, radical hysterectomy or preoperative radiotherapy with bilateral salpingo-oophorectomy and bilateral pelvic-para-aortic lymphadenectomy may be necessary [4]. Consequently, it is important to preoperatively evaluate the cervical involvement when planning treatments.
The purpose of the present study was to evaluate the diagnostic accuracy of MRI for detecting cervical invasion and to analyse the effects of different imaging protocols in patients with endometrial carcinoma.

Literature search
We performed this meta-analysis according to the Preferred Reporting Items for Systematic Reviews-Diagnostic Test Accuracy (PRISMA-DTA) guidelines [32]. A comprehensive literature search of articles pertaining to the accuracy of cervical invasion using MRI in endometrial carcinoma was performed by using the following keywords (including subject word and random word): "endometrial neoplasms", "magnetic resonance imaging", and "cervical". Two authors (G.B., a radiologist with 20 years of experience and Q.B., a radiologist with 5 years of experience) independently conducted the searches on PubMed, Embase, Web of Science, Cochrane Library, and Clinical Trials from January 2000 to May 2020 for English language articles on human subjects. To identify possible missing citations, the reference lists of the relevant articles were manually searched.

Study selection
The two authors who performed the literature search independently reviewed all the titles, abstracts, and full texts to identify potentially eligible articles. Studies meeting the following criteria were included if: (a) the accuracy for the detection of cervical invasion in endometrial carcinoma was evaluated; (b) the histopathological results after surgical resection were used as the reference standard; and (c) sufficient information was presented to allow for reconstruction of 2 × 2 tables. When the data or patient cohorts overlapped in the included studies, we selected the article with the largest number of patients.

Data extraction and processing
The data on the diagnostic accuracy results and additional useful information in the original studies were collected by two researchers (Q.B. and J.W.) who had 5 years of independent experience in data extraction for diagnostic studies. In cases where the researchers had discrepancies, a consensus was reached after discussion with each other. For each study, the following information was extracted: author name, year of publication, nation, patient age, sample size, number of observers, study design, patient recruitment, blinded to reference, magnetic field, manufacturer, sequences of observing cervical infiltration, depth of cervical invasion, interval between MRI and pathology, and the true-positive, true-negative, false-positive, and false-negative values of MRI in detecting cervical invasion in patients with endometrial carcinoma. When two or more observers were in the same study, the most experienced observer was selected; if their experience was not reported, the first observer was selected. The most contemporary MRI scan (e.g., DWI before DCE-MRI) was preferred when different MR pulse sequences were reported in the same study. When the accuracy of any cervical invasion and stromal invasion was reported separately, the latter was preferred.

Assessment of data quality
Quality assessment was conducted by the Quality Assessment of Diagnostic Accuracy Studies-2 [33] (QUADAS-2) by two investigators (G.B. and J.Z., a radiologist with 20 years of experience in pelvic imaging) independently. Any disagreements were resolved by discussion with each other. The QUADAS-2 form is composed of four domains: patient selection (assessing methods of patient selection), index test (assessing the index test and how it was conducted and interpreted), reference standard (assessing the reference standard and how it was conducted and interpreted), and flow and timing (assessing any patients who did not receive the index test and/or reference standard or who were excluded from the 2 × 2 table).

Statistical analysis
The analyses were performed by using Review Texas, USA). The threshold effect was assessed by the Spearman correlation coefficient between the logit of sensitivity and the logit of (1-specificity) [34]. P values < 0.05 were indicated that the threshold effect existed [34]. Heterogeneity for sensitivity and specificity was explored by using the inconsistency index (I 2 value) in forest plots [35]. I 2 values ≥ 50.0% are considered to indicate substantial heterogeneity [35]. A fixed-effects model was used to summarize the overall pooled diagnostic results if homogeneity existed. A random-effects model was utilized if heterogeneity existed. Summary receiver operating characteristic (sROC) curves and the area under the curve were used to elucidate the relationship between the sensitivity and specificity. If heterogeneity existed, meta-regression was performed to assess covariates. Several relevant covariates were as follows: patient age (≥ 60 years or < 60 years), magnetic field (1.5 T or 3.0 T), MR pulse sequences, design (prospective or retrospective), blind to reference (yes or unknown), depth of cervical infiltration (stromal invasion or any cervical invasion), and appropriate interval between MRI and pathology (yes or unknown). Sensitivity analyses were performed on the basis of those potential influencing factors of heterogeneity. The publication bias was assessed by using Deeks'funnel plot with P values < 0.05 [36].

Literature search and data extraction
The detailed flowchart summarizing the literature search and selection is given in Figure 1. A total of 1194 records from January 2000 to July 2020 of English language articles on human subjects were found. Two additional records were identified after manual reference checking. After duplicates were removed, 681 unique citations remained. Based on the screening of titles and abstracts, 602 studies were excluded. The full text of 79 studies was reviewed, and 42 eligible studies involving 4196 patients were included in this meta-analysis. Among these 4196 patients, 739 were confirmed to have cervical invasion by surgical pathology. The mean prevalence of endometrial carcinoma with cervical invasion was 17.6% (range, 4.5%-39.5%). The details of the principal characteristics of every included study are summarized in Table 1.    Figure 2 shows the methodological quality graph of the evaluation of the risk of bias and applicability concerns of the selected studies, according to the evaluation based on QUADAS-2.
The slope coefficients for the Deeks'funnel plot for MRI in assessing cervical invasion in endometrial carcinoma are presented in Figure 3. Publication bias was detected in the funnel plots for the diagnosis of cervical invasion (P = 0.01).

Discussion
This meta-analysis demonstrated high pooled specificity of MRI for detecting any cervical infiltration and stromal invasion in patients with endometrial carcinoma. Sensitivity analyses revealed that magnetic field and MR pulse sequences were helpful in explaining the heterogeneity observed for the sensitivity and specificity of MRI for detecting cervical invasion.
The clinical management and the prognosis of endometrial carcinoma are closely related to cervical invasion [4]. Using a preoperative technique to detect cervical invasion of endometrial carcinoma may be helpful to reduce the scope of operation, minimize costs, and offer fertility-preserving treatment options for young women without cervical invasion [54]. MRI is considered the best non-invasive method for the preoperative staging of endometrial carcinoma [9]. In this meta-analysis, MRI showed low sensitivity (0.58) and high specificity (0.95) for detecting cervical invasion. This finding is similar to that in a previous meta-analysis [55]. Moreover, further sensitivity analyses of magnetic field strength were performed in our meta-analysis. We found that studies with higher field strength (3.0 T) had higher sensitivity (0.74) than studies with a 1.5-T device (0.60) or 1.0-T device (0.51). Hori et al [25] discovered that 3.0-T imaging improved the tumour signal-to-noise ratio by approximately 12% compared with that of 1.5-T imaging. The main reason for this result is that the signal-to-noise ratio is influenced by the magnetic field strength, with higher fields having a better signal-to-noise ratio. Hence, a 3.0-T device can provide better quality MRI and demonstrate higher pooled sensitivity (0.74) for detecting cervical invasion in endometrial carcinoma. However, there are some problems associated with 3.0-T imaging, particularly of the pelvis, such as a larger susceptibility effect and larger chemical shift [25].
These factors may affect the diagnostic accuracy for detecting cervical infiltration in endometrial carcinoma. As a consequence, the pooled specificity was not the highest for the studies using a 3.0-T device (0.96).
T2WI is a conventional MR pulse sequence and one of the best MRI protocols for staging in patients with endometrial carcinoma according to the Updated Guidelines of the European Society of Urogenital Radiology [56]. On T2WI, cervical invasion is defined as a mass within the endocervical canal and/or disruption of the normal cervical stroma [25]. The normal cervical stroma appears hypointense on T2WI due to its rich fibrous tissue, and endometrial carcinoma appears hyperintense, leading to high contrast resolution [31]. Consequently, MRI shows high specificity for detecting cervical invasion. A high specificity means the misdiagnosis rate (false positive rate) is low, and indicates to be good at ruling out cervical invasion. Therefore, if MRI findings do not suggest cervical invasion, that a simple hysterectomy and bilateral salpingo-oophorectomy may be the best surgical option. However, microscopic cervical infiltration may not be observed by using MRI, and only macroscopic cervical invasion could be found, resulting in low sensitivity for detecting cervical invasion in patients with endometrial carcinoma [55]. A low sensitivity means the missed diagnosis rate (false negative rate) is high. If we only rely on T2WI to judge the cervical invasion, many patients will be missed diagnosis, so that no radical surgical resection will be performed, which will affect the prognosis of patients. Therefore, DCE-MRI is necessary for preoperatively detecting cervical invasion.
According to a recent meta-analysis, DCE-MRI can help improve the sensitivity and specificity for detecting myometrial invasion [55] because DCE-MRI provides the observer with obvious contrast resolution between the markedly enhanced normal myometrium and the moderately enhanced tumour. On DCE-MRI, cervical invasion is defined as the interruption of the normal cervical epithelium enhancement [16]. Moreover, delayed DCE-MRI (4-5 min after the injection) is optimal for the detection of cervical invasion [56]. Previous research reported that DCE-MRI improved the detection of cervical infiltration by endometrial carcinoma [11]. Our meta-analysis also found that, compared with T2WI, DCE-MRI could improve the sensitivity (0.80) and specificity (0.96) of detection. DCE-MRI is accepted as the state-of-the-art standard for tumour delineation and as one of the best approaches for the local staging of endometrial carcinoma [56]. However, it is commonly difficult to assess cervical invasion once endometrial carcinoma spreads to the endocervical canal and begins obliterating the interface between the tumour and the cervix [57]. Other MRI functional imaging techniques are needed for accurate preoperative evaluation of cervical infiltration. DWI is a functional technique of MRI that reflects the diffusivity of the water molecules in tumours. DWI offers potential advantages over DCE-MRI because it does not require contrast administration and it entails a shorter imaging time. Recent evidence suggests that DWI improves the evaluation of the myometrial invasion of endometrial carcinoma because DWI is able to determine malignant lesions as a hyperintense area with excellent tissue contrast [10]. To avoid the influence of the T2 shine-through effect, cervical invasion was defined as the appearance of higher signal intensity on high-b-value DWI and low signal intensity on apparent diffusion coefficient (ADC) maps compared with the surrounding normal cervical parenchyma [16,56]. This meta-analysis found that studies that used DWI alone had higher sensitivity (0.86) than studies that used DCE-MRI (0.80) or T2WI (0.73) alone. A significant improvement in the sensitivity was also found in DWI compared with that of DCE-MRI and T2WI for detecting cervical invasion in a previous original study [16]. The false positive rate may increase when cervical mucus is present because mucus shows a high signal on DWI and a low signal on ADC maps, leading to a decrease in specificity. Furthermore, DWI also has other disadvantages, such as limited spatial resolution and image distortions because of susceptibility artefacts. Thus referring to other MR pulse sequences for an anatomical landmark is warranted. As a result, DWI is now routinely used as an adjunct to T2WI and DCE-MRI [56].
There are some limitations in this meta-analysis. First, due to the lack of sufficient supporting literature, sensitivity analyses of other techniques, such as CO2-volumetric interpolated breath-hold examinations, were not performed. Second, some studies did not report the sequences for observing cervical infiltration. In addition, the number of the included studies was limited. It remains to be determined whether combined DWI and T2WI is superior to DCE-MRI and whether a 3.0-T device combined with DWI or DCE-MRI has higher sensitivity and specificity than other devices. Third, publication bias was present among the included studies. One possible reason was that we excluded relevant studies published in other languages. Other possible reasons were that the sensitivity for detecting cervical invasion was low and some articles with negative results might not have been published.
In conclusion, this meta-analysis shows a low pooled sensitivity and high specificity of MRI for detecting any cervical infiltration and stromal invasion in endometrial carcinoma. The studies with a 3.0-T device demonstrated a higher pooled sensitivity than any other study. The higher the field strength was, the higher the pooled sensitivity. Using DWI alone demonstrated higher pooled sensitivity than using DCE-MRI or T2WI only. Studies that used DCE-MRI alone showed higher sensitivity and specificity than those that used T2WI alone.