Diagnostic Performance of MRI to Differentiate Uterine Leiomyosarcoma from Benign Leiomyoma: A Meta-Analysis

Purpose: To perform a meta-analysis comparing the diagnostic performance of increased signal intensity on T1- and T2-weighted magnetic resonance images and apparent diffusion coefficient (ADC) values in differentiating uterine leiomyosarcoma (LMS) from benign leiomyoma (LM). Methods: A systematic literature search for original studies was performed using PubMed/MEDLINE, the Cochrane Library, Embase, and Web of Science. Data necessary for the meta-analysis was extracted from the selected articles and analyzed. Results: Eight studies with 795 patients met our predefined inclusion criteria and were included in the analysis. Increased signal on T1-weighted imaging had a pooled sensitivity of 56.8% (95% CI: 20%–87.4%) for LMS (n = 60) which was significantly higher than 7.6% (95% CI: 2.2%–22.7%) for LM (n = 1272) (p = 0.0094). Increased signal analysis on T2-weighted imaging had a pooled sensitivities of 93.2% and 93.2% (95% CI: 45.7%–99.6% and 42.9%–99.6%) for LMS (n = 90), which were not significantly different from the 54.5% and 53.9% (95% CI: 33.6%–74%, 32%–74%) for LM (n = 215) (p = 0.102 and 0.112). On ADC value analysis, LMS (n = 43) had a weighted mean and standard deviation of 0.896 ± 0.19 10–3 mm2/s, 0.929 ± 0.182 10–3 mm2/s, which were significantly lower from 1.258 ± 0.303 10–3 mm2/s, 1.304 ± 0.303 10–3 mm2/s for LM (n = 159) (p = < 0.0001, < 0.0001). Conclusion: Our meta-analysis demonstrated that high signal intensity on T1-weighted images and low ADC values can accurately differentiate LMS from LM. Although, LMS had a higher pooled sensitivity for T2-weighted increased signal intensity compared to LM, there was no statistical significance.

and LM, 2) human studies, 3) complete original publications, and 4) studies with a histopathologic analysis or imaging features as the standard of reference.

Exclusion Criteria
The exclusion criteria were as follows: 1) studies not published in English; 2) review articles, meeting abstracts, letters, case reports, and articles without sufficient data; and 3) diagnostic techniques other than MRI.

Study Quality Assessment
A single reviewer (M.V.) assessed the quality of all eligible studies using the current Quality Assessment of Diagnos-tic Accuracy Studies (QUADAS)-2 tool. This tool includes four major domains: patient selection, index test, reference standard, and flow and timing. These domains were then further assessed on the basis of the risk of bias, and their applicability was rated as 'high,' 'low,' or 'unclear.' A second reviewer (P.B.) assessed the accuracy of this assessment.

Data Extraction
The following data were extracted from each study: 1) title, author or authors, country in which the study was performed, year published, study type, and MRI scanner; 2) number of patients; 3) study objective; 4) primary findings; and 5) statistical data for analysis.

Diagnostic Performance Analysis
To reduce clinical (pretest probability of malignancy) and methodologic heterogeneity, our primary diagnostic performance analysis included increased signal intensity on T1WI and T2WI and the ADC values of LM and LMS.

Standard of Reference
The standard of reference was histologic confirmation of the lesion (obtained during surgery or biopsy). In the meta-analysis, eight studies used histologic confirmation as the only standard of reference. In the Ando et al. study, histologic confirmation and imaging features served as the reference standard [3].

Statistical Analysis
Data on T1WI or T2WI signal intensity were collected from eight selected studies and included the true number of patients with disease (true positive) and the number detected on T1WI or T2WI. Analyses were performed separately for LMS and LM data, calculating the sensitivities of increased signal intensity on T1WI and T2WI. Sensitivity was estimated with the available study-level data from a random effects model using the DerSimonian and Laird approach [13]. ADC values were compared between LMS and LM. Means and standard deviations were obtained for each study, and an overall pooled mean and standard deviation (SD) were calculated. A t-test was performed using these values to compare LMS and LM. All statistical analyses were performed using R software version 3.6.1. All statistical tests used a significance level of 5%. There were no adjustments required for multiple testing.

Study Selection and Description
The initial database search yielded 60 articles. After we removed all duplicate studies, 45 remained. We reviewed the titles and abstracts to identify articles that were irrelevant to our analysis or were reviews, case reports, letters, or editorials and excluded 31, leaving 14 potentially eligible articles. Of these, five lacked sufficient data and were excluded, leaving a total of 9 ( Table 1). A flow diagram of the study selection procedure is shown in Figure 3.

QUADAS-2 Assessment
The results of the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) are presented in Table 2

Diagnostic Performance
We performed a T1WI high signal intensity analysis and found that its pooled sensitivity for LMS (56.8%) was significantly higher than for LM (7.6%) (p = 0.0094  Figures 4 and 5).

Discussion
We found that high signal intensity on T1WI and low ADC values can be used to differentiate LMS from LM. LMS had higher pooled sensitivity on T2WI than did LM, but this result was not statistically significant.
On MRI, LMs usually present as solitary or multifocal, well-defined masses of variable size, with low signal intensities on T1WI and T2WI. However, LM with red degeneration and lipoleiomyoma demonstrates high signal intensity on T1WI, and cystic and degenerative LM often show high signal intensity on T2WI [1]. LMS usually presents as a heterogeneous and poorly demarcated mass, with variable appearance on T1WI. It may show low or intermediate signal intensity on T1WI, similar to LM, but it frequently demonstrates areas of high signal intensity, corresponding to hemorrhage or necrosis. On T2WI, LMS shows an intermediate to high signal because of its high cellularity. In addition to the imaging discrepancies between LMS and LM, there are discordant results in the literature on the diagnostic accuracy of high signal intensity on T1WI and T2WI [14].
An earlier systemic review by Kaganov et al. reported that there was a significant relationship between T1WI and T2WI signal intensities on and tumor pathologic characteristics (p < 0.05) but not between ADC values and tumor pathologic characteristics (p = 0.18) [15]. The decision tree diagram indicates that low signal intensity on T1WI and T2WI was most commonly associated with LM, whereas high signal intensity was a good indicator of LMS. Our meta-analysis has the advantage of a larger patient cohort. Tanaka et al. reported that a more than 50% high signal intensity on intralesional T2WI, focal high signal on T1WI, and unenhancing regions can be used to distinguish LMS from LM [10]. High signal intensity on T1WI and T2WI had 72.7% sensitivity, 100% specificity, 100% positive predictive value, 80% negative predictive value, and 87.0% overall accuracy. A prospective study conducted by Lin et al. reported that the sensitivity of T2WI was significantly higher than that of T1WI (0.88 vs 0.63, p < 0.05) [6]. They also found that the specificity (0.96) and AUC (0.92) of contrast-enhanced MRI were significantly higher than those of diffusion-weighed imaging, T2WI, or T1WI. Valdes-Devesa et al. reported that a high or inhomogeneous signal on T2WI and poorly defined borders were significantly more common in sarcomas (LMS, carcinosarcoma, and endometrial stromal sarcoma) than in LMS [11]. Contrary to the results of our meta-analysis, Ando et al. reported that LMs had more homogenous high  signal intensity on T1WI, were well-defined, had more frequently T2 hypointense rims, had greater signal intensity ratios of high signals on T1WI (high signal intensity area-to-skeletal muscle signal intensity ratio), and had lower high signal occupying rates on T1WI (high signal intensity on T1-to-whole tumor square measure ratio) than did LMS (p < 0.05) [3]. Lipoleiomyomas and LM with red degeneration exhibit a high signal on T1WI, which is why they were excluded from the study as potential confounding factors.  [9]. The LM cases were subdivided into cellular, degenerated, and ordinary. All cellular LMs had low signal intensity on T1WI and high on T2WI. All ordinary LMs had low signal intensity on T1WI and T2WI; degenerated LMs had high signal intensity on T2WI.
Diffusion-weighted imaging is based on the diffusion motion of water molecules. It is widely used to distinguish   between malignant and benign tumors by measuring the ADC value [16]. In general, a low ADC value is correlated with malignant lesions, as their higher cellularity and total nuclear area restrict water diffusion [17][18]. Previous studies have shown that ADC values can help distinguish uterine sarcoma from LM [5-9, 11, 12]. However, an overlapping of the ADC value between ordinary LMs and LMS has also been mentioned [6,9]. Our findings are in consensus with those of most previous studies, as we found that the weighted mean and SD for LMS (0. Tamai et al. reported that the ADC values of uterine sarcomas (1.17 ± 0.15) were lower than were those of the normal myometrium (1.62 ± 0.11) and degenerated LM (1.70 ± 0.11), with no overlap [9]. The ADC values overlapped with those of ordinary LM (0.88 ± 0.27) and cellular LM (1.19 ± 0.18). Similarly, in the study by Lin et al., the ADC value of the combined LMS and Uterine smooth muscle tumor of uncertain malignant potential (STUMP) (median, 0.89; range, 0.74-1.85) was significantly lower than that of LM (median, 1.19; range, 0.70-2.04; p < 0.05) [6]. Nonetheless, the ADC values overlapped among LMS (mean ± SD, 1.05 ± 0.41), STUMP (0.92 ± 0.13), and LM, including cellular (1.43 ± 0.58), infarcted (1.23 ± 0.50), degenerated (1.17 ± 0.17), and ordinary LM (1.14 ± 0.16).
Our study had some limitations. First, our primary analysis was limited because of the small number of studies (n = 8) included in the meta-analysis. Second, combining data from studies without standardized protocols or techniques may have resulted in bias and yielded results that are difficult to interpret and translate to clinical settings. Third, the included studies provided limited data on lesion size, LM type, uterine sarcoma type, interreader variability in MRI reporting, and temporal parameters that can affect the diagnostic performance of MRI. Despite current advances in imaging, there remains a lack of consensus regarding which MRI features are useful for differentiating LMS from LM. The results of our meta-analysis indicate that increased signal intensity on T1WI and low ADC values can provide accurate differentiation. Although, increased signal intensity on T2WI had higher sensitivity for LMS compared to leiomyomas (93% vs 54.5%, 93% vs 53.9%), there was no statistical significance. We recommend that prospective studies with larger cohorts be carried out to further improve the consensus on the significant MRI features of LMS and LM.