|Home | About | Journals | Submit | Contact Us | Français|
To compare MR and US imaging for uterine fibroid measurement
Eighteen women undergoing hysterectomy for symptomatic fibroids underwent preoperative pelvic US and MR imaging. Resected fibroids were correlated with the images. Weighted kappa agreement statistics and Spearman correlations for patient characteristics were calculated.
MR imaging identified 121 of 151 pathologically-confirmed fibroids, yielding 91% positive predictive value (PPV) (95%CI: 85%,95%) and 80% sensitivity (95%CI: 73%,86%). PPV and sensitivity for US were 97% (95%CI: 89%,100%) and 40% (95%CI: 32%,48%), respectively. Mean diameter-equivalent discrepancies between imaging and pathological measurements were 0.51±0.68 cm for MRI and 0.76±0.88 cm for US. Kappa statistics comparing imaging to pathology showed better agreement for MR than US (k=0.60 vs. 0.36). The number of fibroids detected by MR imaging predicted measurement errors (r=0.76; p=0.0002).
Superior sensitivity and minimal measurement discrepancies suggest MR imaging should be preferentially utilized for assessing fibroids in clinical research studies.
Precise uterine fibroid mapping (localization, measurement, and characterization) is essential for research elucidating the natural history of these tumors and for evaluating therapeutic responses to investigational agents.1 Ultrasonography (US), using transvaginal and transabdominal modalities, has been employed most frequently, due to its accessibility and relatively low cost. While a cost-effective instrument, US has been criticized for its significant operator-dependence, resulting in inferior reproducibility as compared to magnetic resonance (MR) imaging.2-5 MR imaging, while more costly, has been touted as the most sensitive modality for evaluating uterine myomas, particularly for the detection of small fibroids.6
To date, few studies have directly evaluated whether MR imaging should be preferred over US as the more accurate assessment of fibroid tumor burden and size.7 In one study, the accuracy of imaging techniques, both US and MRI, was reported to be associated with the number of fibroids present, with US and MRI demonstrating similar accuracy among subjects with few myomas. However, when greater fibroid tumor volume was encountered, MR imaging seemed to outperform transvaginal US in the accurate detection of uterine fibroids.7 Because of the limited comparisons of the two imaging modalities to pathological specimens, we set out to evaluate the performance characteristics of both MR and US imaging compared to hysterectomy specimens (gold standard) collected as part of a randomized, placebo-controlled therapeutic trial.8 The primary objective of this study was to assess the diagnostic ability of MR and US imaging to locate and accurately measure uterine fibroids as part of clinical investigation.
Informed oral and written consent for an Institutional Review Board approved protocol was obtained from healthy, pre-menopausal women with symptomatic uterine fibroid(s) (clinicaltrials.gov, NCT00290251).8 Within 2 weeks of hysterectomy, subjects underwent pelvic imaging (both US and MR) to document fibroid location, size, and number. MR imaging was performed using non-contrast enhanced pelvic T1- and T2-weighted SE images (Achieva 1.5; Philips Medical Systems, Bothell, WA; or Signa HD 1.5, GE Healthcare, Buckinghamshire, UK). Six mm slices were obtained in the sagittal, coronal, and axial planes. US was performed by a single operator using the iU22 machine (Philips Medical Systems) with a 4-9 MHz transvaginal probe. In cases of large uteri outside of the field of view, a 3-5 MHz transabdominal probe was used. Photographic and video documentation was obtained. Pre-surgical image mapping of both US and MR images was performed by a senior radiologist (AP) prior to hysterectomy, so that each fibroid was assigned a number and measurements of three perpendicular dimensions as noted below. A fibroid was considered to be submucosal in location if it distorted the uterine cavity; whereas an intramural fibroid was one that did not distort the uterine cavity and had <50% of its volume protruding into the serosal surface.9 Lastly, a subserosal fibroid had >50% of its volume protruding out of the uterine serosal surface.9
After hysterectomy, fibroids were correlated with the location predicted by the pre-operative image map and assigned the corresponding number. They were then removed and measured in three dimensions by providers unaware of the imaging size. The outcomes considered were fibroid volume assessment and localization as determined by US, MR imaging, and surgical findings. For each fibroid identified by a modality, its size was recorded in three perpendicular dimensions appropriate for describing a prolate ellipsoid (approximate volume: π × (d1 × d2 × d3)/6)). The ellipsoid method of volumetric determination has been regarded as the method of choice for quantitative assessment of uterine volume response to hormonal treatment.10 Three-dimensional data was available for all fibroids identified by MR imaging and those noted from the surgical specimens. However, five fibroids identified with US distributed among five patients had 1 missing dimension (n=2) and 2 missing dimensions (n=3) for which the volumes were imputed by using the smaller non-missing length or, in cases of 2 missing dimensions, the only non-missing length. When a fibroid was not identified by a modality, the volume was assigned a volume of 0 cm3 (i.e. not present).
The NICHD Clinical Trials Database was used for data entry and management. The data were analyzed using Stata 10.0 (Stata; College Station, TX) and StatXact 8 (Cytel Software Corp., Cambridge, MA). Descriptive statistics including means, standard deviations, medians, frequencies, and proportions were determined. Binomial exact confidence intervals were calculated for proportions. Because of the non-normality of the fibroid volume data, log transformations and the arcsinh function (to handle zero and negative values arising from the differences between imaging and pathologic volumes) were used for analysis. Bland-Altman plots were generated to assess the relationship between predicted fibroid volumes by imaging modality compared to surgically confirmed volumes. For each subject, the number of fibroids identified by each imaging modality was evaluated to determine its performance compared to the gold standard pathologic measurements. To determine the size discrepancy between that predicted by the imaging modality and actual pathological size, absolute differences of the diameter-equivalent measures (i.e. conversion of calculated volume to a mean diameter value) were analyzed. This transformation of calculated volume to diameter-equivalent measures resolved the issue of non-normality associated with volumetric measures.
Diameter-equivalent measures for each imaging measurement also were used to create five ordered categories corresponding to clinically meaningful diameters. The fibroid equivalent diameter categories were defined as the following: 0: diameter=0 (volume=0 i.e. not found by an imaging modality); 1: diameter: 0 cm to 2 cm (volume: 0 cm3 to 4.2 cm3); 2: diameter: 2 cm to 6 cm (4.2 cm3 to 113.1 cm3); 3: diameter: 6 cm to 10 cm (volume: 113.1 to 523.6); 4: diameter >10 cm (volume >523.6 cm3). To further address the agreement between the imaging modalities and pathology, weighted kappa agreement statistics were calculated for each modality (MRI and US) using Fleiss-Cohen weighting (equivalent to the inter-class correlation) classified as the following: 0.75-1.00 (excellent), 0.40-0.75 (fair to good agreement), <0.40 (poor agreement).11
A total discrepancy score for each subject was computed as the sum of the squared differences between the size categories totaled across all fibroids to generate a summary of the overall measurement error. To assess variables predictive of measurement error, further analyses were performed using the discrepancy between size categories predicted by MR or US imaging and the pathologically confirmed measurements for individual fibroids. The rationale for this approach was that summing individual discrepancies across all fibroids was a reasonable summary of the overall error for a woman and that using a weighting scheme of squared discrepancies was in keeping with the Fleiss-Cohen definition of the weighted kappa statistic and with standard statistical approaches of weighting errors by their squares. Characteristics of each woman were analyzed as possible predictors of measurement error using Spearman correlations for each set of predictors to determine the predictive values of each characteristic: race, age, body mass index, gravidity, parity, clinical uterine size, and patient self-reported uterine bleeding categories (0-4; 0=none, 1= very mild, 2=mild, 3=moderate, 4=severe), and pelvic pain categories (0-3; 0=none, 1=mild, 2=moderate 3=severe). Statistical significance was considered to be p<0.05 using 2-tailed tests.
Eighteen study subjects yielded 151 pathologically confirmed fibroids (true lesions) with a wide range of size (median: 6.7 cm3; range: 0.01-4985.7 cm3). MR imaging identified 133 total fibroids; however, 12 reported fibroids were not identified at the time of surgery (spurious), resulting in 121 surgically confirmed fibroids mapped by MR imaging (positive predictive value: 91%; 95%CI: 85%, 95%). Thirty confirmed fibroids were not identified by MR imaging, resulting in 80% sensitivity (95%CI: 73%, 86%). Fibroid location did not seem to affect MR results. Of the 76 intramural fibroids identified surgically, 20 were missed by MR imaging (26%); while 7 of the 34 pathologically-confirmed subserosal fibroids were not identified with MR imaging (21%). The 3 remaining fibroids missed by MR imaging did not have pathological location information available. The sizes of those fibroids that were either missed by or spuriously identified by MR imaging are presented in the Table. Notably, MR imaging tended to miss small fibroids, as 14 of the 30 missing lesions were ≤0.52 cm3 (diameter equivalence: ≤1 cm). MR imaging did not identify 4 fibroids with a diameter equivalence of >3 cm and one was 121 cm3 (diameter equivalence: 6.1 cm). Figure 1 demonstrates Bland-Altman plots depicting the relationship between the volume predicted by MR imaging and the pathological volume noted at the time of surgery (n=151 fibroids). Of those fibroids identified by MR imaging, the mean diameter-equivalent discrepancy between that predicted by MR imaging and the pathological measurement was 0.51±0.68 cm. MR imaging identified more than two-thirds of the total fibroids present in 89% (16/18) of subjects (Figure 2). Furthermore, the weighted kappa statistic comparing MR imaging to pathology yielded a k=0.60 (good agreement).
US imaging identified 62 total fibroids; however, 2 reported fibroids were not identified at the time of surgery (spurious), resulting in 60 fibroids mapped by US imaging that were identified at surgery (positive predictive value: 97%; 95%CI: 89%, 100%). US failed to document 91 fibroids that were identified at surgery resulting in 40% sensitivity (95%CI: 32%, 48%). US missed intramural fibroids most frequently. Of the 76 pathologically-confirmed intramural fibroids, 54 were missed by US (71%). In addition, 20 of the 34 surgically-confirmed subserosal fibroids (59%) and 5 of the 17 confirmed submucosal (29%) fibroids were not identified by US. Twelve fibroids missed by US that were identified at surgery did not have surgical location information available. Additional details regarding the sizes of fibroids misidentified by US are presented in the Table. When excluding small fibroids measuring ≤0.52 cm3 (diameter equivalence: ≤1 cm) from analysis, the sensitivity of US increased to 47% (95%CI: 37%, 57%), while the sensitivity of MR imaging was slightly reduced to 84% (95%CI: 72%, 87%).
Figure 3 depicts Bland-Altman plots demonstrating the relationship between the volume predicted by US imaging and the pathological volume noted at the time of surgery (n=151 fibroids). When considering the relationship between the diameter-equivalent measure predicted by US and the pathological measures of individual fibroids noted at surgery, the mean discrepancy among the 60 fibroids that US identified was slightly greater than that observed with MR imaging (0.76±0.88 cm). Nevertheless, US imaging only identified one-half of the total fibroids present in 56% (10/18) subjects (Figure 4). Furthermore, 24 of the 151 fibroids present at surgery that were not identified by US (16%) were larger than 14.1 cm3 (diameter equivalence: 3 cm), including 4 larger than a 113 cm3 (diameter equivalence: 6 cm). Overall, US identified fewer pathologically confirmed fibroids, while identifying fewer false positive “spurious fibroids”, resulting in a weighted kappa=0.36 (moderately poor).
The aforementioned size discrepancy scoring confirmed the observation that the discrepancy between MR imaging and pathological fibroid size was the smallest when a single fibroid was present (data not shown). This observation was likewise noted when examining the performance of US. Interestingly, total fibroid volume, as predicted by MR imaging, did not predict measurement errors as defined by this approach (r=0.05; p=0.84), whereas the total number of fibroids present predicted measurement error by this modality (r=0.76; p=0.0002). Additional variables including race, age, body mass index, gravidity, parity, clinical uterine size, the presence of uterine bleeding and pelvic pain did not predict measurement errors with MR imaging (data not shown). When evaluating predictors of performance using US, the total number of fibroids present was the best predictor of measurement error (r=0.39; p=0.11); however, due to the poor sensitivity of US, no variables analyzed demonstrated significant predictive capacity.
Accurate assessment of uterine fibroids remains an important component of clinical research endeavors as well as clinical management, because many conservative therapeutic options rely on this information.12, 13 In this study, we set out to assess the diagnostic ability of MR and US imaging to locate and accurately measure uterine fibroids as part of clinical investigation. Hopefully, this information would suggest a preferred modality for therapeutic trials and for studies aimed at elucidating the natural history of uterine fibroids. An optimal imaging modality must have high sensitivity for the detection of these lesions and any discrepancy between the volume predicted by imaging and the actual pathologic specimens should be small. Finally, measurable patient characteristics influencing the accuracy of the imaging modality need to be understood.
In this study, the sensitivity of MR imaging was two-fold greater than US for the detection of uterine fibroids (MR imaging: 80%; US: 40%) using pathological specimens as a gold standard. However, when fibroids were identified, the positive predictive value for MR imaging and US were similar. Even after excluding small fibroids ≤0.52 cm3 (diameter equivalence: ≤1 cm) from consideration, the sensitivity of US remained low (47%). This observation suggests that MR imaging be considered as the best modality for the detection of uterine fibroids in clinical research, especially considering its superior ability to detect smaller lesions.
Previous studies suggested that MRI had excellent reproducibility (k=0.97), while US demonstrated lower, albeit reasonable, reproducibility (k=0.74).5 Nevertheless, in contrast to MR, the reproducibility of transvaginal US was limited. Because our underlying study was an intervention trial over a 3-month interval, initial assessments of fibroid location and size would not accurately reflect the reproducibility of these modalities as both time and the intervention might affect measures. In addition, a sole radiologist assessed all images. As a result, we were unable to examine the intra- and inter-observer reproducibility of each imaging modality. However, it has been observed that substantial disparities are present when results are obtained by different observers using US, whereas MRI produced highly reproducible results for the exclusion of abnormalities 5, further supporting the preferential use of MR imaging for clinical investigation.
Although US sensitivity was poor, the discrepancy between pathological specimens and US measurements was small and similar to that observed with MR imaging. This suggests that for clinical applications, one would have reasonable confidence in US assessment of size of identified fibroids, recognizing that some fibroids may be missed. However, both imaging modalities suffered in the ability to predict fibroid volume at extremely large sizes. This is not altogether surprising given that the tumor volume is a cubed function of the fibroid diameters and small discrepancies in large measures result in large errors in volume.
Lastly, our study demonstrated that the total number of fibroids predicted MR imaging measurement errors, but we were unable to demonstrate the same effect with US, largely due to the observed poor sensitivity. Furthermore, unlike previous studies 7, we did not observe a correlation between total fibroid volume and measurement error for MRI. Additional factors including body composition also did not affect the predictive ability of MR imaging or US to detect accurately fibroid size.
In summary, data derived from a randomized clinical trial afforded the opportunity to examine the ability of US and MR imaging to identify uterine fibroids and to predict pathologic measurements. Our study had the advantage that all subjects underwent hysterectomy rather than myomectomy 6, as myomectomy is a less optimal gold-standard resulting in potentially less accurate estimates of diagnostic accuracy. Furthermore, we were able to assess patient characteristics that might affect the predictive capacity of uterine fibroid imaging modalities. These data are particularly helpful with regard to selecting imaging modalities for clinical research endeavors aimed at characterizing the natural history of uterine fibroid growth and assessing responses to therapeutic agents for uterine fibroids. Because MR imaging demonstrated superior sensitivity and minimal measurement discrepancies, it would be reasonable to recommend MR imaging preferentially for the assessment of fibroid number and growth or regression in clinical studies.
Financial Support: This research was supported, in part, by the Program in Reproductive and Adult Endocrinology, Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health, Bethesda, MD.
Where the work was done: Bethesda, MD
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.