Two- and three-dimensional transvaginal ultrasound with power Doppler angiography and gel infusion sonography for diagnosis of endometrial malignancy

Transcription

1 Ultrasound Obstet Gynecol 2015; 45: Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: /uog Two- and three-dimensional transvaginal ultrasound with power Doppler angiography and gel infusion sonography for diagnosis of endometrial malignancy M.DUEHOLM*,J.W.CHRISTENSEN*,S.RYDBJERG*,E.S.HANSEN andg.ørtoft* *Department of Obstetrics and Gynecology, Aarhus University Hospital, Aarhus, Denmark; Department of Pathology, Aarhus University Hospital, Aarhus, Denmark KEYWORDS: color Doppler; endometrial neoplasms; postmenopause; ultrasonography; uterine hemorrhage ABSTRACT Objectives To evaluate the diagnostic efficiency of two-dimensional (2D) and three-dimensional (3D) transvaginal ultrasonography, power Doppler angiography (PDA) and gel infusion sonography (GIS) at offline analysis for recognition of malignant endometrium compared with real-time evaluation during scanning, and to determine optimal image parameters at 3D analysis. Methods One hundred and sixty-nine consecutive women with postmenopausal bleeding and endometrial thickness 5 mm underwent systematic evaluation of endometrial pattern on 2D imaging, and 2D videoclips and 3D volumes were later analyzed offline. Histopathological findings at hysteroscopy or hysterectomy were used as the reference standard. The efficiency of the different techniques for diagnosis of malignancy was calculated and compared. 3D image parameters, endometrial volume and 3D vascular indices were assessed. Optimal 3D image parameters were transformed by logistic regression into a risk of endometrial cancer (REC) score, including scores for body mass index, endometrial thickness and endometrial morphology at gray-scale and PDA and GIS. Results Offline 2D and 3D analysis were equivalent, but had lower diagnostic performance compared with real-time evaluation during scanning. Their diagnostic performance was not markedly improved by the addition of PDA or GIS, but their efficiency was comparable with that of real-time 2D-GIS in offline examinations of good image quality. On logistic regression, the 3D parameters from the REC-score system had the highest diagnostic efficiency. The area under the curve of the REC-score system at 3D-GIS (0.89) was not improved by inclusion of vascular indices or endometrial volume calculations. Conclusion Real-time evaluation during scanning is most efficient, but offline 2D and 3D analysis is useful for prediction of endometrial cancer when good image quality can be obtained. The diagnostic efficiency at 3D analysis may be improved by use of REC-scoring systems, without the need for calculation of vascular indices or endometrial volume. The optimal imaging modality appears to be real-time 2D-GIS. Copyright 2014 ISUOG. Published by John Wiley & Sons Ltd. INTRODUCTION Transvaginal ultrasound (TVS) is the most cost-effective first-line test in the diagnostic work-up of women with postmenopausal bleeding 1 3 and may identify those at very low risk of endometrial cancer 4,5. In patients with increased endometrial thickness, identification of those at high risk for endometrial cancer at the first-line TVS examination may allow them to be referred for fast-track investigation and staging. This could shorten the time from first investigation to treatment. In experienced hands, evaluation of endometrial pattern on TVS, saline contrast sonohysterography (SCSH) and Doppler 6 9 yield promising results for identification of patients at high risk of endometrial cancer Ideally, first-line investigations should be performed by experienced assessors to ensure fast-track referral, but they are available to perform such procedures only rarely. Assessment at two-dimensional (2D) TVS is usually performed in real-time and requires skill in image optimization and pathology recognition. Modern ultrasound machines and software have simplified image optimization, but recognition of pathology continues to require considerable experience. Offline analysis of three-dimensional (3D) volumes is reliable for assessing Correspondence to: Dr M. Dueholm, Department of Obstetrics and Gynecology, Aarhus University Hospital, Brendstrupgaardsvej 100, 8200 Aarhus N, Denmark ( dueholm@dadlnet.dk) Accepted: 16 May 2014 Copyright 2014 ISUOG. Published by John Wiley & Sons Ltd. ORIGINAL PAPER

2 3D ultrasound and endometrial cancer 735 adnexal masses 13, and offline analysis of 3D volumes or 2D video recordings by experts could compensate for the lack of local experts and could hence be an efficient alternative. Real-time evaluation of endometrial pattern based on a score system at 2D-TVS and gel infusion sonography (GIS) is highly efficient for diagnosis of malignancy 14. The score system may be improved at 3D-volume assessment by inclusion of 3D vascular indices and endometrial volume (EV) measurements Thus, we hypothesize that 3D ultrasound may increase the diagnostic performance of 2D ultrasound for endometrial malignancy. We propose a rethink of the diagnostic approach, with the current practice of local first-line assessment being supplemented by secondary evaluation of 3D volumes by non-local experienced assessors. The aim of this study was to assess and compare the diagnostic efficiency of real-time analysis at 2D-TVS and offline analysis of 2D- and 3D-TVS, power Doppler angiography (PDA) and GIS for discrimination between benign and malignant endometrial conditions in women with postmenopausal bleeding and thickened endometrium and, furthermore, to evaluate optimal image parameters for 3D analysis. METHODS Two investigators, who were blinded to patients identities and to the results of all prior microscopic specimens and diagnoses, performed TVS with Doppler and GIS in 169 consecutive women (93 unselected and 76 referred from other hospitals) with postmenopausal bleeding and endometrial thickness 5 mm, from October 2010 to February 2012, at Aarhus University Hospital, Denmark. 2D clips and 3D volumes (TVS, TVS with Doppler and GIS) were also stored after each examination and later analyzed offline. Both examiners had several years of experience in the staging of endometrial cancer. The exclusion criteria used in the present study have been described elsewhere 14. The Central Denmark Regional Ethics Committee approved the study protocol. The reference standard was histopathological findings at hysteroscopy and/or hysterectomy; as this was standard procedure in the department, the committee found that no informed consent for this procedure was needed. Ultrasound analysis was performed in four rounds, as follows. In the first round, real-time evaluation at TVS with PDA and at TVS with GIS was performed and 2D clips and 3D volumes were stored. After ultrasound evaluation, a standard history with respect to clinical variables was obtained. In the second round, stored 2D clips (2D-TVS, 2D-PDA and 2D-GIS) were assessed. In the third round, 3D volumes (3D-TVS, 3D-PDA and 3D-GIS) were evaluated. In the fourth round, data on vascular indices and EV at 3D-PDA were obtained. Each round of analysis was followed by systematic evaluation of pattern (SEP), i.e. a systematic evaluation of the endometrial pattern, performed according to the terms introduced by the International Endometrial Tumor Analysis (IETA) group 15. This included completion of a standard systematic form with IETA parameters; a subjective impression of malignancy was then formed at the end of each examination (real-time 2D-PDA and 2D-GIS, and offline 2D-TVS, 2D-PDA, 2D-GIS, 3D-TVS, 3D-PDA and 3D-GIS). Included for offline analysis were all patients for whom 2D clips or correctly stored 3D volumes, obtained at any of the three examinations (TVS, PDA or GIS), could be evaluated (2D-TVS (n = 165), 2D-PDA (n = 153), 2D-GIS (n = 159), 3D-TVS (n = 169), 3D-PDA (n = 155) and 3D-GIS (n = 160)). A total of D and 3D offline evaluations were undertaken at TVS, PDA and TVS with GIS. In addition, vascular indices and EVs were calculated in 155 3D-PDA volumes. Real-time imaging Transvaginal ultrasound (TVS) and power Doppler angiography (PDA) Examinations were performed, according to a predetermined scanning protocol, with a Voluson E8 Expert (GE Healthcare Ultrasound, Milwaukee, WI, USA) ultrasound machine, equipped with a multifrequency (5 12 MHz) endovaginal probe. Endometrial structure and thickness, and analysis of vessels on PDA, were evaluated using a standard form meeting the IETA criteria 15, as previously described 14 ;the form comprised the following elements: internal endometrial echo structure (hyperechogenic/hypoechogenic/isoechogenic, cystic (regular/irregular), homogeneity/heterogeneity; endomyometrial border, subendometrial halo (endomyometrial junction): visualization and interruption of subendometrial halo, regular/irregular, homogeneity/heterogeneity, and bright line. A 2D power Doppler gate was activated to assess vascularization from the myometrium and the endometrium. Power Doppler settings were set to achieve maximum sensitivity to detect low-velocity flow without noise, using predetermined, standardized settings (frequency, 6 MHz; power Doppler gain, 50; dynamic range, 10 db; edge, 1; persistence, 2; color map, 1; gate, 2; filter, 3; pulse repetition frequency, 0.6). PDA analysis included visual evaluation of the vessels (Doppler parameters) for 15 : dominance (single or double); origin (focal or multifocal); number (few or multiple); large vessels (yes/no); branching (yes/no); if branching, regularity (regular/irregular); presence of: color splashes, densely packed vessels, circular flow. After TVS and Doppler examinations, evaluation of the image parameters was entered on the standard form. After SEP, the main diagnosis of malignancy was made subjectively in the presence of: heterogenic endometrium with irregular or interrupted endomyometrial junction; indistinct endomyometrial border; multiple vessels on Doppler sonography.

3 736 Dueholm et al. Gel installation sonography (GIS) GIS was performed as described previously 14.Asmall flexible sterile catheter (infant-feeding tube, Unometric no: , Abena A/S, Aabenraa, Denmark) mounted with a 10-mL syringe containing Instillagel (E.Tjellesen A/S, Lynge, Denmark) was introduced into the uterine cavity. During instillation of gel (GIS), the pressure was adjusted manually until sufficient for expansion of the uterine cavity. The distension was observed continuously by TVS, and was continued until the entire uterine cavity was clearly visible. The uterine cavity was evaluated in sagittal and coronal views. The same parameters were evaluated at GIS as had been at TVS. Additionally, endometrial thickness was measured in the sagittal plane, as the sum of the maximum endometrial thickness at the anterior and the posterior walls. A standardized coding sheet was used to record the additional GIS parameters defined by the IETA system 15, as previously described 14 : structure of the endometrial surface (smooth, polypoid, irregular), size of localized and/or diffuse lesions (i.e. lesion < or > 25% of surface), and structure of the surface of local lesions (regular or irregular). In order to avoid missing values, a third category, indefinable, was used for each classification. After SEP, the main diagnosis of malignancy at GIS was made subjectively in the presence of: TVS findings of malignancy and/or irregular surface of a localized or diffuse lesion at GIS. 3D volume acquisition A 3D volume acquisition box was placed over the uterus. The patient was asked to remain as still as possible, and at least two 3D volumes were acquired, with a sweep angle of D-PDA of the uterus was then performed with the previously defined fixed settings. Volume acquisition was repeated when artifacts appeared. Offline analysis 2D clips and 3D volume storage At each examination (TVS, Doppler, GIS), three 2D videoclips were obtained, including two sweeps in the sagittal plane and one in the transverse plane. The system was set for retrospective storage of clips of the last 5 10 seconds of scanning. Two 3D volumes at each of 3D-TVS, 3D-PDA and 3D-GIS were stored for later evaluation on a personal computer using the virtual organ computer-aided analysis (VOCAL) software (GE Healthcare Ultrasound). Evaluation of stored 2D videoclips 2D clips obtained during the real-time investigations were stored (by S.R.) in a separate file using the patients ID numbers. All 169 examinations (2D-TVS + 2D- PDA + 2D-GIS) were then given arbitrarily a new identity number. Three months after the investigation was completed, all 2D clips were evaluated (by M.D.) without knowledge of the patient s identity and pathology. The clips were assessed in random patient order, but for each patient in the order 2D-TVS, then 2D-PDA, then 2D-GIS. On a standard form, the assessor noted the patient s age and entered data on the quality of the images (score, 1 5) and the same morphological endometrial parameters as those obtained at the primary evaluation 14,15. Again, after SEP, a subjective diagnosis of malignancy was given at the end of each evaluation, first for TVS, then for Doppler and finally for GIS. Evaluation of stored 3D volumes The stored 3D volumes from the 3D-TVS, 3D-PDA and 3D-GIS examinations were evaluated using VOCAL software (by M.D.) 3 months after the 2D-videoclip evaluations had been completed. Again, all examinations were given a new identification number and were performed in random patient order, with the examiner blinded to the previous date of examination and the patient s identity. Each patient s examination consisted of six to 10 volume determinations in the order 3D-TVS, then 3D-PDA, then 3D-GIS. For each examination, we opened all stored volumes and used the volume with the highest quality for evaluation, according to the morphological criteria stated above. A diagnosis of malignancy was based on subjective impression after SEP andstatedattheendofeachexaminationandtheresults, including image quality, were entered into a standard form. Analysis of 3D power Doppler indices The 3D-PDA volumes were analyzed for 3D power Doppler indices 4 months later. Using the VOCAL program, we evaluated the endometrial area manually in the coronal or C -plane. Using a rotational technique with a 9 step, as described previously 11,16, we obtained 20 endometrial slices that outlined the endometrium at the endomyometrial junction, from the fundus to the internal cervical opening. The VOCAL program was used to calculate the EV and three 3D power Doppler indices: the vascularization index (VI), flow index (FI) and vascularization flow index (VFI). VI measures the number of color voxels in the volume. It may represent the vessels in the tissue and is expressed as a percentage. FI is the mean color value in the color voxels. It indicates the average intensity of blood flow and is expressed as a number from 0 to 100. VFI is the mean color value in all the voxels in the volume. It represents both vascularization and blood flow and is also expressed as a number from 0 to 100. Doppler and risk of endometrial cancer (REC) scores The most optimal Doppler image parameters at real-time 2D evaluation and at offline 2D videoclip evaluation by two observers were evaluated, and a Doppler score was

4 3D ultrasound and endometrial cancer 737 developed. This score was obtained by simple addition of the following Doppler parameters: vessels, but no single or double dominant vessel (1 point); multiple vessels (more than four or five) (1 point); large vessels (1 point); color-splash/densely packed vessels (1 point) 14. A score system for evaluation of the risk of endometrial cancer (REC score) was designed based on different 2D image parameters analyzed by logistic regression and the most optimal real-time evaluation of 2D imaging in this population 14. The REC-score system included: body mass index (BMI) 30 (1 point); total endometrial thickness 10 mm (1 point); total endometrial thickness 15 mm (1 point); vascularity, but no single/double dominant vessel (present = 1 point); multiple (more than four or five) vessels (present = 1 point); large vessels (present = 1 point); color-splash/densely packed vessels (present = 1 point); interrupted endomyometrial junction (present = 1 point); irregular surface at GIS (present = 1 point). Simple addition of these values constituted the REC score 14. A diagram of parameters in the REC-score system and accompanying ultrasound images is shown in Figure 1. Pathology All patients underwent hysteroscopy and/or hysterectomy; microscopic pathology was the reference standard for these examinations 14. Removal of all focal changes was attempted at hysteroscopy; in cases of large, diffuse or localized changes, resectoscopic biopsies were taken from the area of the endometrium with the largest changes. Three to five biopsies were sampled. In cases of normal hysteroscopic findings, one biopsy was taken from the anterior wall of the uterine cavity, one from the posterior wall of the uterine cavity, and curettage was performed at the end of the hysteroscopy. Two pathologists specialized in gynecological oncology evaluated all specimens. Statistics and analysis Data were analyzed using STATA (Statistic Data Analysis, STATA Corp., TX, USA). Continuous data and normally distributed data are expressed as mean ± SD. Wilcoxon s signed rank test and McNemar s test were used for comparison of the scored quality of the examination. We evaluated and compared the diagnostic performance of real-time 2D imaging vs offline 2D videoclip analysis vs offline 3D volume analysis. The diagnostic performance, with regard to discrimination between benign and malignant endometrium, of the different techniques was evaluated in terms of the areas under the receiver operating characteristics (ROC) curve (AUC), sensitivity, specificity, positive likelihood ratio (LR+) and negative likelihood ratio (LR ). ROC-curve analysis was used to compare the diagnostic performance of the different techniques. ROC-curve analysis was performed for all parameters measured by 3D imaging (pattern variables and parameters calculated by VOCAL program (Doppler indices and endometrial volume)). The sensitivity, specificity, LR+ and LR of the most optimal cut-off points were calculated. Stepwise multivariate logistic regression was performed to build a model that could be used to predict malignancy at 3D evaluation. A maximum of four fitting variables were allowed in the model. Optimal cut-off points were calculated using a 45 tangent line intersection or the smallest sum of residual sensitivity and specificity. The most optimal ROC curve calculated by logistic regression was compared to the REC score developed at real-time 2D evaluation; χ 2 tests were used for discrete data. Statistical tests were two-tailed; P < 0.05 was considered statistically significant. RESULTS The mean age of patients was 64.1 (range, 45 89) years. There was endometrial cancer present in 69 (40.8%) of the patients, hyperplasia in 20 (11.8%) and endometrial polyps in 51 (30.2%). Overall diagnostic efficiency of real-time 2D imaging and offline 2D videoclip and 3D volume evaluations Table 1 shows the diagnostic efficiency of real-time 2D imaging and the offline evaluation of 2D videoclips and 3D volumes. Sensitivities ranged from 73% to 89% and specificities were close to 90% at the different offline 3D and 2D evaluations. Addition of PDA and GIS did not change markedly the diagnostic efficiency for offline 2D and 3D evaluation. Real-time evaluation during 2D-GIS had better diagnostic efficiency than did analysis of both 3D-GIS volumes and 2D-GIS videoclips (153 patients, P < 0.05). Figure 2 displays the diagnostic performance (AUC) obtained by the different real-time and offline 2D and 3D techniques. The figure includes only the 136 patients in whom all evaluations with all methods were completed. There was no significant difference between AUCs for offline 2D-TVS and 3D-TVS or between AUCs for offline 2D-PDA, offline 3D-PDA and real-time 2D-PDA evaluations. Diagnostic efficiency and quality of offline 2D clips and 3D evaluations The scores for quality were significantly better for 2D-GIS videoclips than for 3D-GIS volumes (P < 0.001), with 13.2% of the 2D clips being of low quality, compared with 22.5% for the 3D volumes. When we restricted evaluations to cases for which both 2D videoclips and 3D volumes of sufficient quality had been obtained (Table 1), the diagnostic efficiency of these 2D videoclips or 3D volumes was not significantly lower than the diagnostic efficiency of real-time evaluation during scanning. Cases with consensus between 2D videoclips and 3D volumes had a high diagnostic efficiency. The discrepancies (16% between 2D-TVS and 3D-TVS, 5% between 2D-PDA and 3D-PDA, and 9%

5 738 Dueholm et al. (a) (e) Figure 1 Diagrams and ultrasound images illustrating the nine parameters in the risk of endometrial cancer (REC) score system 14. (a) Diagram representing REC-score components for: (1) body mass index 30 (+1 point), (2) total endometrial thickness 10mm (+1 point), (3) total endometrial thickness 15 mm (+1 point), (4) interrupted endomyometrial junction (+1 point) and (5) irregular surface at gel instillation sonography (GIS) (+1 point). Accompanying ultrasound images illustrate interrupted endomyometrial junction (c) and irregular endometrial surface at two-dimensional GIS (b) and three-dimensional (3D) GIS (d). (e) Diagram representing: (6) vascularity with no single/double vessel (+1 point), (7) multiple (more than four or five) vessels (+1 point), (8) large vessels (+1 point) and (9) colorsplash/densely packed vessels (+1 point). Accompanying ultrasound images illustrate a single vessel (3D endometrial polyp) (f), densely packed vessels (g) and multiple and large vessels (h). Simple addition of the REC-score points constituted the REC score.

6 3D ultrasound and endometrial cancer 739 Table 1 Diagnostic efficiency for endometrial malignancy of real-time and offline analysis of two- (2D) and three- (3D) dimensional transvaginal ultrasound (TVS), power Doppler angiography (PDA) and gel infusion sonography (GIS), in patients with postmenopausal bleeding and endometrial thickness 5mm Sensitivity Specificity N % (95% CI) n/n % (95% CI) n/n LR+ (95% CI) LR (95% CI) AUC (95% CI) AUC (95% CI): Patients who had all tests Real-time evaluation N = 158 2D-PDA (77 94) 60/ (80 94) 88/ ( ) 0.15 ( ) ( ) ( ) 2D-GIS (79 95) 62/ (89 99) 84/ ( ) 0.12 ( ) ( ) ( ) Offline evaluation 2D videoclips N = 142 2D-TVS (69 89) 54/ (81 94) 87/ ( ) 0.22 ( ) ( ) ( ) 2D-PDA (67 89) 47/ (81 95) 84/ ( ) 0.23 ( ) ( ) ( ) 2D-GIS (74 92) 59/ (74 90) 74/ ( ) 0.19 ( ) ( ) ( ) 3D volumes N = 145 3D-TVS (57 80) 48/ (85 97) 92/ ( ) 0.33 ( ) ( ) ( ) 3D-PDA (60 83) 45/ (81 95) 83/ ( ) 0.31 ( ) ( ) ( ) 3D-GIS (67 88) 55/ (79 94) 79/ ( ) 0.24 ( ) ( ) ( ) Investigation with agreement at 2D and 3D (offline analysis) TVS (69 92) 42/ (92 98) 84/ ( ) 0.18 ( ) ( ) ( ) Doppler (71 92) 45/ (87 98) 84/ (6.3 35) 0.18 ( ) ( ) ( ) GIS (77 95) 52/ (82 96) 73/ ( ) 0.13 ( ) ( ) ( ) Image quality 1 3*(offline analysis) 3D-GIS (68 91) 43/ (81 96) 64/ ( ) 0.21 ( ) ( ) ( ) 2D-GIS (78 96) 49/ (77 93) 69/ ( ) 0.13 ( ) ( ) ( ) N = 120 N = 115 Diagnosis was based on subjective impression of malignancy following systematic evaluation of endometrial pattern. Reference standard was histopathology at hysteroscopy or hysterectomy. *Image quality 1 3: quality excellent to intermediate. Comparison of areas under the receiver operating characteristics curves (AUC): 3D-TVS vs 2D-TVS offline (n = 163): NS; 3D-PDA vs 2D-PDA offline vs real-time 2D-PDA (n = 151): NS; 3D-GIS vs 2D-GIS offline vs real-time 2D-GIS (n = 153): P < LR+, positive likelihood ratio; LR, negative likelihood ratio; NS, not significant.

7 740 Dueholm et al. AUC PDA GIS 2D-TVS 2D-GIS 3D-PDA 2D-PDA 3D-TVS 3D-GIS Figure 2 Areas under the receiver operating characteristics curves, with 95% CIs, of two- and three-dimensional transvaginal ultrasonography (2D-TVS, 3D-TVS), power Doppler angiography (2D-PDA, 3D-PDA) and gel infusion sonography (2D-GIS, 3D-GIS) at offline analysis and real-time analysis during imaging (PDA, GIS), for diagnosis of endometrial malignancy in patients with postmenopausal bleeding and endometrial thickness 5 mm; only the 136 patients in whom all evaluations by all techniques were completed are included. between 2D-GIS and 3D-GIS) revealed a small group of indefinable evaluations at offline analysis. 3D power Doppler calculations Table 2 summarizes endometrial thickness, the vascular indices and EV according to different kinds of pathology. ET, EV, VI, FI and VFI were all clearly associated with cancer (P < 0.01). Table 3 shows the diagnostic efficiency of the different criteria used at the different 3D evaluations. EV calculation at 3D analysis did not have a higher diagnostic efficiency than did measurement of the endometrial thickness. The individual, different Doppler parameters (vessel patterns a g) at 3D-PDA had large AUCs that were comparable to those of VI, VFI and FI. Doppler score (comprising patterns c, e, f and g) had an AUC above Addition of or replacement with VI, VFI or FI at multivariate logistic regression did not increase the diagnostic efficiency. The model which included BMI, endometrial thickness, presence of an interrupted endomyometrial junction and Doppler score (Table 3, Model 3), had a diagnostic efficiency (AUC, 0.879) that was higher (P = 0.01) than that of the primary subjective evaluation (SEP) at 3D-PDA (AUC, 0.809; Table 1). Evaluation of 3D-GIS (Model 4, Table 3), with BMI, an interrupted endomyometrial junction, Doppler score and irregular endometrial surface at 3D-GIS, had the highest diagnostic efficiency on multivariate regression, with an AUC (0.908) that was clearly higher than that of the subjective SEP at 3D-GIS (AUC, 0.832; Table 1). Application of the REC-score system at 3D-PDA or 3D-GIS had comparable efficiency compared with their respective models (Models 3 and 4). REC scores also had a higher calculated diagnostic efficiency than that of the subjective SEP at 3D-PDA or 3D-GIS (Table 1; P = 0.01). The REC scores for 3D-PDA and 3D-GIS had higher AUCs than had Model 1 (Table 3). DISCUSSION At offline analysis of 3D volumes and 2D videoclips, endometrial cancers were identified with sensitivities of 73 89%, with very few false-positive cases. However, the diagnostic efficiency of neither 2D videoclips nor 3D volumes reached the high diagnostic level of evaluation at real-time imaging in the hands of an experienced investigator using GIS. Addition of vascular indices and volume analysis did not improve the efficiency at 3D analysis. The parameters included in the REC-score system had the highest diagnostic efficiency at both 2Dand 3D-GIS evaluation 14. A problem with the offline evaluation of 3D volumes and 2D videoclips seemed to be loss of image quality; the diagnostic efficiency was acceptable when restricted to the three-quarters of the images that had high image quality or evaluations in which there was agreement between 2D videoclip and 3D volume determinations. Competence in image optimization is important to obtain images of sufficient quality for later offline analysis. Fast-track investigation of all postmenopausal women with increased endometrial thickness is very costly. Improved selection at first-line TVS seems rational and was effective at real-time evaluation during TVS in experienced hands in this population, as it has been in other studies 6 10,12,17. The main obstacle to implementation Table 2 Endometrial thickness and volume and the vascular indices in 169 patients with postmenopausal bleeding and endometrial thickness 5 mm, according to type of endometrial pathology Pathology N Endometrial thickness (mm) Endometrial Vascular index volume (cm 3 ) VI VFI FI Endometrial cancer ( ) 20.5 ( ) ( ) 4.09 ( ) ( ) Hyperplasia ( ) 5.47 ( ) 4.85 ( ) 1.42 ( ) ( ) Benign uterine polyps ( ) 5.88 ( ) 5.85 ( ) 1.13 ( ) ( ) Other ( ) 5.62 ( ) 3.83 ( ) 1.02 ( ) ( ) Data are given as mean (95% CI). P < 0.01 for cancer groups vs all other groups, for endometrial thickness, endometrial volume, vascularization index (VI), vascularization flow index (VFI) and flow index (FI).

8 3D ultrasound and endometrial cancer 741 Table 3 Diagnostic efficiency of different parameters for prediction of endometrial cancer at three-dimensional (3D) transvaginal ultrasound (TVS), power Doppler angiography (PDA) and gel infusion sonography (GIS), in patients with postmenopausal bleeding and endometrial thickness 5mm Parameter AUC (95% CI) Correctly classified (%) Cut-off Sens. (%) Spec. (%) LR+ LR 3D-TVS (N = 169) ET ( ) EV ( ) Heterogeneous echogenicity ( ) Non-cystic echogenicity ( ) Irregular endomyometrial junction ( ) Interrupted endomyometrial junction ( ) Non-intact endomyometrial junction ( ) D-GIS (N = 160) ET ( ) Interrupted endomyometrial junction ( ) Non-intact endomyometrial junction ( ) Irregular endometrial surface ( ) D-PDA (N = 155) Presence of endometrial flow (a) ( ) No dominant vessel (b) ( ) Vascularity, but no single or double dominant vessel (c) ( ) Multifocal vessels (d) ( ) Multiple vessels (e) ( ) Large endometrial vessels (f) ( ) Areas with densely packed or color-splash vessels (g) ( ) Doppler score (c,e,f,g) ( ) Vascularization index ( ) Vascularization flow index ( ) Flow index ( ) Multivariate logistic regression and REC score Model 1: Age, BMI, ET ( ) Model 2: Age, BMI, ET, interrupted endomyometrial ( ) junction Model 3: BMI, ET, interrupted endomyometrial junction, ( ) Doppler score Model 4: BMI, interrupted endomyometrial junction, ( ) Doppler score, irregular endometrial surface at 3D-GIS REC score 3D-PDA (BMI 30, ET 10mm, ET 15mm, ( ) interrupted endomyometrial junction, Doppler score) REC score 3D-GIS (BMI 30, ET 10 mm, ET 15 mm, interrupted endomyometrial junction, Doppler score, irregular surface at 3D-GIS) ( ) Doppler score (c,e,f,g): vascularity, but no single/double dominant vessels (+1) + multiple vessels (+1) + large vessels (+1) + densely packed or color-splash vessels (+1). REC-score system is presented in Figure 1. Comparison of areas under the receiver operating characteristics curves (AUC): Model 1 vs Model 3 vs Model 4 (n = 145, P = 0.02); Model 1 vs REC score 3D-TVS vs REC score 3D-GIS (n = 145, P = 0.01). BMI, body mass index; ET, endometrial thickness; EV, endometrial volume; LR+, positive likelihood ratio; LR, negative likelihood ratio; REC score, risk of endometrial cancer score; Sens., sensitivity; Spec., specificity. of SEP at the first TVS visit is the required presence of an experienced investigator, which may not be a realistic possibility in general practice at the present time. Thus, an approach in which investigators perform real-time evaluations and acquire 3D volumes and/or 2D videoclips may allow for secondary evaluation by experienced investigators at image conferences. Thereby, a limited group of indefinable evaluations may be identified that require additional imaging, such as those with discrepancies between 2D and 3D evaluations or images of poor quality. At first-line ultrasound, this would split patients into four groups: 1) those with a thin endometrium, in whom endometrial cancer is unlikely; 2) a second, high-risk group, in whom cancer is very likely; these patients should undergo fast-track evaluation; 3) a third, intermediate group in whom benign pathology is likely; and 4) a fourth group in whom additional imaging is needed. It is much simpler and cheaper to evaluate images remotely than to send the patient to a specialized unit. This approach is already used for magnetic resonance imaging; for ultrasound, such a system could be time- and cost-efficient, when patients are grouped as suggested. Prior to this study, association between endometrial cancer and multiple vessel pattern 10 and irregular

9 742 Dueholm et al. branching 8 has been described, but how best to correlate different vascular pattern findings to malignancy remained uncertain. In our study, at SEP of 2D- and 3D-PDA, only the presence-of-multiple-vessels pattern was interpreted as malignancy. Thus, as supported by others 12,TVSbasedon SEP supplemented with 2D- or 3D-PDA did not improve the diagnostic efficiency for malignancy. The diagnostic efficiency of the different endometrial patterns at 3D imaging was comparable to that reported in other studies 6 8,10. Vascular index parameters were not more efficient than was the Doppler score in this study. Other authors have not compared the vascular indices to a vascular pattern score system. The vascular indices had high diagnostic efficiency in some studies 9,11,18,19, but not in others 12, and, while the indices had the advantage of low observer variation 9,18,19, there are shortcomings: their measurement is time-consuming, there is a crucial dependence on image settings and a lack of reproducibility between different categories of ultrasound machines 20,21. However, we found that combining the different vascular patterns into a total vascular pattern score (Doppler score) was more efficient at 3D-PDA than was measurement of the vascular indices. The REC-score system had the highest diagnostic efficiency at 3D evaluation. This system improved the diagnostic efficiency by implementing the Doppler score and adding BMI, endometrial thickness, interrupted endomyometrial junction and irregular surface at GIS. In patients with increased endometrial thickness and limited vascularity, the surface at GIS is an important parameter in the REC-score system. This system also had the highest efficiency in a prior 2D analysis 14. A clear strength of the present study was the use of a reliable reference standard in all patients. Shortcomings are combination in the study group of unselected patients with selected patients referred from other hospitals. Another possible weakness could have been recall bias. We attempted to eliminate this by having gaps of several months between the four rounds of ultrasound analyses, analyzing patients in a different order in each round and blinding observers to the identity of patients. The investigator was also blinded to the pathological diagnosis during all evaluations of 2D videoclips and 3D volumes. This design excluded continuous learning during evaluations, which might have increased the diagnostic performance. The quality of videoclips and 3D volumes was clearly a problem. Videoclips were acquired as sweeps, and several 3D volumes had low quality, especially at GIS. We used gel instead of saline infusion to eliminate motion artifacts due to the constant spill of saline. Although we have several years of experience with SCSH, we experienced a gradual improvement in the quality of the 3D-GIS examinations; for example, the catheter was removed before 3D volume acquisition during the later examinations, which reduced the acoustic noise due to the catheter shadow and hence improved the image quality. Vascular findings are dependent on tumor characteristics 22,23 and might also depend on the sample of cancer cases; they therefore must be evaluated in larger studies. We did not evaluate color scores, which could have improved the evaluation 9. Moreover, it might have been advantageous to add Doppler analysis to GIS and to evaluate 3D volumes and 2D videoclips concomitantly rather than in two separate stages; this approach should be evaluated in future studies. In conclusion, SEP offline, from stored 2D videoclips and 3D volumes at GIS, does not possess the same diagnostic efficiency as does real-time evaluation at 2D-GIS, but such offline analysis may be considered in those patients in whom the stored images and volumes are of sufficient quality. As for 2D ultrasound, the efficiency of 3D ultrasound may be improved by implementation of the REC-score system, but addition of vascular indices and calculation of EV apparently have no benefit. At first-line ultrasound examination, endometrial pattern evaluation would split patients into four groups: 1) one group with a thin endometrium in whom endometrial cancer is unlikely; 2) a second, high-risk group in whom cancer is very likely; these patients should undergo fast-track evaluation; 3) a third, intermediate group in whom benign pathology is likely; and 4) a fourth group in whom additional imaging is needed. The optimal image modality appears to be real-time 2D-GIS. ACKNOWLEDGMENT This study was supported with grants from the Danish Cancer Society. We gratefully acknowledge the help of Elisabeth Melin and the staff at the day surgery unit for conducting the blinding of observers, and Charlotte Møller for obtaining 2D videoclips and 3D volumes. REFERENCES 1. Clark TJ, Barton PM, Coomarasamy A, Gupta JK, Khan KS. Investigating postmenopausal bleeding for endometrial cancer: cost-effectiveness of initial diagnostic strategies. BJOG 2006; 113: Dijkhuizen FP, Mol BWJ, Brölmann HA, Heintz AP. Cost-effectiveness of the use of transvaginal sonography in the evaluation of postmenopausal bleeding. Maturitas 2003; 45: van HN, Breijer MC, Khan KS, Clark TJ, Burger MP, Mol BW, Timmermans A. Diagnostic evaluation of the endometrium in postmenopausal bleeding: an evidence-based approach. Maturitas 2011; 68: Timmermans A, Opmeer BC, Khan KS, Bachmann LM, Epstein E, Clark TJ, Gupta JK, Bakour SH, Van den Bosch T, van Doorn HC, Cameron ST, Giusa MG, Dessole S, Dijkhuizen FP, Ter RG. Endometrial thickness measurement for detecting endometrial cancer in women with postmenopausal bleeding: a systematic review and meta-analysis. Obstet Gynecol 2010; 116: Smith-Bindman R, Kerlikowske K, Feldstein VA, Subak L, Scheidler J, Segal M, Brand R, Grady D. Endovaginal ultrasound to exclude endometrial cancer and other endometrial abnormalities. JAMA 1998; 280: Epstein E, Valentin L. Gray-scale ultrasound morphology in the presence or absence of intrauterine fluid and vascularity as assessed by color Doppler for discrimination between benign and malignant endometrium in women with postmenopausal bleeding. Ultrasound Obstet Gynecol 2006; 28: Epstein E, Skoog L, Isberg PE, De Smet F, De Moor B, Olofsson PA, Gudmundsson S, Valentin L. An algorithm including results of gray-scale and power Doppler ultrasound examination to predict endometrial malignancy in women with postmenopausal bleeding. Ultrasound Obstet Gynecol 2002; 20: Opolskiene G, Sladkevicius P, Valentin L. Ultrasound assessment of endometrial morphology and vascularity to predict endometrial malignancy in women with postmenopausal bleeding and sonographic endometrial thickness > or = 4.5 mm. Ultrasound Obstet Gynecol 2007; 30: Opolskiene G, Sladkevicius P, Valentin L. Prediction of endometrial malignancy in women with postmenopausal bleeding and sonographic endometrial thickness 4.5 mm. Ultrasound Obstet Gynecol 2011; 37:

10 3D ultrasound and endometrial cancer Alcazar JL, Castillo G, Minguez JA, Galan MJ. Endometrial blood flow mapping using transvaginal power Doppler sonography in women with postmenopausal bleeding and thickened endometrium. Ultrasound Obstet Gynecol 2003; 21: Alcazar JL, Galvan R. Three-dimensional power Doppler ultrasound scanning for the prediction of endometrial cancer in women with postmenopausal bleeding and thickened endometrium. Am J Obstet Gynecol 2009; 200: Opolskiene G, Sladkevicius P, Jokubkiene L, Valentin L. Three-dimensional ultrasound imaging for discrimination between benign and malignant endometrium in women with postmenopausal bleeding and sonographic endometrial thickness of at least 4.5 mm. Ultrasound Obstet Gynecol 2010; 35: Alcazar JL, Iturra A, Sedda F, Auba M, Ajossa S, Guerriero S, Jurado M. Three-dimensional volume off-line analysis as compared to live ultrasound for assessing adnexal masses. Eur J Obstet Gynecol Reprod Biol 2012; 161: Dueholm M, Møller C, Rydbjerg S, Hansen ES, Ørtoft G. An ultrasound algorithm for identification of endometrial cancer. Ultrasound Obstet Gynecol 2014; 43: Leone FP, Timmerman D, Bourne T, Valentin L, Epstein E, Goldstein SR, Marret H, Parsons AK, Gull B, Istre O, Sepulveda W, Ferrazzi E, Van den Bosch T. Terms, definitions and measurements to describe the sonographic features of the endometrium and intrauterine lesions: a consensus opinion from the International Endometrial Tumor Analysis (IETA) group. Ultrasound Obstet Gynecol 2010; 35: Raine-Fenning NJ, Campbell BK, Clewes JS, Kendall NR, Johnson IR. The reliability of virtual organ computer-aided analysis (VOCAL) for the semiquantification of ovarian, endometrial and subendometrial perfusion. Ultrasound Obstet Gynecol 2003; 22: Alcazar JL, Galvan R. Three-dimensional power Doppler ultrasound scanning for the prediction of endometrial cancer in women with postmenopausal bleeding and thickened endometrium. Am J Obstet Gynecol 2009; 200: Alcazar JL, Ajossa S, Floris S, Bargellini R, Gerada M, Guerriero S. Reproducibility of endometrial vascular patterns in endometrial disease as assessed by transvaginal power Doppler sonography in women with postmenopausal bleeding. J Ultrasound Med 2006; 25: de Kroon CD, Hiemstra E, Trimbos JB, Jansen FW. Power Doppler area in the diagnosis of endometrial cancer. Int J Gynecol Cancer 2010; 20: Alcazar JL. Three-dimensional power Doppler derived vascular indices: what are we measuring and how are we doing it? Ultrasound Obstet Gynecol 2008; 32: Raine-Fenning NJ, Nordin NM, Ramnarine KV, Campbell BK, Clewes JS, Perkins A, Johnson IR. Evaluation of the effect of machine settings on quantitative three-dimensional power Doppler angiography: an in-vitro flow phantom experiment. Ultrasound Obstet Gynecol 2008; 32: Epstein E, Van HC, Mascilini F, Masback A, Kannisto P, Ameye L, Fischerova D, Zannoni G, Vellone V, Timmerman D, Testa AC. Gray-scale and color Doppler ultrasound characteristics of endometrial cancer in relation to stage, grade and tumor size. Ultrasound Obstet Gynecol 2011; 38: Saarelainen SK, Vuento MH, Kirkinen P, Maenpaa JU. Preoperative assessment of endometrial carcinoma by three-dimensional power Doppler angiography. Ultrasound Obstet Gynecol 2012; 39: