Dynamic contrast-enhanced MRI in endometrial carcinoma identifies patients at increased risk of recurrence

Eur Radiol DOI 10.1007/s00330-013-2901-3

UROGENITAL

Dynamic contrast-enhanced MRI in endometrial carcinoma identifies patients at increased risk of recurrence Ingfrid S. Haldorsen & Renate Grüner & Jenny A. Husby & Inger J. Magnussen & Henrica M. J. Werner & Øyvind O. Salvesen & Line Bjørge & Ingunn Stefansson & Lars A. Akslen & Jone Trovik & Torfinn Taxt & Helga B. Salvesen

Received: 11 February 2013 / Revised: 22 April 2013 / Accepted: 23 April 2013 # European Society of Radiology 2013

Abstract Objectives To study the feasibility of dynamic contrastenhanced magnetic resonance imaging (DCE-MRI) for assessment of tumour microvasculature in endometrial carcinoma patients, and to explore correlations with histological subtype, clinical course and microstructural characteristics based on apparent diffusion coefficient (ADC) values. Methods Diffusion-weighted imaging (DWI) and threedimensional DCE-MRI (1.5 T) with high temporal resolution (2.49 s) were acquired preoperatively in 55 patients. Quantitative modelling allowed the calculation of four independent parameters describing microvasculature: blood flow (Fb), extraction fraction (E), capillary transit time (Tc) and transfer constant from the extravascular extracellular space [EES] to blood (Kep); and four derived parameters: blood volume (Vb), volume of EES (Ve), capillary permeability surface area product (PS) and transfer from blood to EES (Ktrans).

Results Endometrial carcinoma tissue exhibited reduced Fb, E, Vb, Ve, PS and Ktrans compared with normal myometrium. Non-endometrioid carcinomas (n= 12) had lower Fb, and E than endometrioid carcinomas (n =43; P 7 years of experience with pelvic MRI) who was blinded to the tumour stage, histological diagnosis and patient outcome. Structural images were used in combination with images reflecting physiology for verification of anatomical structures and tumour location. All patients had a measurable endometrial mass on the structural MRI. Regions of interest (ROIs) were subsequently drawn manually on one dynamic image volume 2 min post-contrast (which is a time delay considered optimal for discrimination between tumour tissue and myometrial tissue) in each patient (Fig. 1) using the NordicICE software (v.2.3.12; NordicNeuroLab, Bergen, Norway). It was decided to draw the ROIs directly on DCEMRI image volumes to avoid registration artefacts between structural and dynamic images. The ROIs of the large tumours were drawn in some distance from the outer boundaries of the

Fig. 1 Axial contrast-enhanced T1-weighted image (a), T2-weighted image (b), DCE image (c), and ADC map (d) from DWI at the same level of the uterus in a postmenopausal 53-year-old woman with FIGO stage IB endometrioid, grade 1, endometrium carcinoma. The uterine tumour invading ≥50 % of the myometrial wall, is hypointense (arrows) relative to the surrounding normal myometrium on a contrast-enhanced T1weighted image (a), and hyperintense (arrows) relative to the

myometrium on a T2-weighted image (b). On the DCE series (c), three regions of interest (ROIs) were defined for each patient: ROI 1 within the tumour tissue, ROI 2 within presumably normal myometrial tissue, and ROI 3 in a pelvic artery. On the ADC map (d) ADC values were measured within tumour tissue (ROI 1) and within normal myometrium (ROI 2). In this patient the mean tumour ADC value was 0.67×10-3 mm2/s compared with 1.32×10-3 mm2/s for myometrial tissue

Physiological MRI

Eur Radiol

tumour in order to limit the impact of minimal temporal changes in uterine position (due to arterial pulsation and bowel peristalsis) which would otherwise lead to inclusion of normal myometrial tissue in the ROIs analysed. For comparison between tumour and reference tissue, the endometrium from which the tumour originates, was the preferred reference tissue. However, many patients had no visible normal endometrial tissue, and if the endometrium was visible, it was very thin. Thus, a very small ROI would make a quantitative assessment of parameters in the endometrium unreliable because of extensive partial volume effects. For these reasons it was decided to use normal myometrial tissue, which was clearly visible in all patients, as reference tissue. Three ROIs were drawn on the DCE-MRI series for each patient: ROI 1 was placed within the tumour avoiding necrotic or haemorrhagic areas, ROI 2 within the presumably normal myometrium (used as reference tissue), and ROI 3 in a pelvic artery (Fig. 1c). The volumes of ROI 1 and ROI 2 varied owing to the variable size of tumours and of non-neoplastic myometrial tissue. Median volume of the drawn ROIs was 0.25 ml (mean, 0.32 ml; range, 0.05–0.96 ml) in tumour tissue and 0.20 ml (mean, 0.23 ml; range, 0.05–0.66 ml) in myometrial tissue. The quality of the vessel signals in ROI 3 differed substantially between patients, most likely because of turbulence or flow artefacts. Thus, a population-based vessel signal, which was individually optimised in each patient with a blind source separation technique [16, 17], was used. Because of the high temporal resolution in the dynamic series (~2.5 s), a two-compartment model incorporating both capillary flow and permeability was chosen. An in-house implementation of the adiabatic approximation model of Johnson and Wilson (aaJW) [16, 18] was thus used to analyse the contrast concentration curves in the myometrium and tumour regions in each patient. The time courses in each ROI were independently interpolated using a cubic spline interpolation (Matlab v.7.7; Mathworks, Natick, MA, USA) to compensate for the time gap when the high-resolution T1weighted images were acquired. Multiple pre-contrast flip angle acquisitions were used to convert from MRI signal intensities to contrast agent concentrations in time [16]. No temporal filtering or other smoothing was applied to the data. This model allowed calculation of four independent parameters describing the microvasculature: blood flow (Fb), extraction fraction (E), intravascular/capillary transit time (Tc) and transfer constant from the extravascular extracellular space [EES] to blood (Kep); and four derived parameters: blood volume (Vb) [Vb=Fb×Tc], volume of EES (Ve) [Ve=(E×Fb)/Kep], capillary permeability surface area product (PS) [E=1 - exp(−PS/Fb[1 - rHematocrit])] and transfer from blood to EES (Ktrans) (Ktrans=Kep×Ve) [16]. The ADC values were measured in two different ROIs on the ADC map: ROI 1 within tumour tissue, and ROI 2 within presumably normal myometrial tissue (Fig. 1d).

Tumour volume was estimated based on the standard anatomical images with measurements of maximum tumour diameter in three orthogonal planes (x, y and z) using the following equation: tumour volume = x × y × z / 2. Surgical staging and clinical outcome All patients were surgically staged according to the revised FIGO staging system (2009) [2]. All patients were considered primary endometrial carcinomas and underwent primary surgical resection of the pelvic tumour. In cases with suspected cervical stromal invasion radical hysterectomy was the standard therapeutic approach unless found contraindicated by the responsible surgeon. None of the cases had unresectable locally advanced endometrial carcinomas. Surgical specimens were sectioned along the longitudinal plane of the uterus. Depth of myometrial invasion and presence of cervical stromal invasion were estimated grossly and confirmed microscopically according to standard procedures [19]. Routine histopathology reports were generated without knowledge of preoperative MRI findings, and used in the study. The pathologists documented the numbers and size of the metastatic lymph nodes. Follow-up data regarding recurrence, progression and survival have been collected from patient records and correspondence with the responsible primary physicians or gynaecologists. Routine clinical follow-up was scheduled every 3rd month during the 1st year, every 6th month the 2nd and 3rd years, and annually the 4th and 5th years after primary treatment. The date of last follow-up was September 2012 and the mean follow-up for survivors was 21 months (range 1–36). Statistical analysis The non-parametric Kolmogorov–Smirnov test for normality was used to verify the normality assumption for the continuous variables analysed. For comparison of DCEMRI and DWI parameters between endometrial carcinoma tissue and normal myometrium in the same patient, the Student’s paired t-test was used. For comparison between patient subgroups of DCE-MRI and DWI parameters within tumour tissue, the Student’s independent samples t-test, the Kruskal–Wallis test and the Jonckheere–Terpstra test for trend were used. The correlation between continuous variables (DCE-MRI and DWI values) was evaluated using Pearson’s bivariate correlation test. Quartile limits were applied to explore the prognostic value of categories for the DCE-MRI and DWI values. Groups with similar survival were merged. Univariate analyses of time to recurrence (for patients considered cured by primary treatment) or progression (for patients known to have residual disease after primary treatment) were performed using Kaplan–Meier. Differences in recurrence-/progression- free

Eur Radiol

survival were assessed by the Mantel–Cox (log rank) linear trend test. Univariate survival analysis and estimation of the Cox proportional hazards ratio was used to study the effect on the survival of continuous variables. The data were analysed using SPSS, 20.0 (Chicago, IL, USA). All reported P values were two-sided and considered to indicate statistical significance when less than 0.05.

5 % (3/55). Pelvic lymph node sampling was performed in 89 % (49/55) as part of the routine surgical staging procedure. Adjuvant therapy was given to 33 % (18/55); chemotherapy in 24 % (13/55) and external pelvic radiation in 9 % (5/55). Median tumour size assessed by structural MRI was 5.6 ml (mean, 31.4 ml; range, 0.1–611.0 ml). Microvascular and microstructural parameters

Results Patients Median and mean patient age in the study sample (n=55) were 69 and 67 years, respectively (range 41–93), and 89 % (49/55) of the patients were postmenopausal. Applying the FIGO 2009 staging criteria, 44 % (24/55) were stage IA (tumour invading