An ultrasound algorithm for identification of endometrial cancer

摘要

Objective To propose a scoring system to predict endometrial cancer using different ultrasound image characteristics at gray-scale, with and without enhancement by gel infusion, and Doppler transvaginal sonography (TVS) and to evaluate intra- and interobserver variability in assessment of these characteristics. Method Unenhanced TVS, Doppler examinations and gel infusion sonography (GIS) were performed prospectively in 174 consecutive postmenopausal women with endometrial thickness ≥ 5 mm. The reference standard in all women was hysteroscopy or hysterectomy with pathological evaluation of the malignancy. The presence of various ultrasound pattern characteristics indicative of endometrial malignancy and intra- and interobserver variability in their assessment were evaluated. Multivariate logistic regression was used to correlate image and clinical parameters to presence of endometrial cancer. Results A simple Doppler flow score (which considered only presence of vascularity and not presence of single/double dominant vessel, multiple vessels, large vessels, color splash or densely packed vessels) had an area under the receiver-operating characteristics curve (AUC) of 0.83 in the prediction of endometrial cancer. Models including endometrial thickness, Doppler score and interrupted endomyometrial junction on unenhanced TVS predicted endometrial cancer with an AUC of 0.95 (95% CI, 0.92-0.99) and, with addition of irregular surface on GIS, the AUC was 0.97 (95% CI, 0.94-0.99). A risk of endometrial cancer (REC) scoring system based on body mass index, Doppler score, endometrial thickness and interrupted endomyometrial junction on unenhanced TVS and irregular surface at GIS performed very well at identifying endometrial cancer; at a REC-score of ≥ 4 the sensitivity for detection of endometrial cancer was 91% and specificity was 94%. Observers agreed in 82.3% of cases (kappa, 0.63 (0.48-0.78)) when subjective parameters were analyzed in stored videoclips. Conclusion Our observer-dependent proposed scoring system seems to perform well in the prediction of endometrial cancer and should be tested in future studies.