Preoperative prediction of high-risk endometrial cancer by expert and non-expert transvaginal ultrasonography, magnetic resonance imaging, and endometrial histology

Abstract

ObjectiveTo identify women with high-risk endometrial cancers using expert and non-expert transvaginal ultrasonography (TVS) and MRI. Study designMyometrial involvement was prospectively evaluated in patients with atypical hyperplasia or endometrial cancer on ultrasound by non-experts at first visit (non-expert-TVS: n = 266) and experts (expert-TVS: n = 188) at second visit. MRI (n = 175) was performed when high-risk cancer was suspected on non-expert-TVS. Preoperatively, high-risk cancer was defined as myometrial involvement ≥50 %, or preoperative unfavorable tumor histology (grade 3 endometrioid, non-endometrioid tumors, or tumor in cervical biopsies) obtained by endometrial sampling or hysteroscopic biopsies. Preoperative evaluations were compared with final histopathology obtained at surgery, high-risk cancer being defined as unfavorable tumor histology or patients with FIGO stage ≥1b. ResultsPreoperative unfavorable tumor histology was seen in 64 women and correctly identified 63 of 128 high-risk cancers. Preoperative diagnosis of unfavorable tumor histology or myometrial involvement ≥50 %, i.e. judged high-risk, had an area under the curve (AUC), sensitivity, and specificity of 79.5 %, 93.8 %, 65.2 % on non-expert-TVS; 85.5 %, 84.4 %, 86.5 % on expert-TVS, and 85.4 %, 89.6 %, 81.2 % on MRI. AUC values were not significantly different between MRI and expert-TVS, but lower on non-expert-TVS (p < 0.02). However, sensitivity was highest on non-expert-TVS, where a low cutpoint for myometrial involvement was used (included potentially deep and difficult evaluations) in contrast to an exact cutpoint of myometrial involvement ≥50 % used on expert-TVS and MRI.The highest AUC, 88.6 %, was seen when MRI was performed in patients with myometrial involvement ≥50 %, determined on non-expert TVS. Sensitivity was reduced to 85.9 %, while specificity increased to 91.3 %. Thus, MRI was needed for risk classification in only 104 (39 %) patients. ConclusionDiagnostically, expert-TVS and MRI were comparable and superior to non-expert-TVS. However, non-expert-TVS classified all patients with unclear myometrial involvement ≥50 %, and thereby only misdiagnosed 6.2 % of high-risk cases. Non-expert-TVS combined with MRI when myometrial involvement was ≥50 % on non-expert-TVS was a simple and effective method comparable with expert imaging to identify low- and high-risk cancer and select patients for SLND. Addition of MRI to the diagnostic regimen was needed in only 39 % of our patients.