Digital image analysis of gland-to-stroma ratio by CD10 objectively differentiates between low-grade endometrioid carcinoma and atypical hyperplasia on endometrial biopsy.

Abstract

Distinguishing between endometrial atypical hyperplasia/endometrial intraepithelial neoplasia (EAH/EIN) and grade 1 endometrial endometrioid carcinoma (EEC) requires the evaluation of gland-to-stromal ratio, presence of stromal invasion, extent of epithelial proliferation and nuclear alterations. In small biopsies, stromal invasion may not always be sampled, so other features become more important. However, assessing of some of these features may be subjective. Digital analysis improves diagnostic uniformity, and when combined with CD10 immunostain, it can potentially become a useful objective parameter. Endometrial biopsies with a diagnosis of EAH/EIN or EEC matched were retrieved with subsequent hysterectomy for reference diagnosis. CD10 immunohistochemistry was applied to the biopsies, followed by scanning and annotation. Pixel-accurate stromal percentages were deduced from digitised whole-slide images from a test cohort. An optimal stromal percentage cut-off, thus gland-to-stroma ratio, was extrapolated from the receiver operating characteristic curve. A separate cohort was used to validate the diagnostic performance of the determined gland-to-stroma cut-off. Seventy endometrial biopsies were included in the test cohort, comprising 48 grade 1 EECs and 22 EAH/EINs. The mean stromal percentage was 15.69% for EEC and 33.65% for EAH/EIN (all endometrial tissue annotated/analysed) and 14.77% for EEC and 31.88% for EAH/EIN (only lesional tissue annotated/analysed). The corresponding gland-to-stroma ratio was 5:1 for EEC and 2:1 for EAH/EIN. The areas under curve were 0.758 (p=0.001) (all endometrial tissue) and 0.761 (p=0.001) (only lesional tissue), (p=0.001). In the validation cohort, a cut-off of 30% CD10-stained stroma (7:3 gland-to-stroma ratio) was superior in diagnostic performance than the H&E diagnosis (p=0.042). Evaluation of gland-to-stroma ratio using CD10 immunostain and digital image analysis is a robust and objective method for distinguishing between grade 1 EEC and EAH/EIN in small biopsies. A cut-off >7:3 is highly indicative of EEC.