Objective: To develop a karyometric image analysis approach to distinguishing atypical endometrial hyperplasia with and without co-occurring adenocarcinoma.
Study design: Six cases of atypical hyperplasia without and 6 cases with co-occurring adenocarcinoma, 4 cases of normal endometrium and 3 cases of adenocarcinoma were identified. From each case 100 nuclei were measured in representative diagnostic areas identified by an experienced pathologist. Discriminant analyses were performed. An unsupervised learning algorithm was applied to define and characterize different nuclear phenotypes, and those data were used to identify cases with co-occurring adenocarcinoma.
Results: Discriminant analysis showed that nuclei from atypical hyperplasia and atypical hyperplasia with co-occurring adenocarcinoma are statistically different. The unsupervised learning algorithm revealed differences in nuclear subpopulations that can be used to correctly identify an estimated 85% of individual cases.
Conclusion: Nuclei from atypical hyperplasia without and with co-occurring adenocarcinoma have statistically different karyometric characteristics that may facilitate case classification.