Quantitative, microscopical, computer-aided diagnosis of endometrial hyperplasia or carcinoma in individual patients.

Abstract

There are many significant differences, but also considerable overlap between the quantitative histopathological features of mild and marked atypical endometrial hyperplasias and well and moderately differentiated carcinomas, thus preventing its application to individual patient care. To try to overcome this problem, a classification rule for the diagnosis in individual patients, using discriminant analysis has been developed. Utilizing nine quantitative features, all the above four groups can be adequately separated. None of the carcinomas was misclassified as hyperplasia, and only one case of marked atypical hyperplasia was erroneously classified as well differentiated carcinoma, but with a probability of carcinoma 0.75, hyperplasia 0.25. By contrast, the classification probabilities of all the confirmed carcinomas exceeded 0.90. Therefore, using 0.90 as a classification level ('threshold'), a reliable rule is obtained. A slightly more simple classification rule distinguishes between all the hyperplasias and all the carcinomas. In this way, all the cases of the test set were correctly classified. The classification rules can be used to select patients with benign disease for hormone therapy (Kistner 1973) as an alternative to hysterectomy, and can be programmed in an inexpensive microcomputer, The quantitative techniques are relatively easy, and are capable of being performed in most histopathological laboratories.