Incorporating artificial intelligence techniques in decision making concerning the optimal management of postmenopausal women with evidence of endometrial pathology

Introduction and Purpose: Artificial neural networks (ANNs) and Classification and Regression Trees (CARTs) have been previously used for the prediction of cancer in several fields. In our study, we aim to investigate the diagnostic accuracy of three different methodologies logistic regression, ANNs and CARTs for the prediction of endometrial cancer in post-menopausal women with post-menopausal vaginal bleeding or endometrial thickness u2265 5 mm, as determined by ultrasound examination. A classification tree is constructed from many repeated splits related to the target variable, these splits are producing rules of the form: If X is less than value A or Y is within a range [C,D] then the sample is classified as normal with probability p. These splits end when no further split can be done, either because all of their observations belong to the same group, or because the number of observations at the same node is small (according to a predefined value). These represent the terminal nodes of the tree. Artificial Neural Networks are complex computational models inspired by the human nervous system which are capable of machine learning and pattern recognition. They are constructed by interconnected neurons. Each neuron sums the weighted inputs (sum of the product of each input is multiplied with the corresponding weight w) and the summation result is passed from a non-linear non-convex and non-concave module (transfer function or activation function). The output of this function is the artificial neuron output as well.Methods: We conducted a retrospective case control study in the section of Gynecologic Oncology of the 3rd department of Obstetrics and Gynecology of our hospital. A consecutive series of post-menopausal women with vaginal bleeding or endometrial thickness u22655 mm were included in the present study. Among them several patients were diagnosed with endometrial carcinoma, whereas the remaining sample of women had no malignancy on the final pathology report. Women with endometrial hyperplasia with, or without atypia were excluded from the present study, as well as women that received hormone therapy, were treated for any type of cancer or for autoimmune diseases.Results: Overall, 178 women were enrolled. Among them, 106 women were diagnosed with carcinoma; whereas the remaining 72 women had normal histology in the final specimen. The test for proportions showed no statistically significant difference between the sensitivity of the CART (78.30%) and the LR (76.42%) model (p=0.7663); on the other hand sensitivity of the ANN model (86.79%) was higher than both the CART and the LR models (p