A data mining approach for predicting the pregnancy rate in human assisted reproduction

One of the most relevant aspects in human assisted reproduction is to decide if, at a given moment, the endometrium demonstrates maximal receptivity for embryo implantation, in order to increase both patients’ convenience and the cost-effectiveness of the assisted reproduction cycle. To help human experts in taking this decision, we developed an artificial intelligence system based on a data mining approach where data extracted from the endometrium/subendomentrium and their vascularization are evaluated. The proposed system has been tested on a dataset of 62 cycles of intracytoplasmic sperm injection (ICSI) and several machine learning methods are compared for obtaining a high performing system. Particularly interesting is the performance obtained considering only three features: the patient’s age, the subendometrial volume and the endometrial vascularization/flow index; the best system, based on a random subspace of decision tree, obtains an area under the ROC curve (AUC) of 0.85 in predicting the pregnancy rate. These preliminary results show that it is possible to measure in a non invasive way a set of features from a patient, for assisting the decision of making or postponing the embryo transfer.