Utilizing Artificial Intelligence in Cine Magnetic Resonance Imaging Analysis: A Promising Approach for Assessment of Uterine Factors and Prediction of Pregnancy Outcomes in Patients With Recurrent Implantation Failure.

Abstract

Purpose: Recurrent implantation failure (RIF) is a form of refractory infertility that persists despite assisted reproductive technology. Cine magnetic resonance imaging (cine MRI) enables the visualization of uterine peristalsis; however, its use in RIF assessment is limited due to the lack of a standardized application method. This study aimed to develop pregnancy prediction models for patients with RIF and to evaluate the utility of cine MRI image analysis using artificial intelligence (AI).

Methods: We retrospectively analyzed the anonymized clinical data and cine MRI images of 188 patients with RIF and known pregnancy outcomes. Two types of models, based on clinical data only or both clinical data and cine MRI images, were built using the Random Forest model. The best model was identified using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity.

Results: Higher performance was achieved using the Random Forest model integrating clinical data and cine MRI images (AUC/accuracy/sensitivity/specificity: 0.835, 0.754, 0.879, and 0.583, respectively), outperforming models using clinical data only (0.617, 0.596, 0.697, and 0.458, respectively).

Conclusions: AI analysis of clinical data combined with cine MRI data improved pregnancy prediction, suggesting that cine MRI can be used to evaluate uterine factor-related RIF.