Optimizing clinical decisions in reproductive medicine with a hybrid AI predictive model

Infertility affects approximately ۱ in ۶ individuals globally, and while In Vitro Fertilization (IVF) is the most widely used Assisted Reproductive Technologies (ART), its success rate remains around ۳۰%. Accurate prediction of IVF outcomes is therefore critical for personalizing treatment and improving success rates. This study proposes a novel hybrid AI model that integrates deep learning Convolutional Neural Networks (CNNs) and Gated Recurrent Units (GRUs) with traditional machine learning K-Nearest Neighbors (KNN) and Multi-Layer Perceptron (MLP), optimized by LightGBM, to predict clinical pregnancy following IVF. Evaluated on a dataset including ۵۰۰ medical records of couples undergoing IVF treatment at Infertility Centers, the model achieved ۹۴.۶۷% accuracy and outperformed all baseline models across key metrics, including precision (۹۴.۱۲%), recall (۹۲.۱۲%), F۱-score (۹۲.۳۴%), and AUC-ROC (۰.۹۶). Key predictive features included female age, Anti-Müllerian Hormone (AMH) levels, and embryo quality, aligning with established clinical knowledge. These results demonstrate the hybrid model’s potential to serve as a robust, data-driven decision support tool for fertility specialists, enabling more accurate outcome prediction and personalized patient counseling. Future work will focus on external validation and transitioning to live birth prediction.

کلیدواژه ها:

Pregnancy ، Reproductive Medicine ، Machine Learning ، Deep Learning ، Hybrid Model