Advanced KPI framework for IVF pregnancy prediction models in IVF protocols.

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Biostatistics

Background:

• Assisted reproductive technology (ART) generates complex data.
• Neural networks offer advanced data processing capabilities for ART.
• Predicting clinical pregnancy in in vitro fertilization (IVF) remains a challenge.

Purpose of the Study:

• To develop a novel deep neural network model for predicting clinical pregnancy likelihood in IVF.
• To integrate key performance indicators and clinical data for enhanced prediction accuracy.
• To provide a tool for retrospective outcome analysis and prospective treatment evaluation.

Main Methods:

• Retrospective analysis of 11 years and 8732 IVF treatment cycles.
• Feature extraction and model training using deep neural networks, specifically recurrent neural networks.
• Internal validation on 1600 preimplantation genetic testing for aneuploidy embryo transfers and external validation across two independent clinics (>10,000 cases).

Main Results:

• The recurrent neural network model achieved high accuracy in predicting clinical pregnancy likelihood (AUC: 0.68-0.86; test accuracy: 0.78; F1 score: 0.71).
• Model performance demonstrated strong sensitivity (0.62) and specificity (0.86).
• The approach is comparable to time-lapse systems but utilizes a simpler methodology.

Conclusions:

• Deep neural networks provide a powerful and simpler approach to analyzing IVF data.
• The developed model accurately predicts clinical pregnancy chances, aiding in treatment cycle evaluation.
• This method presents a promising tool for quality management programs in fertility centers.