Multimodal intelligent prediction model for in vitro fertilization.

Area of Science:

• Reproductive Medicine
• Artificial Intelligence
• Embryology

Background:

• In vitro fertilization (IVF) utilizes embryo selection and outcome prediction, but multimodal AI approaches are underutilized.
• Predicting specific outcomes like singleton vs. multiple pregnancy and miscarriage vs. live birth remains challenging.

Purpose of the Study:

• To introduce VaTEP (Video and Table model for Embryo Prediction), a multimodal AI model for enhanced embryo selection and IVF outcome prediction.
• To integrate time-lapse system (TLS) videos and tabular clinical data for comprehensive embryo assessment.

Main Methods:

• VaTEP employed pre-training on TLS video reconstruction and embryo developmental phase prediction.
• A multiple frame sampling strategy and multitask learning framework were utilized for model improvement.
• The model estimates probabilities for fetal heartbeat, singleton/multiple pregnancy, and miscarriage/live birth.

Main Results:

• The multimodal approach captures spatiotemporal dynamics and developmental information from TLS videos.
• VaTEP provides accurate predictions for key IVF outcomes, aiding clinical decision-making.
• The model facilitates personalized IVF strategies to minimize risks like implantation failure and multiple gestations.

Conclusions:

• VaTEP offers a data-driven tool for personalized IVF decision-making.
• The model supports safer and more effective reproductive treatments by improving embryo selection and outcome prediction.
• This multimodal AI approach advances the field of assisted reproductive technologies.