Deep learning versus manual morphology-based embryo selection in IVF: a randomized, double-blind noninferiority trial

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Medicine
• Clinical Trials

Background:

• In vitro fertilization (IVF) success depends on optimal embryo selection.
• Traditional morphological assessment is subjective.
• Deep learning offers a potential objective method for embryo evaluation.

Purpose of the Study:

• To evaluate the noninferiority of a deep learning algorithm (iDAScore) compared to standard morphological assessment for selecting embryos in IVF.
• To determine if iDAScore can achieve a clinical pregnancy rate comparable to conventional methods.

Main Methods:

• A multicenter, randomized, double-blind, noninferiority trial involving 1,066 women undergoing IVF.
• Participants were randomized to either standard morphological assessment or iDAScore-guided embryo selection.
• The primary endpoint was the clinical pregnancy rate, with a noninferiority margin of 5%.

Main Results:

• The iDAScore group achieved a 46.5% clinical pregnancy rate.
• The morphology group achieved a 48.2% clinical pregnancy rate.
• The study did not demonstrate noninferiority for iDAScore (risk difference -1.7%, 95% CI -7.7, 4.3).

Conclusions:

• Deep learning (iDAScore) was not found to be noninferior to standard morphological assessment for IVF embryo selection.
• Further research may be needed to optimize deep learning applications in reproductive medicine.
• Current deep learning tools do not surpass established methods for predicting IVF success.