Time will tell: time-lapse technology and artificial intelligence to set time cut-offs indicating embryo incompetence

Area of Science:

• Reproductive Medicine and Biology
• Embryology and Assisted Reproductive Technology (ART)
• Artificial Intelligence in Healthcare

Background:

• Time-lapse technology (TLT) enables continuous embryo observation, but morphokinetic algorithms for predicting pregnancy have shown limited success.
• TLT holds potential for identifying developmentally incompetent embryos, preventing non-productive cycles and improving IVF efficiency.
• Discriminating embryos unable to reach blastocyst stage or those with aneuploidies can reduce patient stress and costs.

Purpose of the Study:

• To determine if combining time-lapse technology (TLT), artificial intelligence (AI), and preimplantation genetic testing for aneuploidy (PGT-A) can create more reliable time cut-offs for embryo developmental incompetence.
• To assess the potential of AI-driven time cut-offs to discriminate between euploid and aneuploid embryos, and those with developmental arrest.

Main Methods:

• A training dataset of 70% of embryos from PGT-A cycles (euploid, aneuploid, arrested) was used to define time cut-offs (tPNa-t8) and integrate maternal age ranges.
• Models were generated by fitting outcomes to timing and maternal age, with ROC curves identifying thresholds for predicting incompetence.
• Validation was performed on out-of-sample, internal, and external datasets using confusion matrices to test the accuracy of the defined time cut-offs.

Main Results:

• Specific time cut-offs for developmental incompetence were identified for different maternal age groups across various developmental stages (tPNa, tPNf, t2, t4, t8).
• Time to pronuclear formation (tPNf) and time to cleavage (t2) showed significant association with chromosomal competence, even after adjusting for maternal age.
• However, cut-offs based on tPNf and t2 performed less effectively and were found to be redundant compared to blastocyst development cut-offs.

Conclusions:

• The study suggests that embryo developmental incompetence can be better predicted by incorporating time cut-offs at multiple developmental stages, considering maternal age.
• This AI-driven approach, if validated, could significantly enhance ART efficiency by reducing unnecessary embryo transfers and cryopreservation.
• Further validation in larger sample sizes and diverse clinical settings is warranted, with potential application in static embryo observation.