Deep learning system for classification of ploidy status using time-lapse videos

Area of Science:

• Reproductive medicine
• Artificial intelligence in healthcare
• Embryology

Background:

• Assessing embryo ploidy is crucial for successful in vitro fertilization (IVF).
• Current methods for diagnosing chromosomal status can be invasive and time-consuming.
• Automating embryo assessment could improve IVF outcomes.

Purpose of the Study:

• To develop a spatiotemporal artificial intelligence (AI) model for predicting euploid and aneuploid embryos.
• To utilize time-lapse imaging data for non-invasive embryo assessment.
• To automate the evaluation of embryo development for improved IVF success rates.

Main Methods:

• An end-to-end AI system was developed using a retrospective study design.
• Convolutional neural networks (CNNs) extracted spatial features from video frames.
• Bidirectional long short-term memory (BiLSTM) layers analyzed temporal dependencies for classification.

Main Results:

• The AI model achieved an accuracy of 73.08% in predicting embryo euploidy.
• A multi-input model with a gate recurrent unit (GRU) module showed superior performance.
• The model demonstrated a precision of 0.8205 for euploidy prediction.

Conclusions:

• The study presents a novel AI solution for prioritizing euploid embryo transfer.
• A non-invasive method for chromosomal status diagnosis using deep learning on time-lapse data was identified.
• The AI approach shows potential for automating embryo evaluation by encoding spatial and temporal information.