AI Model Predicts Embryo Development for IVF Using Limited Time-Lapse Images
An innovative AI system seeks to enhance embryo selection in in vitro fertilization by forecasting blastocyst development from a limited number of daily time-lapse images. This hybrid approach integrates DINOv2, a transformer-based vision model, with an upgraded long short-term memory network that includes multi-head attention. DINOv2 captures significant features from embryo images, which are subsequently analyzed over time by the LSTM to produce predictions. This solution tackles the issue of manually reviewing extensive time-lapse data and addresses the frequent absence of comprehensive video systems in various clinics. The model was evaluated using a dataset of 704 embryos. The research, which highlights the challenges of manual embryo selection in IVF, was published on arXiv with the identifier 2604.16505v1.
Key facts
- AI model predicts blastocyst formation in IVF embryos
- Uses limited daily images from time-lapse recordings
- Combines DINOv2 vision model with enhanced LSTM network
- LSTM includes multi-head attention layer
- Tested on real dataset of 704 embryos
- Addresses challenge of manual embryo inspection
- Many clinics lack complete time-lapse systems
- Published on arXiv as 2604.16505v1
Entities
—