Fig. 1: Overview of FEMI and downstream applications.
From: A foundational model for in vitro fertilization trained on 18 million time-lapse images
1 Input images are preprocessed by segmenting and resizing the embryo. The panel shows an example of a hard embryo time-lapse image. 2 The segmented images are used to train the masked autoencoder. 3 The encoder from the autoencoder is fine-tuned for clinically relevant tasks.
