The development of two models for evaluating static images of denuded MII oocytes is described. The first model segments the oocyte into ooplasm, ZP, and PVS, while the second model classifies the oocyte based on features extracted from the masks to predict blastocyst development. The models were developed using images from fertility clinics and an open-source dataset. The segmentation model achieved high accuracy using a Fully Convolutional Branch TransFormer architecture. The mask model, incorporating patient characteristics, achieved accurate predictions of blastocyst development. An ensemble model combining a LightGBM model and a ConvFormer DL model was also developed. Subgroup analyses by clinic and age group were performed to assess the models’ clinical relevance. External validation on a dataset from a Spanish clinic showed promising results. The models were developed following ethical guidelines and regulations, with approvals obtained from relevant committees. Overall, the models show potential for improving the evaluation of oocyte quality and predicting blastocyst development in assisted reproduction procedures.
Source link
Source link: https://www.nature.com/articles/s41598-024-60901-1
GIPHY App Key not set. Please check settings