A human blastocyst dataset including clinical annotations to benchmark deep learning architectures for in vitro fertilization

The dataset contains static morphology images of human blastocyts on day 5 post insemination, additional annotations, an example prediction of an XCeption model trained on the silver-standard data and a script to evaluate automated predictions. Additional annotations include the expert-annotated scores of the morphological grading system according to Gardner (cell expanstion, quality of inner cell mass and trophectoderm), used to rate blastocyst quality by assessing these morphological parameters as basis for the selection of blastocysts for transfer, and clinical outcomes such as fetal heart beat or healthy live birth. To allow researchers to benchmark novel machine learning-based algorithms trained to predict the Gardner scores, each image is assigned to one of two splits: a training- or a test set split. The training split was annotated by an experienced embryologist with long-time experience on how to apply the Gardner criteria to static morphology blastocyst images. The test set split contains annotations of the three Gardner criteria based on the agreement of an international consortium of experienced clinical embryologists. In summary, the dataset contains blastocyst images, a python script to evaluate model predictions, and four CSV-files including: i. the assignment of images to a training set including silver-standard annotations of all three Gardner criteria, ii.the assignment of images to the Gardner criteria test set including gold-standard annotations of all three Gardner criteria, iii. the assignment of images to the clinical dataset and thus, the images of blastocysts that had been transferred including their clinical annotations and iv. an example prediction by an XCeption-based trained deep learning model.