Embryo dataset

Research Hypothesis
Multimodal fusion of embryo blastocyst images with clinical and semen indicators captures complementary information on embryo quality and patients’ reproductive status, achieving more accurate IVF-ET pregnancy prediction than single-modal data. Identifying key predictive features supports personalized clinical decision-making for embryo selection, and an AI model-based platform optimizes IVF-ET protocols, reduces multiple pregnancy risks and improves ART success rates.

Data Collection & Overview
Collected from Gansu Provincial Maternity and Child-care Hospital (2016.01–2020.08), the dataset includes paired blastocyst images and clinical/semen records of 2542 IVF-ET patients with definitive pregnancy/non-pregnancy outcomes, approved by the ethics committee with informed consent from all participants. The dataset was split into a cross-validation set (2056 samples: 810 pregnant, 1246 non-pregnant) and an external validation set (486 samples: 155 pregnant, 331 non-pregnant), with a total of 965 pregnant and 1577 non-pregnant samples.

Data Interpretation & Usage
This dataset links embryo morphological features and clinical/semen indicators to IVF-ET pregnancy outcomes, with core value in the complementary effect of multimodal fusion and the clinical significance of the 36 key features.