Answers are grounded in this paper and cite their sources.
Systematic review highlights dataset variability affecting AI performance in blastocyst evaluation, suggesting standardization is crucial.
This systematic review scrutinised the datasets used to train Artificial Intelligence (AI) tools developed for automated blastocyst assessment. It is widely acknowledged that the quality of these datasets significantly influences the performance of the AI models. Analysis of datasets from 26 studies highlighted considerable variations in various dataset parameters such as dataset size, data diversity, image quality, image capture mechanism and timing, class distribution, dataset endpoints and metadata usage. Some models incorporate morphokinetic or morphometric annotations and clinical metadata, whereas others rely solely on single-point static images. Many studies lack crucial information, such as image capture timing, embryo transfer strategy, and vital information related to the removal of confounding factors, such as uterine factors, hindering cross-study comparisons. Standardisation of datasets is vital for accurate assessment and comparison of commercially available AI models used for blastocyst assessment. The absence of standardised parameters and the lack of removal of confounding variables emphasise the need for greater transparency and standardisation in dataset creation and reporting. Future research should prioritise constructing a robust gold-standard large dataset that includes diverse imaging data and excludes confounding factors. In the absence of such a dataset, comparison of the AI model becomes very subjective.
No takes yet. Share an insight, caveat, or question.
Pande et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: