Expert-Annotated Embryo Image Dataset with Natural Language Descriptions for evidence-based patient communication in IVF

Le résumé fourni par la source

Embryo selection is one of multiple crucial steps in in-vitro-fertilization, commonly based on morphological assessment by clinical embryologists. Although artificial intelligence methods have demonstrated their potential to support embryo selection by automated embryo ranking or grading methods, the overall impact of AI-based solutions is still limited. This is mainly due to the required adaptation of automated solutions to custom, reliance on timelapse incubators and a lack of interpretability to understand AI reasoning. Moreover, the modern, informed patient is questioning expert decisions, particularly if the treatment is not successful. Thus, evidence-based decision justification in tasks like embryo selection would support transparent decision making and respectful patient communication, for the benefit of both, patients and experts. To support this aim, we hereby present an expert-annotated dataset consisting of embryo images previously published and corresponding morphological description using natural language. The description contains relevant information on developmental stages and morphological features. This dataset enables the finetuning of modern foundational vision-language models to learn to describe embryo morphology using natural language. Predicted embryo descriptions can then be leveraged to automatically extract scientific evidence from literature through advanced solutions such as retrieval-augmented generation systems or AI agent-based access, facilitating well-informed, evidence-based decision-making. This, in turn, enables transparent and justified communication with patients, ultimately enhancing the decision-making process. Moreover, language-grounded vision models are known to be a solid basis for further data-efficient, downstream task training. Our proposed dataset of natural-language human embryo image annotations is thus designed to support research in language-based, interpretable, and transparent automated embryo assessment.This dataset contains original data generated by the authors, as well as a subset of images derived from an external dataset (reused images originate from [Gomez et al., 2022]).Gomez Tristan, Feyeux Magalie, Boulant Justine, Normand Nicolas, Paul-Gilloteaux Perrine, David Laurent, Fréour Thomas, & Mouchère Harold. (2022). Human embryo time-lapse video dataset [Data set]. Zenodo. https://doi.org/10.5281/zenodo.6390798

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.