deepBlastoid: un modelo de aprendizaje profundo para la evaluación automatizada y eficiente de blastoides humanos

Recent advances in human blastoids have opened new avenues for modeling early human development and implantation. Human blastoids can be generated in large numbers, making them well-suited for high-throughput screening. However, automated methods for evaluating and characterizing blastoid morphology are lacking. We developed a deep-learning model-deepBlastoid-for automated classification of live human blastoids using only brightfield images. The model processes 273.6 images per second with an average accuracy of 87%, which is further improved to 97% by integrating a Confidence Rate metric. deepBlastoid outperformed human experts in throughput while matching accuracy in blastoid classification. We demonstrated the utility of the model in two use cases: (i) systematic assessment of the effect of lysophosphatidic acid (LPA) on blastoid formation and (ii) evaluating the impact of dimethyl sulfoxide (DMSO) on blastoid formation. The evaluation results of deepBlastoid using over 10,000 images were consistent with the known drug effects and showed subtle but significant effects that might have been overlooked in manual assessments. The publicly available deepBlastoid model enables researchers to train customized models based on their imaging and protocols, providing an efficient, automated tool for blastoid classification with broad applications in research, drug screening, and in-vitro-fertilization applications.