deepBlastoid: a deep learning model for automated and efficient evaluation of human blastoids

Area of Science:

• Developmental Biology
• Biotechnology
• Artificial Intelligence

Background:

• Human blastoids are valuable models for early human development and implantation.
• Current blastoid characterization relies on manual methods, limiting throughput.
• Automated tools are needed for efficient and accurate blastoid morphology evaluation.

Purpose of the Study:

• To develop a deep-learning model for automated classification of human blastoids.
• To assess the model's accuracy, throughput, and utility in drug screening applications.

Main Methods:

• Development of a deep-learning model (deepBlastoid) using brightfield images.
• Classification of live human blastoids with high accuracy and processing speed.
• Integration of a Confidence Rate metric to improve classification accuracy.

Main Results:

• deepBlastoid achieved 87% accuracy, improving to 97% with the Confidence Rate metric.
• The model processed images at 273.6 images per second, outperforming human experts in throughput.
• Demonstrated utility in assessing drug effects (LPA, DMSO) on blastoid formation, identifying subtle impacts.

Conclusions:

• deepBlastoid provides an efficient, automated tool for human blastoid classification.
• The model has broad applications in developmental biology research, drug screening, and in-vitro fertilization.
• Public availability allows customization for diverse research needs.