MEF-YOLO A deep lightweight model for solving the fine-grained of organoid

Abstract Organoids are miniature simplified in vitro model systems that simulate organ structure and function. Despite their utility, challenges remain in addressing organoid assembly and related data analysis. In the context of integrating organoid technology with deep learning, this study proposes a lightweight deep algorithm to efficiently handle high-throughput, multimodal, and fine-grained organoid images, with a primary focus on intestinal organoid images.The study employs the lightweight YOLOv10n as the baseline model for organoid image analysis, introduces a novel organoid image information fusion architecture, and completes the theoretical and engineering design of a specialized algorithm framework for organoid images. Through rigorous experiments—including model comparison analyses, organoid receptive field visualization, and organoid feature attention distribution studies—and performance comparisons with classical models, this work demonstrates how deep learning overcomes the fine-grained analysis challenge in organoid images. Notably, the approach reduces model complexity while enhancing computational efficiency and inference speed for organoid images. This study achieves state-of-the-art organoid recognition performance with minimized computational overhead, offering a new pattern recognition methodology for organoid morphological evaluation. In conclusion, this research presents an innovative technical tool that integrates superior computational performance with real-time multi-dimensional scientific prediction of organoid morphology.