Combining deep learning and microfluidics for fast and noninvasive sorting of zebrafish embryo

Abstract

Sorting of zebrafish embryos remains a challenging task in biomedical research. There is a need for accurate and efficient methods to distinguish between embryos at Stage 1, the zygote period immediately after fertilization, characterized by the single-cell stage, as well as those at advanced developmental stages (above single-cell stage) and non-viable (Dead) embryos. Manual sorting is labor-intensive, error-prone, and time-consuming. Traditional automated techniques, such as fluorescence-activated cell sorting (FACS) and robotic systems, are often invasive or prohibitively expensive, limiting their accessibility and scalability for routine zebrafish embryo sorting. This paper presents a novel approach that integrates deep learning with microfluidic technology to address these limitations. Our system utilizes a YOLOv8-based deep learning model for real-time embryo classification, whereas a microfluidic chip which is equipped with peristaltic pumps, ensures precise sorting with minimal manual intervention. Computational Fluid Dynamics (CFD) simulations are performed to optimise the flow parameters, and experimental validation demonstrate the system's high accuracy and sorting efficiency. The YOLOv8 model demonstrate a detection accuracy of 97.6 % and a processing speed of 10.5 ms. The sorting experiments demonstrate the system's efficacy, with the Stage 1 class achieving a detection accuracy of 90.63 % and a sorting efficiency of 88.13 %. The Advanced class exhibited enhanced performance, with a detection accuracy of 93.36 % and a sorting efficiency of 91.80 %. The Dead class demonstrate the highest performance, with a detection accuracy of 99.03 % and a sorting efficiency of 96.60 %. The system demonstrate an average sorting rate of 2.92 s per embryo. This approach provides a reliable, cost-effective alternative to traditional methods, significantly improving the speed and precision of embryo sorting.