A microfluidic microalgae detection system for cellular physiological response based on an object detection algorithm

The composition of species and the physiological status of microalgal cells serve as significant indicators for monitoring marine environments. Symbiotic with corals, Symbiodiniaceae are more sensitive to the environmental response. However, current methods for evaluating microalgae tend to be population-based indicators that cannot be focused on single-cell level, ignoring potentially heterogeneous cells as well as cell state transitions. In this study, we proposed a microalgal cell detection method based on computer vision and microfluidics, which combined microscopic image processing, microfluidic chip and convolutional neural network to achieve label-free, sheathless, automated and high-throughput microalgae identification and cell state assessment. By optimizing the data import, training process and model architecture, we solved the problem of identifying tiny objects at the micron scale, and the optimized model was able to perform the tasks of cell multi-classification and physiological state assessment with more than 95% mean average precision. We discovered a novel transition state and explored the thermal sensitivity of three clades of Symbiodiniaceae, and discovered the phenomenon of cellular heat shock at high temperatures. The evolution of the physiological state of Symbiodiniaceae cells is very important for directional cell evolution and early warning of coral ecosystem health. We present a label-free, multi-scale, sheath-less microfluidic microalgae detection system enabled with an improved deep learning algorithm, called MD-YOLO, for tiny cell classification and physiological status monitoring.