基于深度神经网络的芯片上活体虫黄藻检测

为了对活体虫黄藻进行快速、无标签和低成本的检测，笔者结合微流控技术、显微图像处理和深度学习神经网络，提出了基于深度神经网络的活体虫黄藻检测方法，对混合了微球、正常与漂白虫黄藻细胞溶液进行检测，结果表明:使用明场显微图像进行训练的神经网络模型泛化至微流控芯片上细胞检测当中，且能以93.9%的平均识别精确度识别不同生理状态下的虫黄藻细胞与其他目标，说明该方法可从大量复杂、异质的细胞群体中快速准确地识别出目标细胞。

A deep neural network-based method for on-chip detection of in vivo Symbiodinium

Measuring the species and density of Symbiodinium is important for predicting coral bleaching. At present, the rapid detection method of microalgae cells has great limitations. Manual microscopic inspection is time-consuming and laborious, while benchtop automated instruments are not suitable for large-scale real-time detection during field sampling. This research combines microfluidics, microscopic image processing and a deep learning neural network to proposes a deep neural network-based method for the microfluidic detection of in vivo Symbiodinium. The results of the detection of solution samples mixed with polymethylmethacrylate microspheres, normal and bleached methanogens cells show the neural network model trained using bright-field microscopic images is generalized to microfluidic on-chip cell detection, and is able to identify different physiological states of Symbiodinium cells and other objects with an average precision of 93.9%, demonstrating the feasibility, accuracy and sensitivity of the method.