基于卷积神经网络的定量相衬显微技术(特邀)

Quantitative Phase Contrast Microscopy Based on Convolutional Neural Networks(Invited)

定量相衬显微可以在无荧光标记的前提下实现对透明样品的高衬度、定量化相位成像,对活细胞及其动态过程观测具有重要意义。然而,传统的定量相衬显微需要记录3幅相移图像才能获得样品定量的相位图像,耗时较长。提出一种基于双通道卷积神经网络的定量相衬显微相位重建方法。该方法可以利用2幅相移图像获得样品的定量相位图像,将传统定量相衬显微的成像速度提高了1.5倍,重建速度提高了1个数量级。实验中,利用COS7细胞的数据对网络进行训练,该网络可以成功实现对3T3细胞的定量相位成像,说明该网络具有一定的泛化能力。该方法有望为活细胞以及亚细胞器互作网络的动态观测提供有力手段。

Quantitative phase contrast microscopy facilitates high-contrast and quantitative phase imaging of transparent samples,eliminating the need for fluorescent labeling,making it pivotal for observing dynamic processes in living cells.Traditional methods,however,require three phase-shifted interferograms to generate a quantitative phase image,resulting in time-intensive procedures.This study introduces a novel phase reconstruction approach for quantitative phase contrast microscopy,leveraging a two-channel convolutional neural network.This innovative method achieves quantitative phase image retrieval from only two phase-shifted interferograms,enhancing imaging speed by 1.5 times and reconstruction speed by an order of magnitude compared with traditional approaches.In our experimental setup,the network was trained using COS7 cell data.The trained network successfully reconstructed quantitative phase images of 3T3 cells,demonstrating its applicability for accurate and robust phase reconstruction across different cell types.This method holds promise as a powerful tool for real-time,high-resolution observation of dynamic living cells and the interaction networks of sub-cellular organelles.