论文标题
使用具有计算特异性（PIC）的相成像对神经网络的延时研究
Time-lapse Study of Neural Networks Using Phase Imaging with Computational Specificity (PICS)
论文作者
论文摘要
在生命科学中，荧光标记技术用于研究细胞的分子结构和相互作用。但是，这种类型的细胞成像具有其自身的局限性，其中之一是染色细胞的过程可能对细胞有毒并可能损坏它们。我们对活神经元的延时成像特别感兴趣，以研究其生长和增殖。神经退行性疾病的特征在于神经元生长和树皮化的表型差异。 本文提出了一种使用深卷积神经网络的无标签数字染色方法，以通过先前的细胞成像方法解决问题。我们的结果表明，当用正确的荧光标签对相位图像进行训练时，深层神经网络可以正确地学习必要的形态信息，以成功预测MAP2和TAU标签。反过来，这种推论使我们能够在现场未标记的神经元中对树突中的轴突进行分类。
In life sciences, fluorescent labeling techniques are used to study molecular structures and interactions of cells. However, this type of cell imaging has its own limitations, one of which is that the process of staining the cells could be toxic to the cells and possibly damage them. We are specifically interested in time-lapse imaging of live neurons to study their growth and proliferation. Neurodegenerative diseases are characterized by phenotypic differences in neuron growth and arborization. This thesis proposes a label-free digital staining method using the deep convolutional neural network to address the issues with the previous cell imaging method. Our results show that a deep neural network, when trained on phase images with correct fluorescent labels, can correctly learn the necessary morphological information to successfully predict MAP2 and Tau labels. This inference, in turn, allows us to classify axons from dendrites in live, unlabeled neurons.
