BI Biological Intelligence

Deep Learning Microscopy

Time:2024-03-07 View count:


Deep Learning Microscopy

Introduction

Deep learning is a learning model of machine learning, which uses very deep structure of multi-layer neural network activated by non-linear functions to analyze signals and data. Convolutional neural network (CNN), taking advantage of local connections and shared weights, has made major advances in computer-vision community. Recently, deep learning algorithm has also been applied in cross-modelity microscopy, phase recovery, holographic image reconstruction and so on. So we are also inspired by the powerful deep learning and previous work on microscopy and proposed several deep learning-based microscopy algorithm.

Highlights

Deep Convolutional Neural Network for Tomographic Phase Microscopy

  • GPU-based implementation of deep CNN to model the propagation of light beam

    • Layer-wise propagation model based on beam propagation method

    • Complex-valued neural network implemented on Tensorflow & Keras

  • 1-norm constrained loss function

    • 1-norm data fidelity term & ℓ1-norm gradient-domain regularizer

Qiao, H., Wu, J., Li, X., Shoreh, M. H., Fan, J., & Dai, Q. (2018). GPU-based deep convolutional neural network for tomographic phase microscopy with l1 fitting and regularization. Journal of biomedical optics, 23(6), 066003.