Cell image characterization using deep learning algorithm

Abstract

Recently lens-free shadow imaging technology has been successfully demonstrated for the detection and analysis of various biological cells [1-3]. Due to its high throughput, low cost, simple and compact properties, this technology has become a promising modality, especially for point-of-care applications. However, automated characterization for the cell type differentiation in lens-free micrograph has been a challenging issue. Recently, machine learning algorithm has made tremendous progress for automated image classification. At the same time, the advancement in computation resources such as GPU (Graphics Processing Unit) and CPU (Central Processing Unit) has facilitated the implementation of the deep machine learning algorithms. In this work we have implemented deep learning algorithm to classify the lens-free images of different cell lines that were generated by the custom developed lens-free imaging system (see figure 1). For this purpose we have used the AlexNet Neural Network architecture [4] provided by NVIDIA DiGiTS deep learning tool to train the neural network for lens-free shadow images of different cell lines. Using this system we have trained the network for 30 training epochs with validation interval of 1 and base learning rate of 0.01 for over 200 samples of each cell lines of red blood cells (RBC) and microbeads. We have tested the efficiency of the neural network for the differentiation of the imagese a promising modality, especially for point-of-care applications. However, automated characterization for the cell type differentiation in lens-free micrograph has been a challenging issue. Recently, machine learning algorithm has made tremendous progress for automated image classification. At the same time, the advancement in computation resources such as GPU (Graphics Processing Unit) and CPU (Central Processing Unit) has facilitated the implementation of the deep machine learning algorithms. In this work we have implemented deep learning algorithm to classify the lens-free images of different cell lines that were generated by the custom developed lens-free imaging system (see figure 1). For this purpose we have used the AlexNet Neural Network architecture [4] provided by NVIDIA DiGiTS deep learning tool to train the neural network for lens-free shadow images of different cell lines. Using this system we have trained the network for 30 training epochs with validation interval of 1 and base learning rate of 0.01 for over 200 samples of each cell lines of red blood cells (RBC) and microbeads. We have tested the efficiency of the neural network for the differentiation of the images