A Deep Learning Model for Automated Segmentation of Fluorescence Cell images

Abstract

Deep learning techniques bring together key advantages in biomedical image segmentation. They speed up the process, increase the reproducibility, and reduce the workload in segmentation and classifcation. Deep learning techniques can be used for analysing cell concentration, cell viability, as well as the size and form of each cell. In this study, we develop a deep learning model for automated segmentation of fuorescence cell images, and apply it to fuorescence images recorded with a home-built epi-fuorescence microscope. A deep neural network model based on U-Net architecture was built using a publicly available dataset of cell nuclei images [1]. A model accuracy of 97.3% was reached at the end of model training. Fluorescence cell images acquired with our home-built microscope were then segmented using the developed model. 141 of 151 cells in 5 images were successfully segmented, revealing a segmentation success rate of 93.4%. This deep learning model can be extended to the analysis of diferent cell types and cell viability.

Recurrent and convolutional neural networks for sequential multispectral optoacoustic tomography (MSOT) imaging

“…For instance, deep learning for automatic segmentation of optoacoustic ultrasound images [34] used the U-Net architecture [35] to perform image segmentation. U-Net is a well-known convolutional neural network (CNN) for image segmentation, particularly biomedical images [36][37][38][39].…”