Label-Free Nuclear Staining Reconstruction in Quantitative Phase Images Using Deep Learning

Investor logo

Warning

This publication doesn't include Faculty of Economics and Administration. It includes Faculty of Medicine. Official publication website can be found on muni.cz.
Authors

VIČAR Tomáš GUMULEC Jaromír BALVAN Jan HRACHO Michal KOLAR R.

Year of publication 2019
Type Article in Proceedings
Conference WORLD CONGRESS ON MEDICAL PHYSICS AND BIOMEDICAL ENGINEERING 2018, VOL 1
MU Faculty or unit

Faculty of Medicine

Citation
Web http://dx.doi.org/10.1007/978-981-10-9035-6_43
Doi http://dx.doi.org/10.1007/978-981-10-9035-6_43
Keywords Deep learning; Quantitative phase imaging; Cell analysis; Cell nuclei segmentation
Description Fluorescence microscopy is a golden standard for contemporary biological studies. However, since fluorescent dyes cross-react with biological processes, a label-free approach is more desirable. The aim of this study is to create artificial, fluorescence-like nuclei labeling from label-free images using Convolution Neural Network (CNN), where training data are easy to obtain if simultaneous label-free and fluorescence acquisition is available. This approach was tested on holographic microscopic image set of prostate non-tumor tissue (PNT1A) and metastatic tumor tissue (DU145) cells. SegNet and U-Net were tested and provide "synthetic" fluorescence staining, which are qualitatively sufficient for further analysis. Improvement was achieved with addition of bright-field image (by-product of holographic quantitative phase imaging) into analysis and two step learning approach, without and with augmentation, were introduced. Reconstructed staining was used for nucleus segmentation where 0.784 and 0.781 dice coefficient (for DU145 and PNT1A) were achieved.
Related projects:

You are running an old browser version. We recommend updating your browser to its latest version.