Deep Learning for Virtual Histological Staining of Bright-Field Microscopic Images of Unlabeled Carotid Artery Tissue

• Published in: Molecular imaging and biology Vol. 22; no. 5; pp. 1301 - 1309
• Main Authors: Li, Dan, Hui, Hui, Zhang, Yingqian, Tong, Wei, Tian, Feng, Yang, Xin, Liu, Jie, Chen, Yundai, Tian, Jie
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.10.2020; Springer Nature B.V
• Subjects: Artificial neural networks; Cardiovascular diseases; Carotid arteries; Carotid artery; Deep learning; Image analysis; Image processing; Imaging; Machine learning; Medical imaging; Medicine; Medicine & Public Health; Neural networks; Radiology; Research Article; Staining; Tissue analysis; Tissues; Training; Transfer learning; Veins & arteries; Blind evaluation; Conditional generative adversarial network; Bright-field microscopic imaging; Virtual histological staining
• DOI: 10.1007/s11307-020-01508-6
• ISSN: 1536-1632; 1860-2002

Purpose Histological analysis of artery tissue samples is a widely used method for diagnosis and quantification of cardiovascular diseases. However, the variable and labor-intensive tissue staining procedures hinder efficient and informative histological image analysis. Procedures In this study, we developed a deep learning-based method to transfer bright-field microscopic images of unlabeled tissue sections into equivalent bright-field images of histologically stained versions of the same samples. We trained a convolutional neural network to build maps between the unstained images and histologically stained images using a conditional generative adversarial network model. Results The results of a blind evaluation by board-certified pathologists illustrate that the virtual staining and standard histological staining images of rat carotid artery tissue sections and those involving different types of stains showed no major differences. Quantification of virtual and histological H&E staining in carotid artery tissue sections showed that the relative errors of intima thickness, intima area, and media area were lower than 1.6 %, 5.6 %, and 12.7 %, respectively. The training time of deep learning network was 12.857 h with 1800 training patches and 200 epoches. Conclusions This virtual staining method significantly mitigates the typically laborious and time-consuming histological staining procedures and could be augmented with other label-free microscopic imaging modalities.