ThickV-Stain: Unprocessed Thick Tissues Virtual Staining for Rapid Intraoperative Histology

Lulin Shi, Xingzhong Hou, Ivy H. M. Wong, Simon C. K. Chan, Zhenghui Chen, Claudia T. K. Lo, Terence T. W. Wong

Keywords: Virtual staining, deep learning, label-free imaging, unprocessed biological tissues

Abstract: Virtual staining has shown great promise in realizing a rapid and low-cost clinical alternative for pathological examinations, eliminating the need for chemical reagents and laborious staining procedures. However, most of the previous studies mainly focus on thin slice samples, which still require tissue sectioning and are unsuitable for intraoperative use. In this paper, we propose a multi-scale model to virtually stain label-free and slide-free biological tissues, allowing hematoxylin- and eosin- (H&E) staining generation in less than a minute for an image with 100 million pixels. We name this ThickV-Stain model, specifically developed to virtually stain intricated and unprocessed thick tissues. We harness the ability of a multi-scale network to encourage the model to capture multiple-level micromorphological characteristics from low-resolution images. Experimental results highlight the advantages of our multi-scale method for virtual staining on unprocessed thick samples. We also show the effectiveness of ThickV-Stain on thin sections, showing generalizability to other clinical workflows. The proposed method enables us to obtain virtually stained images from unstained samples within minutes and can be seamlessly integrated with downstream pathological analysis tasks, providing an efficient alternative scheme for intraoperative assessment as well as general pathological examination.

Submission Number: 57