Virtual hematoxylin and eosin histopathology using simultaneous photoacoustic remote sensing and scattering microscopy

Abstract

Hematoxylin and eosin (H&E) staining is the gold standard for most histopathological diagnostics but requires lengthy processing times not suitable for point-of-care diagnosis. Here we demonstrate a 266-nm excitation ultraviolet photoacoustic remote sensing (UV-PARS) and 1310-nm microscopy system capable of virtual H&E 3D imaging of tissues. Virtual hematoxylin staining of nuclei is achieved with UV-PARS, while virtual eosin staining is achieved using the already implemented interrogation laser from UV-PARS for scattering contrast. We demonstrate the capabilities of this dual-contrast system for en-face planar and depth-resolved imaging of human tissue samples exhibiting high concordance with H&E staining procedures and confocal fluorescence microscopy. To our knowledge, this is the first microscopy approach capable of depth-resolved imaging of unstained thick tissues with virtual H&E contrast.

Interpreting microscopic structures in virtually stained histological sections for veterinary oncology applications

“…Considering the training data preparation, required for the supervised learning framework implementation, such approach allowed us pixel registration matching between the fluorescence input images and the corresponding brightfield H&E images. In this sense, our used strategy is slightly different from samples preparation workflow reported in 5,12,34,35 permitting avoidance of the sample handling process that involves measurement of the unstained slide first, staining the same slide with H&E, coversliping it and measuring it once more, in order to get a match between the VS tissue structure and stained with H&E. Instead, we used an unpaired image to image translation (refered as to stage 4) . Results of stage 4 VS, using unstained deparaffinized section of a soft tissue sarcoma from a cat onto which transfer learning protocol was applied, produced VS images that partially resembled H&E stained section but had suboptimal contrast between red and blue colors and were blurry.…”

“…Quantitative phase imaging 2 , stimulated Raman scattering imaging 3 , multiphoton (or steady-state) fluorescence imaging 4,5 combined into an optical HP slide-scanning setup are commercially offered today for VS implementations [Phasics 6 , Oxford Instruments 7 ; Brucker 8 ; 3DHistech 9 ]. Through the application of deep neural networks, additional techniques have been proven to be feasible for VS in lab studies, including brightfield microscopy 10 , UV photoacoustic microscopy 11 , photoacoustic remote sensing 12 , FLIM, CARS, and SHG nonlinear microscopy 13,14 .…”

Imaging approaches for monitoring three‐dimensional cell and tissue culture systems

“…There are two main mechanisms of achieving resolution (PAM), either optical-resolution PAM (OR-PAM) or acoustic-resolution (AR-PAM), in which for both cases the point source function is a convolution of the optical excitation and the acoustic detection [153]. In OR-PAM, optical focus (absorption contrast) is the main resolution mechanism for lateral resolution, achieving lateral resolution of 1-5 μm with moderately high NA's approximately 1.0 (while a lateral resolution of 200 nm, approaching the diffraction limited has been achieved using a high-NA (1.23) water-immersion lens [164,166]), however, axial resolution is determined acoustically achieving an axial resolution of approximately 15 μm [153,[167][168][169]. However, due to optical scattering OR-PAM achieves axial penetration depth of approximately 1 mm at 75 MHz, measured with an ultrasound detector [153].…”

Bioimaging of Dissolvable Microneedle Arrays: Challenges and Opportunities

“…Researchers have tried to use conditional generative adversarial networks (cGANs) to accept the autofluorescent of nonstained biological tissue with whole slide images and computationally stain them by learning hierarchical nonlinear mappings between image pairs before and after H&E staining [ 37 ]. A recent study presented the ability to create a label-free virtual H&E image, but it requires physical contact between the ultrasonic transducer and the sample to measure the generated sound waves [ 38 ]. In general, the slice imaging process based on algorithm analysis faces the time and economic pressures.…”