Abstract
Objective and Impact Statement: We present a panel of virtual staining neural networks for lung and heart transplant biopsies, providing rapid and high-quality histological staining results while bypassing the traditional histochemical staining process. Introduction: Allograft rejection is a common complication of organ transplantation, which can lead to life-threatening outcomes if not promptly managed. Histological examination is the gold standard method for evaluating organ transplant rejection status, as it provides detailed insights into rejection signatures at the cellular level. Nevertheless, the traditional histochemical staining process is time-consuming, costly, and labor-intensive since transplant biopsy evaluations typically necessitate multiple stains. Furthermore, once these tissue slides are stained, they cannot be reused for other ancillary tests. More importantly, suboptimal handling of very small tissue fragments from transplant biopsies may impede their effective histochemical staining, and color variations across different laboratories or batches can hinder efficient histological analysis by pathologists. Methods: To mitigate these challenges, we developed a panel of virtual staining neural networks for lung and heart transplant biopsies, which digitally convert autofluorescence microscopic images of label-free tissue sections into their bright-field histologically stained counterparts-bypassing the traditional histochemical staining process. Specifically, we virtually generated hematoxylin and eosin (H&E), Masson's Trichrome (MT), and elastic Verhoeff-Van Gieson stains for label-free transplant lung tissue, along with H&E and MT stains for label-free transplant heart tissue. Results: Blind evaluations conducted by 3 board-certified pathologists confirmed that the virtual staining networks consistently produce high-quality histology images with high color uniformity, closely resembling their well-stained histochemical counterparts across various tissue features. The use of virtually stained images for the evaluation of transplant biopsies achieved comparable diagnostic outcomes to those obtained via traditional histochemical staining, with a concordance rate of 82.4% for lung samples and 91.7% for heart samples. Moreover, virtual staining models create multiple stains from the same autofluorescence input, eliminating structural mismatches observed between adjacent sections stained in the traditional workflow, while also saving tissue, expert time, and staining costs. Conclusion: The presented virtual staining panels provide an effective alternative to conventional histochemical staining for transplant biopsy evaluation. These virtual staining panels have the potential to enhance the clinical diagnostic workflow for organ transplant rejection and improve the performance of downstream automated models for the analysis of transplant biopsies.