The content discusses the generation of virtual H&E staining from label-free FLIM images using a DL model. The method was tested on lung cancer tissue samples and expanded to include colorectal, endometrial cancers, and FFPE lung biopsies. Results showed accurate reconstruction of cellular components, enabling identification of lifetime signatures of different cell types. Comparison of different image formats revealed that IW-FLIM was the most effective for virtual H&E staining. Blind evaluation by pathologists confirmed the quality of virtual staining compared to true H&E images. The study demonstrated the potential of DL models in generating accurate virtual H&E-stained images for clinical use. The method was able to faithfully reproduce morphological and textural attributes of various cell types present in the tissue samples. The study also highlighted the importance of combining intensity and lifetime data for a comprehensive analysis of cellular components. Overall, the results showed promising outcomes in generating virtual H&E-stained images for different tissue samples, providing valuable insights for diagnostic decision-making in pathology laboratories.
Source link
Source link: https://www.nature.com/articles/s44303-024-00021-7
Virtual H&E staining from autofluorescence images using deep learning. #VirtualHistology
