A Large-Scale Internal Validation Study of Unsupervised Virtual Trichrome Staining Technologies on Non-alcoholic Steatohepatitis Liver Biopsies

Abstract

<jats:title>Abstract</jats:title><jats:p>Non-alcoholic steatohepatitis (NASH) is a type of fatty liver disease characterized by accumulation of fat in hepatocytes and concurrent inflammation. It is associated with significant morbidity as it can progress to cirrhosis and liver failure and increases the risk of developing hepatocellular carcinoma. NASH has become increasingly prevalent in the US and is a leading cause of liver transplantation for end stage liver disease and a significant public health burden. Grading and staging NASH involves liver core biopsy, followed by microscopic examination of histologic preparations of the tissue. Tissue sections are stained with hematoxylin and eosin (H&amp;E) to grade NASH activity, and stained with trichrome to help stage fibrosis. Methods to computationally transform one stain into another on digital whole slide images (WSI) can reduce the need for additional physical staining besides H&amp;E, alleviating personnel, equipment, and time costs. Recent computational techniques have shown promise and success for virtual staining of tissue. We conducted a large-scale validation study of the viability of virtual staining technologies for H&amp;E to trichrome conversion on WSI (n=574). Our results demonstrate that virtual trichrome technologies present an immediate software solution that can be readily employed in the clinical setting as a diagnostic decision aid.</jats:p>

Topological Feature Extraction and Visualization of Whole Slide Images using Graph Neural Networks

“…In comparison to prior modeling techniques that use handcrafted features, deep learning applies parameterized filters and pooling mechanisms via convolutional neural networks (CNN) to capture and integrate lower level image features into successively higher levels of complexity 3 . These approaches have been used to automatically stage liver fibrosis 4 , identify morphological features correspondent with somatic alterations 5 , assess urine slides for bladder cancer 6 , and circumvent costly chemical staining procedures 7,8 , amongst many others 9 . Many research groups are developing high-throughput clinical pipelines to take advantage of these healthcare technologies.…”

Topological Feature Extraction and Visualization of Whole Slide Images using Graph Neural Networks

“…In comparison to prior modeling techniques that use handcrafted features, deep learning applies parameterized filters and pooling mechanisms via convolutional neural networks (CNN) to capture and integrate lower level image features into successively higher levels of complexity 3 . These approaches have been used to automatically stage liver fibrosis 4 , identify morphological features correspondent with somatic alterations 5 , assess urine slides for bladder cancer 6 , and circumvent costly chemical staining procedures 7 , 8 , amongst many others 9 . Many research groups are developing high-throughput clinical pipelines to take advantage of these healthcare technologies.…”