A Cloud Based Distributed Tool for Computational Renal Pathology

点击翻译按钮获取中文摘要

Quantitative computational analysis of digital pathology whole slide images (WSIs) has shown increasing promise for precision medicine applications in last decade. In recent years, this progress has extended to renal pathology, while seeing a growing need of objective quantification of deep features from large digital WSIs of renal tissues. The current standard of routine brightfield visual assessment of renal biopsies is unable to elicit and quantify the deep features from large WSIs elicited by machine vision techniques. Existing computational renal pathology tools primarily focus on extraction of renal micro-compartments and computational classification of renal diseases. However, understanding the correlation between deep features of renal micro-compartments and clinical biometrics and correlations with molecular level data remain as opportunities for investigation and discovery. A major gap that needs to be closed is that the tools developed by computational researchers are not in a format that can be easily implemented by pathology end-users. The availability of plug-and-play tools will empower renal pathologists and biologists engaged in kidney research, and offer exponential growth in research studies using increasingly available digital datasets across various kidney diseases via consortia including the Kidney Precision Medicine Project, Nephrotic Syndrome Study Network, Cure Glomerulonephropathy, and Human Biomolecular Atlas Project. To address the above gap, experts from computational imaging (Dr. Sarder), software science (Mr. Manthey), nephropathology/basic science (Dr. Rosenberg), and nephrology (Dr. Han) have teamed up to develop a web-cloud based end-user software for nephropathologists, nephrologists, and basic scientists. The proposed tool emerges from ongoing collaboration between the team members. The proposed software will offer the following functionalities to renal pathology end-users: (i) cloud storage and visualization of digital renal pathology WSIs and associated metadata; (ii) microanatomic/histomorphologic annotation capability in an easy-to-use web-based visualization system, allowing users to collaborate while conducting annotation; (iii) automated plug-and-play plugins that would allow users to segment multi-scale renal structures for a large batch of renal tissue WSIs, and (iv) plugins to refine renal micro-compartmental segmentation in a human-artificial-intelligence-loop set-up where humans and AI system collaborate in the cloud to refine the segmentation models iteratively; and finally, (v) measurement of deep image features on the segmented renal structures to enable diagnostic and prognostic research for the spectrum of renal diseases. The distributed tool will facilitate multi-center studies using federated learning where individual centers will not need to export data with protected healthcare information outside their institutes, while still participating in training the proposed system to improve segmentation on their own institutional data. Finally, the system will offer end- users the ability to integrate spatial transcriptomics molecular data with image data in the same system to allow users to navigate the images as well as molecular data in a web-cloud set-up for new scientific discovery.