PathFlowAI: A High-Throughput Workflow for Preprocessing, Deep Learning and Interpretation in Digital Pathology

Abstract

The diagnosis of disease often requires analysis of a biopsy. Many diagnoses depend not only on the presence of certain features but on their location within the tissue. Recently, a number of deep learning diagnostic aids have been developed to classify digitized biopsy slides. Clinical workflows often involve processing of more than 500 slides per day. But, clinical use of deep learning diagnostic aids would require a preprocessing workflow that is cost-effective, flexible, scalable, rapid, interpretable, and transparent. Here, we present such a workflow, optimized using Dask and mixed precision training via APEX, capable of handling any patch-level or slide level classification and prediction problem. The workflow uses a flexible and fast preprocessing and deep learning analytics pipeline, incorporates model interpretation and has a highly storage-efficient audit trail. We demonstrate the utility of this package on the analysis of a prototypical anatomic pathology specimen, liver biopsies for evaluation of hepatitis from a prospective cohort. The preliminary data indicate that PathFlowAI may become a cost-effective and time-efficient tool for clinical use of Artificial Intelligence (AI) algorithms.

A large-scale internal validation study of unsupervised virtual trichrome staining technologies on nonalcoholic steatohepatitis liver biopsies

“…Samples were fixed and embedded in paraffin blocks and 5-micron sections were stained with H&E and trichrome. Slides were scanned using the Leica Aperio-AT2 (Leica Microsystems, Buffalo Grove, IL) scanner at ×20 magnification, stored in SVS image format (JPEG compression at 70% quality), from which they were extracted and converted to NPY (uncompressed unsigned 8bit RGB numpy array) format via our in house pipeline, PathFlowAI [18].…”

“…Deep learning approaches, heuristics that utilize artificial neural networks (ANN) and convolutional neural networks, circumvent this issue in that they are able to find important shapes and patterns for prediction without human specification by capturing and integrating low level features into higher level abstractions [16,17]. Given the advent of these powerful machine learning tools, the Department of Pathology and Laboratory Medicine at Dartmouth Hitchcock Medical Center (DHMC) has fully embraced digital pathology with the caveat that it must leverage advanced computational technologies to provide significant benefits over traditional glass slides [18][19][20]. To this end, we have been developing and validating various deep learning technologies in histopathology and designing them to clinical scale [18,[21][22][23][24][25][26].…”

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

“…Slides were scanned using the Leica Aperio-AT2 (Leica Microsystems, Buffalo Grove, IL) scanner at 20x magnification, stored in SVS image format, from which they were extracted and converted to NPY format via our in house pipeline, PathFlowAI. 18 We utilized 20 H&E/Trichrome WSI pairs for training the deep learning model. WSI are large images that can span hundreds of thousands of pixels in any given spatial dimension.…”

“…Given the advent of these powerful machine learning tools, the Department of Pathology and Laboratory Medicine at Dartmouth Hitchcock Medical Center (DHMC) has fully embraced digital pathology with the caveat that it must leverage advanced computational technologies to provide significant benefits over traditional glass slides. [18][19][20] To this end, we have been developing and validating various deep learning technologies in histopathology and designing them to clinical scale 18,[21][22][23][24][25] .…”

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

“…Many research groups are developing high-throughput clinical pipelines to take advantage of these healthcare technologies. Validating and scaling these technologies is essential for successful deployment 10 .…”