Skip to content
Joshua Levy edited this page Aug 14, 2019 · 7 revisions

Welcome to the PathFlowAI wiki!

Levy,J. et al. (2019) PathFlowAI: A Convenient High-Throughput Workflow for Preprocessing, Deep Learning Analytics and Interpretation in Digital Pathology. medRxiv, 19003897.

Manuscript URL: https://www.medrxiv.org/content/10.1101/19003897v1

PathFlowAI is a preprocessing and deep learning analytics workflow for histopathology images (Whole slide images, WSI).

On a high-performance computing pipeline, preprocessing on the order of a hundred images can finish in minutes, with minimal impact to the file system, storage space and stored in very flexible file formats. Details can be found in the manuscript above.

This repository is still a work in progress, but this guide will give basic usage of the preprocessing and deep learning commands for classification and segmentation tasks. As some of the non-core elements of the pipeline are undergoing restructuring, we welcome contributions in the form of Pull Requests and Issues, especially for non-functioning components.

PathFlowAI is used for both classification and segmentation tasks on the patch level, the information of which can be aggregated on the slide level. You can consult the API here: https://jlevy44.github.io/PathFlowAI/

Here are the core steps for running a successful WSI deep learning analysis:

  1. Preprocessing the data.
  2. Training the model and making predictions.
  3. Visualizing the results.
  4. Model interpretations and embeddings.
Clone this wiki locally