This repository contains open source tools created by Google Research to make it easy to bring digital pathology imaging to Google Cloud and to serve pathology images from Google Cloud for a variety of applications including: medical imaging AI, interactive visualization, and more.
The cloud pathology repository contains the following components:
Ingest digital pathology images into the Google Cloud DICOM store. The tool supports image format conversions for non-DICOM images, DICOM metadata augmentation, and more. The tool is designed to run at scale in GKE to enable enterprise-scale image ingestion. The transformation pipeline can also be run locally on a Linux system or within a virtual machine for testing and small jobs.
HTTP proxy for Google Cloud DICOM store that adds additional pathology-specific features, e.g., accelerated frame serving, support for DICOM rendered transcoding of JPEG-XL encoded DICOM images, and Server side ICC Color Profile transformation, etc.
Reference zero footprint web viewer for visualization of DICOM digital pathology imaging. Viewer supports interactive visualization and annotation of DICOM pathology imaging.
This repository also contains Google Research's Image Life Cycle Management (ILM) tool. ILM is applicable to all DICOM imaging modalities. ILM automates DICOM assets movement between DICOM store storage classes to minimize Cloud storage cost. The ILM tool monitors the Google DICOM store logs and uses heuristics (rules set by the user) to automate the movement of DICOM images to the optimal storage tier (standard - archival).
Use terraform to configure cloud infrastructure to deploy pathology components and/or ILM.
EZ WSI DicomWeb is a Pip installable Python library that greatly simplifies the accessing digital pathology images from a Cloud DICOM Store. The library provides built in support for frame caching to accelerate image access and provides a client-side interface to easily convert pathology imaging into ML embeddings using Google Research Pathology Foundations ML embeddings.
Path Foundation is a machine learning (ML) model that produces digital pathology specific ML embeddings. The embeddings produced by this model can be used to efficiently build AI models for pathology related tasks, with less data and less compute than traditional approaches; see blog post and example colabs.
See CONTRIBUTING.md for details.
See LICENSE for details.
