Minerva: a light-weight, narrative image browser for multiplexed tissue images

John Hoffer, Rumana Rashid, Jeremy Muhlich, Yu‐An Chen, Douglas Russell, Juha Ruokonen, Robert Krüger, Hanspeter Pfister, Sandro Santagata, Peter K. Sorger, Robert Krueger, Yu-An ChenJ. Open Source Softw 2020 · DOI · DBLP

Abstract

Minerva is an open-source, web-based tool designed to handle the "big data" challenge of highly multiplexed tissue imaging. By using a "narrative" approach, it allows researchers to guide viewers through terabyte-scale images using curated waypoints, making complex pathology data as easy to navigate as a standard web map.

Introduction

Modern multiplexed tissue imaging allows scientists to visualize 10 to 100 different proteins within a single tissue sample at sub-micron resolution. While this provides an incredible window into human biology, it creates a massive data challenge: a single microscopy slide can result in terabyte-sized files containing millions of cells. Existing visualization tools often struggle with this scale or require specialized, heavy-duty software. Furthermore, navigating these images is difficult for non-experts without guidance. Minerva addresses this by providing a lightweight tool for web-based display and expert-led "guided tours" of complex tissue data.

Methods and System Architecture

The authors developed a two-part system to bridge the gap between raw data and user-friendly exploration:

  1. Minerva Author: A desktop application (built with React and a Python Flask backend) where researchers curate "stories."
  2. Minerva Story: A lightweight, single-page web application that renders the final product for the end-user using the OpenSeadragon framework to handle image pyramids.
The Minerva ecosystem workflow, illustrating the transition from raw imaging data to curated stories via Minerva Author and Minerva Story.

The system is designed for static hosting, allowing researchers to host interactive tissue stories on simple services like GitHub Pages or Amazon S3 without needing dedicated database servers.

The Authoring Workflow

In Minerva Author, researchers import heavy raw files (such as OME-TIFF), adjust contrast, and select specific protein channels to highlight. The core of the platform is the creation of "waypoints"—saved states that include a specific location, zoom level, and explanatory text.

The Minerva Author interface showing the sequence of steps for image import, channel grouping, pseudocoloring, contrast adjustment, and waypoint creation.

This narrative metaphor moves away from passive data browsing toward active storytelling, mimicking a pathologist sitting next to a colleague and pointing out specific biological features.

Key Insights and Results

  • Decoupling Data from Display: By converting raw, high-bit-depth data into optimized JPEG pyramids, Minerva enables the viewing of terabytes of data on standard laptops or mobile phones without losing perceived resolution.
  • Linked Navigation: The tool can link single-cell data (like scatterplots or heatmaps) directly to the image. Clicking a data point on a chart can jump the image to that specific cell, and vice versa.
  • Performance: Minerva successfully handles images with approximately 60 spectral channels and 10⁶ cells, providing smooth panning and zooming analogous to Google Maps.
  • Versatility: Use cases include validating unsupervised clustering results, assessing antibody staining quality, and creating educational content for medical trainees.

Limitations and Future Work

Currently, adding advanced interactive data visualizations requires manual JSON configuration, though a native UI for this is in development for Minerva Author. Future iterations aim to integrate Minerva into genomic browsers like cBioPortal to allow side-by-side analysis of genetic data and tissue architecture. Additionally, the team is exploring automated feature detection to suggest waypoints for researchers.

Summary updated

BibTeX
@article{hoffer2020minerva,
  author = {John Hoffer and Rumana Rashid and Jeremy Muhlich and Yu‐An Chen and Douglas Russell and Juha Ruokonen and Robert Krüger and Hanspeter Pfister and Sandro Santagata and Peter K. Sorger and John Hoffer and Rumana Rashid and Jeremy L. Muhlich and Douglas Russell and Juha Ruokonen and Robert Krueger and Sandro Santagata and Yu-An Chen},
  title = {{Minerva: a light-weight, narrative image browser for multiplexed tissue images}},
  journal = {J. Open Source Softw.},
  year = {2020},
  doi = {10.21105/joss.02579},
}
Figure