Abstract
The recent development of highly multiplexed tissue imaging promises to substantially accelerate research into basic biology and human disease. Concurrently, histopathology in a clinical setting is undergoing a rapid transition to digital methods. Online tissue atlases involving highly multiplexed images of research and clinical specimens will soon join genomics as a systematic source of information on the molecular basis of disease and therapeutic response. However, even with recent advances in machine learning, experience with anatomic pathology shows that there is no immediate substitute for expert visual review, annotation, and description of tissue images. In this perspective we review the ecosystem of software available for analysis of tissue images and identify a need for interactive guides or “digital docents” that allow experts to help make complex images intelligible. We illustrate this idea using Minerva software and discuss how interactive image guides are being integrated into multi-omic browsers for effective dissemination of atlas data.
Introduction: The Communication Gap in Tissue Imaging
We are entering an era of "highly multiplexed" tissue imaging, where researchers can visualize between 40 and 100+ different molecular markers on a single tissue section. While this provides a genomic-level depth of information regarding the spatial organization of tumors and tissues, it has created a significant communication gap.
Currently, these gigapixel-scale images are shared in one of two ways: as static, low-resolution "postage stamp" figures in journals—which discard the vast majority of the underlying data—or via massive databases that require specialized software and significant computational power to navigate. Furthermore, the biological significance of these images is often opaque to non-experts without the guidance of a trained pathologist. There is a critical need for tools that make these "tissue atlases" both accessible and interpretable to the broader scientific community.
The Minerva Framework: A "Digital Docent"
To address this, we developed Minerva, an open-source software suite designed to act as a "digital docent." The software allows experts to create interactive, narrated "stories" that guide users through high-dimensional pathology data directly in a standard web browser. The framework consists of two primary components:
- Minerva Author: A desktop application (built with React and Python Flask) that allows an expert to import OME-TIFF images. The expert can pan and zoom to specific regions of interest, toggle channel groups (e.g., switching between "Immune Cell" markers and "Tumor" markers), and write narrative "waypoints" accompanied by text and graphical annotations.
- Minerva Story: A lightweight, web-based viewer that renders these narratives. Unlike traditional image management systems (such as OMERO), Minerva is database-independent and uses a client-side approach. It can be hosted on simple cloud storage (like AWS S3) or GitHub Pages, making it fast, scalable, and compatible with mobile devices.
Bridging Quantitative Data and Morphology
One of the primary challenges in multiplexed imaging is reconciling high-dimensional data—often visualized via UMAP or t-SNE plots—with physical tissue morphology. Minerva enables an "aha moment" for researchers by linking these domains: a user can click a data point on a UMAP cluster and have the viewer immediately zoom to that exact cell in the tissue.
This interactivity serves as a vital quality control (QC) mechanism. By overlaying AI-generated segmentation masks (the outlines of cells), Minerva allows users to visually verify the accuracy of machine-learning models. Experts can easily spot "fusion" or "fission" errors where software has incorrectly merged or split cells, ensuring that the quantitative data remains grounded in biological reality.
Demonstrations and Use Cases
The utility of Minerva has been demonstrated across diverse datasets and scales:
- Cancer Research (LUNG-3-PR): Using a large human lung adenocarcinoma specimen imaged with 44-plex t-CyCIF (cyclic immunofluorescence), Minerva was used to link single-cell quantitative data back to specific pixels, allowing researchers to explore the tumor microenvironment in detail.
- Medical Education: Traditional H&E (Hematoxylin and Eosin) stained heart specimens from a patient with multiple myocardial infarctions were transformed into an interactive textbook. This "story" provides a nuanced understanding of cardiac pathophysiology that exceeds the capabilities of static textbook snippets.
- Atlas Integration: Minerva is currently being integrated into large-scale international projects, including the Human Tumor Atlas Network (HTAN) and HuBMAP, to facilitate the dissemination of atlas data.
Conclusion and Future Directions
Minerva represents a shift in how we communicate complex imaging data, moving from static representation to interactive exploration. The software is designed for a low barrier to entry; it typically takes only a few hours to learn and roughly 30 minutes to create a story—comparable to the time required to generate a high-quality static figure.
While Minerva is a powerful tool for viewing and sharing, it is not an analysis tool; it relies on interoperable software for initial image processing and registration. Future work focuses on managing the costs of data egress for multi-terabyte datasets and integrating Minerva stories directly into genomics platforms like cBioPortal. This will eventually allow researchers to view mutational data and tissue morphology side-by-side, providing a truly multi-omic perspective on human disease.
Summary updated