MITI Minimum Information guidelines for highly multiplexed tissue images

💡 Research Summary

The paper introduces MITI (Minimum Information about highly multiplexed Tissue Imaging), a comprehensive metadata and data‑level standard designed to support the rapidly expanding field of highly multiplexed tissue imaging. As large consortia such as the Human Tumor Atlas Network (HTAN), HuBMAP, and the LifeTime Initiative generate 20‑60 channel images that combine deep molecular profiling with spatial context, there is an urgent need for FAIR‑compatible standards that enable consistent deposition, curation, and reuse of both raw and processed data.

MITI adapts the “minimum information” concept long used in genomics (e.g., MIAME, MIGS, MIBBI) to the imaging domain. It defines a set of required metadata covering biospecimen provenance (tissue type, preservation, section thickness), reagents (antibodies, aptamers, dyes, metal tags), acquisition hardware (optical, mass‑spectrometry, light‑sheet), imaging parameters (channel count, resolution, exposure), analysis pipelines (stitching, registration, illumination correction, normalization, segmentation, annotation), and clinical/genetic context (patient demographics, treatment history, model organism genotype). These fields are harmonized with existing initiatives such as QUAREP‑LiMi, the Resource Identification Initiative, Human Protein Atlas antibody standards, and the GDC data model, ensuring interoperability across omics and clinical datasets.

A central contribution is the definition of five hierarchical data levels that parallel the data‑level system used by the Genomic Data Commons. Level 1 comprises raw raster data directly output by the instrument, analogous to FASTQ files. Level 2 consists of fully stitched, registered, illumination‑corrected, and intensity‑normalized whole‑slide images stored in OME‑TIFF (with future migration to OME‑NGFF for cloud scalability). Level 3 adds quality‑controlled, artifact‑removed images together with segmentation masks, machine‑generated spatial models, and human or algorithmic annotations. Level 4 is a tabular “spatial feature table” that records per‑cell marker intensities, coordinates, and morphological descriptors, enabling direct integration with single‑cell sequencing count matrices. Level 5 provides low‑resolution, tiled image pyramids suitable for web‑based pan‑and‑zoom viewers (e.g., MINERVA), facilitating rapid exploration of terabyte‑scale datasets without downloading full‑resolution files.

File format choices are deliberately aligned with the Open Microscopy Environment (OME) ecosystem. While OME‑TIFF remains the baseline, the consortium anticipates transition to OME‑NGFF to meet the performance demands of cloud‑native storage and analysis. Compatibility with DICOM is also considered, especially for projects that wish to integrate radiology data.

Implementation is demonstrated through a JSON schema hosted on GitHub (https://github.com/miti-consortium/MITI) that follows Schema.org principles for discoverability. The schema has been employed in the NCI HTAN data portal, where metadata are captured via a MAGE‑TAB‑style tabular approach combined with a web‑based relational interface (OMeta). This dual strategy lowers the barrier for data generators while ensuring controlled vocabularies and validation.

The authors discuss practical challenges such as the massive size of multiplexed images (up to 1 TB per experiment), the need for automated stitching of thousands of tiles, and the importance of maintaining patient privacy. They note that, unlike genomic data, current IRB guidance generally permits public release of de‑identified histological images, but anticipate future policy evolution for combined clinical‑genomic‑imaging datasets.

Future extensions include support for three‑dimensional imaging modalities, nucleic‑acid‑based spatial methods like MERFISH, and tighter integration with the upcoming Imaging Data Commons. By providing a clear, extensible framework, MITI aims to standardize data generation, promote reproducibility, and enable cross‑modal analyses that are essential for precision medicine, digital pathology, and systems biology.