Skip to content

HistoImagePlot

Side-by-side visualization of tissue thumbnails and HoVer-Net cell-segmentation output in R/Bioconductor — an Ilaria Billato package (Waldron lab) for importing HoVer-Net nuclear-segmentation results from histopathology images into Bioconductor spatial-data structures, producing colored-by-cell-type segmentation overlays alongside the source H&E thumbnail.

Talk at EuroBioC2025

  • Speaker: Ilaria Billato (University of Padova)
  • Day / session: Day 2 (Thu Sep 18, 2025) — late-morning short-talk session (11:00 AM)
  • Talk title: "A standardized R/Bioconductor framework for integrative analysis of histopathological images with multi-omics data"
  • Companion package: ImageTCGA — Shiny application for visualizing image-derived features alongside TCGA multi-omics data
  • Slides (Zenodo DOI): to verify after publish
  • Video: not recorded
  • Abstract / schedule: eurobioc2025 schedule

What it does

HistoImagePlot bridges the gap between a deep-learning histopathology pipeline (HoVer-Net for simultaneous nuclear segmentation and classification) and a Bioconductor spatial-data workflow. It ingests HoVer-Net's per-nucleus output, lays it back over the tissue thumbnail, and produces side-by-side visualizations colored by predicted cell-type label — making the segmentation inspectable as part of a standard R analysis rather than a separate image-viewer step. The companion ImageTCGA Shiny app extends the same primitives to TCGA cohorts, letting an analyst link image-derived morphology features back to the multi-omics layer for the same sample.

How it works

Core idea. Take the HoVer-Net deep-learning histopathology pipeline's per-nucleus output (segmentation contours + cell-type predictions + per-nucleus features), import it into a SpatialExperiment, and render it as a side-by-side overlay on the source H&E thumbnail. This makes nuclear segmentation and cell-type classification first-class entries in a Bioconductor analysis object instead of artifacts trapped in a separate image-viewer.

Inputs / outputs. Input: HoVer-Net output as either JSON files (cell coordinates, types, contours) or H5AD AnnData files (with computed features like intensity metrics and neighbor distances), plus the source tissue thumbnail PNG. Files are typically fetched from a remote URL with BiocFileCache for caching. Output: a SpatialExperiment carrying the per-nucleus features + spatial coordinates, plus overlay plots from plotHoverNetH5ADOverlay() that show the thumbnail beside the colored-by-cell-type segmentation.

Key innovation. Standardizing the H&E-side data on a SpatialExperiment substrate — the same container OSTA uses for transcriptomic spatial data — so image-derived nuclear features can be joined cell-by-cell with downstream multi-omics layers. The ImageTCGA companion Shiny app demonstrates this by indexing HoVer-Net-processed TCGA WSIs against the same sample identifiers TCGA uses for its omics data, making cross-modal queries trivial.

Parameters worth knowing. outClass in the HoverNet() ingestion ("SpatialExperiment" is the documented target). Input file format (JSON for coordinates-only; H5AD when per-nucleus features are already computed). Thumbnail path for the overlay. TCGA sample identifier when sourcing data through the imageFeatureTCGA dataset.

Canonical example. The vignette walks through loading a HoVer-Net H5AD file from a remote URL via the imageFeatureTCGA dataset, converting it into a SpatialExperiment with HoverNet() + import(), and producing the H&E-thumbnail-plus-segmentation-overlay plot for a TCGA sample. The overlay is colored by HoVer-Net's predicted cell-type label per nucleus.

Where it fits in the corpus

  • AACR 2026: axis = bioinfo / AI methods (digital pathology); HistoImagePlot is exactly the kind of small bridge package that lets a histopathology AI model show up inside a Bioc analysis
  • OSTA (entry) — Billato is an OSTA author; HistoImagePlot is the operational H&E-side analogue to OSTA's transcriptomic-side framework
  • GBCC 2025: Billato also gave a related talk at GBCC2025 Day 3 (Oral Session 3) — "A standardized R/Bioconductor framework for integrative analysis of histopathological images with multi-omics data" (with Marcel Ramos Pérez) — cross-walked storyline
  • Nextflow Summit 2026: not mentioned

Notes

The Bioconductor release of HistoImagePlot anchors the talk in something concrete and inspectable. The standardization framing (ImageTCGA + HistoImagePlot as the visualization layer + a planned SpatialFeatureExperiment-style container for image features) is the longer-arc story; expect more package releases from the same group threading image features into the broader OSTA / SpatialExperiment ecosystem.