Computational PhenomicsPenn State College of Medicine

The Geometry of Life and Disease

See biology whole.
Measure it cell by cell.

We combine 3D X-ray histotomography, AI-assisted segmentation, and statistical shape analysis to turn intact organisms and human tissue into quantitative phenotypes.

See the proof Explore a live dataset
Octopus suckers rendered from an X-ray histotomography volume, arms converging toward the mouth.
Whole-organism histotomography Intact octopus hatchling · cellular-scale detail
10 mm

field of view

0.7 µm

isotropic voxels

300+

structures segmented

Selected results from the whole-octopus acquisition at APS beamline 2-BM.

Computational phenomics, plainly

A phenotype is more than an image. It is a pattern we can measure.

Genomes describe possibility. Phenomes record what biology became: the size, shape, orientation, and relationships of cells and tissues in health, development, exposure, and disease.

  1. 01Image intact biology
  2. 02Identify cells and tissues
  3. 03Measure their geometry
  4. 04Compare phenotypes

01 / Matched-tissue validation

The scan must agree with the slide.

A prostate biopsy was scanned whole and unstained. After sectioning and staining, the physical H&E section was registered back to the corresponding plane in the 3D volume.

Raw phase-contrast microCT plane resampled from the prostate biopsy volume. Registered H and E histology section from the same tissue plane. Registered H&E Raw microCT
The right side is the raw phase-contrast microCT. Drag the divider to reveal either the same pixels false-coloured or the physical H&E section registered back to the volume on the left.

Validation uses the familiar slide. The intact volume preserves everything around it.

Scan
Image the unstained biopsy before sectioning.
Register
Match the physical section back to the digital volume.
Measure
Revisit any plane and follow structures through depth.

02 / The imaging platform

The hard problem is not resolution alone. It is resolution across the specimen.

Submicron soft-tissue microCT usually trades field of view for detail. A custom wide-field detector built for beamline 2-BM at the Advanced Photon Source expands that field to 10 mm at 0.7 µm isotropic voxels—wide enough to image an intact hatchling octopus in two acquisitions.

How the detector works
Wide-field detector at APS beamline 2-BM: objective lens, detector stack and rotation stage, with sample slices.
Custom detector stack and 10 mm objective at APS 2-BM, with acquisition settings, field-of-view grid, and reconstructed slices.
14,192 × 10,640pixel sensor
26 keVbeam energy
9,001projections per acquisition
60 msexposure per projection

Detector family

Three scales for different biological questions.

Field of view, made visible

One field. Far fewer seams.

Choose a detector above. The gold frame shows its footprint on one physical scale; every aqua cell is a matched 2,048 × 2,048 detector footprint. This is a geometric coverage comparison—not a scan-row plan. Counts use a 15% overlap model; the diagram edge-aligns the sequence so no numbered field is clipped. Actual scans still depend on beam height and acquisition protocol.

APS 2-BM wide-field system

10mm active width

Field width vs 10 mm100%
Sampling vs 0.7 µmBaseline
Sampling
0.7 µm
Camera array
14,192 × 10,640
Active footprint
≈ 10 × 7.5 mm
Matched 2K footprint
≈ 1.4 × 1.4 mm
Footprint grid
≈ 9 × 7
2K fields to match
≈ 63

On this coverage model, one selected field replaces≈ 63 stitched 2K fields

The 14,192 × 10,640 array sets a 4:3 footprint: approximately 9 × 7, or 63 matched 2K detector equivalents.

Why wider fields become a data-scale problemProjection + reconstruction

A wider sensor changes more than the image footprint. Every projection contains more pixels, while a reconstructed volume adds an entire spatial dimension.

Projection stackviews × width × height × bytes per pixel
Reconstructed volumewidth × width × height × bytes per voxel

10 mm system example. At 9,001 projections and 14,192 × 10,640 pixels, 16-bit projections are approximately 2.7 TB uncompressed. A full 14,192 × 14,192 × 10,640 reconstruction stored at 32 bits is approximately 8.6 TB.

Decimal estimates are shown to communicate scale. Actual stored output varies with cropping, reconstruction geometry, compression, and processing workflow.

03 / Center programs

Three foundations.
Four places to apply them.

The center plan organizes seven connected programs around one pipeline, from imaging physics to useful evidence in medicine, research, industry, and public understanding.

Foundations

Build the capability

  1. 01

    Whole-organism phenotyping

    Advance sample handling, optics, reconstruction, artifact correction, and visualization toward an automated histotomography pipeline.

  2. 02

    Geometry of Life and Disease

    Identify and segment structures, extract quantitative features, and model morphology and topology across species and conditions.

  3. 03

    Cyberinfrastructure

    Make very large images, workflows, and analytical tools usable through scalable storage, processing, and browser-based access.

Applications

Put it to work

  1. 04

    Precision imaging & cancer diagnostics

    Validate 3D tissue imaging against conventional histopathology and connect morphology with outcomes and molecular data.

  2. 05

    Drug safety

    Use whole-organism readouts to identify structural effects and toxicity earlier in preclinical development.

  3. 06

    Environmental sustainability

    Measure how environmental exposures alter cells and tissues in sentinel organisms.

  4. 07

    Health care & science outreach

    Turn complex anatomy and disease mechanisms into open, interactive resources for learners and the public.

Segmented zebrafish blood cells, wild-type and hht mutant overlays, and plots separating cells by genotype and age.
Blood cells segmented from wild-type and hht mutant zebrafish at 4 and 5 days post-fertilization.

04 / From morphology to statistics

When shape becomes a number, biology becomes a distribution.

Segment each cell, measure its geometry and spatial relationships, then compare those measurements across genotype, age, exposure, and disease. Features that specialists recognize by eye become evidence that can be tested.

  • Volume
  • Shape
  • Orientation
  • Topology

05 / Open the data

Do not take our word for it. Open the volume.

These are interactive datasets, not screenshots. A modern desktop browser can pan, zoom, slice in any plane, and follow structures through the depth of each specimen.

Segmented whole-body render of an octopus hatchling, with organs and nerves shown in colour.

Whole animal · 0.7 µm voxels

Octopus histotomography atlas

Explore a whole hatchling with a color-coded segmentation ontology, manuscript presets, and shareable viewer states.

Launch the atlas
Phase-contrast microCT slice of an unstained human prostate biopsy.

Human tissue

Prostate biopsy

Follow nuclei and glandular structures through the depth of an unstained needle-core biopsy.

Open the viewer

Model organisms

More imaging demonstrations

Explore zebrafish anatomy, teaching slides, and additional examples from the Cheng Lab.

View the collection

The viewers stream large research datasets and work best on desktop with a reliable connection.

Work with us

Bring a specimen, a question, or a hard measurement problem.

Computational phenomics only works across fields. We welcome conversations with researchers, trainees, clinical teams, institutions, and industry partners.

Start a conversation
01

Researchers

Plan pilot imaging, analysis, visualization, or a collaborative proposal.

02

Trainees

Ask about interdisciplinary opportunities spanning imaging, computation, and biology.

03

Partners

Explore work in instrumentation, diagnostics, pharmaceutical safety, agriculture, or AI.