Cloud-native image analysis

Microscopy data analysis
at scale.

Visualize, segment, track, and classify your imaging data in hours, not weeks. Access state-of-the-art AI models on secure cloud GPUs. Forget about expensive infrastructure, Python scripts, and fragmented tools.

ScaleTerabytes, streamed to the browser SpeedRun experiments in hours, not weeks Open formats.czi · .lif · .nd2 · OME-TIFF · OME-Zarr
project / organoid-coculture-02 /day-3_t48h.nd2
5D · 2.3 TB
Layers
T-cells (488)
Organoid (561)
Dead marker
Brightfield
Channels · 488 nm
Dimensions
X × Y2048 × 2048
Z120 slices
T96 frames
XY · Z 60/120
t=42/96 · x=1024 y=998
50 μm
Selection · T-cell #4261
Track length4h 12m
Path328 μm
Contacts7
Kill events2
Phenotype
Serial killer
Confidence 94%
Population · n=412
killer · 38
engager · 127
scanner · 184
static · 63
5D
time · z · channels · xy
10×
faster than desktop pipelines
100%
reproducible — versioned workflows
0lines
of code to get results
/01 — Platform

The tools imaging labs need, in one integrated platform.

Stop cobbling together desktop software, notebooks, and scripts. CellWorkflow unifies visualization, segmentation, tracking, AI training, and quantitative analysis in one cloud-native pipeline — so every step is traceable and reproducible, from raw microscope file to quantified biological insight.

01

Visualize anywhere

Explore 2D and 3D datasets in your browser — any device, any size. Navigate Z-stacks, time-lapses, and multi-channel data at terabyte scale, streamed directly from the cloud.

XY · Z 60/120T 42/96 50 μm
02

Segment organoids and T cells

Automated 3D segmentation of tumor organoids, tumor cells, and individual T cells — with proofreading tools for edge cases. Replaces hand-drawn 3D surfaces and reconciliation across annotators.

RAWBRUSHMASK • 412
03

Train & fine-tune AI

Start with pre-trained segmentation models, or fine-tune them on your own data with a few clicks. No notebooks, no GPU provisioning — training runs in the cloud, progress updates live.

0.80.40.1 TRAIN LOSSVAL LOSS EPOCH 24/30
04

Collaborate & publish

Share datasets with a link, invite annotators, and review results together. Export figure-ready outputs and link from your paper directly into the data — so readers can explore your findings.

RL MK JS AP JS ⎘ cellwf.io/s/a3f9…
Scientific partners
imAIgene Lab
Princess Máxima Center
/02 — Workflow

From raw microscope files to quantified results in hours.

One pipeline, versioned at every step. Upload the files your microscope already exports; get back segmentations, tracks, and quantified phenotypes — plus the figures and tables — that would have taken weeks to assemble by hand, all shareable with collaborators and reviewers at a link.

01

Upload

Your 3D time-lapse, in the formats your microscope already exports. Zeiss .czi, Leica .lif, Nikon .nd2, OME-TIFF, OME-Zarr — we handle conversion and storage in an automated pipeline that you only need to run once.

CLOUD
.czi.lif.nd2OME-TIFFOME-Zarr
02

Analyze

Segmentation, 3D tracking through division and death, and unsupervised classification. The full pipeline runs end-to-end on managed cloud GPUs — typically in hours, not the 2–4 weeks of manual execution.

SegmentTrackClassifyCorrelate
03

Report

Quantitative measurements, population signatures, per-track statistics, and exportable tables — plus publication-ready figures — ready to drop into a paper, a lab meeting, or a grant report.

CSV ↓
MeasurementsPlotsFiguresCSV
/03 — Quantitative analysis

Quantify what your cells are actually doing.

Turn every segmented object and track into a quantitative phenotype — not just a count. Cluster cells by morphology, motility, intensity, or interaction, with confidence scores and editable labels. The example below: an immune-cell killing assay, phenotyped end-to-end with the BEHAV3D classifier.

n = 412 cells · 4h co-culture

Population breakdown

Example — CAR-T against patient-derived colorectal organoid, replicate 03

Serial killers
Engage multiple targets, induce cell death
9.2%
Engagers
Form stable contacts without killing
30.8%
Scanners
High motility, transient contacts
44.6%
Static
No migration, no productive contacts
15.4%
Tracks · phenotype-colored · t 0 → 4h
/04 — Applications

Built for imaging labs across fields.

From developmental biology to high-content screening to immuno-oncology, CellWorkflow replaces weeks of manual analysis with a repeatable, quantitative readout — at terabyte scale, without the infrastructure.

01 — Developmental biology

Morphogenesis and lineage over days

Follow cells through division, migration, and differentiation in developing embryos, organoids, and tissues. Maintain single-cell identity across long 3D time-lapses and quantify proliferation, fate, and movement — no manual frame-by-frame correction.

Lineage tracking · Organoids
02 — High-content screening

Phenotype thousands of conditions

Segment and measure cells across plates and perturbation panels. Score morphology, intensity, and dynamics to rank hits — reproducibly, at terabyte scale, with the same pipeline applied identically to every well.

HCS · Perturbation panels
03 — Immuno-oncology

Immune-cell behavior in co-culture

The platform's origin: classify how T cells engage and kill tumor organoids — serial killers, engagers, scanners, static — with the BEHAV3D classifier, to compare CAR-T, TCR-T, and TIL products by behavioral signature rather than endpoint killing alone.

Co-culture · Cell therapy
/05 — Model Hub

State-of-the-art models, one click away.

Stop waiting months for the next model to reach your institution. We curate, benchmark, and package the latest open-source releases — MicroSAM, CellPose, StarDist, Trackastra, and more — ready to run on any dataset, on cloud GPUs we manage for you. Catalog is regularly refreshed.

Curated for microscopy · Regularly refreshed Segmentation · Tracking · Classification
New
Jug Lab · Human Technopole

Trackastra

Transformer-based cell tracking that links segmented instances across time — robust to division, death, and crowded co-culture conditions. Used to build the tracks that feed behavioral classification.

2D / 3D / T Linker Divisions
Updated · v3.1
Stringer & Pachitariu · HHMI

CellPose

The lab workhorse. Generalist model for cells, nuclei, and organelles across brightfield, fluorescence, and phase — plus Human-in-the-Loop fine-tuning.

2D / 3D Fine-tune F10.91
New
Van Valen Lab · Caltech

CellSAM

Foundation model for cell segmentation, trained on 1M+ annotated cells across tissue, imaging modalities, and species. Strong few-shot transfer.

2D Few-shot F10.84
v0.9
Weigert & Schmidt · MPI Dresden

StarDist

Star-convex polygon detection — the gold standard for dense nuclei segmentation. Pre-trained for fluorescence, H&E, and 3D Z-stacks.

2D / 3D Nuclei F10.89
v1.0
Rios Lab · Princess Máxima Center

BEHAV3D classifier

Unsupervised behavioral classifier for T cell–organoid co-cultures — the original BEHAV3D model from Alieva 2022 / Dekkers 2024. Recovers serial killers, super-engagers, scanners, and static cells from track features.

3D / T Behavioral UMAP
+ more TrackMate · Btrack · Cellulus · Mesmer · DeepCell · nnU-Net · plus your own weights.
— Field note

"The bottleneck was never the imaging — it was running the analysis. Weeks of manual work per experiment, stitched across desktop software, notebooks, and scripts. A cloud platform that productionizes the whole pipeline turns that into part of the experimental loop, not a quarterly project."

SA
Scientific advisors Co-developers of BEHAV3D · Princess Máxima Center / imAIgene Lab
/06 — Scientific team

Built by the people who built BEHAV3D.

CellWorkflow exists because the original BEHAV3D authors wanted the pipeline they pioneered to be usable by every imaging lab — not just the ones that can dedicate a senior analyst to it for two to four weeks per dataset.

Scientific advisor

Dr. Anne C. Rios

Group leader at the Princess Máxima Center for Pediatric Oncology, senior author of the BEHAV3D platform papers, and pioneer of 3D large-scale single-cell imaging of patient-derived organoids and immune co-cultures. Closely involved in shaping the platform's behavioral readouts.

Scientific advisor

Dr. Maria Alieva

First author of BEHAV3D: a 3D imaging-transcriptomics platform reveals heterogeneous cytotoxic T cell behaviors (Nature Biotechnology, 2022) and senior author of the BEHAV3D Nature Protocols paper (Dekkers et al., 2024). Established the behavioral phenotyping framework CellWorkflow productionizes.

CellWorkflow is built in collaboration with — but is independent from — the Princess Máxima Center and imAIgene Lab. Advisors do not endorse the platform's commercial claims.

/07 — Pricing

Start free. Scale when the grant lands.

Every plan includes the full viewer, annotation tools, and access to every model in the Hub. GPU credits and storage scale as your pipelines grow.

Fully managed on our infrastructure — the fastest way to get running. Your data stays encrypted, private by default, and never leaves the region you choose.

Free
$0forever
Sign up in a minute and try CellWorkflow on a pilot dataset.
  • 1 user
  • 100 GB storage
  • 1 workflow
  • Access to all models & tools
  • Shared GPU queueSlower during peak hours
  • Email support
  • Inactive accounts & data deleted after 30 days
Request access
Starter
$2,400/ year
For small labs standardizing on a shared pipeline.
  • 3 users, shared workspace
  • 1 TB storage
  • 5 workflows
  • Access to all models & tools
  • 200 GPU-hours / year includedThen $3 / GPU-hour
  • Priority GPU queue
  • Email support
Request access
Enterprise
Customannual · talk to us
For biotechs, CROs, and pharma running discovery and translational programs at scale.
  • Dedicated GPUs · reserved capacity
  • Unlimited users, storage & workflowsStorage capped at 1 PB
  • Unlimited computeNegotiated rate · best-in-class GPUs
  • Custom SLA · SSO
  • Dedicated support · 24h response
  • On-prem / Private cloud available+$30k–50k / year
Request access

Run your first dataset
by the end of the week.

Upload one file from a recent experiment — any modality, any size. We'll run segmentation, tracking, and analysis end-to-end, and walk you through the results on a 30-minute call with a scientist, not a sales rep.