Research & Recognition

Grounded in peer-reviewed research.
Hardened in production.

Most capabilities in OptiFlow began as published, peer-reviewed research and were proven through deployment in demanding imaging environments — from single-cell optical microscopy to semiconductor electron microscopy. The methods are domain-independent: wherever an image holds an answer, the same rigour applies.

In production at ETH Zürich Ongoing trials at NTU 20+ peer-reviewed publications Featured by CNA & The Straits Times
By the numbers

A decade of evidence.

Not a launch bet. OptiFlow rests on years of published research and shipped systems — a track record you can read, cite and verify.

20+
Peer-reviewed publications
1,200+
Citations of that work
10+
Years advancing imaging AI
15+
Imaging modalities worked across
Why it matters to you

What the research proves.

The subject matter varies, but the findings are general. Each result below is a capability you can put to work — whatever your images are of.

Domain-informed models

Encoding the problem beats generic AI

Incorporating the physics of the imaging problem into the model — rather than reaching for an off-the-shelf network — lifted a published disease-classification result from 92% to 97% accuracy.

Applies to any modality with characteristic image physics — the essence of a domain-aware platform.
Label-free imaging

Answers without labels or markers

Published work identifies cancer cells from label-free images at 95% accuracy in under 35 minutes; the same principle drives virtual staining, now in production at ETH Zürich.

Extract the signal without destructive, slow or expensive sample preparation.
Reference-free detection

Find the anomaly with no template

Anomaly and defect detection on electron-microscopy imagery flags what is wrong without a golden reference — separating known failure classes from entirely novel ones.

Transfers directly to industrial inspection, quality control and any surface others can't pre-define.
Production throughput

Accuracy that still runs at line speed

Selective high-resolution acquisition, on-device latency reduction and real-time embedded processing keep analytical depth without the throughput ceiling of scanning everything at full resolution.

Built to deploy on real instruments and lines — not only to publish.
The track record

Peer-reviewed. Production-proven.

Rigour you can check in the literature, and delivery you can rely on in the field. Both feed the same platform.

Validated in peer review

Published & cited

Selected publications in AI-driven imaging and image-based diagnostics. The complete list of 20+ is on the full record.

Featured · APL Bioengineering, 2021
Machine-learning pH imaging and classification of single cells
Identifying cancer cells from label-free images of intracellular acidity — 95% accuracy in under 35 minutes, no fluorescent markers. Covered by The Straits Times, CNA and international press.
Read paper
AIP Advances, 2023
Deep CNN classifies bladder-cancer cells by pH fingerprint & morphology
A convolutional network separating cancer cell types from their acidity signature and shape — extending single-cell classification to new disease targets.
Read paper
Scientific Reports, 2021
Single-cell chromatin biomarkers for tumour progression
An imaging pipeline that grades breast-tissue biopsies by cancer stage from nuclear and chromatin features — AI-based grading at clinical resolution.
Read paper
Manuscript in preparation
Virtual nuclear staining from transmitted-light microscopy
Vision generative AI that renders virtual staining directly from label-free images — no chemistry, no phototoxicity. The engine deployed commercially at ETH Zürich.
Hardened in production

Built & deployed

Applied engineering on commercial instruments and embedded systems, where accuracy and throughput are equally non-negotiable.

Deployed · SEM inspection
Reference-free wafer defect detection
End-to-end detection on electron microscopy — supervised classification of known defects plus unsupervised anomaly detection for novel failure modes. Running on a commercial production instrument.
Autonomous imaging
Selective high-resolution target acquisition
Scans fast at low resolution, then acquires high-resolution images only at the regions that matter — removing the throughput ceiling of full-field high-res scanning while preserving analytical depth.
Embedded & real-time
Low-latency inference on-device
On-device latency reduction through quantisation, weight sharing and attention design, plus low-footprint plugins for live-stream processing in embedded systems.
Commercial deployment · ETH Zürich
Virtual staining in daily laboratory use
An academic collaboration converted into a commercial deployment: label-free virtual staining developing biomarkers at single-cell resolution in a working lab.
Recognition

In the press.

Founding research in AI-driven imaging, covered by Singapore's flagship broadcast and print media and by specialist AI programmes internationally.

The full record

This is the highlight reel.

The complete record — every peer-reviewed publication, collaboration, conference contribution and press mention — is laid out in full, filterable and linked to source. Read it, cite it, check it.

See whether it applies to your images.

Bring an imaging problem and your own data — from any field. We'll show you what OptiFlow does with it, and if your workflow needs something we haven't built yet, we'll build it.

In production at ETH Zürich Ongoing trials at NTU 20+ peer-reviewed publications Featured by CNA & The Straits Times