GramImage AI — AI bacterial species identification from a single Gram-stain field, with Gram class, genus and a calibrated top-5 in seconds.
GramImageAIDon’t wait for the lab to call back.
Seconds, not days
A calibrated first read the moment you see the smear — while culture and sequencing are still running.
A steadier call
Trained on sequencing-confirmed strains, it’s a second pair of eyes that trims the doubt and the reader-to-reader drift.
Wherever the smear is
Phone to eyepiece. No bench, no login. Night shift, remote clinic, or the teaching scope — it travels with you.
It won’t replace the workup — it gives you a head start on it: the confidence to move hours before the definitive answer lands.
What sequencing knows, vision reads.
Species identity is fixed by molecular methods — Oxford Nanopore NGS of the 16S rRNA gene and the rrn operon. The gold standard, and the high-reliability ground truth every training label is built on. Definitive — but hours to days out.
A self-supervised vision foundation model with a lightweight head, trained across 70+ reference species. It reads one stained field — or pools a few for stability — and predicts the identity sequencing would confirm. On device, in seconds.
We don’t sequence on a phone. The model learns the link instead: every training image is a strain whose identity is confirmed by Oxford Nanopore NGS, so DINOv2 learns to map a Gram-stain field to the answer the lab will reach — a calibrated first read at the very first step, while the definitive workup runs.
Capture
Photograph one field through the eyepiece — phone to ocular, or a slide scan.
Analyze
A DINOv2 vision model reads morphology, arrangement, and stain in real time.
Identify
Species, Gram class, and a calibrated confidence — surfaced on the spot.
- What is GramImage AI?
- GramImage AI is a vision tool that identifies bacterial species from a single Gram-stain microscope field. A self-supervised DINOv2 model reads the morphology, arrangement, and stain of one field and returns a calibrated first read — species, Gram class, genus, and the top-5 candidates. It is informational only and not a diagnostic device.
- How does it identify bacteria from a Gram stain?
- The model is image-only. DINOv2, a self-supervised vision backbone, is trained across 70+ reference species whose identity is confirmed by Oxford Nanopore NGS of the 16S rRNA gene and the rrn operon. It learns to map a Gram-stain field to the identity sequencing would confirm — no sequencing happens on the device.
- How many species can it identify?
- 72 species and resistance or phenotype variants — including ESBL, MR, and AmpC forms — across common Gram-positive, Gram-negative, and yeast organisms. The full list is shown in the web demo.
- Can it tell Gram-positive from Gram-negative?
- Yes. Alongside the species, GramImage AI returns the Gram class (positive or negative) and the genus. The web demo colours the Gram call — crystal violet for positive, safranin pink for negative.
- Does it work from a phone photo through the eyepiece?
- Yes. It is designed for a single field photographed through the microscope eyepiece — phone to ocular — or a slide scan. You can also pool up to five fields for a steadier read.
- Is it a diagnostic tool?
- No. GramImage AI gives a calibrated first read so you can move hours earlier — it does not replace the laboratory workup. Definitive identification comes from culture and sequencing. Do not enter patient-identifiable information.
- Can I try it for free?
- Yes. The web demo at remicrobes.com/demo is free: sign in with Google, upload a Gram-stain field or choose a sample, and get the result in your browser. No install required.
- Which platforms is it on?
- The free web demo runs in any modern browser today. A native iOS app with on-device inference and a free tier is coming soon to the App Store.
The lab’s eye, in your pocket.
Free tier · 20 analyses / month · no account required