DermaScan
The model runs on the device — no cloud

Scan. Analyse. Diagnose.

A handheld scanner photographs the skin, ranks the likely conditions on the device in under 50 ms, and shows the doctor the evidence behind each one — so the diagnosis stays theirs.

< 50 ms
on-device inference
95%
F1 score, our own model
Offline
no cloud, no upload
clinic.dermascan.local / results
The DermaScan doctor portal: a magnified image of the skin, the likely conditions ranked by confidence, and the evidence behind the top match.
The problem

Most skin problems never reach a specialist.

65%of skin conditions are seen by a non-dermatologistA GP is looking at a pigmented lesion and has to choose: reassure the patient, or refer and let them wait.
50%of dermatology diagnoses are inaccurateTwo clinicians examine the same lesion on the same day and write down two different things.
30%of melanoma cases are misdiagnosed initiallyThe one that mattered was called benign the first time somebody looked at it.

This isn't a training problem. It's an instrument problem.

How it works

How a scan becomes an answer.

Press the device. Nothing here is a video — the portal that opens is the product.

01Shown
Device scanTakes magnified photos, then analyses them on the device.
DermaScan 2.0
02Open
Doctor portalThe ranked candidates land on the doctor's screen.
03Press
Doctor decidesThe evidence behind each candidate, then the signature.
01Magnified photos of the skinFive images at 20x through medical-grade optics. Blurred shots are rejected before anything is ranked.
02TinyML inference, on the deviceThe model runs on the hardware in the clinician's hand. Under 50 ms, offline, no upload.
03Result leaves for the portalCaptures and ranked candidates go to the doctor's screen. Nothing goes to a cloud.

No cloud call sits anywhere in that loop. The model runs on the device, and the result travels to the doctor's screen on the clinic's own network.

System architecture

Data. Model. Device.

Everything the device does, in three stages.

01Data

Skin images of the four conditions the device detects.

02Model

Trained to tell them apart, then compressed.

03Device

Runs on the hardware in the clinician's hand — no cloud.

Conditions the device classifiesEczemaPsoriasisMelanomaNormal
The portal

What the doctor sees.

From the scan to a signed report, without leaving the desk. These are screenshots of the real product.

Every scan in the clinic, and which ones are still waiting on a doctor.

clinic.dermascan.local / worklist
DermaScan clinician workstation — Worklist. Every scan in the clinic, and which ones are still waiting on a doctor.
Privacy and clinical safety

Patient images never leave the clinic.

Inference runs on the deviceThe model is on the hardware in the clinician's hand. There is no inference server to call.
Works with the network downA clinic with no connection still scans, ranks and reports.
Captures stay in the clinicImages and results move between the device and the clinic's own workstation. Nothing is uploaded to us.
Nothing is recorded without a nameA scan is not part of the patient record until a doctor signs it, and the report says who signed and when.
It never says diagnosisThe screen says closest match and differential candidates. Recommendations are phrased as suggestions.
Not yet cleared for clinical useSFDA device approval is the next milestone on the roadmap. Deployments today are evaluation only.
Where it stands

Built and patented. Not yet approved.

We would rather tell you exactly which step we are on.

  1. DoneDermaScan 1.0Prototype v1 built and demonstrated.
  2. DonePatent filedFirst-phase acceptance received. SA1020253828.
  3. DoneITEX 2025, Malaysia2nd place globally. 2nd place, Monsha'at university startup competition.
  4. In developmentDermaScan 2.0MVP in development — the device and the portal on this page.
  5. NextSFDA device approvalThe next milestone, and the gate to clinical deployment.
The team
Test it

Try it in your own clinic.

We bring DermaScan 2.0, connect it to a machine in your room, and you scan. Twenty minutes, your cases, your doctors deciding whether the reasoning holds up.

Meet the team at LEAP 2026 — the device is on the stand.