B2B · Teledermatology Infrastructure

Stop reviewing
unusable
patient images.

Dermatolify automatically validates the technical quality of patient-submitted skin photographs — before they reach the dermatologist. Three-stage pipeline. REST API. Fully auditable.

Not a medical device (MDR) DSGVO compliant FHIR R4 ready
40–50%
of patient images are diagnostically unusable
ImageQX Study, PMC/PubMed 2023
1,900
lost per dermatologist monthly
in non-billable review time
96%
sensitivity on independent test set
n=100, held-out, 2026-04-16
<1s
per image validation
all three stages combined
The Problem

Manual image rejection
is killing physician throughput.

In async teledermatology, patients submit photos without clinical supervision. Dermatologists must open, assess, and reject unusable images — a step that generates zero revenue.

📷

Wrong subject

Patients photograph feet, furniture, or pets instead of the affected skin area. No automated system currently catches this before the physician queue.

🌫️

Motion blur & focus issues

Handheld smartphone photography produces a high rate of blurry images — especially with close-up macro shots. Standard quality checks miss this.

🌑

Bad lighting

Overexposed or underexposed images hide the very features a dermatologist needs to assess. Dark skin tones are disproportionately affected.

📐

Insufficient resolution

Low-resolution images uploaded by older devices or compressed by messaging apps fall below the minimum threshold for remote diagnosis.

40–50%
of uploads are unusable
(published studies)
8–12
cases per day per physician
~40% rejection rate
~1,900 €
monthly revenue loss
per dermatologist
Three-Stage Pipeline

From upload to decision
in under a second.

Every image passes three independent validation stages. A rejection at any stage returns an actionable reason to the patient immediately.

1

Subject Recognition

A frozen classification model (DINOv2 + SVM) checks whether the image shows a recognisable skin region before any pixel-level analysis is performed. Non-skin images are rejected immediately.

Privacy: image → embedding → image discarded. No pixel data stored or transmitted.

DINOv2 (facebook/dinov2-small) · 384-dim CLS token · RBF-SVM
2

Pixel-Level Metrics

Five deterministic metrics are extracted with Pillow / NumPy / SciPy: resolution, total blur score, skin-region blur score, brightness, and file size. Fully reproducible — no ML inference.

MetricMethod
ResolutionTotal pixel count
BrightnessMean luminance
Total Blur ScoreBrightness-normalised Laplacian
Skin Blur ScoreSobel P95 within HSV skin mask
File SizeRaw file size
3

DMN Decision Table

Metrics are evaluated against a human-readable OMG DMN 1.3 rule file (FIRST-hit policy, FEEL syntax). Every rejection is traceable to a named rule and a measured value.

Thresholds are updated in the DMN file — no code change required. Fully auditable for MDR and FDA 510(k) reviews.

OMG DMN 1.3 · FIRST-hit policy · FEEL syntax · XML
Validated Performance

Numbers that hold up
on unseen data.

Independently validated on a completely held-out test set — images never seen during training. LOO-CV results confirmed.

Independent Test Set · n=100 · 2026-04-16
Sensitivity 96.0%
Specificity 90.0%
Accuracy 93.0%
50 positive (skin region, good quality) + 50 negative (wrong subject or poor quality). Completely held-out — not used in any training or calibration run.
Training · LOO-CV · n=1,000
Sensitivity 97.4%
Specificity 89.8%
SCIN Public Dataset (Google Research, 2024, CC-BY) · 548 pos / 452 neg · 9 training rounds
Additional Metrics · Independent Test
90.6%
PPV (Precision)
95.7%
NPV
Agreement between LOO-CV and independent test rules out overfitting.
Regulatory

Not a medical device.
By design.

Dermatolify is a quality assurance pre-filter, not a diagnostic tool. This distinction is foundational — documented from day one.

  • Classified as non-MDSW under MDCG 2019-11 — no MDR registration required
  • Does not diagnose, classify skin conditions, or interpret clinical findings
  • All metric decisions are deterministic and traceable to named DMN rules
  • The SVM model is frozen and versioned — inference is fully reproducible
  • Privacy by design: pixel data is never stored or transmitted (DSGVO-compliant)
  • Applicable to Swiss MepV for cross-border deployments — regulatory assessment available

Intended Purpose

Dermatolify is intended to automatically assess the technical image quality and formal suitability of patient-submitted skin photographs prior to clinical review.

The system checks criteria such as resolution, exposure, sharpness, framing, and the presence of a recognisable skin region. It determines whether an image meets the minimum requirements for teledermatological evaluation.

It does not make diagnoses, classify skin conditions, or interpret dermatological findings. Clinical assessment remains exclusively with qualified medical professionals.

DL-IP-001 · DL-RA-001 · MDCG 2019-11

Patient Data Privacy

No patient image is ever stored.

Dermatolify processes images in memory only. The moment analysis is complete, all pixel data is discarded. Nothing is written to disk, no image is cached, no data is transmitted to third parties.

🔒
Image → Metrics → Image discarded
Only five numerical metrics leave the analysis step. The original image never touches permanent storage.
🇪🇺
DSGVO-compliant by architecture
Privacy is enforced in code, not policy. There is no configuration that could accidentally enable image retention.
🏥
No patient identifiers processed
Dermatolify receives an image file. It returns a quality decision. No name, ID, or metadata is required or retained.
Integration

Fits into your existing stack.

Dermatolify is a stateless microservice. Plug it into your upload pipeline in an afternoon.

REST API

FastAPI-based endpoints for single image, batch, and FHIR validation. JSON responses with structured rejection reasons.

POST /validate POST /validate/batch POST /validate/fhir GET /health
🔗

FHIR R4

The /validate/fhir endpoint returns a standards-compliant FHIR R4 Observation resource — including all five metrics and SVM confidence as auditable components.

Native EHR integration. Meets § 371 SGB V interoperability requirements.

🐳

Docker · On-Premise

Fully containerised. The DMN rule file is mounted as a read-only volume — update thresholds without rebuilding the image.

docker compose up

Available as cloud-hosted SaaS or on-premise licensed deployment.

Pricing

Simple, transparent licensing.

Non-exclusive SaaS licence. IP remains with Dermatolify. No code shared without a signed agreement.

Pilot Phase
1,000/month
For teledermatology platforms running a clinical evaluation. Fixed monthly fee, no per-image charges.
  • Full 3-stage pipeline via REST API
  • FHIR R4 output endpoint
  • Regulatory documentation package
  • Direct access to founder for integration support
  • Shared SaaS deployment
Request Pilot

Ready to stop reviewing bad images?

We're currently onboarding pilot partners. Integration takes less than a day. No ML expertise required on your end.