Technology
AI That Knows Its Boundaries
Our AI is built to observe: precisely, consistently, and transparently. It's not built to replace the clinician.
System at a glance
Vision Model
What is visible.
Clinician
What is clinically known.
System Logic
What can be inferred.
Dataset
A Model Built on Real Clinical Data
Epitello's core model is fine-tuned on more than 2,000 diabetic wound images, each paired with structured clinical inputs from real cases. Rather than training on generic image datasets, our model learned from the specific visual language of diabetic wounds: the way tissue types present, how wound edges behave, what early deterioration actually looks like.
This focused, domain-specific training is what allows the model to produce consistent, structured observations instead of vague, one-off impressions. The same wound, assessed the same way, every time.
Dataset composition
ConceptualPaired with
18
Structured clinical variables
2,000+
Clinical cases · wound images
Each case: a wound image paired with 18 structured clinical variables. Figures describe the Epitello development dataset.
Recognition
Recognized by the Global Wound Care Community
Awards since 2023
6
International awards earned
2026
Europe's Best Invention of the Year
Presented at the EWMA Congress
Our AI model was presented at the EWMA (European Wound Management Association) Congress, one of the field's leading international scientific platforms. It's a milestone that reflects rigorous validation and peer engagement, not just internal testing, and one of the clearest signals that Epitello's methodology is built to hold up under real clinical scrutiny.
That recognition has continued on the global stage, with Epitello earning six international awards since 2023, including Europe's Best Invention of the Year in 2026.
See full list below.
2026
Europe's Best Invention of the Year
—
International award
—
International award
—
International award
—
International award
—
International award
Architecture
A Three-Layer Architecture
Three distinct layers, kept deliberately separate so that observation, clinical knowledge, and inference never blur into one another.
Vision Model
What is visible?
- Tissue composition
- Necrosis
- Slough
- Granulation
- Wound-edge findings
- Image quality — each returned with a confidence score
Clinician Input
What is clinically known?
- Infection status
- Perfusion
- Exudate
- Charcot status
- Bone or tendon exposure
Clinical Decision Support
What can be responsibly inferred?
- Visual findings and clinical input are combined using structured, rules-based logic
- Supports next steps — it does not decide them
Never a black-box output the clinician has to simply trust.
The AI observes. The clinician knows. The system supports the decision.
Reliability
Built-In Reliability Checks
Not every photo is good enough to act on, and Epitello is built to know the difference. When image quality is poor or the model's confidence in a finding falls below a safe threshold, that observation is flagged for clinician confirmation rather than pushed forward as if it were certain.
The principle
Schema-valid output is not the same as clinically valid output.
Validation
Validated, Not Just Deployed
Epitello is tested well beyond raw model accuracy. Every release is checked across multiple validation layers, including output structure, reliability thresholds, and behavior consistency against prior versions, so that new capabilities never quietly change how the system already behaves.
Output structure
Every result is checked for valid, complete structure before it is trusted.
Reliability thresholds
Confidence and quality gates are enforced on every observation, not just on average.
Behavior consistency
New releases are compared against prior versions, so capabilities never quietly change how the system already behaves.
Release discipline
Each layer is a gate. A release only ships when every gate before it has passed — the engineering discipline behind a system clinicians can rely on.
AI-Assisted. Clinician-Led.
Support clinical judgment, not replace it
Epitello was not built to replace clinical judgment. It was built to support it with structure, consistency, and traceability. Every observation, every input, and every recommendation can be traced back to its source, so clinicians can trust not just the output, but the reasoning behind it.