Accuracy

How we validate the analysis

Methodology

Model scores were compared with clinicians' average scores on the same scale.

What the figure means

It is agreement with clinicians' assessment — not a share of correct diagnoses and not "diagnostic accuracy". Scanderm assesses the aesthetic condition of skin.

Limitations

The result was obtained on a sample of 88 photographs and does not automatically transfer to any shooting conditions.
Analysis quality depends on lighting, angle and photo quality.
The analysis does not replace a doctor's consultation and is not a diagnosis.

Validation results by module

Facial skin

94,8%

agreement with the average assessment of clinicians

88 new photos, 0–10 scale, panel of 5 doctors (Oct 2025). 94.5% of scores within 3 points, 86% within 2; mean error 1.1 points vs 0.57 spread between doctors; panel ICC 0.914.

Teeth (signs of caries)

91,9% / 77,0%

per-photo sensitivity / specificity

Internal validation: 250 photos, 2,194 teeth, 433 doctor-marked lesions. Per photo: NPV 90.7%. Per tooth: sensitivity 70.8%, specificity 95.6%.

Preliminary screening, not a medical device.

Results by feature group

Calculation and study protocol available on request.