Clinical Evidence and Validation
This page explains how the atopic dermatitis severity scoring technology is validated, what the published evidence shows, and the ongoing clinical validation programme.
Peer-reviewed validation: ASCORAD pilot study
The ASCORAD system was clinically validated in a pilot study published in JID Innovations:
"Automatic SCOring of Atopic Dermatitis using deep learning: A pilot study" Medela, A., Mac Carthy, T., Aguilar Robles, S. A., Chiesa-Estomba, C. M., Grimalt, R. JID Innovations, Volume 2, Issue 3, 100107 (2022) DOI: 10.1016/j.xjidi.2022.100107
The study demonstrated automated SCORAD scoring accuracy and reliability against expert dermatologist consensus, establishing the foundational evidence for automated AD severity scoring from photographs.
How the ground truth is established
Expert dermatologist SCORAD assessments performed independently. Each dermatologist scores the affected area extent (BSA), intensity of six clinical signs (0–3 each), and overall severity. The consensus (ground truth) is computed as the mathematical aggregate of their scores.
Why this approach: SCORAD scoring is inherently subjective, particularly for intensity signs. The best approximation of truth is the consensus of multiple experts. This is the same methodology used for validating ASCORAD across the other conditions in the Legit.Health platform.
Consistency advantage
Like all Legit.Health scoring systems, ASCORAD is perfectly reproducible: the identical image always produces the identical score, at every site, at every visit, with no calibration drift or intra-rater variability.
This consistency is particularly valuable for AD trials because:
- SCORAD's BSA component is the most variable: Manual BSA estimation is the single largest source of inter-rater disagreement. Pixel-level segmentation replaces this with an objective measurement.
- Six intensity signs compound variability: With six independently scored signs, even small per-sign disagreements accumulate into large SCORAD differences. AI scoring eliminates this compounding effect.
- Multi-site trials require cross-site comparability: Consistent scoring across all sites without calibration exercises reduces noise in the endpoint data.
Ongoing clinical validation
aEASI validation study
A prospective post-market clinical study (aEASI_HVN) is ongoing, extending the validation to the EASI (Eczema Area and Severity Index) endpoint:
| Parameter | Detail |
|---|---|
| Study code | aEASI_HVN |
| Endpoint | Automated EASI scoring |
| Design | Prospective, post-market clinical study |
| Objective | Clinical validation of automated EASI calculation |
| Status | Ongoing |
Retrospective image analysis programme
The system has been deployed in retrospective analysis programmes processing 85,000+ clinical images for automated visual sign detection across 22+ dermatological signs. This large-scale deployment demonstrates the system's ability to process high-volume clinical trial data reliably.
Regulatory-grade validation pathway
The clinical evidence follows a structured regulatory pathway:
| Standard | Scope | Application to AD scoring |
|---|---|---|
| IEC 62304 | Software lifecycle processes | The AI scoring pipeline follows a documented development lifecycle |
| ISO 14971 | Risk management | Systematic risk analysis for each scoring component |
| IEC 62366-1 | Usability engineering | Validated for both in-clinic and decentralised patient capture |
| MEDDEV 2.7/1 Rev 4 | Clinical evaluation | Clinical evidence compiled per structured methodology |
| MDR Annex XIV | Clinical evaluation and PMCF | Post-market clinical follow-up |
Related validation studies
The same AI architecture and methodology used for AD scoring has been validated across multiple dermatological conditions, with peer-reviewed publications for ALADIN (acne), APASI (psoriasis), and automated SALT scoring (alopecia). Cross-condition validation strengthens the evidence base for the underlying technology platform.
For the full list of clinical evidence across all indications, see the clinical validation section.