Atopic Dermatitis Severity Endpoints for Clinical Trials
The AI scoring provided by Legit.Health delivers automated severity assessment for atopic dermatitis (AD) clinical trials, supporting both SCORAD and EASI — the two accepted primary efficacy endpoints in AD drug development.
Why automated AD scoring for clinical trials?
Manual SCORAD and EASI require a dermatologist to estimate affected body surface area (BSA), score intensity signs on each affected area, and (for SCORAD) record two subjective symptoms from the patient. Inter-rater variability in BSA estimation and intensity scoring is the primary source of noise in AD endpoints.
The AI automates the objective components of both scoring systems:
- BSA estimation: Pixel-level segmentation of affected skin across all photographed body areas
- Intensity scoring: Six objective signs for SCORAD / four signs for EASI — erythema, oedema/papulation, oozing/crusts, excoriations, lichenification, dryness — each scored 0–3 per the relevant methodology
- DIQA quality gate: Image quality assessed before scoring; substandard images flagged for recapture
For SCORAD, subjective components (pruritus NRS 0–10 and sleep disturbance NRS 0–10) are collected via patient-reported outcome instruments and combined with the AI objective score to produce the total SCORAD.
Endpoint capabilities
The AI provides four distinct endpoints, each configurable per study protocol:
| Endpoint | Definition | AI output | Typical use |
|---|---|---|---|
| Objective SCORAD | BSA × (intensity signs sum / 3) | 0–72 continuous scale | Co-primary |
| Total SCORAD | Objective + subjective (pruritus + sleep) | 0–103 composite | Co-primary |
| EASI | Eczema Area and Severity Index | 0–72 composite | Co-primary |
| IGA-AD | Investigator Global Assessment for AD | 0–4 ordinal | Co-primary |
Both SCORAD and EASI are accepted primary efficacy measures in AD clinical trials. SCORAD includes a subjective component (pruritus + sleep), while EASI is purely objective. The Legit.Health platform supports both. The aEASI validation study is ongoing.
How the AI works
The scoring pipeline processes each set of body area photographs in two stages:
Stage 1: BSA segmentation
A deep learning segmentation model identifies affected skin at the pixel level in each photograph. The segmentation output determines the percentage of the visible body area affected by eczema.
Input: body area photograph
Output: affected area at pixel level
Combined: segmentation overlaid on photograph
Stage 2: Intensity sign scoring
For each affected area, six clinical signs are scored independently on the 0–3 SCORAD scale:
| Sign | AI approach |
|---|---|
| Erythema | Colour analysis and boundary detection |
| Oedema/papulation | Texture and surface topology analysis |
| Oozing/crusts | Specific visual feature detection |
| Excoriations | Scratch mark pattern recognition |
| Lichenification | Thickened skin texture classification |
| Dryness | Surface appearance analysis |
Erythema intensity map
Scratch mark pattern recognition
Thickened skin texture
The per-sign scores are aggregated into the objective SCORAD component (maximum 72), which combines with BSA and subjective symptoms to produce the total SCORAD (maximum 103).