Scoring Methodology

This page provides the technical detail on how Legit.Health computes the ASCORAD (automated SCORAD) and aEASI (automated EASI) scores for atopic dermatitis, from image input through to the final severity value.

The SCORAD formula

SCORAD (SCORing Atopic Dermatitis) is a composite measure comprising three components:

$$\text{SCORAD} = \frac{A}{5} + \frac{7B}{2} + C$$

where:

• $A$ = Extent (affected body surface area, 0–100%)
• $B$ = Intensity (sum of six sign scores, each 0–3; total 0–18)
• $C$ = Subjective symptoms (pruritus NRS 0–10 + sleep disturbance NRS 0–10; total 0–20)

The maximum SCORAD is 103. The objective SCORAD (A/5 + 7B/2) has a maximum of 83, and this is what the AI computes. Subjective symptoms are patient-reported.

BSA estimation (component A)

How the AI measures affected area

A deep learning segmentation model (DeepLabV3+ architecture) analyses each body area photograph and classifies each pixel as:

• Affected skin (eczematous lesion)
• Unaffected skin (normal or uninvolved)
• Non-skin (background, clothing, hair)

The percentage of affected skin relative to total visible skin area is computed per photograph. These per-image percentages are combined across all captured body areas using anatomical weighting to produce the total BSA (0–100%).

Why pixel-level BSA matters

In manual SCORAD, the BSA component is estimated using the "rule of nines" — a coarse visual estimation method. This is the single largest source of inter-rater variability in SCORAD scoring. Pixel-level segmentation replaces this estimation with an objective measurement, eliminating the most variable component of the manual score.

Intensity scoring (component B)

Six clinical signs

Each sign is scored on a 0–3 ordinal scale at the representative affected area:

Score	Meaning
0	Absent
1	Mild
2	Moderate
3	Severe

The six signs and how the AI scores each:

Sign	Description	AI methodology
Erythema	Redness of affected skin	Colour analysis calibrated to skin type, boundary detection
Oedema/papulation	Swelling and raised papules	Surface texture and topology analysis from image features
Oozing/crusts	Weeping or crusted lesions	Specific visual feature detection for wet/crusted surfaces
Excoriations	Scratch marks from pruritus	Linear pattern recognition for scratch marks
Lichenification	Thickened, leathery skin from chronic scratching	Texture classification for thickened skin patterns
Dryness	Xerosis of uninvolved skin	Surface appearance analysis compared to normal skin texture

The intensity total $B$ is the sum of all six sign scores (0–18).

Severity thresholds

SCORAD range	Severity classification
0–24	Mild
25–50	Moderate
>50	Severe

These thresholds are the standard SCORAD severity bands used in clinical trials.

EASI scoring

The Eczema Area and Severity Index (EASI) is a co-accepted primary efficacy measure for AD, alongside SCORAD. EASI scores four signs (erythema, induration/papulation, excoriation, lichenification) across four body regions, each weighted by regional BSA.

$$\text{EASI} = \sum_{r} w_r \times \text{BSA}_r \times \left(\text{erythema}_r + \text{induration}_r + \text{excoriation}_r + \text{lichenification}_r\right)$$

EASI range	Severity classification
0–1	Clear
1–7	Mild
7–21	Moderate
>21	Severe

SCORAD vs. EASI: when to use which

Aspect	SCORAD	EASI
Subjective component	Yes — pruritus + sleep NRS (patient-reported)	No — purely objective
Scale	0–103 (objective 0–83)	0–72
Body regions	Whole body + representative lesion area	4 defined regions (head, trunk, upper/lower limbs)
Validation status	Published (JID Innovations 2022)	Ongoing (aEASI_HVN study)
Regulatory acceptance	FDA and EMA accepted	FDA and EMA accepted

Both SCORAD and EASI are fully supported by the Legit.Health platform. The choice is made during protocol design based on sponsor preference, prior precedent in the indication, and regulatory strategy.

Manual vs. automated scoring

Aspect	Manual (SCORAD/EASI)	Automated (ASCORAD/aEASI)
BSA estimation	Visual "rule of nines" (±15–20% variability)	Pixel-level segmentation (objective)
Sign scoring	Subjective 0–3 per sign	AI-computed 0–3 per sign, calibrated against expert consensus
Reproducibility	Inter-rater variability across sites	Identical score for identical image, every time
Time per assessment	5–10 minutes (manual scoring)	<2 seconds (AI processing)
Subjective symptoms	Patient-reported (SCORAD only)	Patient-reported (unchanged; EASI has no subjective component)
Training required	Calibration exercises across sites	None (AI is the rater)