Scoring Methodology

This page provides the technical detail on how Legit.Health computes the AIHS4 score for hidradenitis suppurativa, from image input through to the final IHS4 value.

The IHS4 formula

The International Hidradenitis Suppurativa Severity Score (IHS4) is a count-based severity measure:

$$\text{IHS4} = (1 \times N_{\text{nodules}}) + (2 \times N_{\text{abscesses}}) + (4 \times N_{\text{fistulae}})$$

where $N_{\text{nodules}}$ is the number of inflammatory nodules and papules, $N_{\text{abscesses}}$ is the number of abscesses, and $N_{\text{fistulae}}$ is the number of draining fistulae (tunnels). The weighting reflects clinical severity: fistulae indicate more advanced disease than abscesses, which indicate more advanced disease than nodules.

Lesion detection pipeline

The AI processes each photograph through a deep learning object detection model that identifies and classifies HS lesions:

Step 1: Lesion localisation

The model scans the image and identifies regions containing HS lesions, drawing bounding boxes around each detected lesion.

Step 2: Lesion type classification

Each detected lesion is classified into one of three types:

Lesion type	Visual features used by AI	IHS4 weight
Papules/nodules	Solid, raised, well-defined borders, uniform colour	×1
Abscesses	Fluctuant appearance, surrounding erythema, irregular surface	×2
Draining fistulae	Linear tracts, surface openings, discharge patterns	×4

Step 3: Count aggregation

Lesion counts are aggregated per anatomical region and globally. The IHS4 score is computed from the weighted sum.

Severity thresholds

IHS4 range	Severity	Clinical interpretation
0–3	Mild	Few lesions, predominantly nodules
4–10	Moderate	Multiple lesions, may include abscesses
≥ 11	Severe	Significant lesion burden, often with fistulae

Hurley staging

In addition to IHS4, the AI can estimate the Hurley stage based on lesion distribution and type:

Stage	Description	AI indicators
Hurley I	Abscess formation, single or multiple, without sinus tracts or cicatrisation	Abscesses without fistulae
Hurley II	Recurrent abscesses with tract formation and cicatrisation, single or multiple, widely separated	Abscesses + limited fistulae
Hurley III	Diffuse or near-diffuse involvement with multiple interconnected tracts and abscesses	Multiple fistulae + dense lesion clusters

Manual vs. automated scoring

Aspect	Manual IHS4	AIHS4
Lesion identification	Visual + palpation	Visual detection from photographs
Type classification	Clinician judgement	AI classification from visual features
Inter-rater ICC	0.47 (literature)	0.727 (M-27134-01 trial)
Reproducibility	Variable across sites	Identical score for identical image
Time per assessment	3–5 minutes	<2 seconds
Training required	Calibration exercises	None (AI is the rater)

The AI's ICC of 0.727 represents a 55% improvement over the manual inter-rater ICC of 0.47, demonstrating that automated scoring substantially reduces measurement noise in HS clinical trials.