Sample Outputs and Deliverables

This page shows the concrete outputs that sponsors, CROs, and investigators receive from each acne severity assessment. The AI generates standardised reports immediately after image submission — no central reader, no delays.

Per-visit assessment report

Each image submission generates a complete severity report, accessible to the investigator within seconds:

What the report contains

Output	Detail
Global IGA score	0–4 integer aligned with the FDA co-primary endpoint
Global ALADIN score	0–10 continuous composite (IGA × 2.5), for secondary/exploratory use
Per-perspective lesion count	Individual inflammatory lesion count (papules, pustules, nodules) per half-face
Per-perspective density	Spatial density ratio (0–1) measuring lesion clustering
Per-perspective local IGA	Local severity score for each captured perspective
Annotated images	Each perspective with bounding boxes around detected inflammatory lesions
Image quality (DIQA)	Quality score per image; images below threshold are flagged for recapture
Timestamp	UTC timestamp of image capture and AI processing

Why annotated images matter

The bounding box overlays allow investigators to visually verify the AI's detections against their clinical observation. This audit trail supports regulatory inspection and gives sites confidence in what the system produces.

Longitudinal severity tracking

Across visits, the platform tracks severity evolution from screening to end of study:

What longitudinal tracking delivers

Score trajectory: IGA and ALADIN at every visit, from screening to end of study
Treatment response flag: IGA ≥ 2-grade improvement from baseline (configurable threshold)
Absolute change: Point change in IGA and ALADIN vs. baseline at each visit
Visual chart: Graphical severity curve across all timepoints

This enables investigators and sponsors to identify responders and non-responders during the trial — not just at database lock.

Data export for EDC integration

All assessment outputs are structured for export to the sponsor's EDC system. Fields exported at each visit:

Field	Description	Format
Patient identifier	Study-specific pseudonymised ID	String
Visit date	Date and timestamp of the assessment	ISO 8601
Global IGA score	IGA severity (0–4)	Integer
Global ALADIN score	Composite severity (0–10)	Float
Global lesion count	Total inflammatory lesions (deduplicated across perspectives)	Integer
Per-perspective IGA	Local IGA score for each perspective	Integer
Per-perspective lesion count	Lesions detected per perspective	Integer
Per-perspective density	Spatial density per perspective	Float (0–1)
DIQA scores	Image quality score per perspective	Float
Severity classification	Clear / Almost clear / Mild / Moderate / Severe	String

Data is transferred automatically via RESTful API or scheduled S3 export. Formats include structured JSON and CSV/Excel. Legit.Health provides IQ/OQ documentation and data mapping specifications.

AI visual outputs

Beyond scores, the AI produces visual artefacts that demonstrate exactly what the system detected and how it reached its conclusion. These are available in every per-visit report.

Lesion detection with bounding boxes

Each inflammatory lesion (papule, pustule, nodule) is identified and highlighted with a bounding box overlay on the original photograph. Investigators can verify the AI's detections at a glance.

Original photograph

Input: standardised facial photograph

Bounding box overlay

Output: each detected inflammatory lesion highlighted

Facial landmark detection and pose estimation

The AI maps facial landmarks (eyes, nose, mouth, jawline) to identify the capture perspective, define region boundaries, and deduplicate lesions across overlapping perspectives.

Landmark mesh overlay

Facial landmark mesh with region boundaries

Region exclusion mask

Perspective-specific exclusion zone for lesion deduplication

Anonymization

All photographs are processed with automatic face anonymization. The system detects facial features and applies irreversible blurring to ensure no patient is identifiable from the stored or exported images.

Before anonymization

Original capture (eyes, nose, mouth visible)

After anonymization

Irreversible face blurring applied

Reliability of the scores

The ALADIN scoring system achieves Cohen's κ = 0.53 against the dermatologist consensus — matching or exceeding the typical inter-rater agreement of individual expert dermatologists (κ = 0.46). This means the AI is at least as consistent a rater as adding a board-certified dermatologist to the panel.

Critically, unlike a human rater, the AI produces the identical score for the same image every time, across every site, without calibration drift or fatigue. There is zero intra-rater variability.

For the full validation evidence: Clinical Evidence →

Per-visit assessment report​

What the report contains​

Longitudinal severity tracking​

What longitudinal tracking delivers​

Data export for EDC integration​

AI visual outputs​

Lesion detection with bounding boxes​

Facial landmark detection and pose estimation​

Anonymization​

Reliability of the scores​