Scoring Methodology
This page sets out, item by item, how Legit.Health quantifies the components of the CLASI, from a single-region photograph through to an aggregated Activity and Damage score.
The CLASI instrument
The CLASI (Cutaneous Lupus Erythematosus Disease Area and Severity Index) scores the skin in 13 anatomical regions, each assessed independently and summed without area weighting:
| Group | Regions |
|---|---|
| Head | Scalp; ears; nose, including the malar area; rest of the face |
| Neck | V-area of the neck; posterior neck and shoulders |
| Trunk | Chest; abdomen; back and buttocks |
| Limbs | Arms; hands; legs; feet |
Two further items are scored on the scalp by quadrant, and two are answered once per patient rather than per region.
CLASI Activity
| Item | Scale | Scored |
|---|---|---|
| Erythema | 0–3 | Each of the 13 regions |
| Scale and hypertrophy | 0–2 | Each of the 13 regions |
| Non-scarring alopecia | 0–3 | Scalp, by quadrant |
| Recent hair loss | 0–1 | Once per patient |
| Mucous membrane lesions | 0–1 | Once per patient |
The erythema scale runs from absent, through faint pink and red, to dark red or violaceous. Scale and hypertrophy distinguishes absent, scaling, and verrucous or hypertrophic change. The Activity score ranges from 0 to 70.
CLASI Damage
| Item | Scale | Scored |
|---|---|---|
| Dyspigmentation | 0–1 | Each of the 13 regions |
| Scarring, atrophy and panniculitis | 0–2 | Each of the 13 regions |
| Scalp scarring | 0, 3, 4, 5, 6 | Scalp, by quadrant |
The dyspigmentation subtotal is doubled when dyspigmentation in that patient typically persists beyond 12 months, a patient-level history question rather than an observation. The Damage score ranges from 0 to 56.
How the AI quantifies each component
Every measurement below is produced from one photograph of one anatomical region. The device returns per-sign values together with pixel-level segmentation masks and a DIQA image quality score.
Erythema
Erythema is the dominant contributor to CLASI Activity, and it is scored by judging colour by eye against a four-point verbal scale. That combination, the heaviest item resolved by the least reproducible kind of judgement, is where automated measurement adds most.
The device returns two distinct quantities from the same photograph. Erythema intensity is a graded severity value, the direct analogue of the CLASI 0–3 item. Erythema extent is the erythematous area, derived from a pixel-level segmentation mask and reported either relative to the photographed area or as an absolute area in square millimetres, as described under lesion area below.
CLASI itself grades intensity only and ignores extent within a region, which is a known coarseness of the instrument: a region with a single small plaque and a region that is confluently involved score identically. The extent figure is therefore reported alongside the CLASI-aligned intensity value rather than folded into it, giving a continuous secondary measure with more resolution than the four-point item, without altering the CLASI arithmetic.
Scale and hypertrophy
CLASI combines two physically different findings into one 0–2 item: surface scaling at level 1, and verrucous or hypertrophic thickening at level 2. The device measures them as two independent signs, desquamation intensity for the scaling component and induration intensity for the thickening component, then resolves them into the single CLASI level. Because thickening dominates the item, a region with measurable induration resolves to level 2 regardless of the scaling value.
Measuring the two separately also preserves information the instrument discards. In a study of hypertrophic or verrucous CLE, the induration trend on its own is a more sensitive response measure than the combined 0–2 item.
Dyspigmentation
CLASI records dyspigmentation as a single present-or-absent item per region, covering both loss and excess of pigment. The device measures the depigmented component as a continuous extent, from its own segmentation mask, reported relative to the photographed region or as an absolute area.
Post-inflammatory hyperpigmentation is not quantified as a separate measurement. In practice this means the CLASI-D dyspigmentation item is satisfied by the measured depigmentation where depigmentation is present, and remains a clinician judgement where the pigmentary change is hyperpigmentation alone. The distinction matters in CLE, where the two carry different prognostic weight, and it is set out in full on the limitations page.
Repigmentation as a derived measure
The CLASI-D item is binary, and the Damage score is constructed to accumulate. Neither property allows the instrument to represent partial repigmentation, where a depigmented plaque refills from its periphery inward over successive visits. Under CLASI-D that lesion holds the same value until it crosses the threshold at which the item flips, and then moves by one point.
A continuous hypopigmentation extent has no such step. The reduction in hypopigmented area from baseline is directly reportable per lesion and per region, at whatever resolution the imaging supports, and it is reported as an exploratory measure alongside the CLASI-D item rather than as a modification of it. Because erythema is measured by a separate model, a repigmenting lesion that is still mildly inflamed does not confound the pigmentation figure.
Lesion area and the physical size reference
Every extent measurement starts as a count of segmented pixels, and a pixel count becomes a clinical quantity in one of two ways.
Relative extent divides the segmented pixels by the pixels of the photographed region. It requires nothing but the photograph. It is also a ratio whose denominator is set by how the photograph was framed, so it is comparable only between images captured at the same distance and angle. A closer capture at a later visit shrinks the denominator and inflates the measurement with no clinical change behind it, which makes relative extent a sound descriptive measure and an unreliable longitudinal one.
Absolute area converts segmented pixels into square millimetres using a known physical reference in the frame. The device performs camera calibration against that reference and then reports area in physical units, independent of working distance. Change from baseline is then a change in the lesion.
| Relative extent | Absolute area | |
|---|---|---|
| Requires | Nothing beyond the photograph | A calibration marker in frame |
| Unit | Proportion of the photographed area | Square millimetres |
| Sensitive to framing | Yes | No |
| Suitable for change from baseline | Only under fixed framing | Yes |
| Typical use | Descriptive, feasibility, retrospective sets | Endpoints, longitudinal analysis |
For any CLE study where lesion extent, pigmentation or repigmentation is an endpoint rather than a description, capture with markers and report absolute area. Marker specifications and placement are on the calibration markers page.
Target lesion measurement
CLE protocols commonly designate one or more target lesions, representative lesions identified at baseline and assessed at every subsequent visit. This is a lesion-level design running alongside the region-level CLASI, and it exists because a single well-characterised lesion tracks treatment response with far less noise than a whole-body sum.
For each designated target lesion the device reports area, erythema intensity, desquamation and induration intensity, and depigmentation extent, at every visit where it is captured. Reported as absolute area, these form a per-lesion trajectory that is independent of how many regions were photographed at that visit.
Target lesions are identified by a stable per-lesion identifier assigned at baseline and carried in the image metadata. Physical markers placed on the skin at capture, where a protocol uses them, are an aid to the photographer and to the reviewer; they are not what binds a measurement to a lesion across visits, and a protocol should not rely on them for that.
Alopecia and scalp involvement
The device measures hair loss percentage, the proportion of the scalp showing hair loss, from scalp photography.
This measures the extent of hair loss. CLASI instead scores its pattern: whether loss is diffuse, focal in one quadrant, or focal in more than one quadrant. The two are related but not interchangeable, so hair loss percentage is reported as a continuous scalp measure alongside, rather than in place of, the clinician's CLASI alopecia grade. The same figure informs both the Activity alopecia item and the Damage scalp scarring item; CLASI separates scarring from non-scarring alopecia and the measurement does not. Both points are detailed on the limitations page.
Diagnosis support
Independently of severity scoring, the device returns a ranked differential diagnosis with associated risk indicators. Cutaneous lupus erythematosus is among the conditions it recognises. In a CLE trial this serves screening and eligibility review rather than the endpoint itself.
Components answered outside the image
Three CLASI components are not observations of photographed skin and are collected through the platform as structured clinician or patient entries, then combined with the measured items:
Mucous membrane lesions (CLASI-A, 0–1): an oral and nasal mucosal examination, outside the scope of the skin photography protocol.
Recent hair loss (CLASI-A, 0–1): whether the patient has noticed hair loss in the preceding 30 days, a history question with no visual correlate at a single timepoint.
Dyspigmentation duration (CLASI-D multiplier): whether dyspigmentation in that patient typically persists beyond 12 months, a history question that doubles the dyspigmentation subtotal.
Scale alignment
A sign name alone does not identify a measurement. The same clinical sign is scored on different scales by different validated instruments, and the device returns each sign on the scale of the scoring system that produced it. An erythema value graded 0 to 3 under one instrument is not the same quantity as an erythema value graded 0 to 4 under another, even though both are called erythema.
Every sign read for a CLE study is therefore qualified by its source instrument and mapped explicitly onto the CLASI item scale during study configuration. The mapping is fixed for the duration of the study and recorded in the study configuration, so that a value reported at the final visit means exactly what the same value meant at baseline. Treating an unqualified sign value as interchangeable across instruments is the most consequential integration error available in this architecture, and the explicit mapping exists to prevent it.
From per-image values to a CLASI score
Aggregation runs in four steps.
- Quality gate: each image is assessed by DIQA before any sign is read. Images below the quality threshold are rejected for recapture and never contribute a value.
- Per-image sign measurement: the device returns intensity and extent values for the signs present in that photograph, with a confidence figure per sign.
- Region assignment: each image contributes to the CLASI region it was captured for. Regions photographed in more than one frame are consolidated to a single per-region value.
- Score assembly: region values, quadrant scalp items and the clinician-entered components are combined into CLASI-A and CLASI-D, with the dyspigmentation multiplier applied.
Step 3 is the one that fails silently. If an image is not attached to a patient, a visit and a named region, steps 1 and 2 still succeed and return perfectly valid per-image values, while step 4 becomes impossible. The result is a dataset of measurements that cannot be assembled into the endpoint the study was designed around. The imaging protocol page specifies the metadata that prevents this.
Manual compared with automated assessment
| Aspect | Manual CLASI | Automated components |
|---|---|---|
| Erythema grading | Visual colour judgement, four-point scale | Measured intensity, plus continuous extent |
| Scale and hypertrophy | Combined 0–2 judgement | Desquamation and induration measured separately |
| Dyspigmentation | Present or absent per region | Continuous depigmentation extent |
| Activity versus damage | Rater separates them by judgement | Measured by distinct models |
| Reproducibility | Varies between and within raters | Identical values for an identical image |
| Time per assessment | Over fifty judgements per visit | <2 seconds per image |
| Site calibration | Rater training and calibration exercises | None; the measurement is the same everywhere |
The practical consequence for a trial is the reduction of measurement noise in the endpoint, which is what determines how large a treatment effect a given sample size can detect. The consequence for a long study is that a damage score accumulated over several years remains comparable with its own baseline even when the rater who recorded that baseline has left the site.