AI-powered eczema app turns skin photos into trackable severity estimates

Delivered solution

The HarDNet-based model separates skin from the surrounding image, identifies areas showing signs of atopic dermatitis, and evaluates four visible signs used in the EASI methodology: redness, thickening or papulation, excoriation, and lichenification. It reports the percentage of visible skin showing recognized signs of atopic dermatitis and a severity estimate limited to the photographed area. The result is not presented as a diagnosis or a clinician-administered whole-body EASI score.

The initial model was trained using more than 10,000 publicly available images of skin areas identified through condition-specific searches. Two certified dermatologists reviewed the images for clinical relevance and annotated the data used in the analysis. Because publicly sourced images vary in lighting, framing, quality, and representation, the model is used for longitudinal photo tracking rather than diagnostic decision-making.

The computer-vision functionality sits within a broader care-support workflow. Users can compare photographs and estimates over time, record a clinician-prescribed care plan, schedule reminders, complete POEM questionnaires, and document suspected triggers. A guided assistant surfaces clinician-approved content and reconnects users with Atopic School, while the administrative tools support content, localization, and user management.

Translating a clinical framework into photo-based tracking

EASI was designed for a clinician examining a whole patient in person. A phone camera sees one patch of skin, at whatever distance and lighting the person uses.

The model handles this by scoring only the skin visible in the photo, which keeps each reading defensible and means the number is never presented as a whole-body EASI score. Users can compare photographs and local estimates of the same skin area over time.

A published study found a modest association between the AI-derived estimate and patient-reported POEM scores (r=0.35; P<.001). Because the two methods measure different aspects of the condition, that association should not be read as equivalence with a clinician’s assessment. Camera distance, lighting, and variability in patient-reported information were identified as possible sources of variation.

Clinical safety boundaries

Atopic App supports self-monitoring and adherence to an existing care plan. It is not presented as an FDA-cleared medical device.

The product does not diagnose atopic dermatitis, recommend medication, or create or modify a treatment plan. Its computer-vision model produces an EASI-based estimate limited to the skin visible in a photograph; it does not generate a clinician-administered whole-body EASI score.

Care plans originate with the user’s healthcare professional. The app helps users record the prescribed routine, schedule its individual steps, and document what they completed. Medical content is approved in advance by dermatologists and allergist-immunologists, while the guided assistant selects and sequences that content without generating individual treatment recommendations.

Photographs, severity estimates, questionnaires, suspected-trigger records, and adherence history are combined to help users document changes and prepare for appointments. Diagnosis and treatment decisions remain with a qualified healthcare professional.

Project results

Atopic App launched on iOS and Android in three languages, giving patients and caregivers one place to document skin changes, follow recorded care routines, and prepare information for appointments.

IQutis Labs reports that more than 10,000 people have used the Atopic App. The mobile products are supported by administrative tools for content and user management and remain free on the App Store and Google Play.

The product has also been examined in published research. A pilot randomized study involving 58 children found significant improvement in SCORAD and POEM scores across all study groups, while differences between the original randomized groups were not statistically significant.

In a secondary analysis by engagement, participants who used the app on at least eight days had a median clinician-assessed SCORAD score of 6.25 after about three months, compared with 17.9 among participants without the app and 13 among less-frequent users. Use on at least eight days also significantly predicted changes in SCORAD and POEM in the study’s regression models, and at the six-month assessment the differences between engagement groups were no longer statistically significant.

These findings provide preliminary evidence of an association between sustained app engagement and better disease-management outcomes; they do not establish that any individual feature caused the improvement. An earlier observational study also found a relationship between education and continued engagement: users who completed Atopic School used the app approximately twice as frequently as users who did not complete the course.

For IQutis Labs, the delivered product extended a time-limited educational program into an ongoing digital experience. For users, it brought photo documentation, symptom history, suspected triggers, care-plan completion, and educational support into one workflow designed for the time between clinical visits.