aceEducational Artificial Intelligence Software to Support Assessment of Atopic Dermatitis Severity
Accurate assessment of the severity of atopic dermatitis is crucial for guiding treatment. However, the Eczema Area and Severity Index (EASI), a widely used clinical assessment tool, relies on subjective visual scoring, which can lead to inter-rater variability. This study aimed to evaluate the performance of convolutional neural network-based artificial intelligence software in assessing atopic dermatitis severity from uncropped, unmasked skin images acquired under uncontrolled conditions. In this prospective,…
"student_ipad_school - 015" by flickingerbrad is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.
Researchers prospectively collected 3676 smartphone images from 1016 atopic dermatitis patients at 16 Japanese institutions and trained a ConvNeXt model to score four EASI signs against dermatologist ratings. The model achieved AUCs of 0.825 to 0.860 for scores >=2 and over 0.900 for score 3, with erythema most accurately detected.
If adopted as an educational aid, the tool could help trainees learn consistent EASI scoring and lessen variability that currently stems from subjective visual assessment. Because the authors state it is not for direct clinical decision-making, its impact on treatment choices, patient outcomes, and generalizability beyond Japanese centers remains to be established.
- Prospective multicenter observational study across 16 Japanese institutions collected 3676 smartphone-captured skin images from 1016 patients under uncontrolled conditions.
- Model predicted four EASI signs: erythema, papulation/edema, excoriation, and lichenification, evaluated against dermatologist assessments via ROC analysis.
- For component score >=2, sensitivities were 0.745-0.873 and specificities 0.675-0.764; for score 3, sensitivity up to 1.000 and specificity up to 0.906.
- Erythema demonstrated the highest accuracy among signs, whereas papulation/edema showed the lowest.
ConvNeXt-based AI trained on 3676 smartphone images from 1016 patients achieved AUCs of 0.825-0.860 for EASI component scores >=2 and >0.900 for score 3, supporting use as an educational tool to standardize EASI scoring.
The rundown
The study used uncropped, unmasked skin images acquired under uncontrolled conditions to reflect real-world use, training a ConvNeXt model to predict severity scores for erythema, papulation/edema, excoriation, and lichenification.
Performance was reported by threshold: AUC remained >0.700 for severity >=1, rose to 0.825-0.860 for >=2, and exceeded 0.900 for score 3, indicating stronger discrimination for more severe signs.
Sources
- Peer-reviewedThe Journal of Dermatology2026-07-30
How should this claim be treated?
The debate