JMIR Publications

ABSTRACT

Background:

Psoriasis is a common chronic dermatological disease that is primarily diagnosed based on clinical manifestation. The application of artificial intelligence (AI), particularly deep learning models, to assist in the diagnosis of psoriasis using clinical images is an emerging area nowaday, with the potential to support early diagnosis by primary doctors

Objective:

To develop a clinical image database of psoriasis and develop an AI model for psoriasis diagnosis based on clinical images, and to evaluate the AI model at two medical centers in Vietnam: the National Hospital of Dermatology and Venereology and the Military Central Hospital 108

Methods:

We collected a psoriasis clinical image database comprising more than 23,000 images collected at the National Hospital of Dermatology and Venereology between 2018 and 2024. Based on this database, we developed a YOLO11 deep learning model for psoriasis diagnosis using clinical images. The YOLO11 model was trained on more than 8000 images from the database. The completed AI model was subsequently evaluated at two medical centers: the National Hospital of Dermatology and Venereology and the Military Central Hospital 108 to diagnosis psoriasis from other skin diseases, including pityriasis rosea, pityriasis rubra pilaris, and atopic dermatitis.

Results:

The YOLO11 model demonstrated good diagnostic performance, with a sensitivity of 93.1% for psoriasis vulgaris and 85.9% for pustular psoriasis from 8000 images. At the National Hospital of Dermatology and Venereology, the model differentiated psoriasis from other skin diseases, including pityriasis rosea, pityriasis rubra pilaris, and atopic dermatitis, with a sensitivity of 92.45%, specificity of 91.48%, an area under the receiver operating characteristic curve (AUC) of 0.947, and a Cohen’s kappa coefficient of 0.839. At Hospital 108, the model differentiated psoriasis from atopic dermatitis with a sensitivity of 88.97%, specificity of 100%, an AUC of 0.971, and a Cohen’s kappa coefficient of 0.891.

Conclusions:

The YOLO11 model demonstrated high diagnostic performance in supporting psoriasis diagnosis based on clinical images. This AI-based model has the potential to assist non-dermatologist physicians in the early recognition and diagnosis of psoriasis. Clinical Trial: The study was approved by the Ethics Committee of the National Hospital of Dermatology and Venereology (Approval No. 10/HĐĐĐ-BVDLTW, dated May 29, 2023) and by the Ethics Committee of Hanoi Medical University (Approval No. NCS2024/GCN-HMUIRB, dated May 15, 2024). This study was conducted as part of the national-level research project entitled “Development of an Artificial Intelligence System to Support the Diagnosis of Psoriasis, Atopic Dermatitis, and Skin Cancer in Vietnam” (Project No. KC-4.0-44/19-25).