JMIR Publications

ABSTRACT

Background:

Most studies on deep learning skin disease classifiers are done with binary classifications, i.e. classifying lesions into malignant and benign. However, dermatology practice involves a large number of inflammatory and infective conditions, which are not easily diagnosed by non-dermatologists physicians.

Objective:

To develop a machine-learning based Smartphone application for multi-class skin disease classification and evaluate its performance in different levels of dermatology practice. Also, to explore similar studies in literature.

Methods:

We developed an AI-driven Smartphone application for 40 common skin diseases and tested it at primary care, tertiary care as well as, in private practice.

Results:

In the clinical study, the overall top-1 accuracy was 75.07% (95% CI = 73.75–76.36), top-3 accuracy was 89.62% (95% CI = 88.67–90.52) and the mean area-under-curve was 0.90 ± 0.07. Table 1 shows top-1 positive predictive value and negative predictive value from clinical study of 35 diseases using mHealth app on patients. In the literature, there are very few studies on image-based deep learning multi-class classification of common skin diseases and none of them are evaluated in actual clinical settings.

Conclusions:

AI-driven Smartphone application has potential to improve the diagnosis and management of skin diseases in patients with skin of color. Non-dermatologist, primary care physicians are likely to be benefitted by such applications.