Testing the performance of a deep learning-based mobile application with non dermatologist-phyiscians in the diagnosis of common skin diseases.

Summary: Abstract Body: Background: Artificial intelligence-driven mobile phone applications had been used recently in the diagnosis of skin cancers to screen for diabetic retinopathy. This study shows that the use of mobile phone applications can effectively diagnose common skin diseases by primary care physicians. This is the first study done on an AI-driven app for the diagnosis of common skin disease, previously it was used for the diagnosis of skin cancers. Objective: To test the sensitivity, specificity, and positive and negative predictive values of AI-based mobile app as compared to dermatologist diagnoses. Methods: Convolutional neural network-based algorithm was trained with clinical images of 40 skin diseases. The result of this app was compared against the dermatologist’s diagnosis. Results: 1,004 patients (675 males and 329 females, age range: 18–74 years), 670 belonged to the group of 40 diseases which was included in the app, and 334 belonged to 41 other diseases not represented in the app training. Only the disease classes with more than 10 patients were included in the analysis. The overall top-1 accuracy of the app at 72.04% was significantly higher than the top1 accuracy of two non-dermatologists at 45% and 34.85%, respectively. The area-under-the-curve (AUC) values for the algorithm for most of the diseases were in the range of 0.8–0.9 except for herpes zoster and urticaria with an AUC of around 0.75. The NDPs were able to diagnose common diseases with distinctive and easily recognizable morphology like acne, alopecia, eczema, and keloid with accuracies comparable to the app. Conclusions: The artificial intelligence-driven app has high diagnostic accuracy compared to NDPs and is, therefore, a useful, point-of-care, clinical decision support tool for physicians to detect a range of common skin conditions. Sandesh Shah¹, Somesh Gupta² 1. Dermatology, Nepal Medical College Teaching Hospital, Kathmandu, Central Development Region, Nepal. 2. Dermatology, All India Institute of Medical Sciences New Delhi, New Delhi, DL, India. Bioinformatics, Computational Biology, and Imaging