An artificial intelligence (AI) system that can identify 134 cutaneous skin disorders, augment diagnostic accuracy of these disorders, and assist in the prediction of treatment options has been developed by researchers from Korea. A description of the AI algorithm was published in the Journal of Investigative Dermatology.
To develop this AI algorithm, researchers collected 220,680 images of 174 skin disorders from patients of Asian and European descent. These images were obtained from 4 datasets to train the convolutional neural networks. The algorithm was validated with the Edinburgh dataset, which included 1300 images across 10 skin disorders, and the Seoul National University (SNU) dataset, which included 2201 images across 134 disorders.
In terms of malignancy detection, the areas under the curves (AUCs) for the Edinburgh and SNU datasets were 0.928±0.002 and 0.937±0.004, respectively. For the SNU, the AUCs for primary treatment suggestion 0.828±0.012 for steroids, 0.885±0.006 for antibiotics, 0.885±0.006 for antivirals, and 0.918±0.006 for antifungals. In 4 doctors, significant improvements were observed in the mean top-1 (7.0±4.5%; P =.045) and top-3 (10.1±2.5%; P =.0020) diagnostic accuracies with the algorithm.
Continue Reading
Use of the AI algorithm was associated with significant improvements in the sensitivity and specificity of malignancy prediction in 21 dermatologists for malignancy prediction. Sensitivity of malignancy prediction improved from 76.7±8.1% to 84.9±8.0% (P <.0001), whereas specificity improved from 93.9±2.0% to 94.9±1.9% (P =.0062), respectively. There was also an 83.8% (P <.0001) increase in the malignancy prediction sensitivity in 23 non-medical professionals.
Limitations of this study included the reliance on images submitted by dermatologists and non-medical professionals, the potential inclusion of images with low resolution, and the omission of several other skin diseases.
The researchers noted that this model may “serve as an ancillary tool that empowers the performance of medical professionals in diagnosing cutaneous skin diseases.”
Reference:
Han SS, Park I, Lim W, et al. Augment intelligence dermatology: deep neural networks empower medical professionals in diagnosing skin cancer and predicting treatment options for 134 skin disorders [published online February 5, 2020]. J Invest Dermatol. doi: 10.1016/j.jid.2020.01.019
This article originally appeared on Dermatology Advisor
