Development and validation of two artificial intelligence models for diagnosing benign, pigmented facial skin lesions

• DOI: 10.1111/srt.12911
• Publication Date: 2020-08-09T05:05:00Z
• Authors: Yin Yang; Yiping Ge; Lifang Guo; Qiuju Wu; Lin Peng; Erjia Zhang; Junxiang Xie; Yong Li; Tong Lin
• Keywords: Residual neural network; Skin lesion

Abstract Objective This study used deep learning for diagnosing common, benign hyperpigmentation. Method In this study, two convolutional neural networks were to identify six pigmentary diseases, and a disease diagnosis model was established. Because the distribution of lesions in original training picture is very complex, we cropped image around lesions, trained network on extracted lesion images, fused verification results overall assess performance identifying hyperpigmented dermatitis pictures. Finally, evaluated recognition converged test set through comparison physicians’ assessments. Results The AUC DenseNet‐96 0.98, whereas ResNet‐152 0.96; therefore, concluded that performed better than ResNet‐152. From AUC, has best performance. achieved comprehensive classification comparable doctors. Conclusions diagnostic benign, pigmented skin based had slightly higher specialists.