A deep learning, image based approach for automated diagnosis for inflammatory skin diseases
A deep learning, image based approach for automated diagnosis for inflammatory skin diseases
复制标题
DOI：
10.21037/atm.2020.04.39
复制
发表时间：
2020-05-01
影响因子：
--
通讯作者：
Lu, Qianjin
中科院分区：
文献类型：
--
作者：
Wu, Haijing;Yin, Heng;Lu, Qianjin
背景资料：随着深度学习时代的蓬勃发展，特别是卷积神经网络（CNN）的进步，CNN已被应用于放射学和病理学等医学领域。然而，CNN在同样基于图像的皮肤病学中的应用非常有限。方法：基于EfficientNet-b4 CNN算法，我们开发了一个针对银屑病、湿疹、特应性皮炎和健康皮肤的人工智能皮肤病诊断助手（AIDDA）。提出的CNN模型基于4，740张临床图像进行了训练，并在专家确认的临床图像上评估了性能，这些临床图像分为3种不同的皮肤科医生标记的诊断分类（HC、Pso、Ecz和AD）。AIDDA的总体诊断准确率为95.80% ± 0.09%，敏感性为94.40% ± 0.12%，特异性为97.20% ± 0.06%。AIDDA对Pso的诊断准确率为89.46%，敏感性为91.4%，特异性为95.48%;对AD和Ecz的诊断准确率为92.57%，敏感性为94.56%，特异性为94.41%.Conclusions：AIDDA已经在炎症性皮肤病的诊断中发挥了重要作用，突出了深度学习网络工具如何帮助推进临床实践。
Background: As the booming of deep learning era, especially the advances in convolutional neural networks (CNNs), CNNs have been applied in medicine fields like radiology and pathology. However, the application of CNNs in dermatology, which is also based on images, is very limited. Inflammatory skin diseases, such as psoriasis (Pso), eczema (Ecz), and atopic dermatitis (AD), are very easily to be mis-diagnosed in practice.Methods: Based on the EfficientNet-b4 CNN algorithm, we developed an artificial intelligence dermatology diagnosis assistant (AIDDA) for Pso, Ecz & AD and healthy skins (HC). The proposed CNN model was trained based on 4,740 clinical images, and the performance was evaluated on experts-confirmed clinical images grouped into 3 different dermatologist-labelled diagnosis classifications (HC, Pso, Ecz & AD).Results: The overall diagnosis accuracy of AIDDA is 95.80%+/- 0.09%, with the sensitivity of 94.40%+/- 0.12% and specificity 97.20%+/- 0.06%. AIDDA showed accuracy for Pso is 89.46%, with sensitivity of 91.4% and specificity of 95.48%, and accuracy for AD & Ecz 92.57%, with sensitivity of 94.56% and specificity of 94.41%.Conclusions: AIDDA is thus already achieving an impact in the diagnosis of inflammatory skin diseases, highlighting how deep learning network tools can help advance clinical practice.