Abstract
Although there have been reports of the successful diagnosis of skin disorders using deep learning, unrealistically large clinical image datasets are required for artificial intelligence (AI) training. We created datasets of standardized nail images using a region-based convolutional neural network (R-CNN) trained to distinguish the nail from the background.We used R-CNN to generate training datasets of 49,567 images, which we then used to fine-tune the ResNet-152 and VGG-19 models. The validation datasets comprised 100 and 194 images from Inje University (B1 and B2 datasets, respectively), 125 images from Hallym University (C dataset), and 939 images from Seoul National University (D dataset).The AI (ensemble model; ResNet-152 + VGG-19 + feedforward neural networks) results showed test sensitivity/specificity/ area under the curve values of (96.0 / 94.7 / 0.98), (82.7 / 96.7 / 0.95), (92.3 / 79.3 / 0.93), (87.7 / 69.3 / 0.82) for the B1, B2, C, and D datasets.With a combination of the B1 and C datasets, the AI Youden index was significantly (p = 0.01) higher than that of 42 dermatologists doing the same assessment manually. For B1+C and B2+ D dataset combinations, almost none of the dermatologists performed as well as the AI. By training with a dataset comprising 49,567 images, we achieved a diagnostic accuracy for onychomycosis using deep learning that was superior to that of most of the dermatologists who participated in this study.
Highlights
Convolutional neural networks (CNNs), which are based on a deep-learning algorithm, have diagnosed diabetic retinopathy and skin cancer with an accuracy that is comparable to specialist clinicians, a large number of clinical photographs are required to train the convolutional neural networks (CNNs).[1, 2] There are several CNN models available, such as AlexNet, Visual geometry group (VGG), GoogLeNet, Inception and ResNet
Convolutional neural networks (CNNs), which are based on a deep-learning algorithm, have diagnosed diabetic retinopathy and skin cancer with an accuracy that is comparable to specialist clinicians, a large number of clinical photographs are required to train the CNNs.[1, 2]
We obtained area under the curve (AUC) results for receiver operating characteristic (ROC) curves and we describe the sensitivity/specificity values that maximize the sum of the sensitivity and specificity
Summary
Convolutional neural networks (CNNs), which are based on a deep-learning algorithm, have diagnosed diabetic retinopathy and skin cancer with an accuracy that is comparable to specialist clinicians, a large number of clinical photographs are required to train the CNNs.[1, 2] There are several CNN models available, such as AlexNet, VGG, GoogLeNet, Inception and ResNet. Since 2012, all models that have won the ILSVRC have been based on the deep learning algorithm. Visual geometry group (VGG) is a CNN model developed by the University of Oxford that showed that deeper networks can give better results.[3] Microsoft ResNet-152 is an extremely deep 152-layer CNN model that can learn features at various levels of abstraction to boost performance.[4, 5] ResNet-152 recently won the 2015 ILSVRC with an error rate of 3.6%, outperforming the person who participated in their experiment.[5] The ResNet-152 architecture surpass AlexNet, VGG-19, and other old architectures by a significant margin of at least 7% in one-crop Top-1 accuracy.[6]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
