Studies on Different CNN Algorithms for Face Skin Disease Classification Based on Clinical Images

Abstract

Skin problems not only injure physical health but also induce psychological problems, especially for patients whose faces have been damaged or even disfigured. Using smart devices, most of the people are able to obtain convenient clinical images of their face skin condition. On the other hand, the convolutional neural networks (CNNs) have achieved near or even better performance than human beings in the imaging field. Therefore, this paper studied different CNN algorithms for face skin disease classification based on the clinical images. First, from Xiangya-Derm, which is, to the best of our knowledge, China's largest clinical image dataset of skin diseases, we established a dataset that contains 2656 face images belonging to six common skin diseases [seborrheic keratosis (SK), actinic keratosis (AK), rosacea (ROS), lupus erythematosus (LE), basal cell carcinoma (BCC), and squamous cell carcinoma (SCC)]. We performed studies using five mainstream network algorithms to classify these diseases in the dataset and compared the results. Then, we performed studies using an independent dataset of the same disease types, but from other body parts, to perform transfer learning on our models. Comparing the performances, the models that used transfer learning achieved a higher average precision and recall for almost all structures. In the test dataset, which included 388 facial images, the best model achieved 92.9%, 89.2%, and 84.3% recalls for the LE, BCC, and SK, respectively, and the mean recall and precision reached 77.0% and 70.8%. INDEX TERMS Deep learning, CNN, facial skin disease, medical image processing.

Automatic Detection of Lupus Butterfly Malar Rash Based on Transfer Learning

“…There are several studies in the literature that explore machine learning-based approaches to the Lupus diagnosis [29], [30], but few studies are related to Malar Rash detection [4], [5] on images.…”

“…Few studies describe computational methods for automatic detection of facial skin lesions that are symptoms of Lupus. [4], [5]. In [4], unsupervised learning was used to detect BMR in images generated artificially by Generative Adversarial Networks (GANS).…”

“…In [4], unsupervised learning was used to detect BMR in images generated artificially by Generative Adversarial Networks (GANS). In [5], Transfer Learning was explored to detect facial Malar Rash with a model trained to detect Lupus skin rashes, but not specifically BMR.…”

“…The face area analyzed is one of the main differences in our approach to the related works. In [5], only the skin lesion area is used. In [4] the entire face area is used.…”

WiRIM: Resolution Improving Mechanism for Human Sensing With Commodity Wi-Fi

“…To achieve a comprehensive analysis, we employ the following metrics widely used in statistics [67], [68]: 1) precision (PR), defined as TP TP+FP , where TP (true positive) means the correctly predicted positive value and FP (false positive) means the wrongly predicted positive value, 2) recall (RE), defined as TP TP+FN , where FN (false negative) means the wrongly predicted negative value, 3) F1-score (F1), defined as 2 * PR * RE PR+RE , is a metric that combines the PR and RE, 4) false positive rate (FPR), defined as FP FP+TN , where TN (true negative) means the correctly predicted negative value. FIGURE 13 shows the statistical results of the four metrics, describes the performance of human activity and location recognition in our experiments.…”