Classification of Pathological Images of Skin Diseases Based on Deep Learning
Conference Publication ResearchOnline@JCUAbstract
Skin cancer is one of the highest incidences of cancer, the incidence of the population covers all ages. However, the diagnosis process of skin diseases is complex, which requires doctors to observe and locate the injured sites first, then slice the living tissue under the microscope, which challenges doctors' timely diagnosis and medical treatment plan. Therefore, a more accurate classification algorithm of skin diseases has essential significance and clinical application value for timely skin cancer diagnosis. In recent years, most studies on dermatology algorithms have focused on the binary classification of benign and malignant dermatoses. However, there are many kinds of dermatoses, and each kind of dermatoses has different pathogenesis and treatment methods. Based on this, this paper applies convolutional neural network to eight classification of dermatosis. Furthermore, the size and shape of skin lesions are not the same, and the presence of artifacts such as hair and veins around the injured sites also make an accurate diagnosis more difficult. Therefore, this paper introduces the attention mechanism on the basis of the original Inception-Resnet-v2 network, and at the same time, enhances the original data. Finally, we uses the method of transfer learning to conduct experiments on the training dataset of ISIC 2019 challenge. The results show that the average classification accuracy of the method used in this paper is more than 85%, and the AUC score of each category is above 0.95, which shows that the classifier has good performance.
Journal
N/A
Publication Name
DOCS 2022: 4th International Conference on Data-driven Optimization of Complex Systems
Volume
N/A
ISBN/ISSN
978-1-6654-5982-2
Edition
N/A
Issue
N/A
Pages Count
6
Location
Chengdu, China
Publisher
Institute of Electrical and Electronics Engineers
Publisher Url
N/A
Publisher Location
Piscataway, NJ, USA
Publish Date
N/A
Url
N/A
Date
N/A
EISSN
N/A
DOI
10.1109/DOCS55193.2022.9967728