ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper Layer Optimized Transfer Neural Network by using Lesion Correlation Learning
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
Munseob Lee, Jung Hyun Han, Min Jung Kim, Jeong Eun Kim, Keo Sik Kim, Gihyeon Min, Dong Hoon Son, Sung Chang Kim
Issue Date
2019-08
Citation
International Skin Imaging Collaboration (ISIC) Challenges 2019, pp.1-7
Language
English
Type
Conference Paper
Abstract
The development of neural network is expected to provide effective assistance to dermatologist. During few years, many papers and ISIC challenges show the state of art that neural network can be useful for early diagnosis of dermoscopy images. In this paper, we introduced the layer optimized transfer neural network by using lesion correlation learning method for classifying nine skin lesion diagnosis. First of all, to overcome the limitation of small dataset and large imbalance, we introduce the color constancy method, change color space and NDCI in data augmentation. For constructing the neural network, we proposed three approach of layer optimization of pretrained network model, the transfer model by using lesion correlation learning, and ensembled network model. The proposed network model achieves the promising balanced accuracy on ISIC challenge 2019.
KSP Keywords
Balanced accuracy, Color Space, Color constancy, Correlation learning, Data Augmentation, Dermoscopy images, Early diagnosis, Learning methods, Network model, Skin lesion, Small dataset