ETRI-Knowledge Sharing Plaform



논문 검색
구분 SCI
연도 ~ 키워드


학술대회 Colorectal Cancer Image Segmentation and Classification with Deep Neural Network Based on Information Theory
Cited 0 time in scopus Download 5 time Share share facebook twitter linkedin kakaostory
김화랑, 김광주, 임길택, 최두현
International Conference on Bioinformatics and Biomedicine (BIBM) 2020, pp.2968-2970
20ZD1100, 대경권 지역산업 기반 ICT 융합기술 고도화 지원사업, 문기영
Colorectal cancer (CRC) is the development of cancer from the colon or rectum. Microsatellite instability (MSI) status can be considered as an indicator to predict the prognosis of CRC. We employ MSI prediction of CRC image by designing a neural network model of which base network is DeepLabv3+ with OctaveResNet. Additionally, we add a channel sort module to divide a feature map along with channel intensity. Then each feature map goes through distinct convolution paths. Each convolution path is designed based on information theory: the most important feature goes through the lightest convolution path, vice versa. By dividing feature map and applying different amount of convolutional operation, the model can extract features efficiently. In the experiment, total model weight is reduced but accuracy increases.
CRC, DeepLabv3+, Information theory, MSI, OctaveResNet
KSP 제안 키워드
Colorectal Cancer, Deep neural network(DNN), Feature Map, Image segmentation and classification, Information Theory, Microsatellite instability, extract features, neural network model