http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Unsupervised Liver Segmentation using Domain Adaptation in MRI
Jiwon Jung(정지원),Ehwa Yang(양이화),Woo Kyoung Jeong(정우경),Kyoung Doo Song(송경두),Jae-Hun Kim(김재훈) 한국통신학회 2022 한국통신학회 학술대회논문집 Vol.2022 No.2
Liver segmentation is an essential procedure in Computer-Aided Diagnosis (CAD), surgery, and volume measurement for radiotherapy. But it is still a challenging task to extract liver tissue parenchyma due to nearby organs with similar intensities. When we segment the liver using supervised deep learning, fully-annotated datasets are needed. However, it is hard to obtain well-annotated labels because of their diverse appearances such as size and shape. Also, it takes expensive costs for labeling. In this paper, we performed unsupervised liver segmentation in unlabeled Magnetic Resonance Imaging (MRI) datasets using deep learning. To generate labels of MRI, the domain adaptation technique is applied with CT images containing well-annotated labels. We trained the segmentation model with the MRI dataset which is transferred from CT images and evaluated the model on real MRI datasets. The performance of our model shows 88% dice similarity coefficient accuracy. This study could be one of the solutions to handle the difficulty to train deep learning models with unlabeled datasets.