Latent feature representation with multi-view long-term recurrent reconstructed slice learning for bilateral analysis in digital breast tomosynthesisDBT 양측유방 분석을 위한 다중뷰 장기적 재귀 단면영상 학습 기반 병변의 잠재적 특징 표현 연구

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 533
  • Download : 0
Breast cancer is the most common type of cancer in women. In clinical studies of breast cancer, it is known that masses appear as asymmetric densities between the left and the right breasts, which show different breast tissue structures. Based on the clinical fact, most of researches focused on the development of hand-crafted bilateral features for classifying breast masses by extracting the asymmetric information in 2-D mammograms. However, in 3-D digital breast tomosynthesis (DBT), due to the increase of image data, more effective bilateral features are needed to detect masses. In this paper, a new latent feature representation is proposed, which is boosted by multi-view long-term recurrent reconstructed slice learning for characterizing masses in bilateral analysis of DBT. The proposed method is designed to encode mass characteristics in two parts: a) bilateral feature representation learning, and b) depth directional long-term recurrent learning. First, bilateral asymmetric characteristics of masses in each DBT slice are encoded as a bilateral feature representation by the proposed Siamese architecture of convolutional neural network (CNN). In the architecture, an objective function is devised to effectively learn asymmetric characteristics between the given region-of-interest (ROI) in main view and the corresponding ROI in contralateral view. Then, depth directional characteristics of masses among the learned bilateral feature representations are encoded by the proposed depth directional long-term recurrent learning. In addition, to further improve the class discriminability of latent feature representation, two objective functions have been devised. Experimental results have demonstrated that the proposed latent feature representation achieves a higher level of classification performance in terms of the receiver operating characteristic (ROC) curves and the area under the ROC curve values compared to the performance with feature representation learned by conventional CNN as well as hand-crafted features.
Advisors
Ro, Yong Manresearcher노용만researcher
Description
한국과학기술원 :전기및전자공학부,
Publisher
한국과학기술원
Issue Date
2017
Identifier
325007
Language
eng
Description

학위논문(박사) - 한국과학기술원 : 전기및전자공학부, 2017.2,[iv, 56 p. :]

Keywords

Breast cancer; Bilateral analysis; Latent feature representation; Deep learning; Siamese architecture; Recurrent neural network; Digital breast tomosynthesis; False positive reduction; 유방암; 양측유방 분석; 잠재적 특징 표현; 딥러닝; 샴 구조; 재귀 신경망; 디지털 유방 단층촬영술; 위양성 감소

URI
http://hdl.handle.net/10203/242019
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=675817&flag=dissertation
Appears in Collection
EE-Theses_Ph.D.(박사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0