24 March 2016 Ensemble lymph node detection from CT volumes combining local intensity structure analysis approach and appearance learning approach
Author Affiliations +
Abstract
This paper presents an ensemble lymph node detection method combining two automated lymph node detection methods from CT volumes. Detecting enlarged abdominal lymph nodes from CT volumes is an important task for the pre-operative diagnosis and planning done for cancer surgery. Although several research works have been conducted toward achieving automated abdominal lymph node detection methods, such methods still do not have enough accuracy for detecting lymph nodes of 5 mm or larger. This paper proposes an ensemble lymph node detection method that integrates two different lymph node detection schemes: (1) the local intensity structure analysis approach and (2) the appearance learning approach. This ensemble approach is introduced with the aim of achieving high sensitivity and specificity. Each component detection method is independently designed to detect candidate regions of enlarged abdominal lymph nodes whose diameters are over 5 mm. We applied the proposed ensemble method to 22 cases using abdominal CT volumes. Experimental results showed that we can detect about 90.4% (47/52) of the abdominal lymph nodes with about 15.2 false-positives/case for lymph nodes of 5mm or more in diameter.
© (2016) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Yoshihiko Nakamura, Yoshihiko Nakamura, Yukitaka Nimura, Yukitaka Nimura, Masahiro Oda, Masahiro Oda, Takayuki Kitasaka, Takayuki Kitasaka, Kazuhiro Furukawa, Kazuhiro Furukawa, Hidemi Goto, Hidemi Goto, Michitaka Fujiwara, Michitaka Fujiwara, Kazunari Misawa, Kazunari Misawa, Kensaku Mori, Kensaku Mori, "Ensemble lymph node detection from CT volumes combining local intensity structure analysis approach and appearance learning approach", Proc. SPIE 9785, Medical Imaging 2016: Computer-Aided Diagnosis, 97852X (24 March 2016); doi: 10.1117/12.2214925; https://doi.org/10.1117/12.2214925
PROCEEDINGS
7 PAGES


SHARE
Back to Top