Share Email Print

Proceedings Paper

Pilot study to generate image features by deep autoencoder for computer-aided detection systems
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

We propose an automatic feature generation by deep convolutional autoencoder (deep CAE) without lesion data. The main idea of the proposed method is based on anomaly detection. Deep CAE is trained by only normal volume patches. Trained deep CAE calculates low-dimensional features and reproduction error from 2.5 dimensional (2.5D) volume patch. The proposed method was evaluated experimentally with 150 chest CT cases. By using both previous features and the deep CAE based features, an improved classification performance was obtained; AUC=0.989 and ANODE=0.339.

Paper Details

Date Published: 27 March 2019
PDF: 5 pages
Proc. SPIE 11050, International Forum on Medical Imaging in Asia 2019, 110501J (27 March 2019); doi: 10.1117/12.2521289
Show Author Affiliations
Mitsutaka Nemoto, Kindai Univ. (Japan)
Kazuyuki Ushifusa, Kindai Univ. (Japan)
Atsuko Tanaka, Kindai Univ. (Japan)
Takahiro Yamada, Kindai Univ. (Japan)
Yuichi Kimura, Kindai Univ. (Japan)
Naoto Hayashi, The Univ. of Tokyo Hospital (Japan)

Published in SPIE Proceedings Vol. 11050:
International Forum on Medical Imaging in Asia 2019
Feng Lin; Hiroshi Fujita; Jong Hyo Kim, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?