| Presentation | 2017-07-06 An Automated Method for Generating Training Data for Image Registration Using Deep Learning Masato Ito, Fumihiko Ino, Kenichi Hagihara, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In this paper, we propose an automated method for generating training data that realizes image registration with deep learning. The proposed method minimizes efforts required for supervised learning by automatically generating millions of training sets from tens of vector fields obtained with actual registration. To automate this procedure, we produce a floating image by applying a vector field $Phi$ to a reference image and obtain the vector field for these images from the inverse vector of $Phi$. In experiments, the proposed method took 33 minutes to produce approximately 170,000 training sets from approximately 670,000 2-D magnetic resonance (MR) images and 30 vector fields generated with a previous registration method. We further trained GoogLeNet with these training sets and performed holdout validation to compare the proposed method with the previous registration method in terms of recall and precision. As a result, the proposed method increased recall and precision from 50% to 80%, predicting deformation vectors more correctly. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Image registration / nonrigid registration / deep learning / training data |
| Paper # | MI2017-28 |
| Date of Issue | 2017-06-29 (MI) |
| Conference Information | |
| Committee | MI |
|---|---|
| Conference Date | 2017/7/6(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Tohoku Univ. |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Medical Imaging, etc. |
| Chair | Kensaku Mori(Nagoya Univ.) |
| Vice Chair | Yoshiki Kawata(Tokushima Univ.) / Yuichi Kimura(Kinki Univ.) |
| Secretary | Yoshiki Kawata(Aichi Inst. of Tech.) / Yuichi Kimura(Nagoya Inst. of Tech.) |
| Assistant | Ryo Haraguchi(Univ. of Hyogo) / Yasushi Hirano(Yamaguchi Univ.) |
| Paper Information | |
| Registration To | Technical Committee on Medical Imaging |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | An Automated Method for Generating Training Data for Image Registration Using Deep Learning |
| Sub Title (in English) | |
| Keyword(1) | Image registration |
| Keyword(2) | nonrigid registration |
| Keyword(3) | deep learning |
| Keyword(4) | training data |
| 1st Author's Name | Masato Ito |
| 1st Author's Affiliation | Osaka University(Osaka Univ.) |
| 2nd Author's Name | Fumihiko Ino |
| 2nd Author's Affiliation | Osaka University(Osaka Univ.) |
| 3rd Author's Name | Kenichi Hagihara |
| 3rd Author's Affiliation | Osaka University(Osaka Univ.) |
| Date | 2017-07-06 |
| Paper # | MI2017-28 |
| Volume (vol) | vol.117 |
| Number (no) | MI-117 |
| Page | pp.pp.11-16(MI), |
| #Pages | 6 |
| Date of Issue | 2017-06-29 (MI) |