Paper Abstract and Keywords |
Presentation |
2020-01-29 16:20
Surgical tool segmentation from laparoscopic images using laparoscopic image syntheses and deep learning Takuya Ozawa, Yuichiro Hayashi, Hirohisa Oda, Masahiro Oda (Nagoya Univ.), Takayuki Kitasaka (Aich Ins. of Tech.), Nobuyoshi Takeshita, Masaaki Ito (NCC East), Kensaku Mori (Nagoya Univ.) MI2019-94 |
Abstract |
(in Japanese) |
(See Japanese page) |
(in English) |
This paper proposes a surgical tool segmentation method from laparoscopic images using image synthesis and deep learning. Forceps and anatomical structure recognition in laparoscopic movies is applicable to develop various surgical assistance and surgical process analysis systems. In the analysis of laparoscopic movies using deep learning, it is difficult to collect massive training data of forceps which are used infrequently during surgery. Recognition accuracy of them may be reduced by lack of training data. This paper solves the problem by generating training data using image synthesis. A pairs of synthetic laparoscopic images and segmentation data is automatically generated by superimposing the 3D forceps models on the actual laparoscopic movies. A deep learning model of forceps segmentation is trained using both the synthetic and the manual segmentation datasets. We applied the trained model to extract forceps in laparoscopic movies. The result showed that recognition accuracy of forceps used infrequently during surgery was improved by using the proposed training data. |
Keyword |
(in Japanese) |
(See Japanese page) |
(in English) |
Laparoscopic video / Surgical tool segmentation / Image synthesis / / / / / |
Reference Info. |
IEICE Tech. Rep., vol. 119, no. 399, MI2019-94, pp. 129-134, Jan. 2020. |
Paper # |
MI2019-94 |
Date of Issue |
2020-01-22 (MI) |
ISSN |
Online edition: ISSN 2432-6380 |
Copyright and reproduction |
All rights are reserved and no part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Notwithstanding, instructors are permitted to photocopy isolated articles for noncommercial classroom use without fee. (License No.: 10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
Download PDF |
MI2019-94 |
|