Title:
LEARNING MODEL GENERATION METHOD AND ENDOSCOPE SYSTEM
Document Type and Number:
WIPO Patent Application WO/2023/153448
Kind Code:
A1
Abstract:
Provided is a learning model generation method, in which an anchor image (IM_1), which is selected from among an N number of endoscopic images (IM_1 to IM_N), and an N-1 number of endoscopic images (IM_2 to IM_N), excluding the anchor image, are input to a learning model (121). The learning model outputs an ith prediction probability distribution (pi) from an ith endoscopic image (IM_i), such that an N number of prediction probability distributions (p1 to pN) including an anchor prediction probability distribution (p1) corresponding to the anchor image and an N-1 number of prediction probability distributions corresponding to the N-1 number of endoscopic images are output. In the learning model generation method, weighting parameters of the learning model are updated using the margin of error between the anchor prediction probability distribution and the correct answer label (TGp1) distribution, and a loss function (LSKD) based on the distance between the anchor prediction probability distribution and a distribution calculated from the N-1 number of prediction probability distributions.
Inventors:
NISHIDA SHUNPEI
TOMODA YOSHIHIRO
HASEYAMA MIKI (JP)
OGAWA TAKAHIRO (JP)
TOGO REN (JP)
TOMODA YOSHIHIRO
HASEYAMA MIKI (JP)
OGAWA TAKAHIRO (JP)
TOGO REN (JP)
Application Number:
PCT/JP2023/004224
Publication Date:
August 17, 2023
Filing Date:
February 08, 2023
Export Citation:
Assignee:
OLYMPUS MEDICAL SYSTEMS CORP (JP)
UNIV HOKKAIDO NAT UNIV CORP (JP)
UNIV HOKKAIDO NAT UNIV CORP (JP)
International Classes:
A61B1/045
Foreign References:
| JP2020078539A | 2020-05-28 |
Other References:
JIANPING GOU; BAOSHENG YU; STEPHEN JOHN MAYBANK; DACHENG TAO: "Knowledge Distillation: A Survey", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, 201 OLIN LIBRARY CORNELL UNIVERSITY ITHACA, NY 14853, 20 May 2021 (2021-05-20), 201 Olin Library Cornell University Ithaca, NY 14853 , XP081949624, DOI: 10.1007/s11263-021-01453-z
YUN SUKMIN; PARK JONGJIN; LEE KIMIN; SHIN JINWOO: "Regularizing Class-Wise Predictions via Self-Knowledge Distillation", 2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), IEEE, 13 June 2020 (2020-06-13), pages 13873 - 13882, XP033805490, DOI: 10.1109/CVPR42600.2020.01389
JAISWAL ASHISH, ASHWIN RAMESH BABU, MOHAMMAD ZAKI ZADEH, BANERJEE DEBAPRIYA, MAKEDON FILLIA: "A Survey on Contrastive Self-supervised Learning", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, ARXIV.ORG, ITHACA, 7 February 2021 (2021-02-07), Ithaca, pages 1 - 21, XP093082821, [retrieved on 20230918]
YUN SUKMIN; PARK JONGJIN; LEE KIMIN; SHIN JINWOO: "Regularizing Class-Wise Predictions via Self-Knowledge Distillation", 2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), IEEE, 13 June 2020 (2020-06-13), pages 13873 - 13882, XP033805490, DOI: 10.1109/CVPR42600.2020.01389
JAISWAL ASHISH, ASHWIN RAMESH BABU, MOHAMMAD ZAKI ZADEH, BANERJEE DEBAPRIYA, MAKEDON FILLIA: "A Survey on Contrastive Self-supervised Learning", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, ARXIV.ORG, ITHACA, 7 February 2021 (2021-02-07), Ithaca, pages 1 - 21, XP093082821, [retrieved on 20230918]
Attorney, Agent or Firm:
TAKEKOSHI Noboru et al. (JP)
Download PDF: