Dynamic performance improvement in computer-assisted detection system for diagnostic imaging by combined application of online learning and transfer learning

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K01325
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Medical systems
• Research Institution: The University of Tokyo
• Principal Investigator: Nomura Yukihiro 東京大学, 医学部附属病院, 特任研究員 (60436491)
• Co-Investigator(Renkei-kenkyūsha): SATO Issei 東京大学, 大学院新領域創成科学研究科, 講師 (90610155) ; MIKI Soichiro 東京大学, 医学部附属病院, 特任助教 (30707766)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 医用画像 / 診断支援システム / 転移学習 / オンライン学習

Outline of Final Research Achievements

In this research, a dynamic performance improvement of computer-assisted (CAD) detection software by considering diversity of image data due to imaging scanner or imaging parameters was investigated. We constructed a training method of CAD software using online transfer learning algorithm, and verified by multicenter data. We also investigated whether a simplified method of gold standards definition can be used as an alternative to gold standards defined by pixel-by-pixel painting for CAD software using voxel-based classification.

Free Research Field

医用画像処理