Reliable melanoma discrimination system offers grounds of the decision

2016 Fiscal Year Final Research Report

• Project/Area Number: 26461666
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Dermatology
• Research Institution: Hosei University
• Principal Investigator: IYATOMI Hitoshi 法政大学, 理工学部, 准教授 (10386336)
• Project Period (FY): 2014-04-01 – 2017-03-31
• Keywords: 自動診断支援 / メラノーマ / 深層学習

Outline of Final Research Achievements

We have investigated to build a reliable melanoma discrimination system capable of providing grounds for the decision. Our model successfully estimates all items defined in commonly used clinical findings for melanoma, i.e. a total 15 items defined in ABCD rule and 7-point checklist. We also developed efficient pre-processing techniques for melanoma discrimination for deep learning approach. Our method reduced time-consuming training time to less than 1/9, but it attained superior performance in discriminating melanomas to expert dermatologists.

Free Research Field

機械学習、画像認識