Presentation | 2023-11-14 Pre-training without natural images for Cystoscopic AI Diagnosis of Bladder Cancer Ryuunosuke Kounosu, Wonjik Kim, Atsushi Ikeda, Hirokazu Nosato, Yuu Nakajima, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | When developing AI models, it is sometimes difficult to collect sufficient training data. In these cases, pre-trained AI models based on large real-image databases are often utilized. However, manual annotation causes mislabeling and collecting images from the internet can pose copyright issues for these real images. When developing AI models for medical images, transparency of training data is crucial, and pre-training data for diagnosis support AI should not contain such problems. Thus, in this study, this study proposes a pre-training technique without real images. The proposed method combined two types of automatically generated image databases using the Formula-Driven Supervised Learning (FDSL) method, taking into account the features of cystoscopic images related to our target task. Through the experiments conducted with real clinical cystoscopic images, the proposed method achieves to improve its classification accuracy. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | artificial intelligence / deep learning / bladder cancer / pre-training / formula-driven supervised learning |
Paper # | MICT2023-34,MI2023-27 |
Date of Issue | 2023-11-07 (MICT, MI) |
Conference Information | |
Committee | MI / MICT |
---|---|
Conference Date | 2023/11/14(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Ryo Haraguchi(Univ. of Hyogo) / Hirokazu Tanaka(Hiroshima City Univ.) |
Vice Chair | Hideaki Haneishi(Chiba Univ.) / Takayuki Kitasaka(Aichi Inst. of Tech.) / Chika Sugimoto(Yokohama National Univ.) / Daisuke Anzai(Nagoya Inst. of Tech.) |
Secretary | Hideaki Haneishi(Yamaguchi Univ.) / Takayuki Kitasaka(NAIST) / Chika Sugimoto(Okayama Pref. Univ.) / Daisuke Anzai(Hiroshima City Univ) |
Assistant | Takeshi Hara(Gifu Univ.) / Kenichi Morooka(Okayama Univ.) / Dairoku Muramatsu(Univ. of Electro & Comm.) / Natsuki Nakayama(Nagoya Univ.) / Ami Tanaka(Ritsumeikan Univ.) / Kun Li(Kagawa Univ.) |
Paper Information | |
Registration To | Technical Committee on Medical Imaging / Technical Committee on Healthcare and Medical Information Communication Technology |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Pre-training without natural images for Cystoscopic AI Diagnosis of Bladder Cancer |
Sub Title (in English) | |
Keyword(1) | artificial intelligence |
Keyword(2) | deep learning |
Keyword(3) | bladder cancer |
Keyword(4) | pre-training |
Keyword(5) | formula-driven supervised learning |
1st Author's Name | Ryuunosuke Kounosu |
1st Author's Affiliation | National Institute of Advanced Industrial Science and Technology/Toho University(AIST/Toho Univ.) |
2nd Author's Name | Wonjik Kim |
2nd Author's Affiliation | National Institute of Advanced Industrial Science and Technology(AIST) |
3rd Author's Name | Atsushi Ikeda |
3rd Author's Affiliation | University of Tsukuba(Univ. of Tsukuba) |
4th Author's Name | Hirokazu Nosato |
4th Author's Affiliation | National Institute of Advanced Industrial Science and Technology(AIST) |
5th Author's Name | Yuu Nakajima |
5th Author's Affiliation | Toho University(Toho Univ.) |
Date | 2023-11-14 |
Paper # | MICT2023-34,MI2023-27 |
Volume (vol) | vol.123 |
Number (no) | MICT-256,MI-257 |
Page | pp.pp.37-40(MICT), pp.37-40(MI), |
#Pages | 4 |
Date of Issue | 2023-11-07 (MICT, MI) |