Development of an intelligent cystoscopic bladder cancer diagnosis system

2019 Fiscal Year Final Research Report

• Project/Area Number: 17K16775
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Urology
• Research Institution: University of Tsukuba
• Principal Investigator: Ikeda Atsushi 筑波大学, 附属病院, 病院講師 (50789146)
• Project Period (FY): 2017-04-01 – 2020-03-31
• Keywords: 膀胱癌 / 人工知能 / 膀胱内視鏡検査 / 転移学習

Outline of Final Research Achievements

Non-muscle-invasive bladder cancer is diagnosed, treated, and monitored by cystoscopy. Artificial intelligence is increasingly used to augment tumor detection, but its performance is hindered by the limited availability of cystoscopic images to form a large training dataset. We developed a tumor-detection tool using deep learning-based step-wise transfer learning with a CNN that was pre-trained with general images and further trained with gastroscopic images to better extract features in cystoscopic images. This model was additionally trained using the cystoscopic images. Our results showed that this step-wise organic transfer learning approach yielded a model with better accuracy in differentiating between images of normal and tumor tissues than models trained with only one or two of these datasets. We further demonstrated that the diagnostic accuracy of the AI system was equivalent to that of urologists.

Free Research Field

泌尿器科

Academic Significance and Societal Importance of the Research Achievements

人工知能（AI)は膀胱癌の診療において極めて有用なツールになる可能性がある。膀胱内視鏡画像における正常と異常を判別する診断レベルは、泌尿器科専門医とほぼ同等であることが示されてた。膀胱内視鏡検査の術者の診断レベルを客観的に評価できる可能性も示され、AIには医師の教育を助ける役割も期待される。医師の習熟度に合わせたアドバイスが可能となれば、膀胱内視鏡検査における効率のよい観察と診断のレベル向上が見込まれる。さらにリアルタイムのAIによる診断支援が加われば、術者の経験による診断のばらつきを解消させることができ、結果として、すべての膀胱癌患者の治療成績向上が期待される。