人工知能（AI）を用いた大腸内視鏡検査の最前線—拾い上げ診断

Artificial Intelligence in Colonoscopy：Exploring Computer-aided Detection Systems

要旨●大腸内視鏡検査時の腺腫検出率（ADR）は大腸癌の発生リスクと相関するため，ポリープの確実な検出と切除を行ううえで重要な指標である．近年deep learning技術の登場や医用画像のデジタル化・高解像度化，汎用計算機の高性能化に伴い，内視鏡画像診断分野においても人工知能によるコンピュータ支援診断技術（CAD）の研究開発が盛んに行われている．中でも大腸領域は複数のCADが販売開始され社会実装されつつあり，今後はリアルワールドでの性能や有用性を検討する段階に入っている．CADはその機能により病変検出支援（CADe）と，質的診断支援（CADx）に大別される．本稿ではこのうち，CADeシステムに焦点を当て，最新の研究成果，製品販売状況，今後の課題と展望について概説する．

　Due to the well-established link between adenoma detection rate and colorectal cancer risk, the endoscopic detection and resection of colorectal polyps are of vital concern. Simultaneous advances have recently occurred within the deep learning technology, endoscopic imaging, and computer performance. These developments have enabled progress on numerous ongoing projects involving the CAD（computer-aided diagnosis）using AI（artificial intelligence）. Traditionally, the CAD systems can be categorized into two main groups as follows：CADe（computer-assisted detection）and CADx（computer-assisted diagnosis）. Recently, several CADe systems for colonoscopy have been released and implemented in the clinical environment. Consequently, it is now possible to reflect on the performance of these CADe systems and estimate their real-world utility. In this paper, we outlined the latest research results, market conditions, and future challenges and prospects within AI-assisted colonoscopy.