—新しい小腸画像診断のトピックス—カプセル内視鏡AI

Artificial Intelligence in Small Bowel Capsule Endoscopy

要旨●消化管を自動撮像するカプセル内視鏡は，画像読影者にとって時間的・精神的負担が大きい．画像認識において人間の能力を凌駕しうる畳み込みニューラルネットワーク（CNN）という人工知能（AI）の手法が登場するまでは，高精度の読影支援システム開発は困難であった．本稿では，小腸カプセル内視鏡読影における諸問題を解決するための，CNNを用いたさまざまな最新システムを紹介する．病変の自動検出のみならず，カプセル内視鏡特有の課題解決へも目が向けられている．急速な研究開発段階を経て，臨床導入に向けた実際の臨床場面での検証へとフェーズが移行しつつある．また，今後の展望も多岐にわたる．

　Capsule endoscopy has major disadvantages for physicians, such as the long reading time and the risk of overlooking abnormalities, due to the automatic capture of the gastrointestinal tract. A satisfactory computer-aided supporting system had not been developed before the introduction of CNNs（convolutional neural networks）methodology that could surpass human ability in image recognition. This chapter introduces a variety of state-of-the-art CNN-based systems for resolving various issues in the small bowel capsule endoscopy reading. Issues unique to this examination are focused on, in addition to the automatic detection of abnormalities. A shift has been taking place from the rapid research and development phase to the actual situation evaluation phase for the clinical implementation of this examination. Moreover, the range of its prospects is wide.