AIを用いた大腸内視鏡スクリーニング検査

AI-assisted Colonoscopy for Screening

要旨●大腸癌スクリーニングにおける内視鏡検査の課題である病変見落としや技術格差の解決策として，深層学習を用いたAIが注目されている．本稿では，世界的な開発動向，臨床的有用性，および課題について概説する．CADeはランダム化比較試験にて腺腫検出率を約8%向上させることが示されたが，実臨床では偽陽性によるアラート疲労やhuman-AI interactionなどの課題が顕在化している．また，教育ツールとしての可能性や，費用対効果，最新のガイドラインにおける推奨の相違についてもふれる．AIが真の臨床的価値を発揮するための展望を論じる．

　Deep learning-based artificial intelligence（AI）has garnered considerable attention as a potential solution to persistent challenges in colonoscopy, including missed lesions and variability in operator performance. This review outlines global trends in AI development, its clinical applications, and associated challenges. While randomized controlled trials have demonstrated that computer-aided detection can increase the adenocarcinoma detection rate by approximately 8%, real-world implementation has revealed limitations such as alert fatigue from false-positive detections and complexities in human-AI interaction. Furthermore, this article addresses the potential role of AI as an educational tool, its cost-effectiveness, and discrepancies among current clinical guideline recommendations. Finally, we address future perspectives required for AI to demonstrate its true clinical value in colonoscopy.