大腸ポリープの質的診断支援AI機器の多施設前向き性能評価試験

Multicenter Prospective Performance Evaluation Trial of an AI Device to Assist Qualitative Diagnosis of Colorectal Polyps

要旨●人工知能（AI）技術を用いたCADxは，病変の鑑別を正確にサポートすることが期待されている．実臨床での診断能については不明な点も多く，筆者らはCAD EYE^TMを用いて大腸病変において多施設前向き性能評価試験を行った．380病変を解析対象とし，neoplasticに対するAIと内視鏡医診断の感度・特異度・正診率を評価した．AIの正診率は非拡大BLIで83%，拡大BLIで89%であり，expertと同等であったが，non-expertより優れていた．AIの診断性能は内視鏡医に匹敵し，拡大の有無や内視鏡医の経験レベルの影響を受ける可能性が示唆された．

　Computer-aided diagnosis（CAD）using artificial intelligence（AI）is expected to support the characterization of colorectal lesions. However, there is a lack of information of its application in clinical settings. We conducted a multicenter, prospective performance evaluation study using CAD EYE^TM for colorectal lesions. Altogether, 380 lesions were analyzed, and the sensitivity, specificity, and accuracy of both AI-assisted and endoscopic diagnosis of neoplastic lesions were assessed. The accuracies of AI-assisted diagnosis with non-magnified and magnified BLI were 83% and 89%, respectively, which were comparable to that of expert endoscopists and superior to that of non-experts. The diagnostic performance of AI for differentiating colonic lesions is comparable to that of endoscopists. However, it can be affected by the use of magnification and the endoscopists' level of experience.