AIによる早期胃癌診断の最前線

Endoscopic Diagnosis for Early Gastric Cancer by Artificial Intelligence

要旨●近年，AI（artificial intelligence）による内視鏡診断が注目されている．AIによる胃癌の通常内視鏡診断では，存在診断と深達度診断の報告があり，前者は感度92.2%，陽性適中率30.6%である．後者はAUC（area under the curve）が0.851と超音波内視鏡検査よりも優れている可能性が示されている．一方，AIによる胃癌の拡大内視鏡診断では，小陥凹型胃病変の鑑別診断において，感度96.7%，特異度95%，正診率96.3%と報告されており，非熟練医と比較して感度，特異度，正診率は有意に高い．範囲診断能は，感度65.5%，特異度80.8%，正診率73.8%と報告されている．

　Recently, endoscopic diagnosis using artificial intelligence has garnered considerable attention. The conventional endoscopic diagnosis of gastric cancer involves differential diagnosis and invasion depth diagnosis. More specifically, the sensitivity and positive predictive value for the differential diagnosis of gastric cancer were reported to be 92.2% and 30.6%, respectively. On the other hand, for invasion depth diagnosis, the area under the curve was reported to be 0.851, which can be superior to that observed for endoscopic ultrasound. Furthermore, the values of sensitivity, specificity, and accuracy in the differential diagnosis of small depressed-type gastric lesions using magnifying endoscopy were found to be 96.7%, 95%, and 96.3%, respectively. Finally, the values of sensitivity, specificity, and accuracy in the diagnosis of cancer delineation were reported to be 65.5%, 80.8%, and 73.8%, respectively.