食道領域のAI診断の最前線—食道癌拾い上げ診断の現状と拡大観察への考察

Recent Advances in AI Diagnosis in the Field of Esophagus Cancer:Current Study of the Cancer Detection and Future Applications of Magnifying Endoscopy

要旨●食道癌は進行癌として診断されると予後が悪く，早期診断が重要である．近年，人工知能（AI）はディープラーニングを用いることにより医療の分野で非常に進歩した．筆者らは食道癌拾い上げ診断におけるAIの診断能を検証した．8,428枚の食道癌の内視鏡画像を用いてAIを教育し，別に準備した1,118枚の画像で検証した．AIは食道癌症例の98％を驚くほどの速さで検出することができ，10mm未満の7病変もすべて検出した．また98％の画像で表在癌と進行癌を判断することができた．偽陽性が多い，拡大観察については十分に検討されていないなどまだ課題はあるが，AI画像診断支援システムの臨床応用の可能性が示唆された．

　When esophageal cancer is diagnosed at an advanced stage, its prognosis is poor. Therefore, early detection is of great importance. In recent years, AI（artificial intelligence）using deep learning has made remarkable progress in various medical fields. Here, we demonstrated the diagnostic ability of AI to detect esophageal cancer. We developed an AI diagnostic system trained with 8428 endoscopic images of esophageal cancer and validated the system using independently prepared 1118 endoscopic images. The AI diagnostic system correctly detected esophageal cancer cases with a sensitivity of 98%. The AI was able to detect all 7 small cancers less than 10mm in size. Also, the AI was able to distinguish superficial esophageal cancer from advanced cancer with an accuracy of 98%. Although false positives are a limitation to this study, future work will improve our knowledge in magnifying endoscopic images. Our results suggest that an AI-based diagnostic system is feasible for applied clinical practice in the near future.