上部消化管内視鏡検査における人工知能支援による腫瘍検出の有効性：ランダム化比較試験のシステマティックレビューとメタアナリシス

Esophagogastroduodenoscopy (EGD) remains the primary diagnostic modality for evaluating lesions of the upper gastrointestinal tract (UGI). This meta-analysis assessed the diagnostic performance of artificial intelligence-assisted endoscopy (AI-EGD) compared with conventional endoscopy (Co-EGD) for the detection of UGI neoplasms.

A systematic search of major bibliographic databases identified randomized controlled trials (RCTs) comparing AI-EGD and Co-EGD in adults undergoing EGD. Primary outcomes included neoplasm detection rate (NDR) per patient. Secondary outcomes included NDR per lesion and stratification by lesion size (<10 mm and ≥10 mm), as well as NDR by histological classification: low-grade intraepithelial neoplasia (LGIN), high-grade intraepithelial neoplasia (HGIN), and carcinoma. Risk ratios (RRs) with 95% confidence intervals (CIs) were pooled using a random-effects model.

Eleven RCTs involving 57,512 participants were included. AI-EGD demonstrated higher detection rates per patient (RR: 1.57; 95% CI: 1.23-2.01) and per lesion (RR: 1.55; 95% CI: 1.33-2.18). The benefit was more evident for lesions ≤10 mm (RR: 2.24; 95% CI: 1.72 to 2.91), while detection rates for lesions >10 mm were comparable. Additionally, AI-EGD achieved higher detection rates across histological subtypes, including LGIN (RR: 1.73; 95% CI: 1.30-2.32), HGIN (RR: 1.61; 95% CI: 1.39-1.87), and carcinoma (RR: 1.54; 95% CI: 1.32-1.79).

AI-EGD demonstrated superior performance compared to Co-EGD in detecting UGI neoplasms, particularly for lesions measuring 10 mm or more less. The higher detection rate of LGIN highlights the potential clinical value of AI support. Further trials are warranted to assess the impact of AI across varying levels of endoscopist experience.