Scholarlink

Background <p>The application of Artificial Intelligence (AI)-assisted techniques in adenoma detection during colonoscopy has increased. However, the effectiveness of AI in improving adenoma detection rates (ADR) and reducing adenoma miss rates (AMR) across varying adenoma sizes has yet to be thoroughly assessed.</p> Methods <p>Relevant literature was systematically retrieved from PubMed, Embase, Cochrane Library, Web of Science, ClinicalTrials.gov, VIP Information Database (VIP), Wanfang Data, China National Knowledge Infrastructure (CNKI), and Chinese Biomedical Literature Database (CBM) for randomized controlled trials (RCTs). A meta-analysis was conducted to assess the impact of AI-assisted techniques on ADR and AMR in colorectal adenomas of different sizes (≤ 5&#xa0;mm, 6–9&#xa0;mm, ≥ 10&#xa0;mm). RevMan 5.4 and Stata 17.0 were utilized for statistical analysis. GRADE methodology was applied to rate the certainty of evidence, complemented by TSA to assess the reliability of the outcomes.</p> Results <p>A total of 19 RCTs comprising 15,462 participants were included in the analysis. In the detection of adenomas ≤ 5&#xa0;mm, the AI-assisted group demonstrated superior performance with a significantly increased ADR (RR = 1.43, 95% CI (1.29, 1.59), <i>P</i> &lt; 0.00001) and markedly reduced AMR (RR = 0.35, 95% CI (0.28, 0.44), <i>P</i> &lt; 0.00001) compared to conventional methods. For 6–9&#xa0;mm adenomas, AI assistance significantly increased the ADR (RR = 1.24, 95% CI (1.12, 1.36), <i>P</i> &lt; 0.0001), though its effect on AMR was not statistically significant (RR = 0.59, 95% CI (0.24, 1.43), <i>P</i> = 0.24). For adenomas ≥ 10&#xa0;mm, ADR was significantly higher in the AI-assisted group (RR = 1.21, 95% CI (1.07, 1.37), <i>P</i> = 0.003), but the difference in AMR was not statistically significant (RR = 0.24, 95% CI (0.05, 1.19), <i>P</i> = 0.08). Publication bias analysis suggested potential bias in ADR results for 6–9&#xa0;mm adenomas. TSA analysis showed that AI-assisted technology significantly improved ADR across all adenoma size categories and effectively reduced AMR for lesions ≤ 5&#xa0;mm.</p> Conclusion <p>AI-assisted techniques demonstrate higher diagnostic rates for adenomas of ≤ 5&#xa0;mm, 6–9&#xa0;mm, and ≥ 10&#xa0;mm, with a notable reduction in AMR for lesions ≤ 5&#xa0;mm. While AI can enhance diagnostic quality during colonoscopy, the overall evidence quality remains low, and potential publication bias warrants caution. Further large-scale RCTs are needed to confirm its clinical significance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Effect of AI-assisted diagnosis on adenomas of different sizes: a meta-analysis with evidence from RCTs and trial sequential analysis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月7日

查看全文
  • Dengchao Wang,
  • Youyun Shi,
  • Jinshan Liu,
  • Anyin Wang,
  • Yong Cheng

摘要

Background

The application of Artificial Intelligence (AI)-assisted techniques in adenoma detection during colonoscopy has increased. However, the effectiveness of AI in improving adenoma detection rates (ADR) and reducing adenoma miss rates (AMR) across varying adenoma sizes has yet to be thoroughly assessed.

Methods

Relevant literature was systematically retrieved from PubMed, Embase, Cochrane Library, Web of Science, ClinicalTrials.gov, VIP Information Database (VIP), Wanfang Data, China National Knowledge Infrastructure (CNKI), and Chinese Biomedical Literature Database (CBM) for randomized controlled trials (RCTs). A meta-analysis was conducted to assess the impact of AI-assisted techniques on ADR and AMR in colorectal adenomas of different sizes (≤ 5 mm, 6–9 mm, ≥ 10 mm). RevMan 5.4 and Stata 17.0 were utilized for statistical analysis. GRADE methodology was applied to rate the certainty of evidence, complemented by TSA to assess the reliability of the outcomes.

Results

A total of 19 RCTs comprising 15,462 participants were included in the analysis. In the detection of adenomas ≤ 5 mm, the AI-assisted group demonstrated superior performance with a significantly increased ADR (RR = 1.43, 95% CI (1.29, 1.59), P < 0.00001) and markedly reduced AMR (RR = 0.35, 95% CI (0.28, 0.44), P < 0.00001) compared to conventional methods. For 6–9 mm adenomas, AI assistance significantly increased the ADR (RR = 1.24, 95% CI (1.12, 1.36), P < 0.0001), though its effect on AMR was not statistically significant (RR = 0.59, 95% CI (0.24, 1.43), P = 0.24). For adenomas ≥ 10 mm, ADR was significantly higher in the AI-assisted group (RR = 1.21, 95% CI (1.07, 1.37), P = 0.003), but the difference in AMR was not statistically significant (RR = 0.24, 95% CI (0.05, 1.19), P = 0.08). Publication bias analysis suggested potential bias in ADR results for 6–9 mm adenomas. TSA analysis showed that AI-assisted technology significantly improved ADR across all adenoma size categories and effectively reduced AMR for lesions ≤ 5 mm.

Conclusion

AI-assisted techniques demonstrate higher diagnostic rates for adenomas of ≤ 5 mm, 6–9 mm, and ≥ 10 mm, with a notable reduction in AMR for lesions ≤ 5 mm. While AI can enhance diagnostic quality during colonoscopy, the overall evidence quality remains low, and potential publication bias warrants caution. Further large-scale RCTs are needed to confirm its clinical significance.