Scholarlink

<p>Artificial intelligence (AI) promises to enhance breast cancer screening. Here we evaluated Google’s mammography AI system (version 1.2) across two phases: a retrospective study using 115,973 mammograms from five National Health Service screening services with 39-month follow-up and prospective noninterventional feasibility deployment at 12 sites (9,266 cases). The primary endpoint was AI sensitivity and specificity versus first reader using a 5% noninferiority margin. The secondary endpoints were performance versus second or consensus readers and breast-level analyses. Retrospectively, AI achieved superior sensitivity (0.541 versus 0.437 for first reader, <i>P</i> &lt; 0.001) and noninferior specificity (0.943 versus 0.952, <i>P</i> &lt; 0.001). Cancer detection rate increased from 7.54 to 9.33 per 1,000 women, with AI detecting 25.0% of interval cancers. Performance was particularly strong for first screens (39.3% fewer recalls, 8.8% higher detection) and invasive cancers. No systematic demographic disparities were observed. Simulated second-reader replacement reduced reading time by 32% while increasing detection by 17.7%. Prospective deployment confirmed technical feasibility but revealed a distribution shift requiring threshold recalibration. Implementation requires adaptive calibration and continuous monitoring to ensure safety and equity.</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

Diagnostic accuracy, fairness and clinical implementation of AI for breast cancer screening: results of multicenter retrospective and prospective technical feasibility studies

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月10日

查看全文
  • Christopher J. Kelly,
  • Marc Wilson,
  • Lucy M. Warren,
  • Richard Sidebottom,
  • Mark Halling-Brown,
  • Lin Yang,
  • Megumi Morigami,
  • Jenny Venton,
  • Reena Chopra,
  • Jane Chang,
  • Jonathan Dixon,
  • Fiona J. Gilbert,
  • Daniel I. Golden,
  • Elzbieta Gruzewska,
  • Lesley Honeyfield,
  • Amandeep Hujan,
  • Delara Khodabakhshi,
  • Emma Lewis,
  • Namrata Malhotra,
  • Rachita Mallya,
  • Della Ogunleye,
  • Charlotte Purdy,
  • Rory Sayres,
  • Marcin Sieniek,
  • Tsvetina Stoycheva,
  • Aminata Sy,
  • Susan Thomas,
  • Dominic Ward,
  • Lihong Xi,
  • Shawn Xu,
  • Shravya Shetty,
  • Ara Darzi,
  • Kenneth Young,
  • Hema Purushothaman,
  • Lisanne Khoo,
  • Mamatha Reddy,
  • Hutan Ashrafian,
  • Deborah Cunningham

摘要

Artificial intelligence (AI) promises to enhance breast cancer screening. Here we evaluated Google’s mammography AI system (version 1.2) across two phases: a retrospective study using 115,973 mammograms from five National Health Service screening services with 39-month follow-up and prospective noninterventional feasibility deployment at 12 sites (9,266 cases). The primary endpoint was AI sensitivity and specificity versus first reader using a 5% noninferiority margin. The secondary endpoints were performance versus second or consensus readers and breast-level analyses. Retrospectively, AI achieved superior sensitivity (0.541 versus 0.437 for first reader, P < 0.001) and noninferior specificity (0.943 versus 0.952, P < 0.001). Cancer detection rate increased from 7.54 to 9.33 per 1,000 women, with AI detecting 25.0% of interval cancers. Performance was particularly strong for first screens (39.3% fewer recalls, 8.8% higher detection) and invasive cancers. No systematic demographic disparities were observed. Simulated second-reader replacement reduced reading time by 32% while increasing detection by 17.7%. Prospective deployment confirmed technical feasibility but revealed a distribution shift requiring threshold recalibration. Implementation requires adaptive calibration and continuous monitoring to ensure safety and equity.