Scholarlink

<p>Artificial intelligence (AI) is revolutionizing electrocardiography with impressive performance metrics. However, these figures can be misleading if classical test statistics—particularly pretest probability—are not considered. Bayes’ theorem demonstrates that even excellent AI algorithms can produce numerous false-positive results when disease prevalence is low. Likelihood ratios provide prevalence-independent assessment; calibration and clinical net benefit complement discrimination metrics such as area under the curve (AUC). Physicians must understand the epidemiological context of their patient populations and interpret AI results critically—not noncritically adopt algorithmic decisions.</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

KI-gestützte EKG-Diagnostik

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月23日

查看全文
  • Wilhelm Haverkamp,
  • Gerhard Hindricks,
  • Nils Strodthoff

摘要

Artificial intelligence (AI) is revolutionizing electrocardiography with impressive performance metrics. However, these figures can be misleading if classical test statistics—particularly pretest probability—are not considered. Bayes’ theorem demonstrates that even excellent AI algorithms can produce numerous false-positive results when disease prevalence is low. Likelihood ratios provide prevalence-independent assessment; calibration and clinical net benefit complement discrimination metrics such as area under the curve (AUC). Physicians must understand the epidemiological context of their patient populations and interpret AI results critically—not noncritically adopt algorithmic decisions.