The unintended consequences of artificial intelligence and high-risk triaging.
The unintended consequences of artificial intelligence and high-risk triaging.
复制标题
人工智能和高风险分类的意外后果。
DOI：
10.1007/s00330-023-10553-y
复制
发表时间：
2024
影响因子：
5.9
通讯作者：
Bahl,Manisha
中科院分区：
文献类型：
--
作者：
Bahl,Manisha
用于乳腺成像的多种人工智能（AI）应用程序已在欧洲和美国获批用于临床，并正在乳腺成像实践中迅速部署[1，2]。用于乳腺X线摄影病变检测和诊断的AI算法提供了具有恶性肿瘤相对概率的病变标记物，以及检查水平和/或乳腺水平的数值评分[3]。支持这些应用的证据，主要来自于使用癌症富集数据集的多读片多病例研究，表明AI算法可以提高我们解释乳房X线照片的准确性和效率[3，4]。然而，人工智能对现实世界临床实践的真正影响仍然是未知的，包括人工智能将如何影响放射科医生的决策，以及人机之间复杂的交互最终将如何发挥作用。在临床实践中，人工智能算法可以作为独立的阅读器，在双重阅读设置中取代其中一名放射科医生;通过基于乳腺癌的可能性对乳房X线照片进行分类，用作分类工具;和/或通过标记可疑病变并提供恶性肿瘤的定性和定量相对概率，为放射科医生提供决策支持。在本期《欧洲放射学》中，Al-Bazzaz及其同事探讨了商用AI算法作为决策支持工具对性能的影响。
Multiple artificial intelligence (AI) applications for breast imaging are approved for clinical use in Europe and the USA and are rapidly being deployed in breast imaging practices [1, 2]. AI algorithms intended for lesion detection and diagnosis on mammography provide lesion markers with relative probabilities of malignancy, in conjunction with numerical scores at the examination level and/or breast level [3]. Evidence in support of these applications, primarily arising from multireader multicase studies with cancer-enriched datasets, suggests that AI algorithms could enhance both our accuracy and efficiency in interpreting mammograms [3, 4]. However, what remains unknown is the true impact of AI on realworld clinical practice, including how AI will influence radiologist decision-making and how the complex interaction between human and machine will ultimately play out.In clinical practice, AI algorithms could function as an independent reader, replacing one of the radiologists in double-reading settings; be used as a triage tool by categorizing mammograms based on likelihood of breast cancer; and/or provide decision support to the radiologist by marking suspicious lesions and offering qualitative and quantitative relative probabilities of malignancy. In this issue of European Radiology, Al-Bazzaz and colleagues explore the impact of a commercially available AI algorithm as a decision support tool on the performance