公众号 / 文章技术与安全

放射科人工智能:入门指南第一部分——从构想到算法

AI for Radiology: A Primer Part I. From Idea to Algorithm.
PubMed 文献 · Radiology · 2025-01-01

本文是面向放射科医生的人工智能素养入门系列首篇,聚焦人工智能算法开发各阶段的实践考量。文章指出人工智能正直接或间接嵌入影像工作流程,放射科医生理解算法基础原理有助于判断其何时可能失效或存在偏倚,从而安全有效地整合人工智能工具。后续文章将探讨交付范式、整合障碍与部署后监测监管。

要点

  • 人工智能已直接或间接嵌入影像工作流程,放射科医生的人工智能素养影响其安全有效整合。
  • 理解人工智能算法的基础工作原理,有助于判断人工智能何时可能失效或易产生偏倚。
  • 本文聚焦人工智能开发各阶段的实践考量,后续文章将讨论交付、整合障碍与部署后监测监管。

核心结论

  • 放射科医生投资于人工智能素养可扩大人工智能实施范围,使其能识别可能被忽视的实际局限,确保人工智能在实践中安全有效整合。

研究设计与样本

  • 本文为面向放射科医生的人工智能素养入门系列的第一篇文章,内容为人工智能开发各阶段实践考量的叙述性介绍,未报告系统检索范围与纳入文献。
  • 文章聚焦从构想到算法的人工智能开发阶段,后续文章将考察向最终用户交付的范式及其与人工智能结果的交互、从影像工作流程各利益相关方视角解释人工智能整合障碍,以及模型采购和部署后的监测与监管考量。

临床应用与阅片要点

  • 原文指出,在基础层面理解人工智能算法的工作原理,是使用人工智能作为工具做出知情决策的关键,并有助于理解人工智能何时可能失效或易产生偏倚。
  • 原文指出,放射科医生投资于人工智能素养可识别可能被忽视的实际局限,从而确保人工智能在实践中安全有效整合。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

Advancements in artificial intelligence (AI) over the past decade have secured its role in and beyond the radiology reading room. AI is increasingly embedded in imaging workflows, either directly by influencing practice with provision of AI results to radiologists for interpretive use cases or indirectly by aiding in noninterpretive use cases for image optimization or workflow efficiency. Although AI solutions show potential to change paradigms for how radiologists practice, the ability to effectively leverage AI will depend on radiologists' AI literacy. Historically, AI implementation has been facilitated by those with prior technical experience, while radiologists accommodate newly deployed tools in existing imaging workflows. However, investing in AI literacy enables wider implementation by empowering radiologists to identify practical limitations that may be overlooked, ensuring safe and effective integration in practice. This article is the first in a primer series providing a foundation in AI literacy for radiologists. "Looking under the hood" of an AI algorithm may be daunting, but understanding how it works at a foundational level is key for making informed decisions using AI as a tool, enabling an understanding of when AI may fail or be prone to bias. This article focuses on practical considerations at each stage of AI development. The following articles in this series will build on this content by examining paradigms for delivery to end users and their interaction with AI results, explaining barriers to AI integration from the perspective of various imaging workflow stakeholders (eg, radiologists, technologists, picture archiving and communication system administrators, imaging informaticists, patients, staff in administrative and executive roles, and others), and detailing postdeployment considerations for AI monitoring and regulation after model procurement and deployment in practice.

来源:Radiology,查看原文。版权归原作者所有。

引用

  1. Tejani AS, Rauschecker AM, Kohli M, et al. AI for Radiology: A Primer Part I. From Idea to Algorithm. Radiology. 2025. PMID: 41117645. doi:10.1148/radiol.250692

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。