Kemna Ruby, Zeeuw J Michiel, Ziesemer Kirsten A, Ali Mahsoem, Bereska Jacqueline I, Marquering Henk, Stoker Jaap, Verpalen Inez M, Swijnenburg Rutger-Jan, Huiskens Joost, Kazemier Geert
Department of Surgery, Amsterdam UMC, Vrije Universiteit, Amsterdam, The Netherlands.
Cancer Center Amsterdam, Amsterdam, The Netherlands.
Oncology. 2025 May 26:1-10. doi: 10.1159/000546572.
Artificial intelligence (AI) is increasingly being researched and developed in the medical field and holds the potential to transform healthcare after successful implementation. For patients with colorectal cancer liver metastases (CRLM), many AI models have been developed, but knowledge about translation of these models in the clinical workflow is lacking. Therefore, this systematic review aimed to provide a contemporary overview of the current maturity status of AI models for patients with CRLM.
A systematic search of the literature until November 2, 2023, was conducted in PubMed,
A total of 117 studies were included. Ninety-seven studies (83%) have been published in the last 5 years. The most common study design was retrospective (96%). Thirty-five studies (30%) utilized a dataset of fewer than 50 patients with CRLM. Internal validation was performed in 63% of the studies and external validation in 17%. The remaining studies did not report validation. Half of the studies were classified as high risk of bias. None of the included studies performed real-time testing, workflow integration, clinical testing, or clinical integration.
Although a rapid increase in research describing the development of AI models for patients with CRLM has been observed in recent years, not a single AI model has been translated into clinical practice.
人工智能（AI）在医学领域的研究和开发日益增多，成功实施后有望改变医疗保健。对于结直肠癌肝转移（CRLM）患者，已经开发了许多AI模型，但缺乏关于这些模型在临床工作流程中转化应用的知识。因此，本系统评价旨在对CRLM患者AI模型的当前成熟度状况提供最新概述。
在PubMed、Embase.com和科睿唯安/科学网核心合集中对截至2023年11月2日的文献进行系统检索，以确定符合条件的研究。使用AI和/或放射组学针对CRLM患者的研究被视为符合条件。收集有关研究目的、研究设计、数据集规模、国家、AI应用类型、验证水平和临床实施状态（NASA技术就绪水平）的数据。使用预测模型偏倚风险评估工具（PROBAST）评估各研究的偏倚风险和适用性。
共纳入117项研究。其中97项研究（83%）在过去5年发表。最常见的研究设计是回顾性研究（96%）。35项研究（30%）使用的CRLM患者数据集少于50例。63%的研究进行了内部验证，17%进行了外部验证。其余研究未报告验证情况。一半的研究被归类为高偏倚风险。纳入的研究均未进行实时测试、工作流程整合、临床试验或临床整合。
尽管近年来观察到描述CRLM患者AI模型开发的研究迅速增加，但尚未有一个AI模型转化为临床实践。
