机器学习在急性阑尾炎CT诊断中的应用进展
本文综述机器学习与CT结合诊断急性阑尾炎的应用、进展与挑战。深度学习模型AppendiXNet检测阑尾炎AUC为0.81,3D卷积神经网络AUC最高达0.95、准确率91.5%;机器学习算法鉴别阑尾炎与憩室炎等相似病变AUC为0.951至0.972;通过影像组学与混合模型区分复杂与简单病例AUC为0.80至0.96。挑战包括人工智能的黑箱性质、临床工作流整合及所需大量资源。
要点
- AppendiXNet检测阑尾炎AUC为0.81,3D卷积神经网络AUC最高0.95、准确率91.5%。
- 机器学习鉴别阑尾炎与憩室炎等相似病变AUC为0.951至0.972。
- 影像组学与混合模型区分复杂与简单病例AUC为0.80至0.96。
核心结论
- 机器学习在急性阑尾炎CT诊断中的自动检测、鉴别诊断和严重程度分层方面取得进展,但黑箱性质、临床工作流整合和资源需求仍是挑战。
研究设计与样本
- 综述,旨在综合机器学习与CT诊断急性阑尾炎的应用、进展与挑战并探讨前景。
诊断标准与关键数值
- 深度学习模型AppendiXNet检测阑尾炎AUC为0.81。
- 3D卷积神经网络检测阑尾炎AUC最高达0.95,准确率91.5%。
- 机器学习算法鉴别阑尾炎与憩室炎等相似病变AUC为0.951至0.972。
- 影像组学与混合模型区分复杂与简单病例AUC为0.80至0.96。
局限性
- 人工智能的黑箱性质。
- 人工智能整合到临床工作流存在挑战。
- 需要大量资源。
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
Acute appendicitis represents a prevalent condition within the spectrum of acute abdominal pathologies, exhibiting a diverse clinical presentation. Computed tomography (CT) imaging has emerged as a prospective diagnostic modality for the identification and differentiation of appendicitis. This review aims to synthesize current applications, progress, and challenges in integrating machine learning (ML) with CT for diagnosing acute appendicitis while exploring prospects. ML-driven advancements include automated detection, differential diagnosis, and severity stratification. For instance, deep learning models such as AppendiXNet achieved an AUC of 0.81 for appendicitis detection, while 3D convolutional neural networks (CNNs) demonstrated superior performance, with AUCs up to 0.95 and an accuracy of 91.5%. ML algorithms effectively differentiate appendicitis from similar conditions like diverticulitis, achieving AUCs between 0.951 and 0.972. They demonstrate remarkable proficiency in distinguishing between complex and straightforward cases through the innovative use of radiomics and hybrid models, achieving AUCs ranging from 0.80 to 0.96. Even with these advancements, challenges remain, such as the "black-box" nature of artificial intelligence, its integration into clinical workflows, and the significant resources required. Future directions emphasize interpretable models, multimodal data fusion, and cost-effective decision-support systems. By addressing these barriers, ML holds promise for refining diagnostic precision, optimizing treatment pathways, and reducing healthcare costs.
引用
- Li J, Ye J, Luo Y, et al. Progress in the application of machine learning in CT diagnosis of acute appendicitis. Abdom Radiol (NY). 2025. PMID: 40095017. doi:10.1007/s00261-025-04864-5
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。