人工智能对肺孤立小结节误诊病例影像学表现的回顾性研究  

Retrospective Study on Imaging Findings of Misdiagnosed Cases of Solitary Pulmonary Nodules with Artificial Intelligence

作　　者：张海宇 彭培立 闫宇涛 于联芳 ZHANG Haiyu;PENG Peili;YAN Yutao;YU Lianfang(Radiology Department,No.988 Hospital of the Joint ServiceLogistic Support Force of PLA,Henan 450042,China)

机构地区：[1]联勤保障部队第九八八医院放射诊断科,河南450042

出　　处：《影像技术》2025年第2期34-38,共5页Image Technology

摘　　要：目的:肺孤立小结节的诊断具有一定难度,误诊情况时有发生。开展人工智能对肺孤立小结节误诊病例影像学表现的回顾性研究,旨在深入了解人工智能在这一领域可能出现的误诊情况及其相关影像学特征,有助于进一步优化人工智能算法,提高诊断准确性。方法:收集2022年5月-2023年6月在本院利用AI诊断误诊的110例患者影像资料,对这些病例的影像学图像进行详细分析,观察结节的大小、形态、密度、边缘特征等;将人工智能的诊断结果与实际诊断结果进行对比,找出误诊的情况;提取与误诊相关的影像学特征,如结节的某些特定表现。结果:研究发现,43例恶性结节被误诊为良性,主要表现为完全实性结节,边界光整清晰,无恶性征象;5例恶性完全磨玻璃结节及4例恶性部分磨玻璃结节被误诊为良性,CT无恶性征象表现。而67例良性结节被误诊为恶性,主要为边缘不规则、具有恶性征象的完全磨玻璃结节。AI诊断结果与病理结果对比,完全磨玻璃结节:特异度、敏感度、阳性预测值、阴性预测值、准确率分别为:86.7%、89.5%、90.7%、84.6%、89.2%;部分实性磨玻璃结节:特异度、敏感度、阳性预测值、阴性预测值、准确率分别为:84.4%、88.1%、88.2%、85.3%、80.0%;完全实性磨玻璃结节:特异度、敏感度、阳性预测值、阴性预测值、准确率分别为:88.5%、92.3%、95.1%、91.4%、93.2%。结论:AI在肺孤立小结节诊断中存在一定误诊率,其中边缘光整的恶性结节和具有恶性征象的良性结节易误诊。因此,临床应用中需结合专家读片以提高诊断准确性。Objective:The diagnosis of solitary pulmonary nodules is somewhat difficult,and misdiagnosis often occurs.A retrospective study was conducted on the imaging manifestations of misdiagnosis cases of isolated pulmonary nodules using artificial intelligence(AI),with the aim of gaining a deeper understanding of the possible misdiagnosis situations and related imaging manifestations that may occur with AI in this field.This helps to further optimize AI algorithms and improve diagnostic accuracy.Methods:Imaging data of 110 patients who were misdiagnosed using AI diagnosis in our hospital from May 2022 to June 2023 were collected.The imaging images of these cases were analyzed in detail,,observing the size,shape,density,edge features,etc.of the nodules.The diagnostic results of AI with the actual diagnostic results to identify misdiagnosis were compared.Imaging manifestations related to misdiagnosis,such as specific manifestations of nodules were extracted.Results:The study found that 43 cases of malignant nodules were misdiagnosed as benign,mainly manifested as completely solid nodules with clear boundaries and no malignant signs;Five cases of malignant complete ground glass nodules and four cases of malignant partial ground glass nodules were misdiagnosed as benign,with no malignant signs on CT.However,67 cases of benign nodules were misdiagnosed as malignant,mainly consisting of completely ground glass nodules with irregular edges and malignant signs.the comparison between AI diagnosis results and pathological results shows that the specificity,sensitivity,positive predictive value,negative predictive value,and accuracy of completely ground glass nodules were 86.7%,89.5%,90.7%,84.6%,and 89.2%,respectively.Partial solid ground glass nodules:specificity,sensitivity,positive predictive value,negative predictive value,and accuracy were 84.4%,88.1%,88.2%,85.3%,and 80.0%,respectively.Complete solid ground glass nodules:specificity,sensitivity,positive predictive value,negative predictive value,and accuracy were 88.5%,92.3%,

关 键 词：人工智能 肺结节 误诊病例 影像学表现 

分 类 号：R737.9[医药卫生—肿瘤] R445.1[医药卫生—临床医学]