胸部LDCT人工智能联合医师诊断在肺结节良恶性鉴别中的应用价值  

Application Value of Thoracic LDCT Artificial Intelligence Combined with Physician Diagnosis in Identify of Benign and Malignant Pulmonary Nodules

作　　者：杨玉丽 石先琼 柯双好 周睿 YANG Yuli;SHI Xianqiong;KE Shuanghao;ZHOU Rui(Huizhou Third People's Hospital,Huizhou 516000,China;不详)

机构地区：[1]惠州市第三人民医院,广东惠州516000

出　　处：《中外医学研究》2024年第20期62-66,共5页CHINESE AND FOREIGN MEDICAL RESEARCH

基　　金：惠州市科技计划项目(221014156941115)。

摘　　要：目的:探讨胸部低剂量CT(LDCT)人工智能联合医师诊断在肺结节良恶性鉴别中的应用价值。方法:回顾性分析2022年1月—2023年9月惠州市第三人民医院收治的225例肺结节患者临床资料。所有患者均行胸部LDCT扫描。分析金标准、胸部LDCT医师诊断、胸部LDCT人工智能联合医师诊断良恶性肺结节检出情况。比较胸部LDCT医师诊断、胸部LDCT人工智能联合医师诊断的诊断价值。分析胸部LDCT医师诊断、胸部LDCT人工智能联合医师诊断与金标准的一致性。比较良恶性肺结节胸部LDCT影像学特征。分析不同成分肺结节胸部LDCT人工智能联合医师诊断恶性结节准确性。结果:225例患者经金标准检出139例恶性。胸部LDCT医师诊断检出134例恶性,占比为96.40%(134/139);胸部LDCT人工智能联合医师诊断检出138例恶性,占比为99.28%(138/139)。Kappa检验显示,胸部LDCT医师诊断、胸部LDCT人工智能联合医师诊断与金标准的Kappa值分别为0.860、0.972(P<0.001)。胸部LDCT人工智能联合医师诊断敏感度、准确度、阴性预测值均高于胸部LDCT医师诊断,差异有统计学意义(P<0.05)。恶性肺结节患者短毛刺征、胸膜凹陷征、血管集束征、分叶征、空泡征占比均高于良性肺结节患者,差异有统计学意义(P<0.05)。225例患者中,实性结节117例,恶性结节54例;部分实性结节60例,恶性结节49例;纯磨玻璃结节48例,恶性结节36例。胸部LDCT人工智能联合医师诊断实性结节恶性准确率为98.15%(53/54);部分实性结节恶性准确率为100.00%(49/49);纯磨玻璃结节恶性诊断准确率为97.22(35/36)。结论:胸部LDCT人工智能联合医师诊断可提高肺结节良恶性鉴别诊断价值,可据此早期制定精准治疗方案,以改善患者预后。Objective:To explore the application value of thoracic low dose CT(LDCT)artificial intelligence combined with physician diagnosis in identify of benign and malignant pulmonary nodules.Method:A total of 225 patients with pulmonary nodules admitted to the Huizhou Third People's Hospital from January 2022 to September 2023 were selected.Thoracic LDCT scan was performed in all patients.The benign and malignant pulmonary nodules detection condition of gold standard,thoracic LDCT physician diagnosis,thoracic LDCT artificial intelligence combined with physician diagnosis was analyzed.The diagnostic value of thoracic LDCT physicians diagnosis and thoracic LDCT artificial intelligence combined with physicians diagnosis was compared.The consistency of thoracic LDCT physician diagnosis,thoracic LDCT artificial intelligence combined physician diagnosis and gold standard was analyzed.The thoracic LDCT imaging features of benign and malignant pulmonary nodules were compared.The malignant diagnosis accuracy of different components of pulmonary nodules thoracic LDCT artificial intelligence combined with physicians diagnose was analyzed.Result:Among 225 patients,139 malignant cases detected by gold standard.There were 134 malignant cases detected by LDCT physicians diagnosis,accounting for 96.40%(134/139).Thoracic LDCT artificial intelligence combined with physicians diagnosis detected 138 malignant cases,accounting for 99.28%(138/139).The Kappa test showed that the Kappa values of the thoracic LDCT physician diagnosis,thoracic LDCT artificial intelligence combined physician diagnosis and gold standard were 0.860 and 0.972 respectively(P<0.001).The sensitivity,accuracy and negative predictive value of thoracic LDCT artificial intelligence combined with physicians diagnosis were higher than those of thoracic LDCT physicians diagnosis,and the differences were statistically significant(P<0.05).The proportions of short burr sign,pleural depression sign,vascular cluster sign,lobular sign and vacuole sign in patients with malignant pul

关 键 词：肺结节 人工智能 胸部低剂量CT 鉴别诊断 

分 类 号：R734.2[医药卫生—肿瘤]