AI对肺占位性病变的诊断价值及影像特征分析  

DIAGNOSTIC VALUE AND IMAGE FEATURE ANALYSIS OF AI IN LUNG SPACE-OCCUPYING LESIONS

作　　者：纪敬斌 张宸瑜 彭垒 矫文捷[1] JI Jingbin;ZHANG Chenyu;PENG Lei;JIAO Wenjie(Department of Thoracic Surgery,The Affiliated Hospital of Qingdao University,Qingdao 266003,China)

机构地区：[1]青岛大学附属医院胸外科,山东青岛266003

出　　处：《青岛大学学报（医学版）》2023年第4期564-567,共4页Journal of Qingdao University(Medical Sciences)

基　　金：山东省自然科学基金资助项目(ZR2020MH234)。

摘　　要：目的探讨深睿医疗人工智能医学影像辅助诊断系统(AI系统)对肺占位性病变两种不同扫描层间距下获得的胸部CT图像的诊断价值,及CT影像学特征和病变大小对AI系统诊断准确度的影响。方法选择于我院胸外科行手术且术后有明确病理结果的肺占位性病变病人821例,按胸部CT扫描层间距不同分为厚层CT组(扫描层间距≥3 mm)和薄层CT组(扫描层间距<3 mm)。以病理诊断结果为金标准,对AI系统在两种厚度CT影像资料的诊断准确率进行评价,并分析两种不同扫描层间距下CT影像学特征和病变大小对AI系统诊断准确度的影响。结果AI系统诊断不同扫描层间距肺占位性病变CT影像资料的灵敏度、特异度、阳性预测值和符合率均较高(全部>90%),接受者操作特征(ROC)曲线下面积(AUC)>0.90。有无胸膜牵拉在两种扫描层间距CT图像上均会影响AI系统的诊断准确度,而肺部病变的长径、短径、边缘是否清晰、有无分叶会影响AI系统对厚层CT图像的诊断准确度(χ^(2)=4.747~123.691,P<0.001)。AI系统对厚层CT图像不同大小肿瘤诊断准确度差异有显著性(Z=-4.237,P<0.05)。结论深睿医疗AI系统对两种不同扫描层间距CT图像上肺占位性病变均有较高的诊断价值,对扫描层间距<3 mm的薄层CT的诊断准确度更高,但人工阅片仍不可取代。Objective To analyze and evaluate the diagnostic value of Deepwise Medical Artificial Intelligence Medical Image-assisted Diagnosis System(AI system)for chest computed tomography(CT)images obtained under two different scanning layer spacings for lung space-occupying lesions,and to explore the influence of CT imaging features and lesion size on the diagnostic accuracy of the AI system.Methods A total of 821 patients with lung space-occupying lesions who underwent surgery in the Department of Thoracic Surgery of our hospital and had definite postoperative pathological results were selected and divided into thick layer CT group(scanning layer spacing≥3 mm)and thin layer CT group(scanning layer spacing<3 mm)according to different layer spacings of chest CT scan.Taking pathological diagnosis results as the gold standard,the diagnostic accuracy of the AI system for CT image data obtained from two kinds of thickness was evaluated,and the effects of CT imaging features and lesion size on the diagnostic accuracy of the AI system were analyzed under two different scanning layer spacings.Results The AI system had high sensitivity,specificity,positive predictive value,and coincidence rate in diagnosing CT image data of lung space-occupying lesions at different scanning layer spacings(all>90%).The area under the receiver operating characteristic curve was over 0.90.The presence or absence of pleural traction affected the diagnostic accuracy of the AI system for CT images obtained at two different layer spacings,while the long diameter,short diameter,whether the edge is clear,and whether lung lesions are lobed affected the diagnostic accuracy of the AI system for thick-layer CT images(χ^(2)=4.747-123.691,P<0.001).For thick-layer CT images,the diagnostic accuracy of the AI system was significantly different among tumors of different sizes(Z=-4.237,P<0.05).Conclusion Deepwise Medical AI System has a high diagnostic value for lung space-occupying lesions on CT images obtained at two different scanning layer spacings and a high

关 键 词：人工智能 胸部 体层摄影术 X线计算机 肺结节 肺疾病 

分 类 号：R816.41[医药卫生—放射医学]