【摘要】 目的 探讨计算机辅助诊断孤立性肺结节(SPN)对人工诊断的意义。方法 搜集经CT引导下肺组织活检穿刺病理证实的SPN 193例(恶性结节144例,良性结节49例),分析21项(实际应用20项)薄层CT指标(部位、长径、短径、形态、边缘、毛刺、晕征、分叶、棘突、空泡、空洞、细支气管气象、与血管关系、与胸膜关系、钙化、脂肪、卫星灶、透亮影、周围肺气肿、胸膜肥厚、密度)。以CT引导下肺活检穿刺的病理结果为金标准,比较神经网络、支持向量机诊断模型及人工诊断3种方法对SPN的诊断效果。结果神经网络模型诊断准确率71.5%,支持向量机诊断模型诊断准确率68.9%,人工诊断准确率80.3%;敏感性分别为84.0%、65.3%、91.0%;特异性分别为34.7%、79.6%、49.0%。结论 神经网络和支持向量机2种诊断模型对SPN的诊断有一定意义,但准确率低于人工诊断,还不能完全代替人工诊断。更多还原

【Abstract】 Objective To study the signification of computer-aided diagnosis of solitary pulmonary nodule(SPN) for artificial diagnosis.Methods 193 cases of proven SPNs with pathology undergone CT-guided lung biopsy(144 cases of malignant nodules and49 cases of benign nodules) were collected in this study.21 findings of the thin layer CT images of SPNs(practical application of 20),including position,length,diameter,morphology,edge,burrs,dizzy,lobulated,spinous,vacuole,cavity,fine bronchioles air sign,relationship with vessel and pleura,calcification,fat satellite,bright,focal around emphysema,pleural hypertrophy and density of the lesions.Based on the pathologic results of CT-guided biopsy,the diagnostic accuracy of SPN with neural network and support vector machine(SVM) model and artificial diagnosis was compared.Results The diagnostic accurate rate,sensitivity and specificity in diagnosing SPN were 71.5%,84.0% and 34.7% respectively with neural network,80.3%,91.0% and 49.0% respectively with artificial diagnosis,68.9%,65.3% and 79.6%,respectively with SVM model.Conclusion Neural network and SVM models in diagnosis of SPN are of certain significance,but diagnostic accuracys are still lower than that of artificial diagnosis,and cannot completely replace the artificial diagnosis.更多还原