人工神经网络在孤立性肺结节良恶性诊断中的应用  

The neural network model used for identified benign and malignant of the solitary pulmonary nodules

作　　者：孙理[1] 张修石[1] 周洋[1] 刘露[2] 

机构地区：[1]哈尔滨医科大学附属第三医院,哈尔滨150081 [2]哈尔滨理工大学自动化学院图像处理与模式识别研究室

出　　处：《实用肿瘤学杂志》2011年第2期126-128,共3页Practical Oncology Journal

基　　金：黑龙江省卫生厅科研课题（2005-66）

摘　　要：目的评估计算机辅助诊断技术在孤立性肺结节（SPN）良恶性诊断中的意义。方法搜集经CT引导下肺组织活检穿刺病理证实的SPN193例（恶性结节144例，良性结节49例），分析薄层CT指标。对比CT引导下肺活检穿刺的病理结果，统计神经网络和人工诊断两种方法对SPN的诊断准确率。结果神经网络模型诊断准确率71．5％，人工诊断准确率80．3％；敏感性分别为84．0％、91．0％；特异性分别为34．7％、49．O％。结论计算机辅助诊断有较好的诊断准确率，神经网络敏感性接近医师人工诊断。对SPN的诊断有一定意义，能够具有较好的诊断价值，但准确率低于人工诊断，还不能完全代替人工诊断，可作为辅助诊断应用。Objective To evaluate the significance of diagnosis of isolation pulmonary nodules（SPN） u- sing computer - aided diagnosis（CAD）. Method 193 cases with SPN（ including 144 cases of malignant nodules and 49 cases of benign nodules） proved pathologically by CT - guided biopsy in this paper, and analyzed the char- acteristics. According to the pathologry results, Compare the accuracy of the neural network and artificial diagnosis of two methods in SPN＇s diagnostic, Result The diagnosis accurancy of neural network model was 71.5%, and the aceurancy of artificial diagnosis was 80.3% ,Sensitivity was 84.0% and 91.0% ,respectively,specificity was 34.7% and 49.0% respectively. Conclusion Computer - aided diagnosis has better diagnostic accuracy rate, and the sensitivity of artificial neural hetwork was closed to physicians diagnosis. It has certain significance in SPN diagnosis,can be a good diagnostic value,but with less accuracy. So it cannot completely replace the artificial diagnnsis ,but can be used as auxiliary diagnostic.

关 键 词：孤立性肺结节 计算机辅助诊断 神经网络 

分 类 号：R445.3[医药卫生—影像医学与核医学]