论文部分内容阅读
目的 研究人工神经网络 (ANN)在 3种常见孤立性肺结节 (SPN)鉴别诊断中的应用。方法 收集经临床和手术病理证实的SPN(直径≤ 3cm) 15 0例 (包括周围型肺癌、肺错构瘤、肺结核球各 5 0例 ) ,全部病例行常规CT扫描 ,病灶行HRCT扫描 (层厚 1.2 5 3mm ,采用骨重建算法 )。选取 10种SPN在HRCT上的征象 (钙化、脂肪、空泡征、空洞、细支气管气象、分叶征、毛刺征、血管集束征、胸膜凹陷征及卫星灶 )作为ANN分析参数。 3类SPN中随机选择各 2 5例作为训练数据 ,与MATLAB6.1中的神经网络工具箱共同搭建一个初级的人工神经网络 (artificialneuralnetwork ,ANN)测试系统。余下SPN作为测试数据 ,利用已训练好的ANN对其进行分析诊断。结果 ANN对 3类SPN(各 2 5例 )的诊断正确率分别为 80 %、80 %、84%,平均诊断正确率为 81.3 %。 3种诊断方法 (常规阅片法、最大似然法及ANN)采用两两配对 χ2 检验 ,P值无显著性差异 (Ρ >0 .0 5 )。结论 ANN是计算机辅助诊断 (com puteraideddiagnosis ,CAD)中人工智能的前沿领域 ,为SPN开辟了一个新的辅助诊断模式。
Objective To study the application of artificial neural network (ANN) in the differential diagnosis of three kinds of solitary pulmonary nodules (SPN). Methods Fifty cases of SPN (≤3cm in diameter) were collected by routine clinical and surgical pathology. The CT scan was performed in all cases, HRCT scan was performed on all the cases (including peripheral lung cancer, pulmonary hamartoma and tuberculosis of 50 cases) Thick 1.2 5 3mm, using bone reconstruction algorithm). The signs of the 10 SPNs on HRCT (calcifications, fat, vacuoles, voids, bronchiole meristem, leaflet sign, burr sign, vascular bundle sign, pleural indentation and satellite lesions) were selected as parameters for ANN analysis. Three kinds of SPN randomly selected twenty-five cases as training data, together with the MATLAB6.1 neural network toolbox to build a primary artificial neural network (artificialneuralnetwork, ANN) test system. The remaining SPN as test data, the use of trained ANN analysis and diagnosis. Results The diagnostic accuracy of ANN for three kinds of SPNs (25 cases) were 80%, 80% and 84% respectively, and the average diagnostic accuracy was 81.3%. Three diagnostic methods (conventional reading method, maximum likelihood method and ANN) using pairwise matching χ2 test, P value no significant difference (P> 0.05). Conclusion ANN is the leading field of artificial intelligence in computer-aided diagnosis (CAD), which opens up a new diagnostic mode for SPN.