DenseNet网络深度学习分析CT图像鉴别肺结节良恶性的价值

  目的  探究DenseNet网络深度学习分析CT图像鉴别肺结节良恶性的价值。  方法  选取2017年2月~2019年5月我院收治的疑似肺结节患者80例，患者均进行CT扫描和DenseNet网络深度学习的人工智能系统诊断其良恶性，以病理结果作为金标准。分析CT图像、DenseNet网络深度学习分析联合CT图像对肺结节良恶性的诊断价值。  结果  CT图像表现肺密度增高影，有云雾状阴影，可清晰显示支气管内血管情况，评估结节良恶性准确率为88.75%，敏感度为76.92%，特异性为94.44%，与病理诊断的Kappa值为0.736（P < 0.001）；DenseNet网络深度学习联合CT评估结节良恶性的敏感度为96.15%，特异性为88.89%，DenseNet网络深度学习联合CT评估准确率高于单纯CT评估准确率（91.25% vs 88.75%），且与病理诊断一致性较好（Kappa= 0.810，P < 0.001）。  结论  DenseNet网络深度学习分析CT图像鉴别肺结节良恶性准确性较高，且与病理结果具有较好的一致性。 

DenseNet network deep learning analysis of CT image in differentiating benign and malignant pulmonary nodules

  Objective  To explore the value of DenseNet network deep learning analysis in differentiating benign and malignant pulmonary nodules.  Methods  Eighty patients with suspected pulmonary nodules in our hospital from February 2017 to May 2019 were selected. All patients underwent CT scan and artificial intelligence system of DenseNet network deep learning to diagnose benign and malignant, and pathological results were taken as gold standard. We analyzed the diagnostic value of CT images, DenseNet network deep learning analysis combined with CT images in benign and malignant pulmonary nodules.  Results  CT images showed increased lung density and cloudy shadow, which could clearly show the situation of bronchial vessels. The accuracy rate of benign and malignant nodules was 88.75%, the sensitivity was 76.92%, and the specificity was 94.44%, Kappa value with pathological diagnosis was 0.736 (P < 0.001). The sensitivity and specificity of DenseNet network depth learning combined with CT in evaluating benign and malignant nodules were 96.15% and 88.89%, respectively. The accuracy of densenet network deep learning combined with CT in evaluating benign and malignant nodules was 91.25%, which was higher than that of CT alone (88.75%). It had good consistency with pathological diagnosis (Kappa=0.810, P < 0.001).  Conclusion  DenseNet network deep learning analysis of CT images in differentiating benign and malignant pulmonary nodules has high accuracy and good consistency with pathological results.