Application of artificial intelligence recognition combined with DenseNet network model CT in diagnosis of subsolid pulmonary nodules
-
摘要:
目的 通过人工智能自动识别结合DenseNet网络模型CT检查手段,探讨其在亚实性肺结节患者诊断中应用价值。 方法 选取2018年6月~2019年12月入本院经CT检查的亚实性肺结节患者98例,其中直径≤10 mm组患者32例,10 mm < 直径≤20 mm组患者33例,直径>20 mm组患者33例,将所有患者实施高分辨的CT诊断,并将CT检查数据录入基于DenseNet网络深度学习的人工智能系统,把控人工智能影像诊断的训练数据集质量,由人工智能组和人工读片组分别对所有患者进行良恶性的诊断,比较分析人工智能对肺结节不同直径患者的CT平扫、增强动脉期及延迟期中的CT值、体积及恶性概率的预测值,测试诊断方面的敏感度、特异性以及符合率。 结果 CT扫描对肺结节不同直径患者CT值、体积以及恶性概率的预测值差异均有统计学意义(P < 0.05);直径≤10 mm患者中,人工智能组的敏感度达94.61%,特异性(93.12%)和符合率(92.08%)均高于传统人工读片组(P < 0.05);10 mm < 直径≤20 mm和直径>20 mm患者中,人工智能组的诊断敏感度与人工读片组间差异无统计学意义(P>0.05),但诊断特异性及符合率均高于人工读片组(P < 0.05)。 结论 人工智能识别结合DenseNet网络模型CT可对肺结节识别的敏感度及特异性较高,平扫CT可辅助预测肺结节恶性概率,可辅助临床医生诊断,提高工作效率。 -
关键词:
- 人工智能 /
- DenseNet网络 /
- 肺结节 /
- 诊断
Abstract:Objective To explore the value of its application in the diagnosis of patients with subsolid pulmonary nodules by means of artificial intelligence automatic recognition combined with DenseNet network model CT examination. Methods Ninty-eight patients with subsolid pulmonary nodules admitted to our hospital from June 2018 to December 2019 were selected and examined by CT, including 32 patients in the group of ≤10 mm in diameter, 33 patients in the group of 10 mm < diameter≤20 mm and 33 patients in the group of > 20 mm in diameter. All patients were diagnosed by high-resolution CT, and the CT examination data were recorded into an artificial intelligence system based on DenseNet network deep learning so as to control the quality of the training data set of artificial intelligence imaging diagnosis. All patients were diagnosed as benign or malignant by the artificial intelligence group and the manual film reading group. The predictive value of CT value, volume and malignant probability in CT plain scan, enhanced arterial phase and delayed phase of patients with different diameters of pulmonary nodules by AI were compared and analyzed to test the sensitivity, specificity and coincidence rate in diagnosis. Results There were statistically significant differences in predictive values of CT value, volume and malignant probability of patients with different diameters of pulmonary nodules (P < 0.05). The sensitivity of the AI group was 94.61%, the specificity (93.12%) and compliance rate (92.08%) were higher than those of the conventional manual reading group (P < 0.05). In patients with diameters 10 mm < diameter ≤20 mm and >20 mm group, there was no statistical difference in the diagnostic sensitivity between the artificial intelligence group and the manual reading group (P>0.05), but the diagnostic specificity and compliance rate were higher than those of the manual reading group (P < 0.05). Conclusion Artificial intelligence recognition combined with Densenet network model CT have higher sensitivity and specificity for pulmonary nodules recognition, and plain CT can assist in predicting malignant probability of pulmonary nodules, which can assist clinicians in diagnosis and improve work efficiency. -
Key words:
- artificial intelligence /
- DenseNet network /
- pulmonary nodules /
- diagnosis
-
图 1 患者右肺上叶亚实性肺结节,人工智能识别结合DenseNet网络模型CT分别在平扫(A)、动脉期(B)以及延迟期(C)图像自动预测恶性概率为89.31%、78.34%以及69.98%
Figure 1. The probability of automatic prediction of malignancy was 89.31%, 78.34% and 69.98% respectively in plain scan (A), arterial phase (B) and delayed phase (C) images by artificial intelligence recognition combined with DenseNet network model CT.
表 1 3组间恶性概率的预测值比较
Table 1. The comparison of predictive value of malignant probability between 3 groups (%)
组別 平扫 动脉期 延迟期 直径≤10 mm组(n=32) 84.16 68.96 87.64 10 mm < 直径≤20 mm组(n=33) 93.54 89.67 92.17 直径 > 20 mm组(n=33) 97.98 96.75 97.82 χ2 11.7 33.83 7.5 P 0.003 < 0.001 0.024 表 2 3组图像间SN的CT值比较
Table 2. Comparison of CT values of SN among the three groups of images (Hu, Mean±SD)
组別 平扫 动脉期 延迟期 直径≤10 mm组(n=32) -614.39±4.58 -574.93±110.24 -578.62±110.28 10 mm < 直径≤20 mm组(n=33) -546.35±72.64 -493.65±90.42 -500.36±90.79 直径 > 20mm组(n=33) -341.06±140.39 -293.67±153.16 -281.58±127.46 F 78.547 46.772 63.378 P < 0.001 < 0.001 < 0.001 表 3 3组图像间SN的体积比较
Table 3. Comparison of volume of SN among the three groups of images (mm3, Mean±SD)
组別 平扫 动脉期 延迟期 直径≤10mm组(n=32) 261.24(130.47, 347.58) 223.52(141.36, 314.93) 264.68(148.31, 319.47) 10 mm < 直径≤20 mm组(n=33) 841.36(388.96, 1654.82) 762.93(330.15, 1424.82) 782.49(357.41, 1547.15) 直径 > 20mm组(n=33) 2136.98(715.64, 3549.58) 1902.37(801.35, 3348.54) 1742.48(792.36, 3874.18) H 32.47 35.21 36.52 P < 0.05 < 0.05 < 0.05 表 4 3组的特异性、敏感度及符合率
Table 4. Specificity, sensitivity and coincidence rate among the three groups (%)
组别 直径≤10 mm组 10 mm < 直径≤20 mm 直径 > 20 mm 人工智能组 人工读片组 人工智能组 人工读片组 人工智能组 人工读片组 敏感度 94.61 89.31* 90.67 89.34 92.04 91.32 特异性 93.12 88.14* 89.67 86.57* 90.37 81.93* 符合率 92.08 87.15* 91.65 85.36* 91.49 80.14* *P < 0.05 vs同直径组人工智能组. -
[1] 鄂林宁, 马大庆. 肺内小结节的影像诊断和处理[J]. 中华放射学杂志, 2009, 43(3): 332-4. doi: 10.3760/cma.j.issn.1005-1201.2009.03.034 [2] 唐威, 王建卫, 吴宁, 等. 计算机辅助检测系统在低剂量肺癌筛查结节检出中的应用价值[J]. 中华放射学杂志, 2012, 46(7): 619-23. doi: 10.3760/cma.j.issn.1005-1201.2012.07.009 [3] 王晓华, 马大庆. 计算机辅助诊断在肺结节中的应用进展[J]. 中华放射学杂志, 2006, 40(4): 443-5. doi: 10.3760/j.issn:1005-1201.2006.04.025 [4] Hinton GE, Osindero S, Teh YW. A fast learning algorithm for deep belief nets[J]. Neural Comput, 2006, 18(7): 1527-54. doi: 10.1162/neco.2006.18.7.1527 [5] Hao X, Zhang GG, Ma S. Deep learning[J]. Int J Semantic Comput, 2016, 10(3): 417-39. doi: 10.1142/S1793351X16500045 [6] 陈疆红, 钟朝辉, 江桂莲, 等. 人工智能肺结节辅助诊断系统预测亚实性肺结节恶性概率[J]. 中国医学影像技术, 2020, 36(4): 535-9. https://www.cnki.com.cn/Article/CJFDTOTAL-ZYXX202004014.htm [7] 戴正行, 胡春洪, 王希明, 等. 基于DenseNet网络深度学习法CT图像人工智能分析技术判断肺结节良恶性[J]. 放射学实践, 2020, 35(4): 484-8. https://www.cnki.com.cn/Article/CJFDTOTAL-FSXS202004020.htm [8] 金晨望, 郭佑民. 人工智能辅助诊断技术在低剂量CT肺结节筛查中的应用及质控[J]. 中华放射学杂志, 2019, 53(1): 6-8. doi: 10.3760/cma.j.issn.1005-1201.2019.01.003 [9] 刘晓鹏, 周海英, 胡志雄, 等. 人工智能识别技术在T1期肺癌诊断中的临床应用研究[J]. 中国肺癌杂志, 2019, 22(5): 319-23. https://www.cnki.com.cn/Article/CJFDTOTAL-FAIZ201905011.htm [10] MacMahon H, Naidich DP, Goo JM, et al. Guidelines for management of incidental pulmonary nodules detected on CT images: from the fleischner society 2017[J]. Radiology, 2017, 284(1): 228-43. doi: 10.1148/radiol.2017161659 [11] Lee KH, Goo JM, Park SJ, et al. Correlation between the size of the solid component on thin-section CT and the invasive component on pathology in small lung adenocarcinomas manifesting as ground-glass nodules[J]. J Thorac Oncol, 2014, 9(1): 74-82. doi: 10.1097/JTO.0000000000000019 [12] Kim DW, Jang HY, Kim KW, et al. Design characteristics of studies reporting the performance of artificial intelligence algorithms for diagnostic analysis of medical images: results from recently published papers[J]. Korean J Radiol, 2019, 20(3): 405-10. doi: 10.3348/kjr.2019.0025 [13] 明佳蕾, 方向明. 基于人工智能的CT肺结节检出临床应用及研究进展[J]. 中华放射学杂志, 2019, 53(6): 522-5. [14] Setio AAA, Ciompi F, Litjens G, et al. Pulmonary nodule detection in CT images: false positive reduction using multi-view convolutional networks[J]. IEEE Trans Med Imaging, 2016, 35(5): 1160-9. doi: 10.1109/TMI.2016.2536809 [15] Cicero M, Bilbily A, Colak E, et al. Training and validating a deep convolutional neural network for computer-aided detection and classification of abnormalities on frontal chest radiographs[J]. Invest Radiol, 2017, 52(5): 281-7. doi: 10.1097/RLI.0000000000000341 [16] Park SH, Han K. Methodologic guide for evaluating clinical performance and effect of artificial intelligence technology for medical diagnosis and prediction[J]. Radiology, 2018, 286(3): 800-9. doi: 10.1148/radiol.2017171920 [17] 王爽, 雷盛, 谷涛, 等. 人工智能肺结节辅助诊断软件在肺CT检查结节分析中的临床应用[J]. 中华肺部疾病杂志: 电子版, 2019, 12(6): 757-9. doi: 10.3877/cma.j.issn.1674-6902.2019.06.019 [18] 李欣菱. 基于深度学习的人工智能胸部CT肺结节检测效能评估[D]. 天津: 天津医科大学, 2019. [19] 吕文晖, 周长圣, 李新宇, 等. 利用深度学习模型判断基线胸部平扫CT肺结节的良恶性[J]. 中华放射学杂志, 2019, 53(11): 957-62. [20] 南岩东, 李玉娟, 刘苗苗, 等. 人工智能在肺结节良恶性鉴别诊断中的价值分析[J]. 中华肺部疾病杂志: 电子版, 2020, 13(6): 760-3. doi: 10.3877/cma.j.issn.1674-6902.2020.06.009 [21] 滕雅琴, 贾文霄, 王云玲, 等. 计算机辅助检测系统在CT筛查肺结节中的应用研究[J]. 中国CT和MRI杂志, 2016, 14(5): 33-5. doi: 10.3969/j.issn.1672-5131.2016.05.011 [22] 刘凯, 张荣国, 涂文婷, 等. 深度学习技术对胸部X线平片亚实性结节的检测效能初探[J]. 中华放射学杂志, 2017, 51(12): 918-21. doi: 10.3760/cma.j.issn.1005-1201.2017.12.006