«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《中华肺部疾病杂志》[ISSN:1006-6977/CN:61-1281/TN]

卷:

期数:

2024年06期

页码:

931-935

栏目:

论著

出版日期:

2024-12-25

文章信息/Info

Title:

Significance of different CT reconstruction algorithms in the diagnosis of pulmonary nodules with artificial intelligence

作者:

蔡玉琳; 牛 丹; 马 丹; 王 爽

400037 重庆,陆军(第三)军医大学第二附属医院放射科

Author(s):

Cai Yulin;  Niu Dan;  Ma Dan;  Wang Shuang.

Department of Radiology,The Second Affiliated Hospital of Army Medical University, Chongqing 400037, China

关键词:

肺结节;  高分辨CT;  重建算法;  人工智能

Keywords:

Pulmonary nodules;  High resolution CT;  Reconstruction algorithm;  Artificial intelligence

分类号:

R563

DOI:

10.3877/cma.j.issn.1674-6902.2024.06.014

摘要:

目的 分析不同CT重建算法对人工智能(artificial intelligence, AI)辅助肺结节识别效能的影响。方法 选取2022年12月至2023年4月我院行肺部高分辨CT(high resolution CT, HRCT)扫描的患者200例为对象,采用肺高分辨算法(lung high-resolution algorithm, Lung)和标准算法(standard algorithm, Stnd)两种算法进行层厚为0.625 mm的薄层重建。将两种重建算法得到的原始数据传送至AI辅助诊断软件进行肺结节自动检测,纪录结节的密度、大小。根据结节大小将实性结节分为<5 mm、≥5 mm,亚实性结节分为<8 mm、≥8 mm,统计两种重建算法下AI识别结节的真阳数和假阳数,计算AI对肺结节识别的敏感度。比较两种重建算法下AI软件识别肺结节效能是否有差异。结果 实性结节<5 mm结节996个,Lung算法下真阳数697个(敏感度69.98%),Stnd算法下真阳数817个(敏感度82.03%); ≥5 mm结节74个,Lung算法下真阳数47个(敏感度63.51%),Stnd算法下真阳数67个(敏感度90.54%)。亚实性结节<8 mm 结节358个,Lung算法下真阳数230个(敏感度64.25%),Stnd算法下真阳数340个(敏感度94.97%); ≥8 mm 结节35个,Lung算法下真阳数35个(敏感度100.00%),Stnd算法下真阳数34个(敏感度97.14%)。在Stnd算法下AI识别2组不同大小的实性结节(<5 mm和≥5 mm)敏感度高于Lung算法。对亚实性结节的识别,结节<8 mm时Stnd算法敏感度高于Lung算法(P<0.05),≥8 mm时两种算法之间无差异(P>0.05)。结论 不同CT重建算法条件下AI辅助肺结节识别效能存在差异,对于肺实性结节的识别应用Stnd重建算法敏感度优于Lung算法。对于肺亚实性结节的识别,结节<8 mm运用Stnd算法敏感度高于Lung算法,但假阳性高,≥8 mm两种算法无差异性有高的敏感度。

Abstract:

Objective To analyze the effects of different CT reconstruction algorithms on the effectiveness of artificial intelligence(AI)-assisted pulmonary nodules identification. Methods A total of 200 patients who underwent lung high resolution CT examination in our hospital from December 2022 to April 2023 were selected and reconstructed with a thin layer thickness of 0.625 mm using lung high-resolution algorithm(Lung)and standard algorithm(Stnd). The raw data obtained by the two reconstruction algorithms were transmitted to the AI-assisted diagnostic software for automatic detection of pulmonary nodules, and the density and size of the nodules were recorded respectively. According to the size of the nodules, the solid nodules were divided into <5 mm and ≥5 mm. The subsolid nodules were divided into <8 mm and ≥8 mm. The true positive and false positive numbers of nodule identification by AI in each group under the two reconstruction algorithms are separately recorded, and the sensitivity of AI for the identification of pulmonary nodules was calculated. To compare whether there is a difference in the efficiency of AI software in identifying pulmonary nodules under the two reconstruction algorithms. Results There were 996 solid nodules <5 mm, the true positive numbers and sensitivity were 697(69.98%)under Lung algorithm and 817(82.03%)under Stnd algorithm. There were 74 nodules ≥5 mm, the true positive numbers and sensitivity were 47(63.51%)under Lung algorithm and 67(90.54%)under Stnd algorithm. There were 358 subsolid nodules <8 mm, the true positive numbers and sensitivity were 230(64.25%)under Lung algorithm and 340(94.97%)under Stnd algorithm. There were 35 nodules ≥8 mm, the true positive numbers and sensitivity were 35(100.00%)under Lung algorithm and 34(97.14%)under Stnd algorithm. The sensitivity of AI to identify 2 groups of solid nodules of different sizes(<5 mm and ≥ 5 mm)under Stnd algorithm was higher than that of Lung algorithm. The sensitivity of the Stnd algorithm was higher than that of the Lung algorithm for the identification of subsolid nodules <8 mm(P<0.05), and there was no difference between the two algorithms for the identification of nodules ≥ 8 mm(P>0.05). Conclusion There are differences in the identification efficiency of AI-assisted pulmonary nodules under different CT reconstruction algorithms. The sensitivity of Stnd reconstruction algorithm is better than Lung algorithm for the recognition of solid pulmonary nodules. For the identification of subsolid pulmonary nodules, the sensitivity of Stnd algorithm for nodules <8 mm is higher than that of Lung algorithm, but the false positives are higher. There is no difference between the two algorithms for nodules ≥8 mm and both have higher sensitivity.

参考文献/References:

1 中华医学会肿瘤学分会, 中华医学会杂志社. 中华医学会肺癌临床诊疗指南(2023版)[J]. 中华医学杂志, 2023, 103(27): 2037-2074.
2 邓传颂, 杜巧玲. 人工智能对肺磨玻璃结节的良恶性分析[J]. 影像研究与医学应用, 2022, 6(19): 83-85.
3 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): 319-323.
4 黄云开. CT高分辨重建联合人工智能在肺结节识别诊断中的应用效果分析[J]. 现代医用影像学, 2021, 30(1): 87-89.
5 陈淑娇, 谢丽萍, 陈 懿. 肺内微小结节CT影像征象与病理表现在早期肺癌诊断中的临床应用[J/CD]. 中华肺部疾病杂志(电子版), 2020, 13(4): 521-523.
6 Peter JM, Louis L. Evaluating the patient with a pulmonary nodule: A review[J]. JAMA, 2022, 327(3): 264-273.
7 谭培兰, 张晓林, 柏 辉, 等. 计算机辅助CT图像特征在磨玻璃结节早期肺癌诊断中的应用[J]. 癌症进展, 2019, 17(16): 1946-1948,1952.
8 申太忠. 胸部高分辨CT在肺结节诊断中的应用价值[J]. 影像研究与医学应用, 2020, 4(6): 218-219.
9 曾文彬. 多层螺旋CT高分辨扫描在肺结节病鉴别诊断中的应用[J]. 医疗装备, 2019, 32(23): 27-28.
10 张 强, 李丽香. 高分辨率CT对肺小结节及早期肺癌的鉴别诊断价值[J]. 癌症进展, 2022, 20(19): 2019-2021, 2025.
11 Liu K, Kang G. Multiview convolutional neural networks for lung nodule classification[J]. Int J Imaging Syst Technol, 2017, 27(1): 12-22.
12 曹恩涛, 范 丽, 肖湘生. CT计算机辅助检测与诊断对肺癌早期诊断的应用与进展[J]. 国际医学放射学杂志, 2016, 39(1): 55-60.
13 Pianykh OS, Langs G, Dewey M, et al. Continuous learning AI in radiology: Implementation principles and early applications[J]. Radiology, 2020, 297(1): 6-14.
14 刘士远, 萧 毅. 基于深度学习的人工智能对医学影像学的挑战和机遇[J]. 中华放射学杂志, 2017, 51(12): 899-901.
15 Ali I, Hart GR, Gunabushanam G, et al. Lung nodule detection via deep reinforcement learning[J]. Front Oncol, 2018, 8: 108.
16 You SK, Choi YH, Cheon JE, et al. Effect of low tube voltage and low iodine concentration abdominal CT on image quality and radiation dose in children: preliminary study[J]. Abdominal radiology(New York), 2019, 44(5): 1928-1935.
17 刘 杰. 低剂量16排螺旋CT重建层厚对肺小结节检出的影响[J]. 深圳中西医结合杂志, 2019, 29(1): 76-77.
18 陈 岩, 于小利, 高希法, 等. 基于全模型迭代重建算法的256排CT极低剂量肺结节筛查的临床研究[J]. 中国中西医结合影像学杂志, 2018, 16(6): 567-569,573.
19 程燕南, 李贤军, 李新雨, 等. CT重建算法和显示窗设置影响肺实性结节检出和测量的研究[J]. 西安交通大学学报(医学版), 2022, 43(3): 476-482.
20 曹 源, 李丹阳, 张 扬, 等. 100kVp管电压不同重建算法对AI辅助检测肺结节效能影响[J]. 放射学实践, 2020, 35(10): 1324-1328.
21 Tran GS, Nghiem TP, Nguyen VT, et al. Improving accuracy of lung nodule classification using deep learning with focal loss[J]. J Healthc Eng, 2019, 2019: 5156416.
22 范卫杰, 张 冬. 影像组学及深度学习在肺结节良恶性鉴别诊断中的新理念[J/CD]. 中华肺部疾病杂志(电子版), 2021, 14(5): 549-553.
23 Wang F, Casalino LP, Khullar D. Deep learning in medicine-promise, progress, and challenges[J]. JAMA Intern Med, 2019, 179(3): 293-294.
24 林耀彬, 林勇斌, 赵泽锐, 等.《人工智能在肺结节诊治中的应用专家共识(2022年版)》解读[J]. 中国胸心血管外科临床杂志, 2023, 30(5): 665-671.
25 Reyes M, Meier R, Pereira S, et al. On the interpretability of artificial intelligence in radiology: Challenges and opportunities[J]. Radiol Artif Intell, 2020, 2(3): e190043.
26 Lee JG, Jun S, Cho YW, et al. Deep learning in medical imaging: General overview[J]. Korean J Radiol, 2017, 18(4): 570-584.
27 Bi WL, Hosny A, Schabath MB, et al. Artificial intelligence in cancer imaging: Clinical challenges and applications[J]. CA Cancer J Clin, 2019, 69(2): 127-157.
28 Venkadesh KV, Setio AAA, Schreuder A, et al. Deep learning for malignancy risk estimation of pulmonary nodules detected at low-dose screening CT[J]. Radiology, 2021, 300(2): 438-447.
29 唐思源, 杨 敏, 白金牛. 基于深度卷积神经网络的肺结节检测与识别[J]. 科学技术与工程, 2019, 19(22): 241-248.
30 Zhao Y, de Bock GH, Vliegenthart R, et al. Performance of computer-aided detection of pulmonary nodules in low-dose CT: Comparison with double reading by nodule volume[J]. Eur Radiol, 2012, 22(10): 2076-2084.
31 Silva M, Schaefer-Prokop CM, Jacobs C, et al. Detection of subsolid nodules in lung cancer screening: Complementary sensitivity of visual reading and computer-aided diagnosis[J]. Invest Radiol, 2018, 53(8): 441-449.

备注/Memo

备注/Memo:

通信作者: 王 爽, Email: fskwangs@tmmu.edu.cn

常用功能

更新日期/Last Update: 2024-12-25