Agreement of manual, semi-automatic, and automatic measurements in measuring diameters and volumes of solid pulmonary nodules
10.3760/cma.j.cn112149-20210203-00094
- VernacularTitle:人工、半自动及自动方法测量肺实性结节直径和体积的一致性研究
- Author:
Xi YANG
1
;
Jieke LIU
;
Yong LI
;
Haomiao QING
;
Changjiu HE
;
Peng ZHOU
Author Information
1. 四川省肿瘤医院·研究所 四川省癌症防治中心 电子科技大学医学院附属肿瘤医院影像诊治部，成都 610041
- Keywords:
Lung neoplasms;
Tomography, X-ray computed;
Diameter;
Volume;
Agreement
- From:
Chinese Journal of Radiology
2022;56(1):43-49
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To assess the agreement of manual measurement, semi-automatic measurement based on computer-aided diagnosis (CAD), and automatic measurement based on artificial intelligence (AI) in measuring diameters and volumes of solid pulmonary nodules.Methods:The clinical and low dose CT (LDCT) data of 165 participants in lung cancer screening of Sichuan Cancer Hospital from July 2018 to April 2020 were retrospectively analyzed. The largest nodule of each participant was selected to analyze, and its diameter and volume were measured by one junior and one senior radiologist using manual measurement, semi-automatic measurement based on CAD, and automatic measurement based on AI. Referring to Lung CT imaging reporting and data system (Lung-RADS, version 1.1), all nodules were classified into Lung-RADS 2, 3, 4A, 4B, 4X categories and low and high risk groups according to the diameter and volume measured by different measurement methods. Repeated-measures analysis of variance and paired t-test were used to compare the differences in the diameter and volume of lung nodules measured by different methods. The consistency of the three methods in measuring nodule diameter and volume was assessed by the correlation coefficient (ICC). Linear weighted Kappa coefficient was applied to assess the consistency of different measurement methods in Lung-RADS classification results; simple Kappa coefficient was applied to assess the consistency of different methods in high and low risk grouping results. Results:Difference in the diameters of pulmonary nodules measured by manual measurement, semi-automatic measurement based on CAD, and automatic measurement based on AI was statistically significant [(14.9±6.3) mm, (17.0±6.7) mm, (15.0±5.7) mm, F=88.39, P<0.001], and the pairwise comparisons showed that there was significant difference between semi-automatic measurement based on CAD and manual measurement method ( t=10.97, P<0.001), semi-automatic measurement based on CAD and automatic measurement based on AI ( t=10.07, P<0.001), but no significant difference between manual measurement method and automatic measurement based on AI method ( t=1.04, P=0.301). There was no significant difference in the measurement of pulmonary nodule volumes between the semi-automatic measurement and the automatic measurement method based on AI ( Z=0.70, P=0.482). The consistency of pulmonary nodules diameter measured by different measurement methods was high (ICC>0.75), and the consistency of semi-automatic and automatic measurement of lung nodule volume was high (ICC=0.927). The consistency of three methods for lung-RADS classification and high-and low-risk grouping according to nodule diameter was good (Kappa>0.80). The agreements between the semi-automatic measurement and the automatic measurement method for Lung-RADS classification and high-and low-risk grouping according to nodule volume were good (Kappa>0.80). Conclusion:In terms of diameter measurement of solid pulmonary nodules, automatic measurement based on AI is more consistent with manual measurement than semi-automatic measurement based on CAD. The agreement between automatic measurement and semi-automatic measurement is high in terms of volume measurement.
