Clinical value of quantitative artificial intelligence imaging parameters for predicting the benign and malignant nature and the risk of recurrence of lung nodules ≤2 cm
- VernacularTitle:人工智能影像量化参数预测≤2 cm肺结节良恶性及复发风险的临床价值
- Author:
Yujie MA
1
;
Zhe PIAO
1
;
Yuhe YOU
1
;
Wenjun CAO
1
;
Hongsheng XUE
1
;
Zhilong ZHAO
1
Author Information
1. Department of Thoracic Surgery, Affiliated Zhongshan Hospital of Dalian University, Dalian, 116601, Liaoning, P. R. China
- Publication Type:Journal Article
- Keywords:
Pulmonary nodules;
artificial intelligence;
quantitative parameters;
pathologic subtypes;
prediction model
- From:
Chinese Journal of Clinical Thoracic and Cardiovascular Surgery
2025;32(10):1384-1392
- CountryChina
- Language:Chinese
-
Abstract:
Objective To evaluate the value of imaging quantification parameters in artificial intelligence (AI) assisted diagnosis systems in clinical decision-making for lung nodules≤2 cm and the diagnostic efficacy of AI. Methods Lung nodule patients admitted to Affiliated Zhongshan Hospital of Dalian University from 2020 to 2023 were included. Imaging parameters of lung nodules were extracted using AI assisted diagnosis systems. Multifactor analysis was used to screen predictors for distinguishing benign and malignant nodules and high-risk predictors for recurrent invasive adenocarcinoma, and a diagnostic model was established and its performance evaluated. The diagnostic efficacy of the AI system was judged according to pathological results. Results A total of 594 patients with lung nodules were included, including 202 males and 392 females, with an average age of (58.75±11.55) years. Volume, average CT value, and 3D maximum diameter of non-solid nodules were independent predictors of malignant nodules, with thresholds of 287.4 mm3, −491 HU, and 12.0 mm, respectively. The area under the curve (AUC) for diagnostic efficacy was ranked from high to low as combined model (0.802), volume (0.783), average CT value (0.749), and 3D maximum diameter (0.714). The average CT value and 3D long diameter of solid nodules were independent predictors of malignant nodules, with thresholds of −81 HU and 17.5 mm, respectively, and AUC values of 0.874 and 0.686, respectively, with the combined prediction AUC of 0.957. The mass of cystic nodules was an independent predictor of malignancy when the mass>180.7 mg. Independent predictors of high recurrence risk of invasive adenocarcinoma in non-solid nodules were consolidation-tumor ratio (CTR), average CT value, 3D long diameter, and volume, with thresholds of 0.14, −386 HU, 15.6 mm, and 1018.9 mm3, respectively, and diagnostic efficacy was ranked from high to low as combined model (0.788), 3D long diameter (0.735), volume (0.725), average CT value (0.720), and CTR (0.697). The accuracy of AI in predicting benign and malignant target nodules was 87.4%, with positive predictive value of 96.6% and negative predictive value of 58.9%. Conclusion In clinical decision-making for lung nodules ≤2 cm, AI assisted diagnosis systems have high application value.
