Performance of radiomics models derived from different CT reconstruction parameters for lung cancer risk prediction

BMC Pulmonary Medicine

Research

Weimin Li¹  Bojiang Chen¹  Rui Zhang²  Jie Shi³  Siyun Liu³ 

[1]Department of Pulmonary and Critical Care Medicine, West China Hospital, Sichuan University, 37 GuoXue Alley, Wuhou District, 610041, Chengdu, Sichuan Province, People’s Republic of China

[2]Department of Pulmonary and Critical Care Medicine, West China Hospital, Sichuan University, 37 GuoXue Alley, Wuhou District, 610041, Chengdu, Sichuan Province, People’s Republic of China

[3]Department of General Practice, West China Hospital, Sichuan University, Chengdu, People’s Republic of China

[4]GE Healthcare, Shanghai, China

关键词: Pulmonary nodules;    Radiomics;    Computed tomography;    Reconstruction parameters;   

DOI  :  10.1186/s12890-023-02366-y

 received in 2022-08-15, accepted in 2023-02-21,  发布年份 2023

来源: Springer

【 摘 要 】

BackgroundThis study analysed the performance of radiomics features extracted from computed tomography (CT) images with different reconstruction parameters in differentiating malignant and benign pulmonary nodules.MethodsWe evaluated routine chest CT images acquired from 148 participants with pulmonary nodules, which were pathologically diagnosed during surgery in West China Hospital, including a 5 mm unenhanced lung window, a 5 mm unenhanced mediastinal window, a 5 mm contrast-enhanced mediastinal window and a 1 mm unenhanced lung window. The pulmonary nodules were segmented, and 1409 radiomics features were extracted for each window. Then, we created 15 cohorts consisting of single windows or multiple windows. Univariate correlation analysis and principal component analysis were performed to select the features, and logistic regression analysis was performed to establish models for each cohort. The area under the curve (AUC) was applied to compare model performance.ResultsThere were 75 benign and 73 malignant pulmonary nodules, with mean diameters of 18.63 and 19.86 mm, respectively. For the single-window setting, the AUCs of the radiomics model from the 5 mm unenhanced lung window, 5 mm unenhanced mediastinal window, 5 mm contrast-enhanced mediastinal window and 1 mm unenhanced lung window were 0.771, 0.808, 0.750, and 0.771 in the training set and 0.711, 0.709, 0.684, and 0.674 in the test set, respectively. Regarding the multiple-window setting, the radiomics model based on all four windows showed an AUC of 0.825 in the training set and 0.743 in the test set. Statistically, the 15 models demonstrated comparable performances (P > 0.05).ConclusionA single chest CT window was acceptable in predicting the malignancy of pulmonary nodules, and additional windows did not statistically improve the performance of the radiomics models. In addition, slice thickness and contrast enhancement did not affect the diagnostic performance.

【 授权许可】

CC BY   
© The Author(s) 2023

【 图 表 】

Fig. 2

13148_2022_1349_Article_IEq10.gif

Fig. 1