|Table of Contents|

Machine learning model predicts benign and malignant pulmonary nodules based on CT features(PDF)

《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]

Issue:
2024年第10期
Page:
1315-1320
Research Field:
医学人工智能
Publishing date:

Info

Title:
Machine learning model predicts benign and malignant pulmonary nodules based on CT features
Author(s):
CONG Yulin XU Xiaohu SHEN Chunlin XU Yachun
Department of Imaging, Haian Peoples Hospital, Haian 226600, China
Keywords:
Keywords: CT feature model construction pulmonary nodule benign and malignance machine learning model
PACS:
R318;R734.2
DOI:
DOI:10.3969/j.issn.1005-202X.2024.10.017
Abstract:
Abstract: Objective To construct a machine learning model for predicting benign and malignant pulmonary nodules based on CT features. Methods A total of 129 patients with single solid nodules on CT from January 2021 to January 2023 in Haian Peoples Hospital were selected. All of them underwent chest CT scan, and the quantitative parameters, morphological features and radiomics features were recorded. The differentiation of benign and malignant pulmonary nodules was carried out according to relevant diagnostic criteria. The cases were divided into the training set and the internal test set. The constructed models included radiomics labels, morphological model, CT model and combined model. Results There were 98 cases in the training set (27 malignance and 71 benign) and 31 cases in the internal test set (7 malignance and 24 benign). Univariate analysis showed that there were significant differences in age, lesion diameter, mean density, burr sign, pleural depression sign, vacuole sign and air bronchial sign between malignant group and benign group (P<0.05). Compared with benign group, malignant group had higher proportions of burr sign, pleural depression sign, vacuole sign, air bronchial sign, and larger lesion diameter and mean density (P<0.05). LinkDocAI intelligent diagnosis system for pulmonary nodules was used to outline regions of interest and from which 1 000 radiomics features were extracted. The feature selection was performed in 98 cases, and 20 features were screened out after standardized treatment and correlation testing, excluding missing features, low importance feature values and highly correlated features. Through LASSO regression and 10-fold cross validation, λ1se was selected as the optimal λ to construct radiomics labels, and the two most meaningful features (LBP_Glszm_ZoneEntropy and Gradient_Shape_MinorAxis) were enrolled. CT model was considered as the optimal model in this study, and it had an area under receiver operating characteristic curve of 0.912 and 0.889 in the training set and the internal testing set, respectively. Conclusion The machine learning model to predict benign and malignant lung nodules based on CT features has good predictive efficiency, and it can realize the differential diagnosis of benign and malignant pulmonary nodules.

References:

Memo

Memo:
-
Last Update: 2024-10-29