CT radiomics model for evaluation on pathologic types of lung adenocarcinoma in situ combined with minimally invasive adenocarcinoma and invasive adenocarcinoma
10.13929/j.issn.1003-3289.2020.09.017
- VernacularTitle: 基于CT影像组学模型预测肺原位腺癌及微浸润腺癌与浸润性腺癌
- Author:
Dingli YE
1
Author Information
1. Department of Radiology, Jilin Cancer Hospital
- Publication Type:Journal Article
- Keywords:
Lung neoplasms;
Pathology;
Radiomics;
Tomography, X-ray computed
- From:
Chinese Journal of Medical Imaging Technology
2020;36(9):1345-1349
- CountryChina
- Language:Chinese
-
Abstract:
Objective: To investigate the value of CT radiomics model for predicting pathologic types of adenocarcinoma in situ (AIS), minimally invasive adenocarcinoma (MIA) and invasive adenocarcinoma (IAC) among lung adenocarinoma. Methods: Data of 542 patients with pathologically confirmed lung adenocarcinoma and clear subtypes were retrospective analyzed. AIS and MIA were classified as group 1 while IAC as group 2. The gender and age were compared between 2 groups. Feature extraction software was used to extract three-dimensional texture feature parameters of each lesion, and the imaging omics features obviously different between 2 groups were retained, then the optimal imaging omics features were selected to build a predictive model. All the data were divided into training set and validation set in a ratio of 2: 1. Six machine learning algorithms were used to classify the five-fold cross-training sets to select the best classifier. Then, the five-fold cross-training data set, training set and validation set were analyzed with the prediction model to obtain the ROC curves of the model in predicting pathological subtypes of lung adenocarcinoma as well as the relative AUC, accuracy, specificity and sensitivity. Results: There were 235 patients in group 1 and 307 in group 2. No statistical difference of gender nor age was found between 2 groups (χ2=0.56, t=-0.19, P=0.63, 0.98). A total of 1 766 three-dimensional texture feature parameters were extracted from the lesions, including 988 imaging omics features significantly different between 2 groups. Finally, 10 optimal imaging omics features were retained to construct the prediction model. Perceptron classifier was the best classifier. AUC of the predictive model in predicting pathological subtypes of validation set was 0.95, and the relative accuracy, specificity and sensitivity was 0.88, 0.87 and 0.84, respectively. Conclusion: CT radiomics medel could effectively predict pathological subtypes of AIS, MIA and IAC among lung adenocarcinoma.
