Low-dose CT radiomics nomogram differentiates pure ground-glass microinvasive adenocarcinoma from invasive adenocarcinoma
WANG Hailin1, 2, LIN Guihan1, 2, CHEN Weiyue1, 2, YING Haifeng1, 2, WENG Yaqin3, FU Weidong3, WENG Qiaoyou1, 2, LU Chenying1, 2, JI Jiansong1, 2.
1.Department of Radiology, the Fifth Affiliated Hospital of Wenzhou Medical University, Lishui 323000 China; 2.Key Laboratory of Imaging Diagnosis and Minimally Invasive Intervention Research, Lishui 323000 China; 3.Department of Radiology,Liandu District People’s Hospital, Lishui 323000 China
WANG Hailin,LIN Guihan,CHEN Weiyue, et al. Low-dose CT radiomics nomogram differentiates pure ground-glass microinvasive adenocarcinoma from invasive adenocarcinoma[J]. JOURNAL OF WEZHOU MEDICAL UNIVERSITY, 2024, 54(1): 7-13,19.
Abstract：Objective: To evaluate the value of low-dose CT image nomogram in the differential diagnosis of pulmonary microinvasive adenocarcinoma (MIA) and pulmonary invasive adenocarcinoma (IAC) in pure ground-glass nodule (pGGN). Methods: A retrospective analysis of the clinical and CT imaging data of 239 lung adenocarcinoma patients with pGGN confirmed by surgery and pathology at the Fifth Affiliated Hospital of Wenzhou Medical University from January 2018 to April 2023, including 93 cases of MIA and 146 cases of IAC.Patients were divided into a training set (n=167) and a validation set (n=72) at a 7:3 ratio using the complete randomization method. The radiomics features of the lesions in low-dose CT images were extracted using Radcloud platform, and the best features were preserved by dimensionality reduction. Subsequently, three machine learning classifiers, including Logistic regression (LR), support vector machine support vector machine (SVM),and random forest (RF), were established to validate the classifier with the highest area under the concentration curve (AUC) as the best radiomics model, with the output results as Rad-score. The clinical and CT features with P<0.05 were included in the multivariate logistic regression analysis to screen out the independent risk factors and
establish the clinical model. Finally, a joint model was constructed based on Rad-score and clinical risk factors,and a nomogram was drawn. The area under the receiver operating characteristic (ROC) curve (AUC), sensitivity,specificity and accuracy were used to evaluate the diagnostic performance of the model. Results: Fifteen radiomics features significantly related to the differential diagnosis of MIA and IAC were obtained by dimensionality reduction. Among the three machine learning classifiers, RF had the best diagnostic performance, with AUC being 0.837 in training set and 0.788 in verification set. Multivariate logistic regression analysis showed that the long largest diameter, irregular shape and spiculate sign were independent risk factors for the identification of MIA and IAC. The results of ROC showed that the linear graph had a good diagnostic performance. The AUC, sensitivity,specificity and accuracy in the training set were 0.913, 87.25%, 81.54% and 84.94%, respectively, while those in the validation set were 0.862, 88.63%, 75.01% and 82.78%, respectively. Conclusion: Low-dose CT radiomics nomogram can distinguish MIA and IAC with pGGN, which can be used to guide clinical operation planning.
