J Cai1, H Liu2, H Yuan1, Y Wu3, Q Xu3, Y Lv3, J Li3, J Fu3, J Ye4. 1. Dalian Medical University, Dalian, China; Department of Radiology, Northern Jiangsu People's Hospital, Yangzhou, China. 2. Yizhun Medical AI, Beijing, China. 3. Department of Radiology, Northern Jiangsu People's Hospital, Yangzhou, China. 4. Department of Radiology, Northern Jiangsu People's Hospital, Yangzhou, China. Electronic address: 18051061289@163.com.
Abstract
AIM: To establish a machine-learning model to differentiate adenocarcinoma in situ (AIS) or minimally invasive adenocarcinoma (MIA) from invasive adenocarcinoma (IAC) appearing as pure ground-glass nodules (pGGNs). MATERIALS AND METHODS: This retrospective study enrolled 136 patients with histopathologically diagnosed with AIS, MIA, and IAC. All pGGNs were divided randomly into a training and a testing dataset at a ratio of 7 : 3. Radiomics features were extracted based on the unenhanced computed tomography (CT) images derived from the last preoperative CT examination of each patient. The F-test and least absolute shrinkage and selection operator (LASSO) logistic regression were applied to select the most valuable features to establish a support vector machine (SVM) model. The performance of the model was evaluated using the area under the receiver operating characteristic curve (AUROC), and the accuracy, sensitivity, and specificity were calculated to compare the diagnostic performance of radiologists and the SVM model. RESULTS: Six significant radiomics features were selected to develop the SVM model and showed excellent ability to differentiate AIS/MIA from IAC in both the training dataset (AUROC=0.950, 95% confidence interval [CI]: 0.886-0.984) and the testing dataset (AUROC=0.945, 95% CI: 0.826-0.992). Compared with two radiologists, the proposed model possessed significant advantages with higher accuracy (90.24% versus 75.61% and 80.49%), sensitivity (91.67% versus 50% and 75%), and specificity (89.66% versus 86.21% and 82.76%). CONCLUSION: A machine-learning model based on radiomics features exhibits superior diagnostic performance in differentiating AIS/MIA from IAC appearing as pGGNs.
AIM: To establish a machine-learning model to differentiate adenocarcinoma in situ (AIS) or minimally invasive adenocarcinoma (MIA) from invasive adenocarcinoma (IAC) appearing as pure ground-glass nodules (pGGNs). MATERIALS AND METHODS: This retrospective study enrolled 136 patients with histopathologically diagnosed with AIS, MIA, and IAC. All pGGNs were divided randomly into a training and a testing dataset at a ratio of 7 : 3. Radiomics features were extracted based on the unenhanced computed tomography (CT) images derived from the last preoperative CT examination of each patient. The F-test and least absolute shrinkage and selection operator (LASSO) logistic regression were applied to select the most valuable features to establish a support vector machine (SVM) model. The performance of the model was evaluated using the area under the receiver operating characteristic curve (AUROC), and the accuracy, sensitivity, and specificity were calculated to compare the diagnostic performance of radiologists and the SVM model. RESULTS: Six significant radiomics features were selected to develop the SVM model and showed excellent ability to differentiate AIS/MIA from IAC in both the training dataset (AUROC=0.950, 95% confidence interval [CI]: 0.886-0.984) and the testing dataset (AUROC=0.945, 95% CI: 0.826-0.992). Compared with two radiologists, the proposed model possessed significant advantages with higher accuracy (90.24% versus 75.61% and 80.49%), sensitivity (91.67% versus 50% and 75%), and specificity (89.66% versus 86.21% and 82.76%). CONCLUSION: A machine-learning model based on radiomics features exhibits superior diagnostic performance in differentiating AIS/MIA from IAC appearing as pGGNs.