Diagnosis of Benign and Malignant Pulmonary Ground-Glass Nodules Using Computed Tomography Radiomics Parameters

Abstract

Objective: To assess the clinical value of a radiomics model based on low-dose computed tomography (LDCT) in diagnosing benign and malignant pulmonary ground-glass nodules. Methods: A retrospective analysis was performed on 274 patients who underwent LDCT scanning with the identification of pulmonary ground-glass nodules from January 2018 to March 2021. All patients had complete clinical and pathological data. The cases were randomly divided into 191 cases in a training set and 83 cases in a validation set using the random sampling method and a 7:3 ratio. Based on the predictor sources, we established clinical, radiomics, and combined prediction models in the training set. A receiver operating characteristic (ROC) curve was generated for the training and validation sets, the predictive abilities of the different models for benign and malignant nodules were compared according to the area under the curve (AUC), and the model with the best predictive ability was selected. A calibration curve was plotted to test the good-of-fitness of the model in the validation set. Results: Of the 274 patients (84 males and 190 females), 156 had malignant, and 118 had benign nodules. The univariate analysis showed a statistically significant difference in nodule position between benign nodules and lung adenocarcinoma in both data sets ( P <.001 and .021). In the training set, when the nodule diameter was >8 mm, the probability of nodule malignancy increased ( P < .001). The results showed that the combined model had a higher prediction ability than the other two models. The combined model could distinguish between benign and malignant pulmonary nodules in the training set (AUC: 0.711; 95%CI: 0.634-0.787; ACC: 0.696; sensitivity: 0.617; specificity: 0.816; PPV:0.835; NPV: 0.585). Moreover, this model could predict benign and malignant nodules in the validation set (AUC: 0.695; 95%CI: 0.574-0.816; ACC: 9.747; sensitivity: 0.694; specificity: 0.824; PPV: 0.850; NPV: 0.651). The calibration curve had a P value of 0.775, indicating that in the validation set, there was no difference between the value predicted by the combined model and the actual observed value and that the result was a good fit. Conclusion: The prediction model combining clinical information and radiomics parameters had a good ability to distinguish benign and malignant pulmonary ground-glass nodules.

Artificial-intelligence-based computed tomography histogram analysis predicting tumor invasiveness of lung adenocarcinomas manifesting as radiological part-solid nodules

“…Beyond the first-order histogram features, the second-and higher-order texture features extracted by radiomics techniques were also integrated in the assessment of invasiveness among SSNs (30)(31)(32). Li et al obtained the CT texture features of 109 SSNs and found that the surface area feature and the extruded surface area feature could be predictors of IACs compared with MIAs (33).…”