A predictive model for identifying pulmonary lesions based on CT features of rimmed signs and satellite lesions: A multicentre study

Abstract

Objectives To verify a differential prediction model based on different lung pathology types utilizing the rimmed sign (RS) and satellite lesion (SL) methods to enhance identification efficiency. Methods From January 2015 to October 2023, the clinical data and chest CT images of 3030 patients with lung lesions were analysed. The lesions were divided into three groups: malignant, inflammatory, and benign. The pairwise identification models were constructed separately with (Models_1-3) or without (Models_1*-3*) rimmed signs and satellite lesions, and the developing and test group were divided by 7:3. Six models were built to distinguish between these groups (Model₁ or Model_1*: malignant vs. inflammatory; Model₂ or Model_2*: malignant vs. benign; Model₃ or Model_3*: inflammatory vs. benign). The curve (AUC) was calculated to evaluate the performance of these models. The Delong test was used to compare the differences between different models. Results In the test group, the AUC and Accuracy of Models_1-3 and Models_1-3* were 0.920/84.8%, 0.990/96.2%, 0.881/76.4%, and 0.900/73.5%, 0.989/90.1%, and 0.869/78.6%, respectively. The Delong test showed no significant difference between Models_1-3 and Models_1-3* (p > 0.05), Accuracy (T₁, T₂) > Accuracy (T_1*, T_2*), Accuracy (T₃) < Accuracy (T_3*). Conclusions The six prediction models in this study effectively differentiated among different types of lung lesions, with the rimmed sign and satellite lesion features improving the accuracy of Model₁ and Model₂.