Forecast model establishment and screening of correlative factors of malignant and benign solitary pulmonary nodules

Objectives: To screen out associated radiological indexes and serological markers based on which assistant diagnosis model was established and then to evaluated their efficacy. Methods: With 260 cases of solitary pulmonary nodules (SPN) confirmed by pathology (145 malignant cases and 115 benign cases) as study subjects, correlative factors of solitary pulmonary nodules were screened out from 9 computerized tomography (CT) radiological indexes and 8 tumor makers by logistic regression analysis. On that basis, support vector machine (SVM) model and back propagation (BP) model were established and be compared with the model including all variables. Results: SVM assistant model and BP assistant model were successfully established. When all variables were included, the accuracy, specificity and sensitivity of SVM model was 75.0%, 85.0% and 65.0%, respectively while that of BP model was 77.5%, 75.0% and 80.0%, respectively; when all screening variables were included, the accuracy, specificity and sensitivity of SVM was 80.0%, 95.0% and 65.0%, respectively while that of BP model was 67.5%, 75.0% and 60.0%, respectively. Conclusions: Compared to BP model, SVM model possessed better efficacy and through filtering feature subsets, the accuracy and specificity of SVM model could be effectively improved, which could be seen as a good assistant diagnosis approach for examination of malignant and benign SPN and offered directions for future SPN forecast model in clinical employment and development.