Radiomics and Clinical Characters Based Gaussian Naive Bayes (GNB) Model for Preoperative Differentiation of Pulmonary Pure Invasive Mucinous Adenocarcinoma From Mixed Mucinous Adenocarcinoma.

Abstract

Objective: To develop and validate predictive models based on clinical parameters, and radiomic features to distinguish pulmonary pure invasive mucinous adenocarcinoma (pIMA) from mixed mucinous adenocarcinoma (mIMA) before surgery. Method: From January 2017 to December 2022, 193 pIMA and 111 mIMA were retrospectively analyzed at our hospital in this retrospective study. From contrast-enhanced computed tomography, 1037 radiomic features were extracted. The patients were randomly divided into a training group and a test group (n = 213 and 91, respectively) in a 7:3 ratio. The least absolute shrinkage and selection operator algorithm was used to select radiomic features. In this study, 9 machine learning radiomics prediction models were applied. The radiomics score was then calculated based on the best-performing machine learning model adopted. The clinical model was developed using the same machine learning model of radiomics. In the end, a combined model based on clinical factors and radiomics features was developed. The area under the receiver operating characteristic curve (AUC) value and decision curve analysis (DCA) were used to evaluate the clinical usefulness of the prediction model. Results: The combined model established by the Gaussian Naive Bayes machine learning method exhibited the best performance. The AUC of the combined model, clinical model, and radiomics model were 0.81, 0.80, and 0.68 in the training group and 0.91, 0.80, and 0.81 in the test group, respectively. The Brier scores of the combined model were 0.171 and 0.112. The DCA curve also showed that the combined model was beneficial to clinical settings. Conclusion: The combined model integration of radiomics features and clinical parameters may have potential value for the preoperative differentiation of pIMA from mIMA.