Nucleolin and nucleophosmin expression patterns in pulmonary adenocarcinoma invading the pleura and in pleural malignant mesothelioma

Development and validation of 18F-FDG PET/CT radiomics-based nomogram to predict visceral pleural invasion in solid lung adenocarcinoma

OBJECTIVES

This study aimed to establish a radiomics model based on 18F-FDG PET/CT images to predict visceral pleural invasion (VPI) of solid lung adenocarcinoma preoperatively.

METHODS

We retrospectively enrolled 165 solid lung adenocarcinoma patients confirmed by histopathology with 18F-FDG PET/CT images. Patients were divided into training and validation at a ratio of 0.7. To find significant VPI predictors, we collected clinicopathological information and metabolic parameters measured from PET/CT images. Three-dimensional (3D) radiomics features were extracted from each PET and CT volume of interest (VOI). Receiver operating characteristic (ROC) curve was performed to determine the performance of the model. Accuracy, sensitivity, specificity and area under curve (AUC) were calculated. Finally, their performance was evaluated by concordance index (C-index) and decision curve analysis (DCA) in training and validation cohorts.

RESULTS

165 patients were divided into training cohort (n = 116) and validation cohort (n = 49). Multivariate analysis showed that histology grade, maximum standardized uptake value (SUVmax), distance from the lesion to the pleura (DLP) and the radiomics features had statistically significant differences between patients with and without VPI (P < 0.05). A nomogram was developed based on the logistic regression method. The accuracy of ROC curve analysis of this model was 75.86% in the training cohort (AUC: 0.867; C-index: 0.867; sensitivity: 0.694; specificity: 0.889) and the accuracy rate in validation cohort was 71.55% (AUC: 0.889; C-index: 0.819; sensitivity: 0.654; specificity: 0.739).

CONCLUSIONS

A PET/CT-based radiomics model was developed with SUVmax, histology grade, DLP, and radiomics features. It can be easily used for individualized VPI prediction.