In this study, we aimed to evaluate the role of 18F-fluorodeoxyglucose PET/computerized tomography (18F-FDG PET/CT)-based radiomic features in the differentiation of infection and malignancy in consolidating pulmonary lesions and to develop a prediction model based on radiomic features.

The images of 106 patients who underwent 18F-FDG PET/CT of consolidated lesions observed in the lung between January 2015 and July 2020 were evaluated using LIFEx software. The region of interest of the lung lesions was determined and volumetric and textural features were obtained. Clinical and radiomic data were evaluated with machine learning algorithms to build a model.

There was a significant difference in all standardized uptake value (SUV) parameters and 26 texture features between the infection and cancer groups. The features with a correlation coefficient of less than 0.7 among the significant features were determined as SUVmean, GLZLM_SZE, GLZLM_LZE, GLZLM_SZLGE and GLZLM_ZLNU. These five features were analyzed in the Waikato Environment for Knowledge Analysis program to create a model that could distinguish infection and cancer groups, and the model performance was found to be the highest with logistic regression (area under curve, 0.813; accuracy, 75.7%). The sensitivity and specificity values of the model in distinguishing cancer patients were calculated as 80.6 and 70.6%, respectively.

In our study, we created prediction models based on radiomic analysis of 18F-FDG PET/CT images. Texture analysis with machine learning algorithms is a noninvasive method that can be useful in the differentiation of infection and malignancy in consolidating lung lesions in the clinical setting.