Role of Chest CT Radiomics in Differentiating Tumorlets and Granulomas: A Preliminary Study

Background: To identify the radiomics features of both granulomas and tumorlets (TL) and to assess the potential role of radiomics in differentiating these two diseases. Methods: From 2013 to 2021, ninety patients who had undergone lung surgery and pre-operative chest CT evaluation, with pathologically proven granulomas or TL, were retrospectively enrolled. Two radiologists, in consensus, manually segmented the lesions on CT images. Radiomic features were then automatically extracted from these segmentations using dedicated software. The performance of CT radiomics features in differentiating TL from granulomas was tested by receiver operating characteristic curves and the areas under the curve (AUCs), calculating sensitivity and specificity. Results: The final population consisted of 55 patients (38 female; mean age 64 ± 14 years), 32 with TL and 23 with granulomas. Significant differences were found in 16/107 radiomic features: 3 Shape, 1 First Order, 2 Grey Level Co-occurrence Matrix (GLCM), 2 Gray Level Dependence Matrix (GLDM), 4 Grey Level Run Length Matrix (GLRLM), and 4 Gray Level Size Zone Matrix (GLSZM). Flatness and Long Run High Gray Level Emphasis showed the best performances in discriminating TL from granulomas (AUC: 0.903; sensitivity: 100%; specificity: 80%; and AUC: 0.896; sensitivity: 92.3%; specificity: 76.5%; respectively; both p < 0.001). Conclusions: Radiomics may be a non-invasive imaging tool for characterization of small lung nodules, differentiating granulomas from TL, and may play a role in preventing TL growth and its possible malignant evolution, avoiding delayed diagnosis.