Clinical significance of FDG PET/BT parameters and radiomics in the evaluation of tuberculosis lesions

Conventional diagnostic methods for tuberculosis may be insufficient in certain patient groups, leading to the initiation of empirical treatment. This study aimed to investigate the diagnostic value of radiomic features extracted from Fluorine-18 Fluorodeoxyglucose Positron Emission Tomography / Computed Tomography imaging in differentiating from malignant lung lesions (adenocarcinoma and squamous cell carcinoma), and to compare empirically treated tuberculosis cases with microbiologically/histopathologically confirmed tuberculosis cases. A total of 102 patients who underwent Fluorine-18 Fluorodeoxyglucose Positron Emission Tomography / Computed Tomography between 2014 and 2019 were retrospectively evaluated. Patients were categorized into four groups: empirically treated tuberculosis with confirmed response (n=27), Tuberculosis diagnosed by conventional methods (n=28), lung adenocarcinoma (n=20), and squamous cell carcinoma (n=27). Radiomic features were extracted using Local Image Feature Extraction software, and statistically significant parameters were used to develop Python-based machine learning models. The analysis revealed statistically significant differences particularly in morphological, intensity-based, and textural parameters. Machine learning models trained with these features achieved high accuracy in differentiating tuberculosis from malignancy, with the best model reaching an Area Under the Curve of 0.95. In the classification between Group 1 and Group 2, the Extreme Gradient Boosting algorithm demonstrated the highest sensitivity. This study demonstrated that radiomic analysis quantitatively evaluates the structural characteristics of lesions and provides significant diagnostic contribution, particularly in tuberculosis cases that are difficult to diagnose.