Parametric features of image textures in ¹⁸F-FDG PET/CT evaluation of lung nodules

Objective: To evaluate the parametric features of image textures on ¹⁸F-FDG PET/CT for the differentiation between malignant and benign pulmonary nodules and compare the diagnostic performance of these parameters with SUV_max. Methods: ¹⁸F-FDG PET/CT images of 170 patients (102 males, 68 females, age range: 29-81 (mean 59) years) with pulmonary nodules were retrospectively evaluated. Eighty-nine pulmonary nodules (230 slices) were malignant and 81 (193 slices) were benign. The pulmonary nodules were contoured on CT images and mapped to the co-registered PET images. Thirteen parameters of textural features were extracted and SUV_max was measured. Logistic regression analysis was used to identify the significant texture parameters and create a regression model. The efficacy of the textural features and SUV_max to distinguish between malignant and benign pulmonary nodules was evaluated by ROC curve analysis. The textural features of squamous cell carcinoma and adenocarcinoma were compared via the Mann-Whitney u test. The sensitivity and specificity of the textural features and SUV_max for the differential diagnosis were compared with χ² test. Results: Logistic regression model identified 4 textural features (skewness (β=1.7058), kurtosis (β=-1.0989), angular second moment (ASM, 3=-4.4140) and strength (β=0.5626); all P<0.05) to have significant correlation with the malignancy of lung nodules. The AUC of ROC curve was 0.775 (95% CI 0.732-0.819; P<0.001) with the sensitivity of 89.6% (206/230) and specificity of 50.8% (98/193). ASM and strength had statistically significant differences between squamous cell carcinoma and adenocarcinoma [ASM: 0.0303 (95% CI 0.0392-0.0724) vs 0.0594 (95% CI 0.0721-0.0947); strength: 2.4714 (95% CI 2.4632-4.1050) vs 1.5945 (95% CI 1.9003-2.4652); u=3082.0 and 3115.0, both P<0.01]. The AUC of SUV_max-based diagnosis was 0.757 (95% CI 0.711-0.802; P<0.001) with the sensitivity of 80.9% (186/230) and specificity of 50.3% (97/193) at a cut-off value of 2.5. The sensitivity of the textural features was superior to SUV_max in differentiating malignant from benign pulmonary nodules (χ²=6.903, P<0.01). Conclusions: Image textural parametric features extracted from ¹⁸F-FDG PET/CT are more sensitive to differentiate between malignant and benign pulmonary nodules compared to SUV_max. They might also be useful to discriminate between different pathological types of lung cancers. (authors)