A novel computational CT image analysis method for classifying nodules from normal thyroid tissue

Purpose:To investigate the feasibility of utilizing texture features to classify nodule from normal thyroid tissue in Computed Tomography (CT) images. Materials and Methods: Group A (negative) includes 152 normal thyroid CT images from 55 patients healthy controls enrolled in the study. Group B (positive) includes 134 thyroid images with nodules (50 malignant, 84 benign) of 58 patients undergone thyroid surgery and final diagnoses were confirmed by histopathology. Regions of interest (ROIs) from axial noncontrast CT images were delineated manually and 31 texture features including the gray level co-occurrence matrix (GLCM), the gray level gradient co-occurrence matrix (GLGCM), average intensity, contrast and coherence were extracted. Support Vector Machine (SVM) was used in data classification. Leave one out cross validation (LOOCV) strategy was utilized to take full advantage of the samples. To evaluate the performance of the proposed method, accuracyrate, sensitivity, specificity and area of under receiver operating characteristic (ROC) curve (AUC) etc. were calculated. Results: the accuracy-rate, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) are 0.8895±0.0186, 0.8265±0.0448, 0.9467±0.0141, 0.9340±0.0146 and 0.8586±0.0300 respectively, and the AUC is 0.9520±0.0089. Conclusion: Texture features can help radiologists to classify the nodule from normal thyroid tissue.