Automated Segmentation of Cardiac Fats Based on Extraction of Textural Features from Non-Contrast CT Images

One of the most important features of the cardiac that are associated with several cardiovascular risk factors such as myocardial ischemia, coronary artery stenosis, metabolic syndrome, atrial fibrillation, and others, are epicardial and pericardial visceral adipose tissues around the cardiac. Therefore, automatic detection, quantification, and segmentation of cardiac fats can be used as an additional feature for medical imaging and visualization in clinical routine in order to save time and a reliable tool for cardiovascular risk assessment. In this paper, we propose an automated method for the segmentation of cardiac epicardial and pericardial adipose tissues in non-contrast CT images. The proposed method includes the pre-processing step using thresholding in the fat range and contrast enhancement using histogram analysis, feature extraction step based on texture features extracted using Gabor filter bank based gray-level co-occurrence matrix (GLCM) and pixel information, and the cardiac fats segmentation step are based on pixel labeling and pattern recognition classification algorithms. The experimental results also indicate a good performance of cardiac fat segmentation compared to the manual segmentation obtained by expert. Experiments showed that the accuracy obtained algorithm designed for the segmentation of cardiac fats was 99.0% with a sensitivity of 90.2% and a specificity of 99.7%. In addition, the Dice similarity index for this algorithm was 91.8%.