B-Scan Ultrasound Image Feature Extraction for Quantitative Grading of Fatty Liver Severity

To reduce the strong subjectivity and large deviations of common existence in the clinical B-scan ultrasonic grading of fatty liver severity, a4kinds of features and the corresponding automatic extraction methods are proposed for quantitatively grading fatty liver severities on B-scan ultrasound images．First, aafter eliminating the influences of gray levels and contrasts of the ultrasound images, a the number and size of speckles, awhich represent the near-field echo characteristics, aare calculated using the difference of Gaussian operator．Then, athe near-field vs．far-field mean intensity ratio and the near-field vs．far-field standard deviation ratio, awhich represent the far-field energy attenuation, a are calculated．Finally, aa feature vector selection step is carried out based on the feature set．The experimental results show that the feature vector composed of the above four kinds of features can effectively discriminate among normal liver, amild, amoderate and severe fatty liver．