Subramanya M B, Kumar Vinod, Mukherjee Shaktidev, Saini Manju
Department of Electrical Engineering, Indian Institute of Technology Roorkee , Roorkee, Uttarakhand , India .
J Med Eng Technol. 2015 Feb;39(2):123-30. doi: 10.3109/03091902.2014.990160. Epub 2014 Dec 19.
The present study proposes a computer-aided diagnosis (CAD) system for the diagnosis of grades of fatty liver disease, namely mild, moderate and severe fatty liver along with normal liver tissue. Fifty-three B-mode ultrasound images consisting of 12 normal, 14 mild, 14 moderate and 13 severe fatty liver images are used. Based on the visual interpretations by the radiologists, region of interests (ROIs) from within the liver and one ROI from the diaphragm region are considered from each image. The texture features of these ROIs are combined in three ways to form ratio features, inverse ratio features and additive features. The sub-sets of optimal features are obtained by a differential evolution feature selection (DEFS) algorithm and a support vector machine (SVM) has been used for the classification task. The Laws ratio features have shown better performance with an average accuracy and standard deviation of 84.9±3.2. Hence, the CAD system could be useful to the radiologists in diagnosing grades of fatty liver disease.
本研究提出了一种用于诊断脂肪肝疾病分级的计算机辅助诊断（CAD）系统，该疾病分级包括轻度、中度和重度脂肪肝以及正常肝组织。使用了53幅B超图像，其中包括12幅正常肝脏图像、14幅轻度脂肪肝图像、14幅中度脂肪肝图像和13幅重度脂肪肝图像。根据放射科医生的视觉解读，从每幅图像中选取肝脏内部的感兴趣区域（ROI）以及膈肌区域的一个ROI。这些ROI的纹理特征通过三种方式进行组合，以形成比率特征、反比特征和相加特征。通过差分进化特征选择（DEFS）算法获得最优特征子集，并使用支持向量机（SVM）进行分类任务。Laws比率特征表现出更好的性能，平均准确率和标准差为84.9±3.2。因此，该CAD系统对放射科医生诊断脂肪肝疾病分级可能有用。
