Classification for Ultrasound Tomography of Liver Function by Hybrid Machine Learning Algorithms

• Main Authors: Shih Yen Hsu, 許士彥
• Other Authors: Tai-Been Chen
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/71837020965349044656

碩士 === 義守大學 === 資訊工程學系 === 101 === Machine learning methods had widely used for classification of images. However, a single machine learning method was hardly to present classified performance for more than two groups in dataset due to the properties of texture and intensity distribution in images. Hence, the hybrid machine learning methods was proposed to classify more than two groups in dataset (ultrasound images) with high accuracy and performance.The effective liver ultrasound images were including 36 normal, 51 fatty and 48 cancers in this study. The proposed depth correction method was applied to images at first. Three region of interest (ROI) were manually drawing on images. Next, five features of ROI were calculation. Finally, the hybrid logistic regression and support vector machine method was ready to classify ultrasound images.The accuracy, Kappa statistics, and mean absolutely error (MAE) of proposed method were 87.5%, 0.812, and 0.119 higher than those of logistic regression with 75.0%, 0.548, 0.280 or support vector machine with 75.7%, 0.637 and 0.293. Therefore, the hybrid method is more accuracy and performance and less error than those of single method.The hybrid method not only provides higher accuracy in three group ultrasound, it is also an effective classifier for multiple group dataset.