Machine learning Classification based on Principal Component Analysisand Multilinear Principal Component Analysis for Study of CardiologyUltrasound in Left Ventricle

• Main Authors: Bo-Jen Tsai, 蔡博任
• Other Authors: Mong-Na Lo Huang
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/48kkc3

碩士 === 國立中山大學 === 應用數學系研究所 === 102 === In this work, we study heart diseases related to diastolic and systolic of heart in left ventricle. Following Hung (2012) and Liu (2012), the differences between the gray-scale values of diastolic and systolic in left ventricle are used to evaluate the function of heart. In the past, when we studied image data, we usually used principal component analysis for dimension reduction because it is highdimensional data. Furthermore, we also use the multilinear principal component analysis by Hung et al. (2012) for dimension reduction and the way of dimension reduction is mapping the highdimensional space to lower dimensional space. Then we discuss the data after dimension reduction by principal component analysis and multilinear principal component analysis, and combine different machine learning classification methods such as regularized discriminant analysis, support vector machine, and random forest to probe the correct rate of classification. In this work, the heart ultrasound image data is collected from Kaohsiung Veterans General Hospital. On the above basis, we make an empirical study on applying the PCA, MPCA, machine leaning classification method for the heart ultrasound image data, and expect to find out the most effective method for classification, and recognize the important areas that influence heart function.