Applications of Machine Deep Learning in MR Images

• Main Authors: Ming-Chi Wu, 吳明基
• Other Authors: James Cheng-Chung Wei
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/sdv992

博士 === 中山醫學大學 === 醫學研究所 === 106 === Background and Purpose: Machine deep learning is the use of a large amount of training data to learn the characteristics of the data, and further use these features to accomplish specific purposes, such as detection, classification, or prediction. The purpose of this study is to develop a system to automatic recognition of the specified structure and classification from MR imaging and pattern recognition technology using machine learning algorithm. Materials and Methods: The imaging processing with image contrast enhancement, image binarization and noise processing to obtain the maximum circumscribed rectangle. To find candidate regions of the basal ganglia nuclei, we extracted features of the basal ganglia using multiple methods of texture feature extraction, including GLCM, Law''s Mask, and the Hu''s texture feature extraction method. In addition, the Neural-fuzzy-based Adaboost algorithm will be our classifier to train and test the extracted features. Results: Regarding auto recognition of basal ganglion, the results of our hybrid classifier algorithm of this study is 92.3% accuracy. And the performance metrics for nasopharyngeal mass classification accuracy in the training, validation and testing datasets were 93.44% and 92.78%. Conclusion: Hybrid Classifier algorithm combing the advantages of SONFIN and adaboost that is a feasible system method and has the high recognition rate for basal ganglia. And our machine learning combining neuro-fuzzy system and adaboost algorithm accurately classification form benign and malignant mass of nasopharynx.