Pattern recognition for magnetic resonance knee imaging using convolutional neural network / Zhang Xinyu

• Main Author: Zhang, Xinyu
• Format: Thesis
• Published: 2018
• Subjects: R Medicine (General); T Technology (General)
• Online Access: http://studentsrepo.um.edu.my/8545/4/Resaech_Report_KQB160003.pdf; http://studentsrepo.um.edu.my/8545/

Nowadays, knee osteoarthritis is a popular disease all over the world. Cartilage degeneration is the performance of osteoarthritis. It is important to research on the characteristic of cartilage. Magnetic resonance imaging provides prominent result in the assessment of osteoarthritis disease. In this project, convolutional neural network was used to identify the region of knee cartilage. 9600 magnetic resonance images were used as dataset where 3440 images were cartilage and 6160 images were background. Each image is 100*100 pixels. GoogLeNet model was the selected CNN model for training data. Nvidia digits was the platform under the Linux system for training data. After training, trained model was imported in OpenCV doing localization. Another 40 images were used for testing model. Then, manually cropping of cartilage was done in MATLAB. At last, the confusion matrix of accuracy of CNN recognition came out.