부산대학교 도서관

Quantitative analysis of many neurological diseases depends on automated and accurate segmentation and classification of structures. Nowadays, the deep learning based image classification and segmentation methods have gained interest of research because of their self-learning capabilities over huge amounts of dataset. This paper focuses on the use of Convolutional Neural Network which takes the feature maps preprocessed in Curvelet domain to classify the MRI brain image datasets. Curvelets provide better sparse representation and the features extracted are more accurate than traditional wavelet transform due to its multi-directional capability. Next, the segmentation methods to study the anatomical structures and localization of brain tumors is dealt and finally the performance of the CNN is discussed. Comparing with the wavelet transform and classification using traditional classification methods like SVM, PNN, the feature extraction in Curvelet domain and CNN provides an increase in accuracy