Welcome to a master thesis presentation with the title "Deep learning methods for MRI brain image analysis: 3D convolutional neural networks for Alzheimer's disease detection and brain tumor classification" presented by Mahmood Nazari and Karl Bäckström, MPCAS.
When? 15:15 on Friday, 15 September, 2017
Where? Landahlsrummet (room 7430), Hörsalsvägen 11, 7th floor, Chalmers
Examiner: Irene Gu
In this thesis work, we investigate deep learning methods for medical image analysis and classification. In particular, our studies have been conducted on using a large amount of MR brain images for two diseases: (a) detection of Alzheimer disease (AD); (b) classification of brain tumor, glioma. Our study is focused on employing deep learning methods, more specifically, 3D Convolutional Neural Networks (CNNs) with different architectures, for detection/classification of the above two diseases. The first part of the study is focused on the detection of Alzheimer disease. For our study on AD, MR images were obtained from a public available dataset ADNI. We propose a novel AD detection scheme by employing 3D CNNs with a simple architecture that is effective, after numerous tests. In the scheme, a set of pre-processing on the MR images is also employed, as it is shown to be essential for the detection performance. The proposed scheme has yielded a good performance, with a AD detection rate of 98,7% on the case where training and testing images were randomly partitioned, and 85.7% in the case where training and testing images were partitioned according to subjects. Performance evaluation and comparison were also conducted. The second part of the study is on brain tumor (glioma) classification. We have studied the same deep learning method, 3D CNNs, however, with different architectures for the brain tumors. The dataset is obtained from “MICCAI BRaTS competition 2017” consisting of high-grade glioma (HGG) and low-grade glioma (LGG) MR images. Our preliminary study on classification of low/high grade gliomas has shown promising results with a classification accuracy of 85.96%. Further work is required for a depth study of brain tumor classification.
Everybody is welcome!