Detection of Early Changes in White Matter Degeneration Using Texture Analysis of Magnetic Resonance Images
| dc.contributor.advisor | Ikonomidou, Vasiliki | |
| dc.contributor.author | Biswas, Debosmita | |
| dc.creator | Biswas, Debosmita | |
| dc.date | 2013-07-24 | |
| dc.date.accessioned | 2014-05-22T16:51:02Z | |
| dc.date.available | 2014-05-22T16:51:02Z | |
| dc.date.issued | 2014-05-22 | |
| dc.description.abstract | In the absence of permanent cure, early detection and diagnosis of neurodegenerative diseases are of utmost importance to make use of palliative measures for enhancing the quality of life of millions of Americans. However, a large number of people are not diagnosed at an early enough stage where medications can delay the full onset of the diseases (NIA 2013). While various techniques to analyze brain images of subjects have been proposed to address this challenge, none of the techniques provide a robust and reliable solution. In this thesis, we present a novel technique using texture analysis of T2magnetic resonance (MR) images to lay the foundation for an effective solution, using Alzheimer's disease (AD) as the case study. The technique consists of the following four steps. First, we utilize the textural property of the MR images to obtain a set of features that encode statistically meaningful information about the spatial distributions of the gray tone variations. Second, we compute texture feature maps (a feature value stored at every image voxel) on the white matter regions of the images that xii are segmented into regions of interest (ROIs) based on the anatomical structure of the brain. Third, we identify the subset of relevant and uncorrelated features from our initial feature set by using statistical measures like mean, coefficient of variance, and mutual information. These features yield statistically different values in the different ROIs and also in the different subjects for the same ROI, and the variations in the values are independent of each other. Thus, they are expected to afford better predictive powers in terms of detecting early signs of AD than the complementary set of features. Last, we validate the utility of the relevant features by carrying out statistical hypothesis tests on two groups of subjects, where the first group consists of subjects who have the APOE ε4 genes that are often found in AD patients, and the other group comprises of subjects who do not have the APOE ε4 genes. Results show that the entropy-type features yield promising results and are able to distinguish between the two types of subjects in many cases. It is hypothesized that the lack of statistical differences for certain subjects belonging to the two groups is due to the non-advent of neurodegeneration in those subjects. Hence, we believe that this technique provides a valuable first step towards early detection of neurodegenerative diseases without requiring genetic information and functional imaging modalities. Further work will involve more effective feature set generation and extensive validation and verification using ground truth information and long-duration trials involving monitoring of subjects who are predicted to have early symptoms of the diseases. | |
| dc.identifier.uri | https://hdl.handle.net/1920/8689 | |
| dc.language.iso | en | |
| dc.subject | White matter | |
| dc.subject | Early detection | |
| dc.subject | White matter degeneration | |
| dc.subject | Alzheimer's diease | |
| dc.subject | Neurodegeneration | |
| dc.title | Detection of Early Changes in White Matter Degeneration Using Texture Analysis of Magnetic Resonance Images | |
| dc.type | Thesis | |
| thesis.degree.discipline | Electrical Engineering | |
| thesis.degree.grantor | George Mason University | |
| thesis.degree.level | Master's | |
| thesis.degree.name | Master of Science in Electrical Engineering |