عنوان پایان‌نامه
کدگشایی حالت های مغز با استفاده از گراف های ارتباط کارکردی داده ی fMRI
- رشته تحصیلی
- مهندسی برق- مهندسی پزشکی - بیوالکتریک
- مقطع تحصیلی
- کارشناسی ارشد
- محل دفاع
- کتابخانه دانشکده برق و کامپیوتر شماره ثبت: E1968;کتابخانه مرکزی -تالار اطلاع رسانی شماره ثبت: 50075
- تاریخ دفاع
- ۲۸ شهریور ۱۳۹۰
- دانشجو
- فاطمه مختاری
- استاد راهنما
- غلامعلی حسین زاده دهکردی
- چکیده
- تصویر برداری تشدید مغناطیسی کارکردی MRI یکی از آخرین پیشرفت های تصویر برداری سیستم های عصبی است طی هر حالت شناختی نواحی متفاوتی از شبکه های مجزا با هم دارای تقابل و ارتباط کارکردی هستند یک روش جالب که اخیرا" برای شناسایی شبکه های کارکردی مورد توجه قرار گرفته است ا ستفاده از تئوری گراف می باشد.تحقیقات زیادی روی توصیف توپولوژی مغز با استفاده از تئوری گراف صورت گرفته است
- Abstract
- Functional magnetic resonance imaging (fMRI) is one of the latest advances of nueroimaging. During each brain’s cognitive state, different regions interact and connect together. As an interesting approach, graph theory is recently used to identify the functional networks of the brain. In present study, we have gone beyond this goal, and tried to apply the functional connectivity graphs to decode five different brain’s cognitive states. These states are: 1) Fixation to a dot presented in the middle of the screen; 2) detection of a single stimulus; 3) perceptual matching; 4) attentional cueing; and 5) a delayed match-to-sample test of working memory. In the analysis of fMRI data, at first, common preprocessing methods are applied to data. Then, those regions which have shown activation within rest and task are choosen. The functional connectivity graphs are then obtained for different states using the cross correlation method. These graphs are mapped to the reduced-dimensional vector or graph space applying three feature conditioning algorithms including principal component analysis, 2-dimentional singular value decomposition and backward edge elimination. In the last algorithm, we have also designed a support vector classifier enriched by a sub-tree kernel to classify the above mentioned states. By comparing the results of these three algorithms and algorithm proposed in previous related study, it is observed that using backward edge elimination algorithm with sub-tree kernel raise the performance significantly, in both aspects of final number of informative connectivities and regions and machine performance criterion. This algorithm leads to correct classification rate of 0.93. Regions including Insula, Cerebellum, Post cingulate cortex, Left middle superior frontal gyrus, and Left ventromedial prefrontal and connectivities between Insula and Cerebellum, and Left ventromedial prefrontal and Post cingulate cortex play the most important roles in discriminating between different states. Keywords: Decoding brain states, fMRI connectivity graphs, Principal component analysis, 2-Dimensional singular value, Backward edge elimination, Sub-tree kernel
