Loading...
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 56246 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Fatemizadeh, Emadeddin; Amini, Arash
- Abstract:
- In recent years, due to the structural need of most medical data for graphic models such as the graphic model of patients and the loss of data correlation in previous methods, graphic methods have been designed and developed. On the other hand, with the growing presence of magnetic resonance imaging devices in various medical centers, a large amount of functional magnetic resonance images of healthy and sick people have become available to researchers. In this study, our goal is to use a new method in the field of graphic modeling so that we can extract functional connectivity graphs from functional magnetic resonance images and measure the performance of these graphs in different groups of people. In this thesis, we first extracted the activity of the regions of interest from the pre-processed magnetic resonance images using different time series atlases. In the following, we extracted functional connection graphs using the smooth graph learning method. In these graphs, each vertex is an interested region of the magnetic resonance image, and our goal is to find functional connections between the regions. In the following, we also extract other classical graphs such as correlation graphs and evaluate the performance of graphs in the classification test with various graph approaches and extracting various graph features from the extracted graphs structure. In the ABIDE dataset, we seek to identify healthy people and autism, in the Development dataset, we seek to separate children from adults using their functional connectivity graph structure, and in the CNP dataset, we categorize the four healthy groups, schizophrenia, hyperactivity, and bipolar. Finally, the totality of the results shows the superiority of our extracted graphs compared to classical methods. In the ABIDE dataset, the smooth graph extraction method can compete with the results of non-deep methods of articles, and in the Development dataset, the extracted graphs can distinguish children from adults
- Keywords:
- Graph Learning ; Graph Signal Processing ; Functional Magnetic Resonance Imaging (FMRI) ; Machine Learning ; Brain Effective Connectivity ; Neuroimaging
-
محتواي کتاب
- view
- مقدمه
- مفاهیم کلی
- مروری بر کارهای انجام شده
- مقدمه
- کارها با رویکرد ریاضی پردازش سیگنال گرافی
- دستهبندی انواع تحلیل دادگان پزشکی با رویکرد گرافی
- کارهای حوزه تحلیل اتصال عملکردی بر اساس تصاویر تشدید مغناطیسی عملکردی
- کارهای حوزه تحلیل مبتنی بر ساختار الکتریکی
- کارهای حوزه تحلیل ساختار آناتومیک
- کارهای حوزه بخشبندی ساختار آناتومیک
- کارهای حوزه مدل شبکهی بیمار
- جمعبندی
- روشهای پیشنهادی
- نتایج
- جمعبندی
- مطالب تکمیلی
- دیتاست ABIDE
- دیتاست development
- دیتاست CNP
- دیتاست TADPOLE
- دیتاست DEAP
- دیتاست DREAMER
- دیتاست MASS
- دیتاست HFECGIC
- دیتاست ADNI
- دیتاست ODIR
- دیتاست UKBB
- نمودار t-SNE
- GCN
- شبکه GCN پویا
- شبکهی گراف MCI
- GAT
- GIN
- مدل شبکه کانولوشنال نمودار زمانی-تطبیقی
- استخراج عضویت در گراف
- یادگیری گراف نهفته
- مدل ST-GCN
- مدل EA-GCN
- مدل EV-GCN
- مدل InceptionGCN
- مدل MMGL
- مدل MGNN
- مدل GMAN
- مدل MLWGAT
- مدل PETNet
- نتایج تکمیلی دیتاست ABIDE
- نتایج تکمیلی دیتاست Development
- نتایج تکمیلی دیتاست CNP