Our brain is a very complex and highly related network. Its inside, there are not just structurally connected regions but also functionally connected components. This means, components that are not connected directly can show unpredictably excessive correlation. So, highly connected structures with each other cannot be determined just focusing structural connectivity of the brain regions because of active functional communication system of the brain. Complex communication system in the brain shows that while spatially close regions display uncorrelated or anticorrelated fluctuations, structurally unconnected regions can display highly correlated fluctuations. Fransson reported that while precuneus/posterior cingulate cortex (precuneus/PCC) region and included medial and dorsolateral parts of prefrontal cortex, angular gyrus, thalamus and pons shows positive correlation, premotor cortex bilaterally, dorsolateral prefrontal cortex, supplementary motor cortex, inferior parietal lobe, occipital cortex, and insula bilaterally shows negative correlation with precuneus/PCC. Thus, structural connectivity is not the only representation to determine connection between regions, functional connectivity analysis between different regions, also, as important as structural connectivity analysis to define map of the brain.
In this project, novel mathematical computer methods for quantification of the structure of brain white matter tracts that are extracted from Diffusion Magnetic Resonance Imaging (MRI) are developed. First, a new fiber clustering method for automatic anatomical clustering of white matter fibers that traverse the braintem is proposed. The five major pathways of the brainstem (Corticospinal Tract - CST, Medial Lemniscus - ML, Middle Cerebellar Peduncle - MCP, Superior Cerebellar Peduncle - SCP, Inferior Cerebellar Peduncle- ICP) are labelled by the proposed method. A fiber registration technique that estimates the deformation between the pre-operative and post-operative CST and ML tracts, which are extracted from the Diffusion MRI, is developed. Finally, in order to quantify the tract changes due to surgical operation, two types of fiber measures are designed. The internal fiber measures are based on the shape, volume, and orientation features of pre-operative and post-operative tracts independently. The second type of measures are based on the features of the deformation field estimated between the pre-operative and post-operative tracts. A total of 14 constructed measures are calculated over the dataset obtained from the clinical partner of the project, compared to the patient clinical scores, and a correlation between some of the designed measures and clinical findings is observed.
The aim of this project is to great a backend of an image classification application for visually impaired people to help them recognize the Turkish lira banknotes. Convolutional Neural Network architectures will be used for object detection and classification.
Project aims to help Coast Security to detect and classify ships.