At the Computer Science department office
Appointment on Visitation important
Topic: Computer Vision And Medical Image Analysis
Description: My current research area is the development of a robust computerised parcellation technique for parcellating (dividing) ageing brains accurately, objectively and in a timely manner with minimum human intervention into lobar sections. This work involves the development of computer algorithms for image analysis and implementation of such algorithms on the study population.
|1.||M.Sc (Artificial Intelligence)||Department of Computer Sciences, University of Lagos, Akoka, Lagos||1986|
PARCELLATION OF MRI BRAIN IMAGES USING HYBRIDIZED ATLASES SELECTION MODEL
INTRODUCTION: Multiple atlas-based parcellation model has been demonstrated to perform better than single atlas-based parcellation model in terms of accuracy of the parcellation of human brain. However, its accuracy level is limited if used for ageing brain due to the presence of age-related changes such as atrophy.
AIM: The aim of this study is to develop a novel multiple atlases selection model that ensures improved accuracy for the parcellation of ageing brain by combining cost functions with similarity metric and atrophy measure for atlases selection.
MATERIALS AND METHODS: Ten high-resolution anatomical brain MRI images and ten atlases that accompanied the images were used. An MRI image is used one at a time as the target image while the remaining nine MRI images were used as the source images. Using each target image, consensus parcellated images were obtained for all the cost functions, similarity metrics and atrophy measure. Atlases of the model that combines cost function, similarity metrics and atrophy measures were selected and the consensus parcellated image was constructed. The similarity metric between each consensus parcellated image and the atlas of the target image was computed. The mean of these similarity metrics for each atlases selection model was also computed.
RESULTS: The mean were: Combination of Cost function, Similarity and Atrophy = 0.5999810, Normalised mutual information = 0.5978935, Mutual information = 0.5970887, Similarities = 0.5867609, Correlation ratio = 0.5825561, Normalised correlation = 0.5822807, Atrophy = 0.5751072, Least squares = 0.5750365, Single atlas = 0.5187616.
CONCLUSION: The model that combines Cost function, Similarity and Atrophy performs better, in terms of accuracy, than the other benchmark models.
CONTRIBUTION TO KNOWLEDGE: Atlases selection was carried out using similarity, atrophy and a combination of cost function, similarity and atrophy. According to literature, none of these methods has been used for atlases selection.
OWATE PATRICK is a Lecturer II at the Department of Computer Science
OWATE has a M.Sc in Artificial Intelligence from Department of Computer Sciences, University of Lagos, Akoka, Lagos