In the domain of computational neuroscience, scientists are making continuous efforts to understand healthy brain structure which helps to identify and explain damages and brain diseases disrupting its functions. Noticeably, magnetic resonance image (MRI) technology and brain morphometry methods opened noble opportunities for further study of biological brain structure in neuroscience using machine learning techniques. Among these, the association between cortex structural change and cognitive ageing are getting more focus in current researches. There is a great demand to realize why cognitive functioning decline differently in normal ageing. In addition, many research works indicated that a cognitive reserve can be improved by lifestyle factors such as education, job satisfaction and leisure activities at older ages and these factors are useful to reduce risk of brain pathology like Alzheimer disease. In this regard, the importance of education to delay cortical thinning due to normal ageing recently taking researchers’ attention in the domain. In other words, there is an interest of realizing the hypothesis stating that the mean cortical thickness of an individual might have different rate of change at older ages with respect to individual’s further educational attainment.
In light with these, the main objective of this dissertation is to analyze predictive modeling approaches that can be used to assess the associations between demographic attributes of a subject and its brain cortical structure. The study focuses on the investigation of the two demographic attributes chronological age and educational level and their effect on brain cortex structure. We design robust features selection method that identifies the most important explanatory features among the macrostructure of cortex region interests (ROIs) and confounding factors of demographic properties. The proposed model is tested using 2,911 cognitively normal subjects (age 45-91 years) which are collected at a single medical center and acquired their brain magnetic resonance images. All images were acquired using the same scanner with the same protocol.
Accordingly, first the modeling of brain age prediction from cortical thickness data gathered from large cohort brain images is presented. This study proposes to first apply Sparse Group Lasso (SGL) for feature selection by utilizing the brain’s anatomical grouping. Once the features are selected, a non-parametric non-linear regression using the Gaussian Process Regression (GPR) algorithm is applied to fit the final age prediction model. Experimental results demonstrate that the proposed method achieves the mean absolute error of 4.05 years, which is comparable with or superior to several recent state of the art methods.
In the second study, the framework of investigating the association between education and elderly cortical thickness is presented. The study is tested by further sub setting the dataset representation into three situations based on the levels of education measured in years of study. Among this, we obtained an optimal accuracy of 90% by using Deep Learning model on the most important predicting variables selected using Sparse Group LASSO when the sub dataset is represented containing unschooled (0 years of study) and post graduate levels of participants.