From Brain Imaging to Graph Analysis: a study on ADNIs patient cohort

ABSTRACT In this paper, we studied the association between the change of structural brain volumes to the potential

development of Alzheimer’s disease (AD). Using a simple abstraction technique, we converted regional cortical and

subcortical volume differences over two time points for each study subject into a graph. We then obtained

substructures of interest using a graph decomposition algorithm in order to extract pivotal nodes via multi-view

feature selection. Intensive experiments using robust classification frameworks were conducted to evaluate the

performance of using the brain substructures obtained under different thresholds. The results indicated that compact

substructures acquired by examining the differences between patient groups were sufficient to discriminate between

AD and healthy controls with an area under the receiver operating curve of 0.72.

INTRODUCTION Brain functionality and decreasing cognition is known to be associated with the progression of Alzheimer’s disease

(AD). The difference between asymptomatic and symptomatic AD can change over time, and this period between

being a healthy individual to having clinically present AD is referred to as mild cognitive impairment (MCI). This

functional and cognitive decline is marked with memory lapses, poorer executive function, and increasing

complexities associated with common activities of daily living that can last years ​[1]​. The standard diagnosis of AD

patients typically begins with a series of neuropsychological tests, clinical assessments, followed by various imaging

tests. Over time, patients undergo multiple brain imaging visits at different points which allow physicians to link

physical manifestations of AD to lab and clinical measurement data. This wealth of information enables researchers

to take advantage of these multi-visit data to look at the AD neurodegeneration process (through observing cognitive

decline) both cross-sectionally and longitudinally. Recent studies have used many methods look at AD from a

different perspective. Researchers use techniques such as hierarchical classification ​[2]​, convolutional neural

networks ​[3]​, tract-based spatial statistics ​[4]​, in addition to a whole series of multi-modality data ​[5] to improve AD

diagnostic performance. Further, studies have looked at various combinations of phenotype classification, typically

incorporating both stable and converted MCI patients ​[6]​. This allows for a rich understanding of AD progression

and of biomarkers that may be readily available to drive an improved and quicker AD diagnosis. In this work, we

aim to leverage the advantages of graph-based approaches while discarding some of the disadvantages that come

with diffusion MRI based techniques. Specifically, we used structural volume data to uncover how different regions

of the brain are inter-connected during AD progression. We also incorporated longitudinal information which

accounts for structural changes in a patient’s brain over time. These inter-region network effects will be used for

phenotype classification to demonstrate their discrimination power.

RELATED WORK Graph theory and graph-based approaches have been studied for many years in the neuroimaging community. These

networks have been created using several different modalities including structural magnetic resonance imaging

(MRI), diffusion tensor imaging (DTI), and positron emission tomography (PET) for various diseases such as AD.

Typically, these approaches look at the differences between normal controls (CN) and those with the studied

disease. The subsequent analysis allows for interpretation of nodes, hubs, a

…(Full text truncated)…