A Large-scale Comparison of Cortical and Subcortical Structural Segmentation Methods in Alzheimer’s Disease: a Statistical Approach

Abstract

Alzheimer's disease (AD) is a neurodegenerative disease that leads to anatomical atrophy, as evidenced by magnetic resonance imaging (MRI). Automated segmentation methods are developed to help with the segmentation of different brain areas. However, their reliability has yet to be fully investigated. To have a more comprehensive understanding of the distribution of changes in AD, as well as investigating the reliability of different segmentation methods, in this study we compared volumes of cortical and subcortical brain segments, using automated segmentation methods in more than 60 areas between AD and healthy controls (HC). A total of 44 MRI images (22 AD and 22 HC, 50% females) were taken from the minimal interval resonance imaging in Alzheimer's disease (MIRIAD) dataset. HIPS, volBrain, CAT and BrainSuite segmentation methods were used for the subfields of hippocampus, and the rest of the brain. While HIPS, volBrain and CAT showed strong conformity with the past literature, BrainSuite misclassified several brain areas. Additionally, the volume of the brain areas that successfully discriminated between AD and HC showed a correlation with mini mental state examination (MMSE) scores. The two methods of volBrain and CAT showed a very strong correlation. These two methods, however, did not correlate with BrainSuite. Our results showed that automated segmentation methods HIPS, volBrain and CAT can be used in the classification of AD and HC. This is an indication that such methods can be used to inform researchers and clinicians of underlying mechanisms and progression of AD.

Diagnosis of early mild cognitive impairment using a multiobjective optimization algorithm based on T1-MRI data

“…Our method was based on volumetric data: brain images were segmented into smaller brain areas over the cortex (volBrain and CAT) as well as hippocampus (HIPS) using standardised brain atlases. Therefore, the individual components involved in the optimisation algorithms reflect the size of each brain area, which is extremely interpretable 26 . This is in contrast to optimisation methods based on deep neural networks and support vector machines that are mostly considered as black boxes 42 .…”

“…Hence, there is no need for particular expertise to use these tools, which makes them more accessible and more practical in clinical settings. The validity of these methods, however, is still to be fully studied 17 , 26 , 60 – 62 . In particular, their level of accuracy in segmentation of brains with different disorders (such as those with atrophy) is less clear.…”

“…In this method, T1-MRI data of healthy participants (CN; n = 56) and patients (EMCI; n = 54) are extracted from the ADNI database. A similar segmentation method was used as in an earlier report 26 , see also Supplementary Methods section. We obtained volumes of different brain areas using volBrain, CAT and HIPS.…”

“…Hippocampus subfield Segmentation (HIPS) 22 , volBrain 23 , and Computational Anatomy Toolbox (CAT) 24 , 25 are some of the commonly used fully-automated methods. The segmented brain areas are used as features in the classification methods 26 , 27 .…”

Classification of early-MCI patients from healthy controls using evolutionary optimization of graph measures of resting-state fMRI, for the Alzheimer’s disease neuroimaging initiative

“…Further, in two studies, we showed that T1-weighted MRI (structural MRI; sMRI) can be used in classification of AD and MCI. Indeed, the majority of early studies looking at classification of AD and HC was done on sMRI [ 22 ]. This is mostly due to costs and accessibility of sMRI data [ 23 ].…”