Conventional MRI

It is widely accepted that there is an association between the neurofibrillary pathology and the atrophy displayed in structural MRI (Whitwell et al. 2008), which is the most common marker used for diagnosing and monitoring the progression of AD, while there is no association between the beta-amyloid burden and the extent of brain atrophy (Josephs et al.

2008).

In a standard conventional MRI protocol 3 types of sequences are usually applied: coronal 3D T1-weighted, transverse T2-weighted and transverse fluid-attenuated inversion recovery (FLAIR). Transverse T2* imaging, such as susceptibility weighted imaging, sometimes is included as a way to detect MBs. These sequences provide a minimum set of information to answer two questions: the extent and pattern of brain atrophy and the

degree of vascular damage (Barkhof et al. 2011). T1 describes structural changes, T2 and FLAIR reflects pathological changes and T2* is sensitive to haemorrhages. Many factors should be considered in evaluating an imaging protocol for investigating dementia (Barkhof et al. 2011, Bhogal et al. 2013): (1) are any reversible causes of dementia syndrome present (e.g. normal pressure hydrocephalus and subdural haematoma); (2) the presence, extent and location of small cerebrovascular diseases; (3) the presence and distribution of MBs; (4) degree and pattern of general cortical atrophy; (5) focal atrophy, especially in the hippocampus, medial temporal lobe, frontal lobes, posterior cingulate, precuneus, cerebellum and pons.

Visual ratings provide a qualitative study of the images; they are fast with some scales which are recognized for being specific in evaluating the changes in dementia patients.

There are several scales available: global cortical atrophy, medial temporal lobe (MTL) atrophy using the Scheltens scale (Scheltens et al. 1995) and ARWMC using the Fazekas scale (Fazekas et al. 1987). The assessment of the MTL atrophy in coronal T1 weighted MRI is considered to be the gold standard for the diagnosis of AD, as the hippocampal atrophy rate varies as MCI progresses to AD (Jack et al. 2008).

The problem with these ratings is the subjective component inherent in the final decision;

in addition, experience is needed in using these scales. Furthermore it is not easy to monitor the progression with respect to time.

In AD, there is usually a marked atrophy of the hippocampus (Scheltens 2009) (Figure 2).

It has been reported that there is an annual atrophy rate of around 1.55% in controls and 4%

in AD patients (Jack et al. 1998). The amygdala also undergoes atrophy in AD (Cavedo et al.

2011). Structural MRI assessments of the entorhinal cortex and the hippocampus, particular the CA1 region, are associated with a higher risk of developing AD (Stoub et al. 2005, Apostolova et al. 2006). The atrophy starts in the entorhinal cortex, progresses to the hippocampus and medial temporal lobe and finally to the parietal cortex (Jack 2012).

BvFTD can usually detect a marked atrophy in the frontal lobes and anterior temporal lobes, and this can be evaluated visually (Bhogal et al. 2013). Hippocampal atrophy may be detected also in bvFTD cases (Figure 2), thus it is not a phenomenon restricted to AD (van de Pol et al. 2006). FTD patients have a higher rate of whole-brain atrophy in comparison with AD cases (Chan et al. 2001).

Figure 2. From left to right, coronal T1-weighted MRI scans of: control case (Scheltens score = 0), AD case (Scheltens score=4) and bvFTD case (Scheltens score = 2; prominent frontal atrophy). Courtesy of Dr. Yawu Liu, Kuopio University Hospital.

A visual assessment can also identify other changes such as lacunes, MBs or the enlargement of the ventricles, all of which can be altered in the presence of normal ageing and dementia diseases.

Manual volumetry (outlining) (Figure 3) is useful, but it is time consuming. Furthermore there is no standard technique for delineating the hippocampus in the MRI images.

Figure 3. Visual rating system for assessing hippocampal atrophy: on the left, the hippocampus outlined in red shows mild atrophy (score=2). Courtesy of Dr. Yawu Liu, Kuopio University Hospital.

The use of automatic and semi-automatic techniques is intended to exclude the subjective component that biases the decision of individual clinicians. Moreover in automatic techniques, one can select specific regions-of-interest (ROIs) and quantify their volumes in order to monitor the rate of atrophy with respect to time, thus it is possible to study more regions of the brain. A large number of different methods exist for quantifying atrophy, as reviewed (Soininen et al. 2012a). Five of these automatic methods are as follows: automatic hippocampal volumetry (HV), tensor-based morphometry (TBM), voxel-based morphometry (VBM), manifold-based learning (MBL) and cortical thickness (CTH).

HV displays the total volume of the left, right and whole hippocampus (Wolz et al. 2011).

Hippocampal atrophy is not a phenomenon restricted to AD, although the rate, amount and pattern atrophy can differentiate AD from FTD (van de Pol et al. 2006, Frisoni et al.

1999). Furthermore HV can evaluate specific regions in the hippocampus, such as conducted in one study comparing the accuracy between global hippocampal atrophy and selected atrophy in CA1, which found maximum atrophy in the CA1 in amnestic MCI and AD patients (La Joie et al. 2013).

TBM calculates the average Jacobian of atrophic voxels within a ROI, weighted based on voxel-wise p-values (Wolz et al. 2011). In other words, the voxel-wise proportion of variance in volumes is computed for specific ROIs for one disease versus another disease (Brun et al. 2009). One study using support vector machines and linear discriminant analysis in ADNI data was able to differentiate healthy controls from AD patients with an accuracy of 87%, and stable-MCI (SMCI) from progressive-MCI (PMCI) with an accuracy of 64% (Wolz et al. 2011). Brambatti et al., compared FTD cases and controls after 1 year follow-up, and in the whole brain analysis, a significant atrophy change was detected in the anterior cingulated and paracingulate gyri (Brambati et al. 2007). Another study observed different atrophy patterns SD and PNFA (Lu et al. 2013).

VBM involves a voxel-wise comparison of the local concentration of gray matter between two groups of subjects (Ashburner, Friston 2000). It has been widely used in the field of dementia. One study indicated that in contrast to controls, MCI cases displayed significant

unilateral atrophy in the medial temporal lobe (Pennanen et al. 2005). In addition, one study surveyed the conversion from amnestic MCI to AD after an 18 month follow-up; it found a significant GM loss in converters relative to non-converters in the hippocampal area, temporal gyrus, posterior cingulated and precuneus (Chetelat et al. 2005). A meta-analysis revealed GM changes in the frontal-striatal-limbic brain areas in patients with bvFTD compared to controls (Pan et al. 2012). One study using statistical parametric mapping detected decreased GM volume in the frontal and anterior temporal lobes in a FTD group while in AD there were decreased volumes bilaterally in the posterior cingulated gyri and parietal lobules (Kanda et al. 2008).

MBL estimates the coordinates of a subject in a low-dimensional manifold space learned from pairwise image similarities (Wolz et al. 2011). It has been shown to differentiate controls and AD cases with a high degree of sensitivity (90%) (Wolz et al. 2011).

CTH computes the average thickness within a ROI defined based on a group-level statistical analysis (Wolz et al. 2011). One ADNI study revealed that normalized thickness could distinguish AD patients from controls with an accuracy of 85% and predicted conversion from MCI to AD in 76% of the subjects (Querbes et al. 2009). When applied to compare FTD with MCI and controls, if one determined cortical thinning in the frontal and temporal poles, then no statistical differences were found in the AD vs. FTD group comparison (Hartikainen et al. 2012a).