The participants were divided into three age-range matched groups, which consists of 40 healthy normal controls (NC, Male/Female = 22/18, aged 65.4 ± 6.2 year old), 41 patients with aMCI (Male/Female = 20/21, aged 68.3 ± 6.7 year old) and 28 patients with AD (Male/Female = 11/17, aged 68.5 ± 7.9 year old). All subjects were screened by a medical history review, physical examination, and neuropsychological testing. The physical examination, and neuropsychological testing was performed within 2 weeks of the MR examinations. Four key cognitive domains were assessed: memory, executive function, language and visuospatial skills. The neuropsychological assessments performed in this study including the Mini-Mental State Examination (MMSE), clinical dementia rating (CDR); word sequence learning-recall, visual reproduction, logic memory II in the memory function evaluation; semantic association of verbal fluency, Wisconsin Card Sorting Test (WCST) completed categories, Wechsler Adult Intelligence Scale-III (WAIS-III) digit symbol-scaled, WAIS-III digit span-scaled in executive function evaluation; 3-D block construction Models in visuospatial evaluation, and object naming test in the language function evaluation. Subjects with depressive symptoms were excluded by using either the Hamilton Depression Rating Scale (NC) or the Cornell scale for depression in dementia (AD and aMCI). The general exclusion criteria for MRI examination also applied. The Apolopoprotein E status were also examined if available.

The diagnosis of AD was made by using NINCDS-ADRDA criteria (McKhann et al., 1984). The MCI patients were determined by the criteria of Petersens et al. (Petersen, 2004). An aMCI patient was defined as patient with MCI and memory impairment. The healthy controls were identified as cognitively normal, judged by clinical assessment and neuropsychological testing. The demographic data are summarized in Table 1.

Images were acquired using a 3T MR scanner (MAGNETOM Trio a TIM system, Siemens, Erlangen, Germany). T2 fluid-attenuated inversion recovery and T₁-weighted magnetization-prepared rapid acquisition gradient echo (T1-MPRAGE) images were acquired to rule out the concomitant neurological disorders. T1-MPRAGE images were subsequently used in post-processing for data normalization. The following imaging parameters were used: repetition time (TR)/echo time (TE) = 2000 ms/2.63 ms, 160 axial slices of voxel size = 1 × 1 × 1 mm³, inversion time = 900 ms, flip angle = 9 degree. The single acquisition time was 4 min 08 s.

Diffusion weighted images were acquired using spin-echo echo planar imaging (EPI) sequence with the following parameters: TR/TE = 7324 ms/83 ms, 64 axial slices of voxel size = 2 × 2 × 2 mm³, b-value 1000 s/mm² with the diffusion weighted gradient applied along 64 non-collinear directions. The single average acquisition time was 8 min 47 s.

Among the groups (NC, aMCI, and AD), no significant difference was noticed in gender, age and education level (Table 1). Compared to NC, AD patients had significant declination in all neuropsychological tests (p < 0.001). Compared to aMCI patients, AD patients showed reduced cognitive function in all tasks except for the learning-recall test of memory function, the completed categories test of executive function and the object naming test of language function. In addition, patients with aMCI had disturbed functions in the mean global and memory functions, semantic association of verbal fluency and the complete categories of Wisconsin card sorting test. No Apolipoprotein ε4 (ApoE4) carriers were found in NC. Nine of 41 aMCI patients and 4 of 28 AD patients were ApoE4 carrier.

Compared to NC, aMCI showed increased cortical MD in the right fusiform. In contrast, wide spread regions (Figure 1A, violet colored, 39 AAL regions) with significant increase MD can be found in AD patients, mainly located within the temporal lobe and cingulate gyrus. When comparing AD patients to aMCI, significantly increased MD was noticed in 10 cortical regions (Figure 1B, violet colored), including the bilateral hippocampus, bilateral posterior cingulum, right anterior cingulum, right amygdala, right caudate and left cuneus.

Both patients with aMCI and AD performed volumetric loss in cortical regions. The atrophy regions in aMCI were found in bilateral hippocampus and right parahippocampal gyrus, when compared to NC. In contrast, atrophy regions were found in AD patients mostly in temporal, parietal and frontal lobes, including the hippocampus, insula, fusiform, superior temporal gyrus, olfactory, and the inferior frontal triangularis (Figure 1A, warm colors). In addition, AD patients showed atrophic regions mainly hippocampus, fusiform gyrus, insula, olfactory gyrus, middle temporal gyrus and inferior frontal triangularis when compared to aMCI (Figure 1B, warm colors).

The increased MD pattern among patients with aMCI and AD was showed in Figure 2. Initial MD increase can be observed in aMCI patients only located in the right fusiform (green color), when compared to NC. Compared to aMCI patients, AD patients showed expended regions of increased MD (yellow color), such as hippocampus, parahippocampus, amygdala, inferior temporal gyrus, cuneus, precuneus, anterior cingulum and posterior cingulum. In the later stage, increaed MD regions can be observed in every lobe of cortical brain in AD patinets (red color), compared to NC. Significant difference in the MD values were found in different cortical regions among the groups in comparison (Figure 3, stars: between AD and NC; pounds: aMCI and NC; plus: aMCI and AD). Although both patient groups have significantly increased MD in cortical regions, the amount of change and the affected regions from the aMCI patients is always less than AD. The left hippocampus was the site with the most significant increase (p < 0.00001, 12.51% increase in AD patients), when compared to NC. In the affected regions, MD increased by 7.52% in AD patients on average.

To identify the difference in the involvement of white matter, the difference of TBSS in between aMCI and NC was shown in Figure 4. Significant changes can be noticed in right superior longitudinal fasciculus (SLF), internal and external capsule (Figure 4A). A general increase of MD was noticed throughout the whole brain (Figure 4B). However, no significant difference in white matter was found between the aMCI and AD patients.

Table 2 summarizes the cortical regions with significant correlation between MD and neuropsychological scores. Hippocampus and its neighborhood in the temporal lobe, such as fusiform gyrus and inferior temporal gyrus, are the main regions with significant correlation the functions of interest.

MD in the olfactory cortex and amygdala were significantly correlated with scores of Mini–Mental State Examination (MMSE) and Clinical Dementia Rating (CDR). The learning recall score of memory function was significantly correlated in regions located in the inferior frontal gyrus pars triangularis. The regions with correlation to semantic verbal fluency appeared to be scattered in the whole brain. The Wechsler Adult Intelligence Scale (WAIS-III) digit symbol-scaled score were significantly correlated with regions located in calcarine and lingual gyrus. In the visuospatial test, the 3-D block construction score significantly correlated with the cingulum, middle occipital, cuneus, lingual and angular gyrus. In the language function test, the object naming score was correlated with inferior parietal and angular gyrus. No significant correlation was found between education level and MD.

The current study reported an increased number of affected cortical regions and with growing difference, as the disease progressed from aMCI to AD. We measured the cortical mean diffusivity from two patient groups in comparison to the NC. Significant differences between AD and aMCI groups can be noticed in two aspects: the spatial pattern of the affected regions and the magnitude of increase. Although the atrophic pattern in AD suggested the whole brain involvement, the current study provided image based in vivo evidence that the microstructural damage might begin at an earlier stage. The spatial distribution suggested the involved regions might extend beyond the temporal lobe. The increase in mean diffusivity, a reflection of the severity in damage, indicated a widespread cortical impairment from MCI to AD. This spatial distribution pattern of regions with increased MD is similar to the post mortem pathology of amyloid deposition in AD patients (Braak and Braak, 1991). Furthermore, the affected region is consistent with that from the volumetric atrophy. Because the increase of MD is often attributed to an increase from the extracellular volume (Syková, 2004), as a result of the potential cell death, our observation might suggest a microstructural damage due to cellular degeneration or neural demyelination (Basser and Pierpaoli, 1996). Besides, significant correlations between the MD from the affected regions and the neuropsychological scores were observed.

In the aMCI patients, the current study observed significant atrophy located in the bilateral hippocampus, with increased MD in the peripheral region, noticeably fusiform gyrus. Only the right fusiform showed significant increase in MD (p = 0.0005, percentage increase of 4.35%). Similar increase of MD was noticed in bilateral hippocampi (Left side: p = 0.001, increase = 4.98%; Right side: p = 0.03, increase = 3.37%). Although the difference did not reach statistical significance, the percentage increase from left hippocampus is even higher than that in the right fusiform.

The white matter damage indicated that a more distant region could be affected such as SLF (Figure 4A). Post-mortem study showed that the brain degeneration in aMCI (Braak and Braak, 1991) started from the hippocampus and spread into the surrounding area in the later stages. The increased cortical MD in right fusiform might be related to the functions of word recognition and within-category identification (McCarthy et al., 1997). The finding is consistent with the reduced performance in verbal fluency and completed categories tasks in this specific group of patients. The gray matter atrophy pattern in the hippocampus might be related to the potential damage in memory formation and storage (Burgess et al., 2002), as observed in the impaired memory functions.

Increased MD of the right SLF might suggest declination in the microstructural integrity within this region. The SLF structurally connects with the temporal, occipital and frontal lobes, which is functionally associated with the memory formation and higher cognitive functions such as verbal fluency (Kamali et al., 2014). In addition, the loss of integrity of white matter tracts which connected to gray matter with memory function has been associated to AD-risk (Gold et al., 2012). Increased MD in SLF could be potentially more sensitive to neurodegeneration than the volumetric atrophy.

Increased MD was observed in widespread cortical regions of the AD patients. The spatial distribution of the involved regions was similar to that in beta-amyloid deposition and neurofibrillary tangle (Braak and Braak, 1991). This similarity might suggest possible association between changes in MD and the disruption within the neuronal environment. This spatial pattern of increased MD is also consistent with gray matter atrophy, similar to the previous report (Burton et al., 2009). The region with the most significant increase of MD and the most severe volumetric atrophy are both located in the hippocampus. Because hippocampus and the entorhinal cortex are involved in consolidating information from short-term memory into long-term memory (Schwindt and Black, 2009), the dysfunction in these regions are the hallmark of the AD pathology (Braak and Braak, 1991; Du et al., 2001). The increase of MD in hippocampus might attribute to the memory loss in our study (Table 1).

The affected white matter tracts in AD mainly located in bilateral uncinate fasciculus, SLF, and cingulum (Figure 4B). These affected tracts involves in the connection between the frontal, temporal and occipital lobes in neocortex (Friederici, 2011). It can be noticed that the cortical MD in the surrounding area was significantly increased. This observation might suggest that a more extensive neural network might be involved, as reflected in the increase of MD. In addition, the degeneration of the cingulum involved the development of cognitive impairment in AD (Bozzali et al., 2012), which might refer to the significant cognitive dysfunction in our AD patient group.

Compared with aMCI, more extensive cortical regions were involved in AD patients (one cortical region in aMCI patient and 40 in AD patients). Both the increase of MD and the involved regions suggest that the aMCI patients in the current study are in a transition state between the NC and AD patients. More severe brain degeneration in the more severe brain degeneration can be observed in AD patients when compared to aMCI. However, in term of white matter involvement, no significant difference between AD patients and aMCI can be found. This observation might support the fact that white matter changes could be secondary to the neurodegeneration in AD type dementia.

The current study noticed an overlap between the regions of gray matter atrophy and the more extensive and widespread areas of increased MD (Figure 1). The neurodegeneration process (Tripathi, 2012) begins with the loss of neural integrity, which subsequently evolve into massive cell death and shrinkage, and thus ultimately resulted in volumetric reduction and brain atrophy. The increase of MD is often attributed to the underlying microstructural damage. Because the spatial distribution pattern of the involved regions, as indicated by the increase of mean diffusivity, is similar to the post-mortem pathology of amyloid deposition in AD (Braak and Braak, 1991) and is more extensive than the atrophy map. It might suggest an increased sensitivity in the changes of MD than tissue atrophy. This observation certainly warrants new interest in further investigation in the future.

Hippocampus, fusiform and inferior temporal gyrus were mainly negative correlated with the cognitive evaluation scores. The previous studies reported that the hippocampus atrophy associated with not only loss of memory formation but also executive dysfunction such as non-memory cognitive impairment in early AD and aMCI patients (Burgess et al., 2002; Nagata et al., 2011). The current finding suggested that the degeneration of hippocampus and its surroundings were not only the landmark of aMCI and AD, but also involved the reduction of clinical cognitive performance in every aspect of evaluation including memory executive, visuospatial and language functions.

In addition to the hippocampus, different regions of cortical MD with significant correlation were noticed in different tasks for cognitive performance scores. Specifically, the memory function (the learning recall task) test showed negative correlation to MD in inferior frontal gyrus pars triangularis on left side, which has been known as a critical role in memory retrieval (Badre and Wagner, 2007). A scattered spatial pattern of cortical MD was also correlated with the semantic verbal fluency task of the executive function tests. The damage in cortical regions might interrupt the network of executive function including the ability of attention, planning and execution (Zhang et al., 2009; Nagata et al., 2011).

Deficits in visuospatial function in patients with aMCI and AD may suggest the loss of integrity of visual processing network. The middle occipital, cuneus and angular gyrus were noticed with significant correlation to the 3D block construction of the visuospatial test, which are critical roles involved the visuospatial information process (Bone, 2001). Interestingly, the regions with significant correlation between visuospatial function and cortical MD were mainly found on the left side. Although a lateralized brain network in visuospatial function were reported (Thiebaut de Schotten et al., 2011), the association between laterality of visuospatial function and cortical brain degeneration might need for further investigation.

Apolipoprotein ε4 (ApoE4) carriers could be found in patients with MCI and AD. The ApoE4 is the most widely known genetic risk factor for sporadic AD in different ethnic groups (Sadigh-Eteghad et al., 2012). However, Wai et al have reported a mild impact of the ApoE4 status on the extent of white matter disruption in the aMCI patients (Wai et al., 2014). The present study did not found significant trend between of the ApoE4 carrier and aMCI/AD. Number of ApoE4 carriers in aMCI/AD patient groups were smaller than non-carriers. Interestingly, the previous study has showed that the ApoE4 is less common in Chinese people than Caucasians or Japanese (Wong et al., 2005), which could explain the observation of leak of association between ApoE4 and AD/aMCI.

The major limitation in the current study is a compromised spatial resolution in a template-based parcellation approach, each of which consisted of multiple voxels. This might lead to the potential statistical type I error. In a template-based parcellation approach, the MD value is an average from all voxels within the region of interest. If the observed effect within the region is heterogeneous, the statistical significance might be affected. Furthermore, the inhomogeneous nature and limited number of aMCI participants might reduce the sensitivity of detection. To establish a relationship between MD and the amyloid deposition is an interesting and important issue, which will have to be investigated in the animal study in the future because biopsy is impossible in human neurodegeneration study. Secondly, the diagnosis remains clinical, based on psychometric evaluation (Petersen, 2004). In the study design, only age and gender in the participants were matched. This data inhomogeneity, as shown in MMSE, is one major limitation in the current study, which to an extent reflected the difficulty in the diagnosis of the neurodegenerative disease, specifically Alzheimer Disease. Objective evidences such as neuroimaging-based and genome-based might improve the accuracy of clinical judgment. Finally, because the APOE status remained uncertain for a significant amount of the participants because they did not provide the blood sample for examination, this might lead to concerns in the reliability of the result for the effects of APOE in neuroimaging.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.