Background

Alzheimer’s disease (AD) is the most common cause of dementia in the elderly population. It is progressively deteriorating different cognitive and behavioral domains causing serious effects on the quality of life of patients and their family members, and eventually, AD contributes to death. The diagnosis of AD usually involves clinical assessments, cognitive tests, and brain imaging. Early and accurate diagnosis of AD can help to delay the decline of the cognitive functions of the patients for months or years and also help in early initiation of treatment [1].

AD is characterized pathologically by the presence of abundant extracellular amyloid neuritic plaques and intracellular neurofibrillary tangles (NFTs) [1,2,3], the hippocampus and entorhinal cortex were found to be the first sites of pathological changes [4], and these changes are present years or even decades before the onset of the clinical manifestations of the disease.

The most widely used clinical criteria for the diagnosis of AD are the National Institute of Neurological and Communicative Disorders and Stroke Alzheimer’s Disease and Related Disorder Association (NINDS-ADRDA) criteria and Mini Mental Status Examination (MMSE). Developments in magnetic resonance imaging (MRI) have greatly improved the understanding of AD. MRI enables us to visualize structural, metabolic, and functional changes. Medial temporal lobe atrophy can be associated with the disease [5].

Unfortunately, this has been found to have limited utility both in early disease and in the differentiation of AD from other causes of dementia and from normal aging [6]. In other words, by conventional MRI means, atrophy of the hippocampus and entorhinal cortex can be assessed using simple validated scales [7]; however, there is an overlap with age-related atrophy cases that makes the distinction of AD from normal aging more problematic in the elderly to achieve an accurate and reliable assessment using a human rater [8].

Volumetric analyses of brain structures have become increasingly common for diagnostic purposes and for identifying disease progression. It has been demonstrated that atrophy of the medial temporal lobe (MTL) memory structures, where the pathological changes of AD in terms of neurofibrillary tangles and neuronal loss first occur [2], can be detected even in subjects with mild cognitive impairment (MCI) [9, 10].

Accurate regional segmentation on MR images is necessary to obtain the volumetric measurements for various structures as the conventional approach based on signal intensity alone is not sufficient to distinguish between different structures as observed and characterized by Fischl et al. [11].

The hippocampus has a critical role in declarative and episodic memory, as well as a focus of structural change in normal aging [12] and diseases such as epilepsy and Alzheimer’s disease (AD) [13].

The hippocampal formation consists of a number of distinct, interacting subfields (Figs. 1 and 2). Despite its internal complexity, limited MRI resolution previously forced researchers to model the hippocampus as a single, homogeneous structure in neuroimaging studies of aging and AD [14].

Fig. 1
figure 1

Four coronal slices for hippocampal head and body with corresponding manual annotations. The slices are ordered from anterior to posterior. Red = CA1, blue = subiculum, dark purple = presubiculum, yellow = parasubiculum, cyan = granule cell layer of the dentate gyrus (GC-DG), light green = hippocampus-amygdala-transition-area (HATA)

Full size image
Fig. 2
figure 2

Consequent stages during the processing of 3D T1-W images in FreeSurfer software. a Conventional coronal image at the hippocampal body level showing mild widening of the choroid fissure and temporal horns. b Axial image showing removal of non-brain tissue followed by subcortical segmentation where there is the identification and outlining of different brain structures including here the hippocampus as a whole organ. c Axial image showing segmentation of the hippocampus into subfields. d Coronal image showing filling of different hippocampal subfields and annotation in different colors. The table is showing calculation of the hippocampal subfield volume done by quantification of the number of voxels

Full size image

Even though studies have shown that whole hippocampal volumes derived from segmented MRI scans are powerful biomarkers for AD [11, 15], treating the hippocampus as a single entity disregards potentially useful information about its subfields (Fig. 3); they are also thought to play different roles in memory and learning [16, 17] and to be affected differently by AD and normal aging—as indicated by ex vivo, histological studies [18].

Fig. 3
figure 3

Consequent stages during the processing of 3D T1-W images in FreeSurfer software. a Conventional coronal image at the hippocampal body level showing moderate widening of the choroid fissure and temporal horns. b Axial image showing removal of non-brain tissue followed by subcortical segmentation where there is identification and outlining of different brain structures including here the hippocampus as a whole organ. c Axial image showing segmentation of the hippocampus into subfields. d Coronal image showing filling of different hippocampal subfields and annotation in different colors. The table is showing calculation of the hippocampal subfields volume done by quantification of the number of voxels

Full size image

Studies have shown that atlas-based volumetric analysis found that even hippocampal subfields have specific changes and neurodegenerative atrophy patterns associated with AD and can distinguish different diagnostic groups.

Methods

This study was conducted on a total of 45 subjects: 15 controls (9 men, 6 women) with mean age 62.4 years (± 6.2), ranging from 56 to 69 years old, and 30 Alzheimer’s disease patients (17 men, 13 women) with mean age 64.5 years (± 3.3), ranging from 58 to 70 years old; this was done in the time period from 2016 to 2018. The patients were referred to the Radiodiagnosis Department and the controls were among their relatives and caregivers. Then approval of Research Ethics Committee (REC) of Tanta University and written consent were obtained from all studied participants. Inclusion criteria were patients who have been diagnosed with Alzheimer’s disease by clinical examination, after fulfilling the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA), Diagnostic and Statistical Manual of Mental Disorders IV (DSM IV) criteria, MMSE score ranged from 9 to 23, and CDR score from 1 to 3. All individuals of the study were subjected to full history taking, thorough clinical examination, and routine laboratory investigations in the form of CBC, liver and kidney function test, fasting and 2-h postprandial blood glucose, ESR, LDH, and CRP to exclude diseases that affect the conscious level and memory function. Exclusion criteria included dementia of other types other than Alzheimer’s disease, systemic illness or organ failure, and patients unable to do MRI who have a cardiac pacemaker or metallic prosthesis.

Image acquisition and post-processing

The images were acquired by using 1.5 Tesla GE closed configuration whole-body scanner using a standard quadrature head coil. The patients were entered head first in supine position. All patients were subjected to the following conventional MRI protocols: sagittal 3D T1-weighted spoiled gradient (SPGR) utilizing the following parameters—a repetition time (TR) of 7.2 ms, an echo time (TE) of 120 ms, a slice thickness of 1.2 mm, and FOV = 256 × 256 mm. A high spatial resolution (isotropic) three-dimensional (3D) T1-weighted spoiled gradient echo (SPGR) of 1.2 mm slice thickness with no spacing was obtained in sagittal plane, with preparation time of 500 ms and an imaging duration of 5–7 min per acquisition; a repetition time 7.2 ms and minimum full echo time; a flip angle of 10° (to partly compensate inhomogeneity in the radiofrequency field); 160 slices, encoding frequency 16 kHz, a field of view of 30 cm, and an acquisition matrix size of 192 × 192 (FH AP RL; FH = foot-to-head; AP = anteriorto-posterior; RL = right-to-left); and frequency DIR S/I, NEX 1.0 with active automated shimming. Phased-array uniformity enhancement (PURE) was applied and surface coil intensity correction (SCIC) was off. This protocol resulted in high-resolution images with high gray-white matter contrast, used for measuring of specific gray matter (hippocampus) region of interest.

MRI data preprocessing

The preprocessed images are lodged into the automated brain segmentation software FreeSurfer, then algorithm initiation and calculation of hyperparameters for subcortical segmentation.

Data processing technique

A sequence of steps is then performed by the software as following: removal of the skull and any remaining background noise from T1 images to produce a skull-stripped corrected volume that is used in the segmentation algorithm; spatial normalization process is then done to the volume to correct the displacements, rotations, and scales and bring the person’s volume into Talairach space which is a standard morphological space; identification of the boundary of the gray/white matter interface as well as the pial surface; segmentation of the different brain anatomical structures, where different measures can be made; and the segmentation of the subcortical white matter and deep gray matter volumetric structures, such as the hippocampus. These structures can then be filled in different label colors.

Measurement of the hippocampal subfield volume ROI analysis

Further segmentation of the hippocampus is done and different subfields can be extracted to be labeled differently, and calculation of volume is then measurable and generated for each of the segmented hippocampal subfields (Fig. 4). The conventional images are interpreted by two neuroradiologists and one neurologist who have a medical doctorate degree and are experienced in neuroimaging for more than 10 years duration, and then the final volume measurements are the results of an automated processing which has no human rater involved.

Fig. 4
figure 4

Coronal T1-weighted images through the hippocampus head (nearly the same level) showing hippocampal subfields segmented and annotated in different colors, a in two different healthy control subjects and b in two different Alzheimer’s disease patients, during the processing. Red = CA1, blue = subiculum, dark purple = presubiculum, yellow = parasubiculum, cyan = granule cell layer of the dentate gyrus (GC-DG), light green = hippocampus-amygdala-transition-area (HATA)

Full size image

Statistical analysis of the data

Data were provided for the computer and analyzed using IBM SPSS software package version 20.0. To assess both the principal impacts of diagnosis (ADversus-control) factors on the hippocampal subfield volume, this work used analysis of covariance with total intracranial volume, education, sex, and age as nuisance variables.

Qualitative data were described using number and percent. Quantitative data were described using range, mean, standard deviation, and median. Comparison of MRI findings with normal values was done using one sample t test. Significant test results are quoted as two-tailed probabilities. The significance of the obtained results was judged at the 5% level. P value < 0.05 was considered significant.

Results

This study included 45 subjects: 15 controls (9 men, 6 women; mean age 62.4 ± 6.2 years) who were proved to be healthy and 30 Alzheimer’s disease patients (17 men, 13 women; mean age 64.5 ± 3.3 years). The demographic data and cognitive scores for control and Alzheimer’s disease cases are shown in Table 1. All patients were diagnosed by clinical examination, conventional MRI, and then hippocampal subfields volumetric measurements after segmentation with FreeSurfer (Fig. 4). The subiculum and CA1 subfield volumes in patients with Alzheimer’s disease were significantly reduced on both sides in relation to age-matched controls (P < 0.05), and there was no significant difference found between both sides (Table 2).

Table 1 Demographic data and cognitive score
Full size table
Table 2 Subiculum and CA1 subfields mean volumetric measurements in patients and control groups
Full size table

They were found to have a positive correlation with the MMSE score which proved to be statistically significant when compared with normal elderly subjects; correlation coefficient (r) was 0.971, 0.957, 0.98, and 0.987 for right CA1, right subiculum, left CA1, and left subiculum, respectively (Table 3).

Table 3 Correlation between the volumes of CA1 and subiculum in mm3 in AD patients and MMSE and CDR scores
Full size table

Meanwhile, they showed a negative correlation with the CDR score which was also statistically significant when compared with normal elderly subjects; correlation coefficient (r) was − 0.892, − 0.879, − 0.917, and − 0.871 for right CA1, right subiculum, left CA1, and left subiculum, respectively (Table 3).

Other studied subfields that were decreased could not be proved to be statistically significant. There was no significant correlation between group differences in age, sex, or years of education. Sensitivity, specificity, PPV and NPV of CA1, and subiculum volume reduction in AD relative to healthy elderly control were high, and the sensitivity of the study is increased by combination between volumes and the clinical exam scores.

Discussion

Alzheimer’s disease is the most common cause of dementia in elderly people, and it is a progressive disease and its prevalence is predictable to increase as populations continue to age around the world. It is associated with neurofibrillary tangles and neuritic plaques, along with other changes as well as amyloid angiopathy and age-related brain atrophy which may contribute to cognitive impairment [19].

The hippocampus was found to have a special affinity for AD pathology and considered an early site for involvement, so hippocampal atrophy is used to be an imaging marker of AD and included in the diagnostic criteria [20].

Visual rating methods used in the assessment of the volume of the medial temporal lobe and the size of the surrounding CSF spaces showed obviously that they are lacking the accuracy in anatomical segmentation, as a result has low sensitivity and specificity for distinguishing AD from normal subjects with senile changes. In other words, there were dementia subjects with minimal MTA score who have only increased width of the choroid fissure without the temporal horn dilatation and were found to have already established regionally pronounced hippocampal atrophy by volumetric means. It was noted also that cases showing some degree of temporal horn dilatation in conventional imaging could have normal volumetric studies without any significant atrophy [21].

This study aimed to identify a specific regional atrophy pattern characteristic for Alzheimer’s disease in comparison to normal age-matched control, and this could enable us to better understand the course of the disease and help in its early effective treatment. FreeSurfer software was used for automated hippocampal segmentation using the preprocessed images, after initiation of the segmentation algorithm and calculation of hyperparameters and then using the output to calculate their volumes. Volumes were normalized by the intracranial volume and the gray matter volume [22].

One of the advantages of FreeSurfer software is that it is applicable for images produced by any MRI machine. Patients were classified according to MMSE and CDR clinical examinations into mild, moderate, and severe groups. The mean age of each group was 62.4, 65.7, and 66 years for mild, moderate, and severe groups, respectively. The number of years of education was included in the criteria, and their means for the three groups were 12.2, 11.2, and 12.5 years, respectively. The patients also had mean number of years since first diagnosed as AD: 3.5, 4.7, and 6.25 years for the three groups, respectively. They had presented variable medial temporal lobe atrophy-MTA scores and white matter lesion FAZKEAS scale.

After the segmentation process of the hippocampal subfields, it was found that both subiculum and CA1 on both sides had significantly reduced volumes in AD patients relative to the control group. This finding is matching with other previous studies as Kerchner et al. [23] , Zhao et al. [24], and Trujillo-Estrada et al. [25] which suggested the involvement of CA1 and subiculum subfields in particular.

These findings were correlated with the MMSE and CDR scores. All results showed significant P value and correlation. Correlation to age was determined by Pearson’s correlation coefficient. The other subfields showed no significant volume differences between AD and controls.

This work is also matching with recent studies; Hett et al. [26] have presented in their study that CA1 and subiculum are the subfields which show the most significant atrophy in AD. Also, Zhao et al. [24] has found that significant atrophy is seen in CA1, subiculum, presubiculum, molecular layer, and fimbria subfields among subjective mild cognitive decline, amnestic mild cognitive impairment, and Alzheimer’s groups; they in turn had suggested that this could be used as a biomarker in early stages of AD. Using the same image analysis approach, Hanseeuw et al. [27] showed some difference as their study reported significant volume losses in the subiculum and CA2-3 subfields in a small group of 15 amnestic MCI subjects and 15 healthy controls.

Mueller et al. [28] evaluated correlations between subfield volumes and two different memory scores based on the California Verbal Learning Test in a mixed group including cognitively normal healthy controls with a subjective memory complaint and patients with cognitive impairment, and it was found that atrophy of CA subfields appear to be related to associative memory dysfunction observed in patients with cognitive impairment. In a similar way, analysis was run on 490 individuals (including controls, MCI, and AD patients), and the authors did not find any specific correlates in healthy controls, while regions corresponding to CA1 and subiculum were associated with delayed recall performances in patients [29].

An automated hippocampal shape-analysis method by using a pattern-recognition algorithm revealed a positive correlation between CERAD delayed recall scores and hippocampal deformation in the CA1 and subiculum [30].

The earlier postmortem pathologic studies showed that degeneration of the CA1 and subiculum found to be more severe as compared with other hippocampal components in early stages of AD, and these findings are correlated with this study. Human autopsy studies suggest that volume loss of subiculum and CA1 is related to the number of neurofibrillary tangles in these areas, neuronal loss, loss of dendritic arbor, or afferent innervation. Previous studies suggested that marked degeneration of the perforant path, that is providing input from layer III of entorhinal cortex to CA1 and subiculum, was a characteristic feature of AD [29].

However, CA3 and the DG, in contrast to subiculum and CA1, are not affected by the formation of plaques, tangles, and neuronal loss until a later stage in AD [31]. Mueller et al. [28] found that atrophy of CA subfields, as grouped together in a single region, appear to be related to associative memory dysfunction observed in patients with cognitive impairment and correlate with incapability of the patients to benefit from semantic processing during encoding new information. In a similar way, analysis was run on 490 individuals (including controls, MCI, and AD patients from the Alzheimer’s disease neuroimaging initiative), and the authors did not find any specific correlates in healthy controls, while regions corresponding to CA1 and subiculum were associated with delayed recall performances in patients. A reliable result appears to determine the role of CA1 atrophy in memory dysfunction in MCI and AD [29]. The results of this study reveal that sensitivity, specificity, PPV and NPV of CA1, and subiculum volume reduction in AD in relation to healthy elderly control were high, and the sensitivity of the study is increased by combination between volumes and the clinical exam scores. Limitations of this study are firstly that it is not possible to separate the temporal relation between subfield volume changes and the clinical symptoms because of its design as a cross-sectional study, and secondly, the relatively small number of studied subjects. Future longitudinal study and larger sample size combined with other biomarkers will be needed to determine which hippocampal subfields show the earliest atrophy in the disease process and to confirm the findings.

Conclusion

This work has proposed a new approach to better capture Alzheimer’s disease. Reduced CA1 and subiculum volumes is seen in patients with Alzheimer’s disease. Further studies are needed to integrate these findings in the routine workup in evaluating patients with AD.