Magnetic Resonance Imaging and Anatomical Correlation of Human Temporal Lobe Landmarks,in 3D Euclidean Space: A Study of Control and Alzheimer’s Disease Subjects

INTRODUCTION

The temporal lobe is a region involved in a variety of cognitive functions. These include sensory processes such as vision and hearing, as well as higher cognitive abilities such as declarative memory [1] and social and emotional behavior [2, 3]. Neural structures (amygdala, hippocampal formation, and parahippocampal region) are located in the medial region of the temporal lobe (MTL).

In a rostrocaudal direction, the temporal pole cortex (TP) is the rostralmost structure at the middle cranial fossa, and the most anterior component of the parahippocampal region [4, 5]. It corresponds to Brodmann’s area 38 [6] or to von Economo and Koskinas area TG [7]. Atrophy of the TP is associated with frontotemporal dementia syndrome, which entails a memory deficit and impaired perceptual processing [8]. The TP is continuous with the perirhinal cortex (Brodmann’s areas 35 and 36) and is bordered by the collateral sulcus laterally and by the limen insulae (li) or frontotemporal junction in the anterior part. Depending on the depth of the collateral sulcus, the perirhinal cortex borders the entorhinal cortex (EC) [9, 10]. Shortly after the li, the EC or Brodmann’s area 28 begins, which extends as far as the posterior limit of the uncus and the anterior limit of the lateral geniculate nucleus. The EC is a periallocortical structure with a special lamination in six layers that do not correspond to the neocortical layers [10–12]. The EC can be parceled into eight cytoarquitectonically distinct subfields, homologous to the nonhuman primate EC. This area is the gateway of cortical information to the hippocampus, which makes this region critical for memory formation. It has been shown that the EC is early affected in AD [13, 14]. Another structure that appears at the level of the li is the amygdala; its posterior boundary coincides withthe gyrus uncinatus. The amygdala partially overlies the head of the hippocampus. This region is responsible for a variety of emotional behaviors, such as fight and fear [15–17]. Caudally, the beginning of the hippocampus proper (Hp) protrudes on the temporal portion of the lateral ventricle (temporal horn, LV_t). This protrusion is part of the head of the hippocampus, which appears as an oval cross-section of gray matter, ventral to the amygdala. It is followed by the hippocampal body and ends in the hippocampal tail, which is related caudally with the splenium of the corpus callosum. The hippocampus is made up of three subfields: Gyrus dentatus (DG), Cornu ammonis fields (CA3, CA2, and CA1), and subiculum,which is followed by the presubiculum andparasubiculum [18].

Technological advances enabled the improvement of neuroimaging techniques, which allows precise anatomical examination of MTL structures, bothin vivo and ex vivo. Structural and functional magnetic resonance images (sMRI and fMRI), as well as positron emission tomography (PET), have been instrumental in the knowledge of this brain region in a non-invasive way. These techniques have improved the in vivo exploration and quantification of morphological changes that occur as a result of neurological diseases.

The hippocampal formation is the target region for several neurodegenerative and psychiatric diseases [19–21]. Brain volume and atrophy rates are sensitive markers to the neurodegenerative processes in the hippocampal formation. Structural neuroimaging is a fundamental part in the clinical evaluation of patients with suspected Alzheimer’s disease (AD) [22, 23]. Structural changes in presymptomatic AD are changing the diagnosis of the disease, and influence its future treatment. The atrophy of the MTL is now considered as a valid sign for the diagnosis of mild cognitive impairment [24–26]. Likewise, studies that compare MRI and histological measurements in both control and AD cases show a high correlation in the volume determined in the two types of studies [27, 28]. Despite great technological efforts, it is still difficult to delimit the cytoarchitectonic borders in this region with neuroimaging. New 7T or 9.4T scanners have opened a wide field of research where small anatomical details are detectable. Furthermore, they support other computational models for establishing atlases that describe anatomical details only available to the microscopic analysis of tissue[29–34].

In this sense, our goal is to associate the most common anatomical landmarks in a geometric space using MR imaging (1.5T and 3T) and histological sections of the MTL. We aim to identify the parameters that best separate control and AD groups, and to establish, if possible, the linear combination of parameters that best differentiate between these groups. Our contribution aims to determine the interrelationship in length of the different anatomical structures of the MTL in a 3D space, based on distances, and to create a visible anatomical network. The anatomical landmarks correspond to structures essential for cognitive functions that are severely affected in AD. These measurements follow a known mathematical analysis (Euclidean distances in 3D space between two points or centroids) that is seldom used in neuroanatomy.

MATERIALS AND METHODS

RESULTS

Table 3 shows the Euclidean distances between different anatomical landmarks in both CO and AD cases. The analysis of all 83 measured distances in the 3D space (including distances in a plane, rostrocaudal distances, and Euclidean or real ones) in 1.5T and 3T MR studies, and histological sections showed no significant differences between the cerebral hemispheres in the CO group (in AD cases only one hemisphere was studied). In the MRI study, statistical differences between hemispheres were found only in measurements between the corpus callosum and other anatomical structures.

With regard to sex, only the TP/Hp_e distance was significantly different (p < 0.01), particularly when the measurement is made in the same plane (Pl) and when 1.5T MR is used (26.7 mm and 17.2 mm, for men and women, respectively). We did not find any other significant difference in terms of sex in the distances between neuroanatomical landmarks.

Given the high inter-rater reliability in the identification of anatomical references, the distances could be easily replicated for both CO and AD groups (Fig. 3 shows the reproducibility of the measurements of the amygdala and hippocampus, specifically RC measurements between A_b/A_e and Hp_b/Hp_e). In both cases, there is clear biological variability among different cases showing correlation (Pearson r coefficients near one). In this regard, the highest correlation value (r = 0.98) is found when histological sections were used to identify landmarks and measure distances. These values were slightly lower for 3T MR data (r = 0.97), whereas the lowest correlation values were obtained for 1.5T MRI (r = 0.92).

The most representative anatomical landmarks in the MTL are shown in Fig. 4. The images from 3T MR, 1.5T MR, and histologic sections are compared at different levels showing a high coincidence for both size and morphology. We also include the distances in the three methods of study, in both, CO and AD groups, indicating the average in the Pl, RC, and Re measurements (Fig. 5). In percentage terms, more changes were found in RC distances. The TP/DG distance showed the largest reduction in distance between CO and AD cases (10.8%, 32.3%, and 9.1%, for Pl, RC, and Re measurements, respectively).

The difference between CO and AD was more relevant in the study of histological sections than in the measurements in the RC axis relative to the other two parameters (Pl and Re). The TP/DG distance showed the most significant differences when CO and AD groups were compared. The AD group showed a distance reduction of 33% (Fig. 6). The analysis including the three groups (CO < 65 years, CO > 65 years, and AD cases) produced similar results, showing statistical differences between CO and AD groups. However, no significant differences were found between the CO < 65 years and CO > 65 years groups.

The discriminant analysis revealed that a single variable resulted in a clear differentiation (82% of cases correctly classified) between four groups (CO, AD mild, AD moderate, and AD severe groups) using the TP/DG in RC & Hist distance (in rostrocaudal dimension and histological sections). More discrimination was obtained when we added the TP/li in Pl & Hist, and the li/hf in Re & Hist; in this case, only one AD moderate case was misclassified as AD mild, with a global percentage of correct classification greater than 95%. The full differentiation between CO and AD groups occurred when we added the TP/A_e in RC & 1T study. Figure 7 shows how Fisher’s classification function successfully distributed the cases. When AD mild and AD moderate cases were grouped together, a 91% of correct classification was reached with the TP/DG in RC & Hist distance (Fig. 8 Top); if TP/li in Pl & Hist distance was added, only a case was misclassified (Fig. 8 Middle). The 100% correct classification occurred when li/hf in Re & Hist, and TP/A_e in RC & Hist were added to Fisher’s classification function (Fig. 8 Bottom).

The TP/DG in RC & Hist measures allowed us to differentiate all CO and AD cases (100% accuracy, Fig. 9). The multivariate study of our three groups (CO < 65 years, CO > 65 years, and AD cases) showed that, when the selection was limited to a single distance, the most accurate parameter was the TP/DG in RC & Hist, which provided a 76% of correct classification (Fig. 10 Top). Adding the TP/A_e distance (both in RC & 1T, and in Pl & 3T), the cc/li in RC & 3T, and the hf/A_e in RC & Hist distances, we reached a 90% rate of correct classification with two CO cases misclassified (Fig. 10 Middle). We observed that a 100% rate of correct classification occurred when up to 11 distances were included in the study (Fig. 10 Bottom).

DISCUSSION

MTL changes may be essential for an early diagnosis of neurodegenerative diseases such as AD. In particular, specific changes in certain particular structures or layers may modify the global environment, and thus lead to variations in the surrounding structures. This variation modifies the distances between MTL structures. Numerous studies reveal the existence of changes in the distribution of MTL components and the marked atrophy of this region, visible even macroscopically [39–42]. Certainly, when it comes to this degree of macroscopic detail, the disease is already in an advanced stage, and symptoms are conclusive. The identification of these changes in the early stages of the disease may enable us to provide an earlier, faster, and more effective treatment of the disease.

In-depth brain structure descriptions allow the precise location of different brain anatomical landmarks, especially by means of rostrocaudal studies in different atlases [43]. In recent years, atlases based on MR imaging have become relevant in establishing early changes in neurodegenerative diseases [44–47]. This process of standardization in the location of different brain structures is increasing in such a way that patterns of brain morphology in humans have been created and compared with those of non-human primates. These studies are based on the analysis of the Euclidean distance matrix on collections of MR images, establishing 3D coordinates of the different anatomical structures [48]. These authors state that the relative connectivity between neural structures reveals a special morphology, which is closely related to functionality, and denote the role of the expansion of the temporal lobe in human evolution, highlighting the importance of the structures in the MTL.

Our research provides an assessment of the relationship between MR imaging (1.5T and 3T) and histological sections, which resulted in a high correlation between these two types of studies in identified structures (landmarks), with a high variability between cases of study. However, the variability between raters was lower than the variability among cases. In this regard, we have not only made these measurements in the traditional rostrocaudal axis (used routinely to describe the position of the different structures of MTL), but we have also applied Euclidean distances for the separation of these references, both in a plane (if both references were in a single plane) and the global Euclidean distance or actual linear path between these reference points. Comparison of our results with the results published by Insausti and Amaral [18] and Frankó et al. [49] is crucial for determining the distances at which different structures lie, and in order to search for modifications in distances among MTL structures in cases of neurodegenerative or other neurological diseases. Our goal is to strengthen the effectiveness of rostrocaudal descriptive data reported by these authors, considering the successive appearance of anatomical items along this axis. In our case, we found a broad agreement with the results reported by Insausti and Amaral [18]. In addition, MR and histological section duality highlights the need for further studies toestablish MR atlases correlated with histology and their quantitative description (the latter having a significant and predominant clinical value due to its precision in the identification of anatomical structures). One of our goals was to gather more quantitative data to enhance the morphological description of this rostrocaudal axis along the MTL, in particular through new measurements in 3D space. However, we must conclude that our results reinforce the great importance of the rostrocaudal axis highlighted in other studies [10, 50]. We must also state the added difficulty of an undetermined degree of subjectivity in the serial record of the different anatomical references in histological sections.

Our results demonstrate efficacy in differentiating control brains and those with AD. Many statistically significant parameters have been described in order to clarify existing changes in AD. The most important ones were those including distances between the TP and other anatomical landmarks (specially the DG). The identification of these elements is not always easy or accessible in routine MR studies, and it is still more necessary as its applicability could be immediate. Moreover, despite the high number of control cases analyzed, we must take into account the high variability among individuals; hence, the justification for relevant changes to differentiate neurodegenerative lesions. However, this study has the added value of using the same material for both MR (ex vivo) and histological sections, thus providing higher accuracy by combining histology and the MRI identification of landmarks. Also, it is important to note that our study entailed sectioning the brain in blocks of approximately one centimeter prior to performing the MRI, which prevented a perfect fit of all these blocks. Moreover, this fact is comparable to the precise adjustment that is required for histological sectioning with a microtome.

This research examined, for the first time, MTL distances in 3D space and applied multivariate discriminant analysis to attempt to explore the ideal choice of distances between anatomical landmarks that would allow a more precise classification of CO cases and AD patients. This discrimination is strong in histological sections with a linear single variable, i.e., the rostrocaudal distance between the TP and the beginning of the DG. In addition, we can also conclude that good discrimination is reached between CO and AD groups even considering three stages of AD cases (mild, moderate, and severe).