Volumetry of Olfactory Structures in Mild Cognitive Impairment and Alzheimer’s Disease: A Systematic Review and a Meta-Analysis

Olfactory decline is an early symptom of Alzheimer’s disease (AD) and is a predictor of conversion from mild cognitive impairment (MCI) to AD. Olfactory decline could reflect AD-related atrophy of structures related to the sense of smell. The aim of this study was to verify whether the presence of a clinical diagnosis of AD or MCI is associated with a volumetric decrease in the olfactory bulbs (OB) and the primary olfactory cortex (POC). We conducted two systematic reviews, one for each region and a meta-analysis. We collected articles from PsychNet, PubMed, Ebsco, and ProQuest databases. Results showed large and heterogeneous effects indicating smaller OB volumes in patients with AD (k = 6, g = −1.21, 95% CI [−2.19, −0.44]) and in patients with MCI compared to controls. There is also a trend for smaller POC in patients with AD or MCI compared to controls. Neuroanatomical structures involved in olfactory processing are smaller in AD and these volumetric reductions could be measured as early as the MCI stage.


Introduction
Alzheimer's disease (AD) is the main cause of dementia in older adults [1]. According to a large meta-analysis that included 119 studies, the overall point prevalence of dementia due to AD among individuals 60 and older is 40.2 per 1000 persons in community settings [2]. AD is a neurodegenerative pathology characterized by the accumulation of amyloid-β plaques and tau neurofibrillary tangles in the brain that leads to dementia [3]. The neuropathology of AD begins 20 years or more before first cognitive symptoms appear [4]. At the behavioral level, it has been proposed that the first manifestation of AD neuropathology is a complaint of a recent cognitive change known as Subjective Cognitive Decline (SCD) [5] before the manifestation of cognitive deficits known as Mild Cognitive Impairment (MCI) [6]. Although these early clinical stages are useful to better predict who is at risk of developing dementia related to AD, it would be important to find earlier and more specific markers. For instance, only 14% of individuals with SCD and 34% of those with MCI are expected to convert to dementia at least [7,8]. Neurobiological damage related to AD that appears during a silent phase preceding SCD and MCI stages [4,[9][10][11] first occurs in the transentorhinal and limbic regions [12,13], which are involved in memory and olfactory processing [14][15][16][17].
The neuropathology of AD is driven by two processes: an extracellular accumulation of amyloid-β proteins (amyloid plaques) and an intracellular accumulation of hyperphosphorylated tau proteins (neurofibrillary tangles) [3,18]. Thal et al. [19] suggest five phases of amyloid-β accumulation: it appears (1) first in the neocortex, (2) then in the allocortex

Materials and Methods
This study has been conducted following PRISMA guidelines [58]. The protocol of this study was not registered.

Eligibility Criteria of the Selected Studies
Eligible studies were required to meet the following criteria: (1) contain an MRI measurement of OB and/or POC volumes, (2) include a clinical group (AD dementia or MCI) and a control group of cognitive healthy participants, and (3) both title and abstract had to be written in English.
Patients from AD groups had to meet the criteria for a clinical diagnosis of AD, characterized by a significant and progressive decline in two or more cognitive domains typically lead to memory deficits, behavioral symptoms, impairment of activities of daily living, and dementia [32,59].
Patients from MCI groups had to meet the criteria for a clinical diagnosis of MCI, characterized by the presence of cognitive or memory complaints, objective cognitive impairment, and a preserved independence in functional abilities that exclude dementia [60,61]. Participants from the control groups were cognitively normal individuals.

Outcome
In each eligible study, total volume of both OB and /or POC had to be calculated from MRI scans by a manual or automatic segmentation from T1-or T2-weighted sequences.

Search Strategy and Information Source
We searched for studies published up to February 2021 in PubMed, PsychNet, and Ebsco databases. Unpublished theses were found using the ProQuest Dissertations and Theses database. The following keywords were used: "Alzheimer", "mild cognitive decline", "MCI", "MRI", "volum*", "thickness", "olfactory bulb", "olfactory cortex" using the following syntax: ("Alzheimer" OR "mild cognitive impairment" OR "MCI") AND ("MRI" OR "volum*" OR "thickness") AND ("olfactory bulb" OR "olfactory cortex"). We also used the snowballing method and examined reference lists from eligible studies found in databases. After excluding duplicated studies, we reviewed 93 studies for OB comparison and 39 studies for POC comparison (See Figure 1). We then excluded reviews, case studies, qualitative papers, and off-topic studies (e.g., animal studies, no MRI data, absence of control group, etc.). As a result, 31 potentially eligible studies were identified for OB and 24 for POC.

Study Selection and Risk of Bias in Individual Studies
Based on the eligibility criteria mentioned above, the first author (BJ) evaluated all of the selected studies. The first author then sent the list of potentially eligible studies to a

Study Selection and Risk of Bias in Individual Studies
Based on the eligibility criteria mentioned above, the first author (BJ) evaluated all of the selected studies. The first author then sent the list of potentially eligible studies to a research assistant who was blind to the purpose of the study. Articles were included if they were approved by both evaluators based on the risk of biased assessment.
The risk of bias of the selected studies was assessed using the Newcastle-Ottawa Scale (NOS) [62] as recommended [63]. The NOS is a tool to evaluate the quality of non-randomized case-control studies included in meta-analyses. Criteria were based on the evaluation of participants' selection, the comparability between groups, and the ascertainment of the quality of methods used to measure OB or POC volumes. It was agreed that the most conservative result would be selected when disagreements would emerge between both evaluators. No major disagreement emerged, and no studies were excluded following this evaluation. However, it has to be mentioned that both evaluators were unable to assess the risk of bias for two studies, as they were written in Chinese [64,65]. These two studies were included in the eligible studies, as all relevant data for the metaanalysis were present in the abstracts written in English.

Analysis
We used Meta-Essential [66] to perform analyses. We calculated Hedges' g to obtain a standardized effect size for each comparison using the mean volumes and standard deviations reported in the eligible studies. When a study reported two volumes from the same structure (e.g., left and right volumes reported separately), a single effect size was calculated using the standard and recommended procedures [67] in order to avoid assigning more weight to studies with multiple outcomes. In this case, the effect size is computed as the mean of the left and right structure effect sizes: Mean effect size = Hedges g1 + Hedges g2 / 2 The variance of this mean is: In this equation, r is the correlation coefficient that describes the extent to which left and right structure volume co-vary.
Then, we calculated a combined effect size when the number of studies was appropriate (≥5) [67]. We followed recent guidelines for interpreting combined effect sizes as small (g ≥ 0.16), medium (g ≥ 0.38) and, large (g ≥ 0.76) in geriatric populations [68]. We used the more conservative random effects model to compute the significance level of the mean effect sizes for each study.

Risk of Bias across Studies
We qualified the presence of heterogeneity using Cochrane's Q-statistic and generated I 2 to quantify the degree of heterogeneity among effect sizes [69]. We assumed heterogeneity if P Q was significant at p < 0.05. When heterogeneity was assumed and the number of included studies was sufficient [70], we then tested the effect of potential moderators such as age, sex, scanner type, software used to perform analyses, MRI sequences, and the type of view (sagittal VS coronal) used.
We qualified publication bias using the Rosenthal's failsafe-N test that gives the number of potential unpublished studies that are required to turn the combined effect size statistically insignificant or to change the conclusions of the meta-analysis [71].

Study Selection and Characteristics
After analyzing full-text articles, six studies met the criteria for a total of 152 patients with AD and 166 controls (See Table 1).
The Rosenthal's failsafe-N was 180 which is large and suggests no publication bias.

Study Selection and Characteristics
For the comparison between patients with MCI and controls, we found three different studies for a total of 104 patients with MCI and 108 controls (See Table 2).
The Rosenthal's failsafe-N was 180 which is large and suggests no publication bias.

Study Selection and Characteristics
For the comparison between patients with MCI and controls, we found three different studies for a total of 104 patients with MCI and 108 controls (See Table 2).
All studies comparing OB volume between patients and controls used well-known clinical criteria to select their participants and the majority used a manual segmentation technique to measure OB volume. One study used an automatic parcellation of OB volumes [73]. Most studies measured OB volume controlling for factors such as total intracranial volume, age, sex, and education, except for one study that did not control for these factors [74].

Volumetry of the Primary Olfactory Cortex
Four studies met the criteria, but two studies used the same sample, leading to a total of three eligible samples. This prevented us from carrying out a formal meta-analysis. Again, all studies comparing POC volume between patients and controls used well-known clinical criteria or a clinical rating scale, to selected their participants. One study used an Brain Sci. 2021, 11, 1010 9 of 16 automatic to segmented the POC volume and two studies used a manual segmentation method. Each study measured OB volume controlling for factors such as total intracranial volume, age, sex, or education.
A general trend for smaller structures in both AD and MCI groups compared to the control groups is observed (See Table 3). Indeed, three studies compared POC volumes between patients with AD and controls. Two studies found a significantly smaller volume in patients with AD, one study reporting a more important decrease in the left POC for those with AD [77]. Among these three studies, two studies included MCI groups. Both studies found smaller POC volumes in patients with MCI compared to controls, but only one comparison was significant [53]. One study compared patients with early MCI to those with late MCI and reported a smaller volume for the early MCI group [78].

Discussion
This meta-analysis and systematic review examined neuroanatomical structures involved in primary olfactory processing in both MCI and AD. We found a lower OB volume in both clinical groups compared to those in the control groups. When looking at the POC, despite the small number of the studies included in the present meta-analysis, a trend for lower volume is also found in both clinical groups compared to those in the control groups. These results are consistent with the hypothesis of a progressive atrophy of brain structures involved in olfactory processing in the course of AD. Volumetric reduction of olfactory brain structures is measurable as early as the MCI stage and is still more severe at the dementia stage.
The volumetric reductions in olfactory brain structures are in line with post-mortem studies that showed the presence of amyloid-β plaques and neurofibrillary tangles in both OB and POC of patients with AD [51,52,80]. Kovacs et al. (2001) demonstrated and argued that OB damage occurs very early in Braak's staging (i.e., stage 0 or I) before AD pathology spread through the central olfactory system [80]. Our results regarding the volumetric reduction of OB and POC in patients with MCI or AD support this hypothesis. Volumetric reduction of these structures might have resulted from neurodegeneration due to the accumulation of amyloid-β plaques and neurofibrillary tangles [18]. Thus, amyloid-β plaques and neurofibrillary tangles are hypothesized to cause early damage in OB and POC of patients with AD and could result in a volumetric reduction of these structures that is measurable from MRI scans. However, it is important to note that this meta-analysis and systematic review included studies based on clinical criteria for both AD and MCI rather than on specific neuropathological measurements of AD. Therefore, future studies should include measurements of AD-pathology, such as CSF amyloid-β, amyloid PET, CSF phosphorylated tau, and tau PET, in order to verify that damages to olfactory structures are the direct expression of AD pathology. This consideration is particularly important in studies involving MCI patients since it is only a portion of MCI patients that will convert to dementia (≈34%), and more specifically, ≈31% of MCI patients that will convert to Alzheimer dementia type [8].
Neurodegeneration could explain olfactory deficits found in AD. The disease affects main olfactory functions such as odor detection threshold, discrimination of different odors, with a more severe deficit in higher-order olfactory tasks such as identification and recognition of odors [41]. One study found that left hippocampus volume reduction is related to poorer olfactory identification, which requests both olfactory and memory abilities [81]. The current study shows that the volumetric reduction observed in the course of AD is not specific to hippocampal structures and is found in other brain structures related to olfactory functions, i.e., the OB and POC. OB volume was found to be related to some specific olfactory functions such as odor identification [82] and odor detection threshold [83,84]. Thus, impaired performance of patients with AD on these functions might have resulted from neurodegeneration that occurred in OB. POC volume was also found to be related to some specific olfactory functions. One of the POC structures, the piriform cortex, is responsible for encoding odor objects [85]. Deficits in olfactory functions such as odor identification were found to be strongly correlated with tau and amyloid deposition within this structure [86][87][88]. Another structure of the POC, the entorhinal cortex, seems to be involved in olfactory functions as this structure plays a role in the transmission of olfactory information to the hippocampus [17,49]. However, the specific role of the entorhinal cortex in olfactory functions remains unclear, as very few studies have investigated this question. On a structural level, Petekkaya et al. [73] showed a significant and positive correlation between the volume of the entorhinal cortex and the OB. On a behavioral level, Devanand et al. [89] did not find any correlation between the entorhinal cortex volume and scores of odor identification. More behavioral and neuroimaging studies are needed to better understand the role of these structures and to better qualify the consequences of OB and POC volume reduction on olfactory functions in AD.
From a clinical point of view, neuroimaging techniques allow the quantification of brain structures and thus provide the possibility to detect in vivo cerebral atrophy, which can be used as a marker of neurodegeneration. Our results suggest that a volume reduction of OB and POC can be observed early in the course of the disease and can be detected from the MCI stage. Thus, OB and POC volume reduction might be new interesting biomarkers of AD. However, olfactory dysfunctions and atrophies in olfactory-related structures are not specific to AD and are also present in other neurodegenerative diseases such as Parkinson's disease [90]. Therefore, we propose to combine OB and POC volume reduction with more traditional biomarkers such as hippocampal atrophy to enhance the specificity of the early diagnosis of AD. As a result, using this new combining approach, we might increase the detection of those with MCI that will convert to AD.
At a methodological level, although we found a global effect size in favor of an OB volume reduction of patients with AD compared to healthy older controls, it was statistically heterogeneous. When analyzing the clinical and methodological diversity among studies, they were all very similar. Since our research question was precise and because the studies included in the meta-analysis shared many similarities, we concluded that the combination of different effects sizes was appropriate. The small number of studies included (n = 6) prevented us from conducting moderator analysis, which is what is recommended when heterogeneous effect sizes are found [69]. Factors such as the hardware/software used, the type of scanner or sequence used to measure the OB volume could explain such heterogeneity [91,92]. OB volume is also known to have a large interindividual variability and this variability could also explain the heterogeneity [80]. Another factor that could explain heterogeneity is that not all studies controlled for total intracranial volume, which is an important covariate to take into consideration when analyzing volumetric data. Finally, heterogeneity could be explained by the fact that the majority of studies included used a manual segmentation technique instead of an automatic segmentation technique to obtained OB volumetric data. Futures studies should focus on the development of automatic segmentation methods of the OB [93].
This meta-analysis and systematic review have certain limitations. The most apparent is the small number of studies included. In fact, one of the main results of this study is that there is a lack of scientific literature for studies that have examined brain structures related to olfactory functions in the course of AD. Therefore, with only four studies resulting from the systematic review process that compared POC or OB volume between patients with MCI and healthy elderly controls, we were unable to conduct a meta-analysis using a random-effects model, as is typically recommended [94]. Regarding the selection bias of the studies included in the reviews, we were unable to evaluate the quality of two studies [63,64] as only the title and abstracts were written in English (full texts were in Chinese and we received no response from the authors). However, we decided to include these two studies in the meta-analysis, since all pertinent information was accessible in the abstracts and the studies were published in peer-reviewed journals. For all the included studies except the last two, we used the NOS tool to assess the risk of bias as recommended [62]. No studies were excluded following the risk of bias assessment. Finally, a close examination of the included studies showed some divergence on the structures included in the POC. Several models of the POC have been conceptualized [95][96][97], but they generally included common structures such as the piriform cortex, the anterior olfactory nucleus, the amygdala, the periamygdaloid cortex, and the anterior performed substance. Nevertheless, there is a need for a better classification of the structures included in the POC, especially if POC volume is used as an early biomarker of AD.
Our results indicate a volumetric reduction of both OB and POC in patients with AD and results of studies from the systematic review show that this reduction is also present in patients with MCI. New studies are needed to better characterize the degree of volume reduction of both OB and POC in patients with MCI or those that are in an earlier stage of the disease, for instance those with a SCD [5]. Second, no studies included the distinction between amnesic and non-amnesic MCI groups. Future studies should compare olfactory structures between these subgroups since amnesic MCI has been associated with a greater olfactory impairment compared to non-amnesic MCI patients [98]. Third, there is a need to encourage longitudinal studies that focus on volume reduction of olfactory-related structures in the course of AD. Results from these studies could support the hypothesis that the volume of neuroanatomical structures involved in olfactory processing decrease as the disease progresses. Longitudinal studies with larger samples of cognitively healthy participants at baseline could also lead to the analysis of the predictive value of these volumetric measurements on the development of AD-related cognitive and olfactory decline. Lastly, future researches should focus on a better characterization of the POC and on the development of fully automatized segmentation methods of these structures.

Conclusions
To conclude, a volumetric reduction of the neuroanatomical structures involved in olfactory functioning is present in patients with AD and can be measured as early as the MCI stage. Combining this neuroanatomical finding with more traditional biomarkers of AD, such as the hippocampal atrophy, volumetric reduction of OB and POC could increase the specificity of the early diagnosis of AD.