A Systematic Review of Intervention Studies Examining Nutritional and Herbal Therapies for Mild Cognitive Impairment and Dementia Using Neuroimaging Methods: Study Characteristics and Intervention Efficacy

Neuroimaging facilitates the assessment of complementary medicines (CMs) by providing a noninvasive insight into their mechanisms of action in the human brain. This is important for identifying the potential treatment options for target disease cohorts with complex pathophysiologies. The aim of this systematic review was to evaluate study characteristics, intervention efficacy, and the structural and functional neuroimaging methods used in research assessing nutritional and herbal medicines for mild cognitive impairment (MCI) and dementia. Six databases were searched for articles reporting on CMs, dementia, and neuroimaging methods. Data were extracted from 21/2,742 eligible full text articles and risk of bias was assessed. Nine studies examined people with Alzheimer's disease, 7 MCI, 4 vascular dementia, and 1 all-cause dementia. Ten studies tested herbal medicines, 8 vitamins and supplements, and 3 nootropics. Ten studies used electroencephalography (EEG), 5 structural magnetic resonance imaging (MRI), 2 functional MRI (fMRI), 3 cerebral blood flow (CBF), 1 single photon emission tomography (SPECT), and 1 positron emission tomography (PET). Four studies had a low risk of bias, with the majority consistently demonstrating inadequate reporting on randomisation, allocation concealment, blinding, and power calculations. A narrative synthesis approach was assumed due to heterogeneity in study methods, interventions, target cohorts, and quality. Eleven key recommendations are suggested to advance future work in this area.


Introduction
Dementia is a syndrome comprising over 100 diseases and is characterised by a decline in cognition that interferes with function and independence [1]. Over 46.8 million people worldwide have a diagnosis of dementia [2], and currently there is no cure. Dementia has a heterogeneous pathophysiology, with multiple mechanisms thought to play a role in 2 Evidence-Based Complementary and Alternative Medicine the various types. For example, there are several hypotheses on the pathogenesis of Alzheimer's disease (AD) alone (the most common type of dementia, making up approximately 60-80% of all cases [3]) including the amyloid-beta peptide hypothesis, the inflammation hypothesis, the tau hypothesis, and the cholinergic hypothesis [4]. Oxidative stress, hypoxia, calcium imbalance, abnormal metal accumulation, amyloidbeta peptide accumulation within mitochondria, and brainspecific insulin signalling deficiencies are all thought to play a role in the complex pathophysiology of AD [5,6]. Because of this, first-line single target pharmacological therapies for AD, acetylcholinesterase (AChE) inhibitors (e.g., donepezil) and N-methyl-D-aspartate (NMDA) receptor antagonists (e.g., memantine), are not particularly effective, boosting cognitive function in the early disease stages only, and are unable to slow or stop the disease progression [7,8].
In the absence of effective pharmaceutical options for dementia, complementary medicines (CMs) have been thoroughly explored. Randomised-controlled trials (RCTs) have been conducted on a range of CMs for dementia, cognitive decline, and mild cognitive impairment (MCI), with many studies currently ongoing. This research has largely focused on nutritional and herbal medicine interventions (e.g., resveratrol, anthocyanins, fish oil, vitamins B and E, Ginkgo biloba, Curcuma longa, Bacopa monnieri, and multiherb formulas such as Sailuotong [SLT]), dietary interventions (e.g., ketogenic and Mediterranean diets), mind-body interventions (e.g., mindfulness, yoga, tai chi, and other types of physical activity), and manual therapies (e.g., acupuncture), and has yielded mixed results due to a range of methodological inconsistencies. Therapies that show potential as adjunct treatments for dementia, or prevention methods, should be thoroughly investigated with the most rigorous and objective measures to reduce sources of bias.
Neuroimaging techniques can provide an objective, precise, and noninvasive measure of neuronal function and are particularly useful in the assessment of complementary therapies for dementia. Popular functional techniques applied in CM research include electroencephalography (EEG), functional magnetic resonance imaging (fMRI), positron emission tomography (PET), magnetoencephalography (MEG), single photon emission computed tomography (SPECT), and functional near-infrared spectroscopy (f NIRS). Structural magnetic resonance imaging (MRI) and diffusion tensor imaging (DTI) can also be used to assess changes in morphology following longer interventions. As detailed in Table 1, depending on study characteristics such as the sample's degree of cognitive impairment, intervention type and duration, neurocognitive function of interest, and reasons for using neuroimaging, these methodologies have a range of advantages and limitations that should be considered carefully before a specific technique is applied in a CM dementia research study.
Neuroimaging, in particular functional neuroimaging, can be utilised in dementia CM research as a sensitive measure of neurocognition, with the capacity to record changes that cannot otherwise be detected by standard pen-andpaper neuropsychological tests. This is useful given the small effect sizes often reported in CM research, particularly acute studies, and that any proposed intervention for cognitive decline is effectively fighting an uphill battle against neurodegenerative pathophysiology. Furthermore, some techniques can be used to explore the mechanisms of action of a therapy, which is particularly useful in psychopharmacological studies (e.g., nutritional and herbal medicines).
The aim of this systematic review was three-fold: (1) provide a comparison and critical evaluation of the characteristics of studies assessing nutritional and herbal medicines for MCI and dementia; (2) evaluate their use of structural and functional neuroimaging methods; (3) summarise intervention efficacy. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [9] was followed during the planning, conduct, and writing of this review.

Eligibility Criteria.
Several initial scoping reviews were conducted to determine the eligibility criteria and review scope. Eligibility criteria were determined in line with the PICO principles for systematic reviews [10]: Peer-reviewed studies were included if they reported a herbal or nutritional intervention for MCI or dementia and either structural or functional neuroimaging as an outcome measure. It should be noted that the search strategy was intentionally kept broad and also included both mind-body (e.g., yoga) and manual treatments (e.g., acupuncture); due to the large volume of results, only studies assessing nutritional and herbal interventions were included. Reviews, commentaries, conference proceedings, editorials, preclinical (in vitro and in vivo), and acute clinical studies were excluded, as were studies that were not published in English, or when the full text could not be retrieved.

Search Strategy.
The research team and an experienced librarian reviewed the search strategy before systematic searching commenced. Six databases were searched for studies published in peer-reviewed journals. Abstracts were retrieved from PubMed, ScienceDirect, Web of Science, ProQuest, Scopus, and PsycINFO ranging from databases' dates of inception to August 28, 2016. A full list of keywords and an example of the search strategy for the Scopus database are detailed in Supplementary Material available online at https://doi.org/10.1155/2017/6083629 (Table S1). Similar searches were carried out in the other five databases, with only minor modifications to permit changes in the use of searching symbols. Reference lists of key articles were also searched for other eligible studies. Captures changes in cognitive and/or sensory function that can have their source localised in the brain.
SPECT Quantifies changes in brain blood flow and metabolism.
Nuclear gamma camera captures a gamma-emitting tracer being absorbed by brain tissue at the same rate as blood flow.
Relatively cheap compared to other functional imaging methods (e.g., PET, fMRI).
Administration of radioactive isotope (usually injection) and exposure to gamma radiation.
Useful in assessing interventions for dementia as SPECT can differentiate dementia pathologies (e.g., vascular dementia versus Alzheimer's disease).
Allows for the mechanisms of action of CM therapies to be explored.
PET Typically assesses regional brain glucose metabolism by detecting gamma rays emitted by a positron-emitting tracer.
Regional CBF.  Limitations when trying to measure activity in subcortical tissue.
CM intervention-associated changes in CBF can be ascertained. Mechanism of action can be explored due to the modulation of haemoglobin.
MRI Structural MRI images the anatomy of the brain using magnetic fields, radio waves, and field gradients.
Most frequently used measures are voxel-based morphometry and ROI analyses.
No functional information available.
Volumetric changes in brain regions (or whole brain) can be investigated. Any changes are best explored in chronic studies.
DTI Measures diffusion of water in order to provide information on tissue microstructures so that white matter pathways within and between brain regions can be explored.
Exploration of brain networks is becoming increasingly popular within the field. Better resolution with more angles (e.g., 61 direction scan).
No functional information available.
White matter integrity and structural network connectivity can be explored.
Any changes are best explored in chronic studies.
Evidence-Based Complementary and Alternative Medicine 5 Note. Items rated as "yes" were scored as 1. Items rated as "no" or "unable to determine" were both scored as 0. Higher scores indicate a lower risk of bias.

Data Extraction and Appraisal.
One reviewer examined the titles and abstracts of each article. If there was any doubt regarding the eligibility of an article, the full-text was retrieved for clarification. Articles deemed eligible by one reviewer were further assessed by two other independent reviewers to ensure inclusion criteria were met. Any disagreements were resolved by reviewing the full papers and a subsequent discussion. Study characteristics were extracted from each full-text article. Data extracted included title, authors, publication date, aim, study type, disease focus, study population characteristics, number of participants (target cohort and controls), age (mean/median and SD), gender ratio, participant recruitment, diagnostic criteria, neuroimaging technique and analysis method, neuropsychological test battery, definition and dosage of CM, length of intervention, follow-up, and findings.
An assessment of methodological risk of bias in individual studies was conducted. A 10-item scale was constructed to suit the relevancy of studies in this review. The scale was informed by the Cochrane Handbook [11] and the Quality Checklist for Healthcare Intervention Studies [12] (detailed in Table 2) to capture major sources of bias including selection bias, internal and external validity bias, reporting bias, and statistical bias. For each study, the following elements were assessed: random sequence generation, allocation concealment, sampling, blinding, intervention description, neuroimaging methodology, validity and reliability of outcome measures, selective reporting, adverse events, and statistical power. Each of the 10 items on the scale were rated as yes (scored as 1), no, or unable to determine (both scored as 0), allowing higher scores to indicate a lower risk of bias. Studies with total scores ≥ 9 were considered to have a low risk of bias.

Participants.
Across all the included studies (taking into account the 3 studies on the same RCT [16][17][18]), the total sample size was = 1,055 (476 males, 569 females, 1 study with 10 participants did not specify sex [26]; mean age = 70.9, SD = 6.8 years), with individual studies ranging from 8 to 179 participants, and 3 studies with less than 20 participants [21,23,26]. Sample size was determined with a priori power calculation in 3 studies [15,18,20], all of which achieved target sample size.

Intervention Efficacy in Low Risk of Bias Studies.
Four of the 21 studies were reported particularly well and demonstrated a low risk of bias (scoring ≥ 9) [15][16][17][18]. Three of those studies were reporting on findings from the same randomised, double-blind, placebo-controlled trial (VITACOG) [16][17][18] investigating the effects of 2 years of high dose vitamin B treatment for people with MCI, and the other was a randomised, double-blind, placebo-controlled 24-week international multisite clinical trial [15] on Souvenaid5 for AD. All 4 studies incorporated a relatively comprehensive neuropsychological test battery, rather than just a simple global measure of cognition (e.g., ADAS-cog, MMSE). One of those studies reported on EEG network connectivity [15], and the other 3 reported structural MRI: regional grey matter volume [16] and whole brain atrophy [17,18]. One study found a reduction in EEG beta network integrity in the placebo, but not the intervention group, indicating counteraction of network decline after 24 weeks of 125 mL/day Souvenaid in people with AD [15]. The other 3 studies showed a reduction in regional grey matter and whole brain atrophy after 2 years treatment with high dose vitamin B (0.8 mg/day folic acid, 20 mg/day vitamin B6, 0.5 mg/day vitamin B12) for people with MCI, compared to placebo [16][17][18].
Three of the 4 studies reported associations between cognitive test scores and neuroimaging outcome measures [15,16,18]. In 1 study, an association between EEG beta activity and memory performance ( -score across NTB; see Table 3) was reported at midpoint in the Souvenaid group only [15]. An association between rate of atrophy and both final MMSE scores and baseline Telephone Interview of Cognitive Status-Modified (TICS-M) scores was reported in one of the high dose vitamin B studies [18]. There was also an association between increased grey matter loss and lower MMSE and CDR-SOB scores and poorer delayed recall and category fluency performance [16] in another of the vitamin B studies.
One MRI study showed significantly increased whole brain volume after 26 weeks of the target multimodal intervention (see Table 3 for details) and reduced volume for the control group [22], and another study showed no difference in whole brain volume between treatment (52 weeks of 1 g/day citicoline) and placebo [14]. One fMRI study reported both increased BOLD response in the right putamen and reduced BOLD in the right middle temporal gyrus when participants completed an episodic memory encoding task after 1.2 g/day Bushen for 12 weeks [31].
One CBF study reported an increase in white matter CBF with stable xenon CT after 12 weeks of a traditional Chinese medicine [26], and one TCD CBF study reported increased blood flow velocity to the middle and anterior cerebral arteries after 12 weeks of 19.2 mg/day EGb 761 standardised ginkgo extract and 75 mg/day aspirin [33].

Efficacy on Cognition across All Studies.
Across all studies, 13 reported positive effects on cognition [13, 19-21, 23-26, 29-33], 4 studies reported negative results [14,22,27,28], and 4 did not report on cognition findings alone [15][16][17][18]. As detailed above in Section 3.4, the key patterns of results for the commonly used neuropsychological tests are detailed below, with further information available in Table 3. Two studies [13,24] reported improvements (a reduction) in ADAS-cog scores in the intervention group (12 weeks of 10 g/day fuzhisan [13]; 4 weeks of 10 mL/day Cerebrolysin [24]), and 1 in the control group (26 weeks of 10 mg/day donepezil [27]), and another showed a significant deterioration in ADAScog scores following treatment in the CM arm (26 weeks of 2000 IU/day Vitamin E) but noted improvements in the other two parallel arms (5 mg/day donepezil and 1.5 mg/day rivastigmine) [28]. 16

Associations between Neuroimaging and Cognitive Measures.
Six of 21 studies reported associations between measures of cognition and neuroimaging markers [15-18, 31, 32]. One study showed a relationship between activation in the right putamen during an episodic memory task and Stroop performance, and reduced middle temporal gyrus deactivation with AVLT performance [31]. Another study showed that greater posterior cingulate cortex deactivation was associated with improved MMSE and digit span scores [32].
Four studies did not report neuropsychological test battery findings alone as they had already been published previously [15][16][17][18]. For example, one of those studies reported an association between memory performance and EEG beta band activity at the midpoint of the 24-week Souvenaid trial (125 mL/day) [15]. Please see Table 3 for more detailed information on studies reporting associations between clinical and neuroimaging findings.

Participants.
The majority of studies reported information on how a diagnosis of MCI or dementia was made or confirmed and included sufficient inclusion and exclusion criteria to allow replication [13-18, 20-25, 27, 31-33]; one study did not detail cognitive status of the control group [30]. Important demographic information, such as years of education, a factor known to significantly increase the risk of dementia [37], was missing from other studies [19,28]. In order to meaningfully assess the efficacy of an intervention, it is essential that the tested cohort is as homogeneous as possible. This can be done by closely following guidelines stipulating the most up-to-date diagnostic criteria for MCI [38], dementia (relative to the type; e.g., McKhann et al. [39]), and subjective cognitive complaints [40], and by carefully recording and reporting all relevant participant demographics and baseline characteristics. Care must also be taken to match participant characteristics between active and control groups, with one study not detailing information on the cognitive status of the control group, making comparison impossible [30].

Study Setting.
The majority of studies recruited from memory clinics [14,15,22,29], hospitals [13,23,33], the community [16][17][18], or a combination of those settings [31,32]. However, 5 studies did not report the recruitment setting [19,24,[26][27][28]. The recruitment setting for dementia studies has been shown to dramatically influence the participant characteristics and health outcomes. For example, participants with MCI recruited from a memory clinic have been shown to have an annual conversion rate to dementia that is 10% higher than participants recruited from the community [41]. Thus, future work in this field should ensure that the recruitment setting is carefully considered in study design and reported adequately in the published results.

Intervention
. The majority of studies tested a Chinese herbal medicine [13,19,21,23,26,[30][31][32][33] or a vitamin [16-18, 25, 27, 28] intervention, with most using a tablet or capsule for oral administration [14, 16-22, 27, 28, 31-33]. Only just over half the studies reported enough detail for the intervention to be replicated [13-18, 20, 23, 25, 27, 28, 33]. The main difficulty here was that, for herbal medicines, standardisation did not occur [26,30], or the details were not supplied. For the latter, this included missing information on the particular standardised formula used (e.g., EGb 761 for Ginkgo biloba), missing information on dose or dosing regimen, and/or inadequate information on commercially available extracts (e.g., brand/manufacturer) [19,21,22,26,[29][30][31][32]. Quality control and quality assurance (Good Manufacturing Practice [GMP]) is required for psychopharmacological research, and the absence of complete information on intervention formulation makes results near impossible to replicate. This problem is further compounded when multi-herb formulas are used, as a greater degree of preclinical work is required to develop standard operating procedures (SOPs) for extraction methods, and to optimise ratios of individual constituents. It should also be noted that treatment duration varied (1) Study Characteristics: (a) Research needs to follow the most recent guidelines stipulating diagnostic criteria for subjective cognitive complaints, MCI, and dementia, and should be closely adhered to when formulating protocols to ensure that the study population is as homogeneous as possible.
(b) Essential baseline characteristics, particularly ones known to increase the risk of dementia, should be reported.
(c) Recruitment setting needs to be carefully considered and always reported. (2) Methodologies: (a) All information on neuroimaging data collection, pre-and post-processing pipelines, and quantification needs to be reported to ensure that the results can be adequately scrutinised and replicated.
(b) Optimal analytic techniques should be utilised for the quantification of neuroimaging data.
(c) Standardised neuropsychological tests that are appropriate clinical trial endpoints for the level of cognitive impairment should be used.
(3) Intervention Efficacy: (a) Standardised herbal extracts should be used to reduce the variability between studies.
(b) Multi-herb formulas require substantial preclinical research to optimise ratios of active components, and determine their efficacy and safety as a formula.
(c) Dosage should be kept similar to other research (unless there is a rationale for adjusting dose) to reduce variability between studies.
(d) The length of trials should be carefully determined and have a rationale to allow for greater comparability between studies.
(e) An appropriate control group, such as a placebo matched to colour, shape, taste and smell of the active treatment, should always be included. substantially between studies from 4 weeks [24,25] to 2 years [16][17][18], adding further complexity to comparisons between studies.

Study Design.
Although the majority of studies included a control group [13-20, 22, 24, 27-33], four studies did not [21,23,25,26], rendering a high risk of bias. A control group, such as a placebo, should always be incorporated to establish whether a true relationship between the treatment and outcome actually exists. In the context of herbal medicine research, appropriate placebos are often difficult to establish because they need to match the active treatment on taste, smell, look, and feel. Herbal medicines can be pungent and have a distinctive taste so additional care needs to be taken when matching to a placebo [42].

Structural and Functional Neuroimaging Methods.
Most studies incorporated functional neuroimaging methods [13, 15, 19-21, 23-28, 30-33], largely EEG [15, 19-21, 24, 26-30]. There were large differences in the tasks and analytic methods described in these studies, but the majority of EEG papers assessed auditory oddball P300 ERP component amplitudes and latencies [20,[26][27][28][29][30]. The P300 has been widely explored in ERP literature and has been associated with a range of cognitive processes including memory [43], the orienting of attention [44], decision-making [45], and expectancy [46,47]. The studies assessing P300 in this review largely reported baseline-to-peak quantification methods (when quantification was described at all), despite this being an ineffective approach for disentangling the multiple subcomponents comprised within the monolithic P300 peak (i.e., P3a, P3b, Novelty P3, and Slow Wave) that represent a range of cognitive processes [48]. Given that effect sizes from CMs can be small [49], and that interventions may affect various cognitive domains, it is imperative that optimal analytic methods are employed to maximise the chance of detecting an effect. Alternative component quantification methods, such as Principal Components Analysis (PCA), should be adopted for future CM ERP studies [50].
Neuroimaging data acquisition, pre-and postprocessing pipelines, and analyses were adequately reported in only 12 of 21 studies [13-18, 22, 23, 25, 28, 31, 32]. There was insufficient information on how the data were collected (e.g., recording parameters, task details including length of resting state condition, and stimulus delivery) [19-21, 24, 26, 30, 33], inadequate reporting of pre-and postprocessing techniques that are in line with widely accepted best practice (e.g., artefact rejection) [21,29], and missing data quantification details (e.g., Fast Fourier Transformation [FFT] parameters, quantification of P300) [24,26]. Given the potential limitations of some neuroimaging techniques (as outlined in Table 1), it is imperative that future work describes all data acquisition, processing, and analytic techniques to ensure that variability in results between studies can be adequately accounted for.
Although the majority of studies reported positive results [13,[15][16][17][18][19][20][21][22][23][24][25][26][29][30][31][32][33], as noted above, the quality of reporting in most of these studies was relatively poor, indicating a high risk of bias. The results and conclusions from those studies should be viewed with a degree of caution. Given that functional neuroimaging methods are often more sensitive than standard pen-and-paper tests, it is even more important that high quality data, analyses, and interpretations are reported.

Measures of Cognition.
The majority of studies utilised the MMSE [13-18, 20-28, 30-33], ADAS-cog [13,24,27,28], and tested verbal learning [14-18, 22, 31, 32]. Similar to the neuroimaging results, most studies reported positive effects on cognition [13, 19-21, 23-26, 30-33], even though the risk of bias assessment indicated that only 13 studies used appropriate outcome measures [13-18, 20, 22-24, 28, 31, 32]. For example, it has been argued that the MMSE is not appropriate for cognitive assessments in people with MCI due to its low sensitivity (18%) in that cohort [51]. However, all but 1 [17] of the 7 MCI studies included here reported MMSE scores. These shortcomings make it challenging to meaningfully interpret the efficacy on cognition of the CMs reviewed here. The 4 studies that scored a low risk of bias utilised comprehensive neuropsychological test batteries [15][16][17][18] and did not report on the efficacy of these cognitive outcome measures as they had already been reported previously when the complete results of those RCTs were published elsewhere. Future work should also utilise a comprehensive neuropsychological test battery and use outcome measures that are appropriate clinical trial endpoints for the level of cognitive impairment of the target cohort [52].

Study Quality and Risk of Bias.
The majority of studies assessed in this systematic review were at high risk of bias [13,14,[19][20][21][22][23][24][25][26][27][28][29][30][31][32][33]. One of the most common (and significant) issues was that a power calculation was not reported in the majority of studies (Table 4). Most studies had a relatively small sample size and were consequently at risk of Type II error (false negative). The 3 studies that did conduct a power calculation all achieved their recruitment target [15,18,20]. Bias also came from a lack of reporting on how randomisation and allocation concealment were carried out. Most studies were randomised trials [13-18, 20, 24, 27, 28, 31-33]; however, only a small number of these actually reported on the randomisation procedure [15][16][17][18]28] and an even smaller number on how allocation was concealed [15][16][17][18]. Randomisation allows for the distribution of participant characteristics to be left to chance. Without adequate randomisation, it cannot be assumed that the null hypothesis (that participant groups have been drawn from the same population) is true [53]; this jeopardises internal validity. In relation to allocation concealment, given that most studies utilised an oral intervention, there is no reason that similar future work should not report how allocation was concealed and who was blinded. It must be acknowledged that this is not always the case in some physical activity interventions [22], where allocation concealment can be challenging. A further source of bias came from the lack of reporting of adverse events, which was done by only 12 studies [15-18, 20-23, 26, 27, 31, 32]. Future work should always report adverse events that may have been due to the intervention as it ensures the safety of participants.

Intervention Efficacy.
The focus of the 4 high quality studies that scored a low risk of bias [15][16][17][18] was to report detailed analyses of neuroimaging secondary outcome measures. Of those four studies, 3 reported that 2-year treatment for MCI with high dose vitamin B (0.8 mg/day folic acid, 20 mg/day vitamin B6, and 0.5 mg/day vitamin B12) reduced whole brain and regional grey matter atrophy, compared to placebo [16][17][18], and 1 found that 24 weeks of 125 mL/day Souvenaid maintained EEG beta network integrity in people with AD, where this declined in the placebo group [15].
Three of those studies also reported an association between cognitive test scores and neuroimaging outcome measures [15,16,18]. It was found that lower MMSE, CDR-SOB, delayed recall, and category fluency scores were associated with accelerated grey matter loss in one of the high dose vitamin B studies [16]. Baseline TICS-M and final MMSE scores were associated with rate of atrophy in another high dose vitamin B study [18], and midpoint memory performance was associated with beta activity in the Souvenaid study [15]. In terms of clinical use, the above studies indicate that 2 years of high dose vitamin B or 6 months of 125 mL/day Souvenaid have potential clinical utility as an adjunct therapy for people with MCI or Alzheimer's disease, respectively.

4.5.
Recommendations. This systematic review has identified a number of consistent shortcomings in CM neuroimaging research into cognitive decline. In an effort to improve the rigour and validity of this important and developing field, the authors suggest 11 key recommendations emerging from the 3 review aims that future work should adhere to. These are detailed in Box 1.

Strengths and Limitations.
This systematic review focused on studies reporting a chronic intervention only. Acute studies may necessarily utilise a different range of neuroimaging methods than those reported here. For example, structural MRI is not appropriate for acute treatment administration as structural brain changes take longer than a few hours to be detected. Future research should systematically summarise and critically appraise acute CM studies [54,55] to provide a more comprehensive overview of the field. Furthermore, the heterogeneity of the interventions and neuroimaging techniques employed made meta-analyses impossible here. Future work (with a different aim) could consider focusing on only one intervention or neuroimaging modality in order to quantify efficacy. It should also be noted that the authors of included studies were not contacted by the authors of this review.
This review not only focused on efficacy but also on summarising the characteristics of studies, intervention efficacy, and methods utilised. Particular consideration was given to identifying risks of bias. Neuroimaging and CM are a rapidly evolving area of research; thus the findings reported here highlight a number of significant strengths and weaknesses in this field that can be addressed in future work in an effort to improve the evidence base.

4.7.
Conclusions. This systematic review summarised and critically appraised CM research on people with cognitive decline, MCI, or dementia that incorporated neuroimaging as an outcome measure. It was found that most studies focused on people with AD, utilised a herbal medicine intervention that was on average 12 weeks long, and used EEG or structural MRI as neuroimaging outcome measures. Nearly all studies reported positive results, despite the majority having a high risk of bias. The most common issues were a lack of reporting on randomisation, allocation concealment, blinding, and the lack of a power calculation. Eleven recommendations to improve future neuroimaging CM research on people with MCI and dementia have been highlighted in the recommendations box. The authors hope that the pragmatic approach taken to this systematic review will lead to an uptake of these recommendations and a subsequent increase in the quality of CM neuroimaging research on people with MCI or dementia.