Cross-sectional results on CI-related GM changes in PD patients remain heterogeneous, and three meta-analyses were conducted to identify the most consistent GM alterations associated with PD-CI (Xu et al., 2016; Mihaescu et al., 2019; Zheng et al., 2019). A meta-analysis study in 2016, using five studies compared 92 PD-MCI patients with 192 PD without CI (PD-NCI) patients, and from ten studies, compared 168 PDD patients with 233 PD without dementia (PD-ND) patients. The authors have reported significantly reduced GM volume in the left superior frontal gyrus (SFG), left superior temporal lobe (STL), and left insula in the PD-MCI group, and in the left SFG, and bilateral STL extending to the hippocampus in the PDD group (Xu et al., 2016). A meta-analysis study in 2019 compared 405 PD-MCI patients with 559 PD-NCI patients from fifteen studies, and compared 178 PDD patients with 278 PD-ND patients from eleven studies. A significantly higher GM atrophy in bilateral dorsolateral prefrontal cortex, left angular gyrus, right supramarginal gyrus, left insula, and midcingulate cortex in the PD-MCI group, and bilateral insula and right hippocampus in the PDD group was reported (Mihaescu et al., 2019). Another meta-analysis study in 2019 consisting of 504 PD-MCI patients and 554 PD-NCI patients from twenty studies reported a drastically remarkable GM atrophy in the left anterior insula extending to the inferior frontal gyrus, and orbital part in the PD-MCI group (Zheng et al., 2019). GM atrophy in the insula and frontal-limbic-temporal regions are robust features of CI in PD, and unilateral-to-bilateral development of structural atrophy is a potential indicator for progression of PD-MCI to PDD. In particular, several studies attempted to analyze GM changes across various stages of cognitive decline by comparing a prospective sample of PD-NCI, PD-MCI, PDD patients, and healthy subjects (HCs). These typically report a pattern of linear and progressive cortical changes (reduced GM volume or cortical thinning) with increasing levels of CI (Song et al., 2011; Melzer et al., 2012; Pagonabarraga et al., 2013).

Notably, several structural MRI studies focused on assessing the differences in subcortical GM structures or ROIs crucial in the cognitive functions of PD patients. The volume of the hippocampus and substantia innominata (SI), a major source of cholinergic input to the cerebral cortex, progressively decreases corresponding to the cognitive status (Weintraub et al., 2011; Choi et al., 2012). Recent advances in the imaging technology and software analysis tools have made the study of hippocampal subregions in vivo a possibility (Iglesias et al., 2015). Smaller cornu ammonis 1 (CA1) and hippocampal-amygdaloid transition area (HATA) volumes have been observed in PD-MCI patients relative to the PD-NCI patients (Becker et al., 2021). In addition, as compared with the PD-NCI group, the PD-CI group exclusively exhibits atrophy in the entorhinal cortex (ENT) (Goldman et al., 2012; Jia et al., 2019), amygdala (Schulz et al., 2018), thalamus (Mak et al., 2014; Chen et al., 2016; Schulz et al., 2018), accumbens nucleus (Mak et al., 2014), and nucleus basalis of Meynert (Schulz et al., 2018).

Recent longitudinal studies described the progression patterns of GM changes in PD patients with CI, which can facilitate a better understanding of the mechanism of CI in PD. Patients with PD-MCI show increased global atrophy within 18 months relative to both PD-NCI patients and HCs (Mak et al., 2017). Other studies have also attempted to investigate the longitudinal changes in the cortical and subcortical GM and reported a higher rate of cortical thinning in the frontal/supplementary motor area, parietal-temporal, occipital cortices and a significant decrease in the volume of the amygdala and accumbens nucleus in PD-MCI patients relative to both PD-NCI patients and HCs (Hanganu et al., 2014; Mak et al., 2015). Some longitudinal studies have examined the differential progressive GM loss in the specific cortical regions. A 4-year follow-up study reported that over time, as compared to the PD-NCI patients, those with PD-MCI had significantly greater progression of cortical thinning in the posterior regions of the brain related to visuospatial and visuoperceptual (VS/VP) changes (Garcia-Diaz et al., 2018). Another 4-year follow-up study held the opinion that the hippocampal subfields have different vulnerabilities to the degenerative processes related to aging as well as PD. PD patients and HCs have similar longitudinal changes in CA1, while both PD-NCI and PD-MCI patients have a more severe decline in the anterior and posterior hippocampal segments, implying its predictive value for assessing memory dysfunction (Uribe et al., 2018).

Researchers have examined structural alterations in the brain to predict cognitive decline in PD patients. PD patients with normal cognition at baseline and who develop PD-MCI during follow-up are classified as “PD-MCI converters,” while those who remain cognitively asymptomatic during the follow-up are classified as “non-converters.” Relative to the non-converters, PD-MCI converters have a significant reduction in GM volume of temporal regions and the amygdala/hippocampus structural covariance network at baseline, along with evidently progressive frontal lobe atrophy and longitudinal white matter (WM) volumetric reduction (Wen et al., 2015; Zhou et al., 2020; Chen et al., 2021). Some studies have further investigated the pattern and progression of cortical thickness and subcortical volumes in PD-MCI converters, and reported a higher degree of anterior cingulate cortex, parietal, temporal, and occipital cortices thinning, atrophy of accumbens nucleus at baseline, and progressive atrophy in the thalamus, caudate nucleus, accumbens nucleus, and CA2-3 of hippocampal subregions (Mak et al., 2015; Foo et al., 2017a,b; Filippi et al., 2020; Gorges et al., 2020).

Non-demented PD patients who progress to PDD during follow-up are classified as “PDD converters.” At baseline, relative to the non-converters, PDD converters have lower GM volume in uncus, global hippocampus, and hippocampal subfields (presubiculum, subiculum, and CA1) (Gee et al., 2017; Low et al., 2019), lesser thicknesses of the precentral gyrus and superior frontal gyrus (RSFG) and anterior cingulate (Compta et al., 2013), and larger ventricular volume (Camicioli et al., 2011). Moreover, over time, PDD converters have higher GM atrophy in the global hippocampus, hippocampal subfields [parasubiculum, presubiculum, granule cell layer of the dentate gyrus (GC-DG), and fimbria] and parahippocampal gyrus (Gee et al., 2017; Low et al., 2019). Another study classified PD-MCI patients as PDD converters or PD-MCI non-converters based on whether they are subsequently diagnosed with PDD during follow-up. Relative to PD-MCI non-converters, the PDD converters were found to have lower GM density in prefrontal areas, insular and caudate nucleus, smaller normalized SI volume, and a lesser cortical thickness extending from the posterior cortical area into the frontal region, and the frontotemporal cortices (Lee et al., 2014; Chung et al., 2019; Filippi et al., 2020).

Baseline volumes of global WM, global hippocampus, and hippocampal subregions (parasubiculum, CA4, GC-DG, and HATA), thalamus, and accumbens nucleus are significant neuroimaging predictors for determining the conversion from PD-NCI to PD-MCI; the baseline global GM volume cannot significantly predict the conversion to PD-MCI (Kandiah et al., 2014; Wen et al., 2015; Foo et al., 2017a,b). Moreover, GM volume in the global hippocampus and hippocampal subregions (presubiculum, subiculum, and CA1), cortical thickness in the medial superior frontal/olfactory cortices, superior frontal/anterior cingulate and precentral regions are significant predictors for the progression to PDD (Compta et al., 2013; Kandiah et al., 2014; Chung et al., 2019; Low et al., 2019).

Growing evidence highlighted that brain regions undergoing neurodegeneration may overlap synergistically with the pathological processes for PD and Alzheimer’s disease (AD) (Masliah et al., 2001; Clinton et al., 2010). To determine its association with CI, a validated AD-spatial pattern of brain atrophy (spatial pattern of abnormality for recognition of early AD index, SPARE-AD) was quantified in PD patients with a range of cognitive abilities. The findings support the involvement of hippocampus and parietal-temporal cortex with CI in PD (Weintraub et al., 2012), and affirm the importance of AD biomarkers as effective predictors for CI in PD (Tropea et al., 2018). In addition, autopsies have identified degeneration of the cholinergic basal forebrain, especially in the nucleus basalis of Meynert, causing cortical cholinergic depletion, and eventually leading to the CI in patients with PD and AD (Liu et al., 2015). GM volume in the nucleus basalis of Meynert was a critical cholinergic biomarker to predict CI in PD-NCI patients (Ray et al., 2018; Schulz et al., 2018).

As a consequence of cerebral small-vessel disease (SVD), white matter hyperintensities (WMHs) in the brain can reflect pathological WM tissue ischemia (Erten-Lyons et al., 2013). PD patients with greater WMH have more severe cognitive deficits and a higher annual rate of change in global cognition (Dadar et al., 2018; Chahine et al., 2019; Chang et al., 2021). The periventricular WMH burden is significantly associated with PD-MCI and severe WMH with PDD; both are independent of cardiovascular risk factors (Huang et al., 2020; Tsai et al., 2021). In addition, heavier temporal WMH burden is associated with worse verbal memory (Chahine et al., 2019), and higher periventricular WMH burden correlates with greater decline in the executive and visuospatial functions (Huang et al., 2020). Therefore, WMH is strongly correlated with CI in PD and may serve as a surrogate marker for CI in PD (Kandiah et al., 2014; Fang et al., 2021). Visible perivascular space (PVS) has also been considered as an imaging biomarker for SVD. A longitudinal MRI-visible PVS study reported higher PVS severity in basal ganglia (BG) as a predictor of cognitive conversion of PD-NCI patients as well as for cognitive decline in PD-MCI patients (Park et al., 2019). Summary findings from other MRI techniques correlated with cognitive dysfunction in PD can be found in Table 3.

Neuromelanin-sensitive MRI (NM-MRI) is a promising technique that depicts characteristics of tissue containing neuromelanin, a dark polymer product of noradrenaline metabolism (Nakamura and Sugaya, 2014). The substantia nigra (SN) of the midbrain and the locus coeruleus (LC) within the pons contain large clusters of neuromelanin-producing cells, and the loss of these neuromelanin-containing cells within the two regions is a primary pathological diagnostic criterion for PD. The NM signal intensity of the LC in PD-MCI patients is significantly lower than that of the control subjects, and may serve as an imaging marker for assessing cognitive dysfunction in PD (Li et al., 2019; Prasuhn et al., 2021).

Quantitative susceptibility mapping (QSM) is another new MRI technique that enables the quantification of susceptibility-changing materials. Iron accumulation has been proposed as a significant pathomechanism for PD, which can be systematically examined using the QSM method. PD-MCI patients have significantly higher QSM values in the orbitofrontal cortex, cuneus, precuneus, fusiform gyrus, and head of the caudate nucleus relative to the PD-NCI patients (Uchida et al., 2019). Meanwhile, PDD patients have significantly higher QSM values in the left hippocampus as compared to non-demented PD patients (Li et al., 2018). These findings suggest the association between CI in PD and cerebral iron burden, which may be an auxiliary biomarker for early detection of CI in PD.

Diffusion tensor imaging (DTI) is an in vivo surrogate measure for the microstructural integrity of WM tracts. Using this technique, the rate of 3-dimensional water molecule diffusion through a pulsed-field gradient can be measured. The directionality of movement is quantified as fractional anisotropy (FA) and its total magnitude is measured as mean diffusivity (MD). The decrease in FA and increase in MD within the central nervous system have been proposed as sensitive imaging indicators for detecting early microstructural cortical damage in patients with neurodegenerative disorders.

The PD-MCI patients have a distributed pattern of WM abnormalities in the anterior and superior corona radiata, genu, and body of the corpus callosum, anterior inferior fronto-occipital, uncinate, and superior longitudinal fasciculi relative to PD-NCI patients and HCs (Agosta et al., 2014). Lower FA values in frontotemporoparietal regions, hippocampus, amygdala, thalamus, BG, and higher MD values in hippocampus, ENT, amygdala, thalamus, and nucleus basalis of Meynert relative to PD-NCI patients have been reported in PD-MCI patients (Galantucci et al., 2017; Schulz et al., 2018). PDD patients have significantly lower FA values in the prefrontal WM, hippocampus, and the genu of the corpus callosum relative to the non-demented PD patients (Kamagata et al., 2013; Chen et al., 2015), and lower FA values in both the anterior and the posterior cingulate fiber tracts as compared to HCs (Kamagata et al., 2012). Microstructural abnormalities in WM of PD patients with variant cognitive statuses (PD-NCI, PD-MCI, and PDD) have been investigated to elucidate the pathological processes of CI in PD. The loss of microstructural WM integrity is spatially restricted in PD-NCI patients, while these alterations increase with progressive cognitive deterioration (Hattori et al., 2012; Deng et al., 2013; Melzer et al., 2013).

In a previous study, dynamic MD changes were specifically assessed by a longitudinal approach in newly-diagnosed drug-naïve PD patients. A higher 1-year MD increase in frontal and occipital cortices was found in PD patients relative to HCs, which correlated with changes in the cognitive measures (Sampedro et al., 2019). Moreover, the increased MD values in the nucleus basalis of Meynert can predict the conversion from PD-NCI to PD with CI (Schulz et al., 2018). Thus, further longitudinal investigations are required to establish the clinical utility of DTI as a sensitive biomarker for the pathology of CI in PD.

Emerging evidence supported that cerebrovascular abnormalities including altered cerebral blood flow (CBF), cerebral blood volume (CBV), and blood–brain barrier permeability are linked with pathophysiology in CI. Using the arterial spin labeling (ASL) technique, brain perfusion can be quantitatively assessed in terms of CBF. A pattern of decreased CBF at the posterior part of the brain, mostly in the lateral and medial parietooccipital cortices, summarized as “posterior hypoperfusion,” has been reported in the PD-MCI group relative to HCs (Arslan et al., 2020), which may be a suitable early biomarker for CI in PD. Parietal CBF correlates to visuomotor skills and also may serve as a potential early biomarker for CI in PD (Pelizzari et al., 2020). Using the inflow-based vascular-space-occupancy (iVASO) MRI approach, potential arteriolar CBV (CBVa) abnormalities reflecting the homeostasis of the most actively regulated segment in the microvasculature can be calculated. A decreasing trend of CBVa in the substantia nigra, caudate nucleus and putamen, and an increasing trend of CBVa in the pre-supplementary motor area and intracalcarine gyrus have been reported in PD-CI patients relative to HCs (Paez et al., 2020). These exploratory findings provide important information for efforts toward developing potential biomarkers for the risk of CI in PD.

Resting-state functional MRI (rs-fMRI) can provide a reliable and reproducible tool to examine the functional activity in different brain networks by measuring intrinsic blood oxygen level-dependent (BOLD) low-frequency signal fluctuations. There is a growing interest in using fMRI to investigate the neural basis for CI in PD.

A voxel-based meta-analysis consisting of 289 PD-NCI, 222 PD-MCI, 68 PDD, and 353 HC subjects from seventeen studies showed that PD patients with CI had reduced functional connectivity (FC) in specific brain regions that are parts of the default mode network (DMN) (Wolters et al., 2019). Moreover, notably, increased FC may be the initial manifestation of altered brain function preceding cognitive deficits (transition from PD-NCI to PD-MCI), and these functional disruptions may be considerably associated with a cognitive decline in the progression of PD-related pathology (Gorges et al., 2015; Zhan et al., 2018; Zarifkar et al., 2021). A novel graph-theoretical analysis of functional networks indicates that the topological properties of brain networks are disrupted in PD-MCI patients with reduced long-range connections and enhanced local interconnectedness (Baggio et al., 2014). Moreover, a recent study suggested that PD-MCI patients had an increased random organization of brain networks and altered nodal centralities as compared with HCs (Hou et al., 2020). These results provide a new perspective on the functional basis of CI in PD.

In an exploratory longitudinal study, FC between the medial prefrontal cortex (mPFC) and posterior cingulate cortex (PCC) within DMN at baseline were found to be the best putative predictor for CI in PD (Zarifkar et al., 2021). PD patients with a longitudinal decline in cognitive function have a significant decline in dorsal attention-frontoparietal internetwork (DAN-FPN) strength, which may serve as a prognostic marker for assessing the development of dementia in PD (Campbell et al., 2020).

Proton magnetic resonance spectroscopy (¹H-MRS) owing to a high natural abundance and magnetic sensitivity of protons in the brain allows for in vivo measurements of four major metabolites, that reflect the integrity of different elements in the brain, including N-acetylaspartate (NAA), choline compounds (Cho), creatine (Cr), and myo-inositol (MI) (Nie et al., 2013).

The PD-CI patients have higher the choline to creatine (Cho/Cr) ratio in the occipital lobe, reduced concentration of NAA in the middle and superior frontal gray matter regions, and reduced concentration of total Cho and total Cr in the frontal white matter regions relative to PD-CN patients and HCs (Nie et al., 2013; Chaudhary et al., 2021). According to the differential status of CI, levels of NAA in the dorsolateral prefrontal cortex are reduced in the early stage with progression to the hippocampus in PDD patients (Pagonabarraga et al., 2012). Frontal neuro-metabolism may serve as a promising biomarker for characterizing cognitive deficits in patients with PD (Chaudhary et al., 2021).