All studies were included that met the following criteria: (1) studies that reported PSD using structural and functional MRI methods (PET, SPECT were also included); (2) swallowing screening and/or clinical swallowing assessment and/or instrumental assessment with clear signs of dysphagia or aspiration due to dysphagia; (3) there were no restrictions on language, stroke side or age of participants for inclusion in the study. All relevant studies published up to November 2022 were included.

A systematic search of the PubMed, Cochrane Library, MEDLINE, Embase, PsycINFO, Google Scholar, Web of Science, and CNKI databases was conducted. The search terms were (“stroke” or “ischemic stroke” or “hemorrhagic stroke”) and (“aspiration” or “abnormal swallowing” or “dysphagia”) and (“magnetic resonance imaging” or “lesion symptom mapping” or “functional magnetic resonance imaging” or “voxel-based image analysis”) and equivalent MeSH terms (see Supplementary Material 2 for the search process). Reference lists of selected studies and related reviews were also retrieved for more comprehensive inclusion. The study selection was carried out independently by Tang and Liu, disagreements were resolved through discussion and negotiation, and a third reviewer, Qiu, was added when they could not be resolved. A total of 340 studies were initially searched and screened, and a total of 30 studies met the inclusion criteria by evaluating the titles and abstracts. Read the full article for further evaluation, which includes a meta-analysis of fMRI of poststroke dysphagia by Liu et al. (2017). After the studies included in their analysis, a new functional ALE analysis was performed on this basis to increase the stability of the results. After including the studies in their analysis, 35 studies were finally included. This included 20 structural neuroimaging and 14 functional neuroimaging studies, and 1 (Mihai et al., 2016) reported both structural and functional neuroimaging (see Figure 1 for a flow chart of the review).

Data extraction for included studies was performed for author information, year of literature publication, number of participants, gender, age, stroke side, imaging method, time from onset to swallowing assessment, and peak coordinates of the Montreal Neurological Institute (MNI) or Talairach reported in the article. If articles reported peak coordinates from the whole brain MNI or Talairach, they were integrated together for the ALE meta-analysis, and studies that did not report coordinates or literature that reported coordinates but region of interest (ROI) studies were included in the narrative analysis. If two or more different tasks are used in the same article, the coordinates of both tasks are included in one dataset. Data extraction was performed independently by Tang and Liu and input into a form designed by the authors. Disagreements were resolved by negotiation through Qiu.

Activation likelihood estimation is a coordinate based quantitative analysis technique used in neuroimaging. The spatial probability distribution is computed at the center of a given coordinate, and ALE mappings are obtained according to the activation probability of each voxel, distinguishing true convergence of focus from random clustering of focus (i.e., noise) by testing the null hypothesis of random spatial correlation between experiments (Eickhoff et al., 2009, 2012, 2016; Moher et al., 2009; Müller et al., 2018). The ALE meta-analysis method can address the small sample size of a single study, sensitivity to differences between conditions, and limitations such as low reliability, by integrating multiple studies to derive consistency in the location of neuronal changes in the brain (Eickhoff et al., 2009). In this review, structural lesions and functional changes in relevant brain regions were analyzed using the ALE meta-analysis method. The peak coordinates of significant brain regions extracted from all included studies were recorded separately into the specified txt file using Ginger-ALE 3.0.2¹ (Eickhoff et al., 2009, 2012; Turkeltaub et al., 2012; Müller et al., 2018) software, and the MNI standard space was selected using Lancaster conversion to convert the Talairach coordinates to MIN coordinates. A cluster-level family wise error (FWE) based threshold was selected in reference to Eickhoff et al. (2016), Moher et al. (2009), and the width of the Gaussian probability distribution was determined separately for each experiment based on the peak coordinates of brain regions, based on empirical estimates of inter-subject variability and taking into account the number of subjects in each experiment. The threshold alignment was set to 1,000, the cluster-level FWE value was set to 0.05 and the peak p-value threshold was 0.001. Structural MRI ALE analysis was chosen to select larger masks to expand the focal points at the spatial brain boundaries. Functional MRI ALE analysis was chosen to select conservative small mask to limit the experimental effect of adjacent focal points. For the results of the ALE analysis, we examined the contribution of each cluster, and we defined the studies in which the ALE analysis provided coordinates to form clusters as the contributing clusters of studies. The results are shown selected for visualization using Mango 4.1² software, superimposed on the MNI template file at brainmap.org.³

As can be seen from Table 2 there were 21 studies included in the narrative analysis. The total sample size was 2,447, which included 1,546 patients with dysphagia and 901 controls.

The results of the ALE analysis showed that two clusters were associated with PSD compared to the control group (see Table 1 for contributing clusters of studies). The brain regions involved are the right lentiform nucleus, the right caudate nucleus and right thalamus (see Figure 2 and Table 3).

As can be seen from Tables 1, 4, a total of 15 studies were included in the narrative analysis of functional imaging MRI. The sample size of the included studies was 715 cases, of which 386 were patients with PSD and 329 were controls.

Activation likelihood estimation analysis was performed separately for all included studies reporting coordinates of hyperactivation and hypoactivation compared to controls. The results of the ALE analysis showed three clusters of significant hyperactivation compared to controls (see Figure 3A and Table 5). All three clusters had only one peak each and involved the brain regions of the right culmen of the anterior cerebellar lobe, left anterior cingulate gyrus, and right insula. There was one cluster with significant hypoactivation in the left thalamus (see Figure 3B and Table 5).

Based on the narrative and ALE analyses, it is clear that stroke patients with dysphagia are associated with structural lesions in the right lentiform nucleus, and structural lesions and functional abnormalities in the thalamus. The lentiform nucleus, a component of the basal ganglia, is an important relay nucleus of the inferior cortical conduction tract and plays an important role in maintaining normal tone of the swallowing muscles and in coordinating the swallowing muscle groups (Suntrup et al., 2012). Injury to the lentiform nucleus can lead to damage to the corticomedullary tract, cortico-neostriatum-thalamocortical and neostriatum- substantial nigra loops, resulting in impaired inhibition of the medulla oblongata (Wan et al., 2016). This eliminates the reflex relaxation of the cricopharyngeal sphincter and the sphincter exhibits sustained hyper-reflex contraction which in turn leads to dysphagia. Kim J. et al. (2022) confirmed cricopharyngeal dysfunction and correlated it with the severity of lesions in the lentiform nucleus. In addition, swallowing is a complex sensorimotor process in which the thalamus involves visual, auditory, olfactory, and gustatory sensors and serves as an ingestion center that can be involved in appetitive, endocrine regulation (Toogood et al., 2017). When food enters the mouth, olfactory, and visual responses combine with appetitive responses to activate the brainstem swallowing center (Jie, 2007). During swallowing, the cerebral cortex activates downstream fibers that reach the medulla oblongata swallowing center through the internal capsule and thalamus to produce swallowing and oral mastication movements. Therefore, the cortico-basal ganglia-thalamic circuit (BTC) plays an important role in swallowing, and disruption of this circuit by lesions in any of its components (e.g., structural lesions of the thalamus) may reduce the activation of the thalamus and lead to the development of dysphagia. In addition, related studies have identified thalamic lesions that cause impaired anterior hyoid movement (Wilmskoetter et al., 2019) or delayed pharyngeal swallowing (Wan et al., 2016) leading to dysphagia, and the thalamus may be associated with the prognosis of dysphagia (Galovic et al., 2017).

The cerebellum monitors the execution of swallowing and adapts to planned movements, effectively comparing the expected body movements with the actual behavior and adjusting the movement plan accordingly (Moon et al., 2022). In addition, the cerebellum plays a key role in ensuring accurate, fluid, and coordinated muscle activity (Roostaei et al., 2014) and can effectively coordinate muscle activity associated with swallowing. Several studies on swallowing function in healthy subjects have found various activation patterns in the superior cerebellar hemispheres related to swallowing function (Zald and Pardo, 1999; Suzuki et al., 2003; Malandraki et al., 2009; Grabski et al., 2012). The cerebellum can assist the cerebral cortex in information processing by increasing connections to the cortex and linking to the brainstem to form a cerebellar-cortical pathway (Amore et al., 2021). During voluntary swallowing, the cerebellum acts as a neural response enhancer, modulating or reinforcing swallowing cortical firing, exhibiting functional connectivity with the primary motor cortex, inferior frontal gyrus, basal ganglia, and thalamus (Jayasekeran et al., 2011). These rich functional connections may allow the cerebellum to complement cortical and brainstem control of swallowing after stroke, coordinate cortical and brainstem output, and participate in feedforward and feedback control (Sasegbon et al., 2020).

White matter connections including the corticomedullary tract and the white matter regions through which it passes are involved in motor conduction of swallowing. The corticomedullary tract is the neurotransmission pathway involved in swallowing (Galovic et al., 2017; Jo et al., 2017; Fandler et al., 2018), with fibers from the premotor and primary motor cortex crossing the corona radiata, internal capsule, and pons, and finally projecting to the bilateral cranial nuclei V, VII, IX, X, XI, and XII. Thus, the swallowing muscles are innervated by corticomedullary projections from both hemispheres (Hamdy et al., 1996). Any lesion in this pathway may lead to dysphagia (Jo et al., 2017; Moon et al., 2019; Emos and Rosner, 2022), and post-stroke damage to the corticomedullary tract interrupts the autonomic control of mastication and food mass transport (Cola et al., 2010; Jang et al., 2017), and the more severe the damage, the more difficult is the prognostic recovery of swallowing (Im et al., 2020; Jang et al., 2020).

Physiological studies have shown that the insula plays a role in regulating sensory and motor aspects of digestive tract function involving the oropharynx, esophagus and possibly other regions of the gastrointestinal tract (Binkofski et al., 1998; Kern et al., 1998; Martin et al., 2001) and autonomous oral motor control (Dronkers, 1996) and that the transition area between the inner surface of the frontal insula and the insula is the primary gustatory cortex (Kobayakawa et al., 1996) and that additional gustatory fields may be located in the insula, temporal lobe and the peduncle of the precentral and postcentral gyrus (Cerf et al., 1998; Faurion et al., 1998; Martin et al., 2001). In addition, the insula receives and projects to brain regions associated with swallowing [including somatosensory-motor cortex, premotor areas, supplementary motor areas, thalamus, anterior cingulate gyrus, and isolated nuclei of the brainstem] (Augustine, 1996; Daniels and Foundas, 1997; Leopold and Daniels, 2010; Wilmskoetter et al., 2020), so the insula is often considered the central hub of the “swallowing network” (Leopold and Daniels, 2010; Babaei et al., 2013; Yuan et al., 2015; Wilmskoetter et al., 2020). Dysphagia due to insula lesions may be physiologically disrupted and associated with disruption of sensory-motor integration (Mosier and Bereznaya, 2001), and there is support for the use of insula lesions in predicting dysphagia (Daniels and Foundas, 1999; Im et al., 2018; Hess et al., 2021). Thus, the insula may be a key region driving neuronal plasticity and effective recovery from dysphagia stroke (Galovic et al., 2017).

This review also suggests that functional abnormalities of the cingulate gyrus contribute to PSD, which is consistent with the study by Liu et al. (2017). The cingulate gyrus is involved in emotion and reward functions, as well as hunger and taste regulation (Babaei et al., 2010; Long et al., 2019), and is part of the default network, executive control network and the salience network. Therefore, the cingulate gyrus is not only involved in the execution and control of swallowing, but activation of the cingulate gyrus can modulate the negative emotions associated with dysphagia after stroke (Toogood et al., 2017). However, many scholars have also interpreted the involvement of the cingulate gyrus in terms of higher-order motor processing or attention (Hamdy et al., 1999a,b; Martin et al., 2004; Toogood et al., 2005; Malandraki et al., 2009). Toogood et al. (2017) found that brain activation in the anterior cingulate cortex was significantly greater during swallow preparation than during swallow execution, and activation within the bilateral insula and left dorsolateral central cortex was greater during swallow execution than during swallow preparation. Therefore, hyperactivation of the cingulate gyrus after stroke may contribute to an effective recovery of swallowing function (Li et al., 2009; Liu et al., 2017).

This review has several limitations. First, the patients included in this review were at different stages of stroke, and the neuropathic mechanisms of dysphagia in acute or subacute vs. chronic stroke may not be the same. Second, the types of stroke included in this review included both ischemic and hemorrhagic strokes, and the location of the lesion was not restricted to a particular brain region, so the neuropathic mechanisms may be different for different stroke types and lesion locations. Furthermore, the number of structural imaging and functional imaging studies included in the ALE analysis was relatively small. When the number of included studies is small, there is a greater likelihood that the data will be driven by a small number of studies and the robustness of the presented results will not be strong enough. We performed a Jackknife sensitivity analysis on the results of the ALE analysis to verify the replicability of the results. We also recognize that the results may change in the future as additional studies on PSD are published. Finally, because ALE meta-analyses can only be performed for studies with whole brain analyses because of the inherent bias in studies of regions of interest, yet the importance of small regions used as ROIs cannot be ignored, future studies need to summarize the main findings of ROI-based studies and acknowledge the possible contribution of these regions. In the future, as larger studies emerge, the inclusion of larger sample sizes for ALE analysis could yield more robust results or allow for subgroup analysis of time period of onset, stroke type, and lesion location; as well as analysis of studies with similar ROIs to provide more insight into the mechanistic contributions of these ROI regions.