Hepatic encephalopathy (HE) is a neuropsychiatric syndrome, which is featured by a range of clinical manifestations, such as psychometric changes, numbness, and coma. Minimal HE (MHE) is a subset of HE in which symptoms are undetectable by routine examination and can only be diagnosed by neuropsychological measurements and neurophysiological tests (Tao et al., 2013). As reported, MHE can reduce the quality of life and the working ability (Bajaj et al., 2009), and MHE is prone to progress to overt HE (OHE) without proper treatment. Patients with cirrhosis with OHE usually develop poor prognosis and a loss of learning ability (Umapathy et al., 2014). Thus, even patients recovering from the disease can still have abnormal neuropsychiatric function and high recurrence rates (Sharma et al., 2010).

The voxel-based morphometry (VBM) method has many advantages over the region of interest (ROI) analysis ability (Pereira et al., 2010), such as automation, assumption-free, operator independence, and whole-brain gray matter abnormalities. With VBM, MRI morphometry studies have focused on structural abnormalities of gray matter and white matter. However, the structures to be evaluated need to be determined in advance, because of the major limitations of ROI-based brain morphological change measurement techniques (Busatto et al., 2008). Because of the different sample sizes of participants and different research methods, the results of these studies were contradicted (Costafreda et al., 2009). In comparison, VBM is a more accurate method than the manual volume approach and can overcome the limitations of the ROI approach. It is important to confirm unanimous results of VBM studies on gray matter volume (GMV) in those with HE based on a meta-analysis (Wang et al., 2015).

At present, several VBM studies of GMV in HE have been published. However, the results are diverse. For instance, one study found that regional gray matter atrophy was mostly confined in bilateral frontal regions, the bilateral temporal pole, and the cerebellum (García-García et al., 2017), while another study suggested that the loss of gray matter was observed in the precuneus, the bilateral insular cortex, and the caudate nucleus (Chen et al., 2012).

Seed-based d mapping (SDM) is a statistical software for meta-analysis, using neuroimaging technology to detect differences in brain activity. The SDM, combining positive and negative differences, has proven to be better than methods such as the activation likelihood estimation and the multilevel kernel density analysis (Radua et al., 2012). In addition, to assess the robustness and heterogeneity of the results, SDM enables several complementary analyses, such as jack-knife and subgroup analysis (Radua et al., 2012). However, SDM has not been applied to a meta-analysis comparing patients with HE with normal controls. In this study, we have used SDM to evaluate the published VBM studies on patients with HE and the control group for identifying consistent regional gray matter abnormalities.

To the best of our knowledge, this is the first study to analyze the changes in brain GMV of patients with HE with SDM software and three subgroup analyses. Both the comprehensive analysis and the subgroup analysis have confirmed that patients with HE had extensive symmetry changes in gray matter. Our research may help to understand the potential neurodegenerative process of HE. An important finding of subgroup analysis is that, in patients with different degrees of HE, the changing area of brain GMV is similar, and the range of symptoms increases with the aggravation.

The current voxel meta-analysis mainly showed that GMV increased in the bilateral thalamus, the bilateral hippocampus, and the bilateral lingual gyrus extending to the calcarine fissure and the precuneus. In addition, the GMV decreased in the bilateral insula, the basal ganglia, the anterior cingulate gyrus extending to the superior frontal gyrus, and the cerebellum. The results of the sensitivity analysis remained largely unchanged and the results were robust and highly reproducible.

Several neuroimaging findings suggested that changes in the cortico-striato-thalamic pathway could play a crucial role in HE (Qi et al., 2012, 2013b,2015). As information from the cortex is returned to the thalamus indirectly or directly through the striatum—white matter system, the thalamus may act as a filter for sensory input (Nakajima et al., 2019). The basal ganglia are considered to be playing an important role in the pathophysiology of HE (Qi et al., 2013a). In previous studies, basal ganglia indicated functional disconnection with other brain regions and abnormalities in metabolism (Sharma et al., 2010; Chen et al., 2012), which resulted in the disinhibition of the thalamic activity and inputs more inhibitory information into the cortex, leading to neurocognitive dysfunction (Grillner and Robertson, 2015; He et al., 2020). In addition, manganese is known to deposit within intracranial structures, mainly the globus pallidus and the pituitary gland (Dietemann et al., 1998; Rovira et al., 2008). Previous studies have shown that manganese deposition is also one of the reasons for the loss of basal ganglia neurons (Kano and Morishima, 2021). Moreover, manganese can colocalize in such intracranial structures with other elements, such as iron or gadolinium. These elements showed cellular toxicity in neurons, which may also be the reason for the reduction of GMV in basal ganglia (Mallio et al., 2020). An increase in thalamic volume may indicate hypertrophy or hyperplasia of neurons or glia. It is speculated that the increase of thalamic volume is a compensatory effect of basal ganglia dysfunction, but the increase of thalamic volume cannot completely improve brain function (He et al., 2020).

The results showed that patients with HE also suffered from obvious increase in some brain regions, namely, the bilateral precuneus, the lingual gyrus, and the calcarine fissure and its surrounding cortex. Notably, these areas were linked to the visual functions (Rao et al., 2007; Rita Machado et al., 2017), which are relative to impaired visual information processing in patients with cirrhosis. The increase in the cortical thickness observed in patients with cirrhosis seems to contradict previous works of research (Montoliu et al., 2012; Wernberg et al., 2019). The reason for such inconsistency remains unclear. However, a previous study’s results were similar to ours, which speculated that the increase of cortical thickness was related to mild brain edema in patients with HE (Wu et al., 2015). With the progression of the disease, the scope of the increase in the volume of related brain regions is reduced. We speculated that the progressive atrophy of the cortex caused by dysfunction weakens the increase in the GMV. The related neuropathological mechanism is not clear and is worthy of further study.

Our study found that, in comparison with the healthy control group, patients with HE had wider GMV reduction areas with the progress of the disease. These findings have been shown in previous studies (Garcia-Martinez et al., 2011; Lin et al., 2019). The bilateral insula is the most significant area of GMV reduction, the decreased insula volume has been revealed in HE, which correlates with cognitive alterations (Chen et al., 2017; García-García et al., 2017). It is found that the resting-state neuronal activity represents abnormal regional uniformity in the insula of patients with HE (Ni et al., 2012; Chen et al., 2016). The insula integrates internal and external stimuli and plays an important role in episodic memory processing. The anterior cingulate cortex is a significant region for attentional control, response suppression, and error detection (Kerns et al., 2004). The anterior cingulate cortex and the insula participate in the detection and the location of external stimuli and internal events (Flook et al., 2020) and is also a key brain region of the default mode network (DMN; Zhang et al., 2013; Wertz et al., 2020). Studies have shown that DMN has high metabolic activity at rest. Therefore, decreased gray matter voxel of the anterior cingulate cortex may be the neurobiological basis of cognitive impairment. In addition, we also noted that, without HE, the volume of the anterior cingulate cortex region did not change in patients with cirrhosis, which maybe because of the early stage of the disease.

The cerebellum is a regulator of motor function, but a link between cerebellum changes and impaired executive function has been established (Kühn et al., 2012). Our results showed that the area of reduced cerebellar volume increased significantly with the progression of the disease. Studies have shown that the cerebellum was more susceptible to the harmful effects of hyperammonemia than the cerebral cortex (Rodrigo et al., 2010). Our study also found that the volume of the bilateral striatum and the thalamus changed with the increase of plasma ammonia, which could be caused by an increased blood flow of related brain regions in patients with cirrhosis (Felipo et al., 2014). Cerebral blood flow in patients with HE has regionally selective changes, such as decreased blood flow to cortical regions but significantly increased blood flow to the basal ganglia, the cerebellum, and the thalamic structures (Butterworth, 2019). These regions are more susceptible to plasma ammonia. New animal models using ammonia as a precipitation factor also confirmed cerebellar neuron loss (García-Lezana et al., 2017). Gray matter loss reflects ammonia-related neurological changes.

The discovery and measurement of HE are challenging in clinical practice. Nowadays, artificial intelligence and machine learning have been increasingly applied to disease diagnosis, especially medical image recognition (Supekar et al., 2022). We reviewed previous articles and obtained a relatively unified result of GMV change in patients with HE, which provides a basis for delineating ROIs in machine learning. EEG is a common tool for monitoring HE (Amodio and Montagnese, 2015). However, EEG signals may be damaged by background noise, which leads to inconsistent diagnostic accuracy (Mutanen et al., 2018). Machine learning algorithms can analyze records containing a large number of variables and can find complex linear or non-linear relationships between variables (Gazda et al., 2021). Machine learning and deep learning methods have been used to analyze EEG to predict various chronic psychiatric illnesses (Natu et al., 2022; Paul et al., 2022). Using artificial intelligence and machine learning to comprehensively analyze MRI and EEG to diagnose HE more accurately may be the direction of future research.

This study still has some limitations. First, due to the sample size, the power of the results of this study may be limited, which is similar to the early voxel-based meta-analysis. Second, there are still some deviations in the VBM method, especially in areas with large anatomical variation; it may excessively show population differences, which is due to its relative insensitivity to more spatial changes (Bookstein, 2001; Tisserand et al., 2002). Another issue is that the meta-analysis is related to coordinates in published studies, leading to less accurate results (Zhou et al., 2014). Moreover, only 9 studies met the inclusion criteria for the quantitative analysis in our research. A small number of studies would reduce the credibility of the results, especially the subgroup analysis should be interpreted with caution as they were driven by less than 10 studies. In addition, the lack of a protocol with a priori methods, the inclusion of English language papers only, the absence of searches for gray literature, and the use of a non-validated quality assessment checklist may have introduced a range of biases.

Based on the available imaging works of literature, this meta-analysis revealed widespread and significant changes in GMV in patients with HE. The subgroup analysis suggested that, in patients with different degrees of HE, the changing area of GMV was similar and the range of symptoms increased with aggravation. The change of GMV in patients with HE was significantly related to plasma ammonia.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

JL, JW, and CZ: conceptualization, methodology, and project administration. LZ, WZ, and LC: investigation. TS and YR: resources and data curation. LZ and YC: writing—original. BS and LC: draft, formal analysis, and visualization. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.