Explore the language and the level of consciousness in “vegetative” patients

One recent study published in the journal Nature Neuroscience explored the ability of language processing and the residual consciousness in “vegetative” patients. This research is a collaborative work from Dr. WANG Liping’s lab of the Institute of Neuroscience (ION), Center for Excellence in Brain Science and Intelligent Technology (CEBSIT) of the Chinese Academy of Sciences (CAS), and from Drs. MAO Ying / WU Xuehai’s medical team of the Department of Neurosurgery, Huashan Hospital, Fudan University. In this study, researchers applied high-density electroencephalography (hdEEG) recordings with a novo language paradigm, to investigate the neural tracking of hierarchical linguistic structures and the dynamics of brain states during the processing of language in patients with disorders of consciousness (DOC). Together with technics of machine learning, these EEG-derived neural indices could help with diagnoses and prognoses in personalized medicine (Figure 1).

Figure 1. Researchers combined a hierarchical linguistic paradigm with the hdEEG recording to help clinicians to test residual consciousness in patients with DOC and to predict future outcomes in individual patients.

Nearly 100,000 patients annually in China fall into a coma due to trauma, stroke, anoxia and other diseases, and turn out to be in different clinical DOC conditions, mostly unresponsive wakefulness syndrome (UWS) or minimally conscious state (MCS). Generally, the MCS patients have higher residual consciousness and higher possibility of rehabilitation. However, the clinical diagnostic assessment of patients’ conditions is mainly based on behavioral scales rated by experienced clinicians, which is subjective and has a misdiagnosis rate that can reach up to 40%. The study that combines language processing and the state of consciousness is important -- on the one hand, studying the ability of language processing in different states of consciousness and investigating the underlying neural mechanism are valuable in science, on the other hand, the use of such language processing related neural indices as biomarkers of residual consciousness in patients with DOC has extremely important clinical and social significance.

Although neuroscientists have not yet thoroughly understood the neural basis and computing mechanism of the processing of linguistic hierarchy in humans, the present understandings could already provide translational references. Previous studies showed that, the cortical activities in healthy subjects exhibited tracking to hierarchical linguistic structures in perceived speech sequences, such as words, phrases and sentences (Figure 2a). Neural tracking of word, phrase and sentence hierarchies was reflected by inter-trial phase coherence (ITPC) peaks at corresponding frequencies.

Figure 2. a) Paradigm and demonstrative neural tracking of hierarchical linguistic structures. b) Demonstrative brain state maps. c) Research process.

On this basis, the researchers hypothesized that residual consciousness in patients with DOC could be reflected by the strength of speech-tracking responses, especially neural tracking of higher-level linguistic structures. Three kinds of linguistic sequences different hierarchies were used, including word lists, phrase sequences and sentence sequences. The bedside hdEEG were simultaneously recorded in UWS and MCS patients. At the group level, the results showed a progressive increase in the strength of ITPC from UWS to MCS, alongside the healthy control group for comparison, matched the increasing level of behavioral ratings. It is worth noting that, at the individual level, fifteen patients exhibited significant ITPC at 1 or 2 Hz, and six of them (40.0%) showed significant improvement of clinical diagnosis 100 days after the EEG recordings. This finding suggested that the neural tracking of higher linguistic hierarchies could indicate residual consciousness in patients with DOC. Results from machine learning also showed successful classification of patient groups, particularly when listening to phrase and sentence sequences.

From the perspective of neural mechanism, consciousness is not a static brain function but a real-time evolutional process with dynamic change, self-preservation, and whole-brain cooperation. Based on the global workspace theory of consciousness, the researchers further hypothesized that, brain activities associated with higher level of consciousness will mainly stay in the prefrontal-parietal cortical information loop, while for brain activities with lower level consciousness, sensory cortical loops are involved more. That is, for the present paradigm, the more hierarchies in the linguistic sequence, the more high-level brain areas are involved. The team systematically investigated the temporal dynamic of global brain states (EEG microstates) of the three groups when they perceived the three kinds of linguistic sequences with different hierarchical structures, and compared them with the state in resting state. The EEG microstates are obtained from the electroencephalographic topography across time, which have various parameters such as duration and occurrence. These parameters can reflect the temporal and spatial characteristics of the brain state (Figure 2b). Group level comparison confirmed that, higher cognition related brain states occurs more and sustained longer in healthy subjects than in patients with DOC and vice versa. The results also revealed a correlation of the difference of the dynamic of brain states between the patient’s groups and the linguistic hierarchy in the sequence (Figure 3). In addition, the decoding model (Figure 2c) derived from the neural tracking and brain state indices can successfully characterize the level of residual consciousness in patients with DOC.

Figure 3. Comparison of EEG microstates parameters between the MCS and UWS groups.

Importantly, despite of the better diagnostic performance of the EEG-based classification model than that based on behavioral assessments, the EEG-based prognosis model can be used to predict the outcome of individual patients after 100 days, with an accuracy of 80%. Most crucially, to further test the external validity and the generalization ability of these models, the researchers separately tested them on an independent dataset. Only the EEG- but not the behavior-based classifier trained before exhibited a high predictive accuracy in both outcome-positive and -negative groups across the two sample sets. This finding suggested that, the team may have developed a general ‘biomarker’ for assessing the level of consciousness of human brain, which could be potentially applied in other circumstances, like sleep or anesthesia. Future work will focus on the optimization of the present protocol bonding with multimodal assessments and finally providing experimental evidences and theoretical basis for scientific studies in patients with DOC.

This work entitled “Assessing the depth of language processing in patients with disorders of consciousness” was published online in Nature Neuroscience on May 25, 2020. The research was jointly directed by WANG Liping of the CEBSIT (ION) of the Chinese Academy of Sciences and WU Xuehai of Huashan Hospital of Fudan University, conducted by GUI Peng, JIANG Yuwei, ZANG Di, QI Zengxin, TAN Jiaxing, TANIGAWA Hiromi, JIANG Jian, WEN Yunqing and XU Long, with the direction of ZHAO Jizong, MAO Ying, POO Mu-ming, DING Nai and DEHAENE Stanislas. This work was also supported by the Primate Physiology Research Platform Core Facility of ION and founded by CAS, NSFC and STCSM.

https://www.nature.com/articles/s41593-020-0639-1

Keywords: language processing, disorders of consciousness, brain state, bedside EEG

Author contact:

WANG Liping

Institute of Neuroscience, Center for Excellence in Brain Science and Intelligent Technology, Chinese Academy of Sciences, Shanghai, China

E-mail: liping.wang@ion.ac.cn

WU Xuehai

Huashan Hospital of Fudan University, Shanghai, China

E-mail: wuxuehai2013@163.com