Severe brain injury can lead to disorders of consciousness (DOC), such as a coma. Some patients regain consciousness after injury, while others do not. Those who do not recover are unable to communicate or move in purposeful ways, and need long-term care. It can be difficult for physicians to predict which patients will mend. This is mainly based on their observations of the patient’s behavior over time. But such perceptions are subjective and vulnerable to errors. More accurate and objective methods are needed.

Several studies suggest that the cause of the injury, the age of the person at the time of injury, and how long the person has had a DOC may predict recovery. Recent studies have shown that using a brain-imaging tool called resting state functional magnetic resonance imaging (fMRI) to measure communication between different parts of the brain may help to calculate the likelihood of recovery.

Now, Song, Yang et al. show that combining resting state fMRI with three pieces of clinical information may help to better predict who will improve. Song et al. created a computer model that forecasts recovery from DOC based on fMRI results, the cause of the person’s injury, their age at the time of injury, and how long they have had impaired consciousness. The model could tell which patients would regain consciousness 88% of the time for 112 patients from two medical centers. It also identified several patients who got better despite initial predictions from doctors that they would not.

The experiments show that combining multiple types of information can better predict which patients with DOC will convalesce. Larger studies are needed to confirm that the computer model is reliable. If they do, the model may one day help physicians and families to better plan and manage patients’ care.

Severe brain injury can lead to disorders of consciousness (DOC). Some patients recover consciousness after an acute brain insult, whereas others tragically fall into chronic DOC. The latter cannot communicate functionally or behave purposefully. Most patients remain bedridden, and require laborious care. The medical community is often confronted with an inability to meet the expectations of the chronic DOC patients' families. The social, economic, and ethical consequences are also tremendous (Bernat, 2006). In parallel, although more validations are required, recent pilot studies have proposed new therapeutic interventions, which challenge the existing practice of early treatment discontinuation for a chronic DOC patient (Schiff et al., 2007; Corazzol et al., 2017; Yu et al., 2017). However, before using these novel therapeutic interventions, clinicians first need to determine whether the patient is a suitable candidate. The availability of an accurate and robust prognostication is therefore a fundamental concern in the clinical management of chronic DOC patients, as medical treatment, rehabilitation therapy and even ethical decisions depend on this information.

To date, the prognostication for a DOC patient is based on physician observation of the patient's behavior over period that is sufficient to allow determination of whether there is any evidence of awareness. On the one hand, a patient's motor impairment, sensory deficit, cognitive damage, fluctuation of vigilance and medical complications could give rise to misjudgments; on the other hand, for the assessor, a lack of knowledge regarding DOC, poor training and non-use of adequate behavioral scales are additional elements that may contribute to a high possibility of mistakes. Consequently, careful and repeated behavioral assessments are considered to be particularly important for a precise diagnostic and prognostic judgment (Wannez et al., 2017). Nonetheless, behavioral assessments are inevitably subjective and vulnerable to a variety of personal interferences (Giacino et al., 2009). Physicians and scientists have therefore been seeking accurate and objective markers for diagnosis and prognosis (Demertzi et al., 2017; Noirhomme et al., 2017).

Several pioneering studies have suggested that the etiology, incidence age and duration of DOC are important indicators for prognosis (Multi-Society Task Force on PVS, 1994). Specifically, patients who have non-traumatic brain injury are expected to have a worse functional recovery than traumatic brain injury patients, and young patients were considered more likely to have a favorable outcome than older ones. During the recent decades, some pilot prognostic models have also been explored that are based on features of neurological examination (Zandbergen et al., 1998; Booth et al., 2004; Dolce et al., 2008), abnormalities detected with electroencephalogram (EEG) and evoked potentials (Steppacher et al., 2013; Kang et al., 2014; Hofmeijer and van Putten, 2016; Chennu et al., 2017), anatomical and functional changes identified with brain computed tomography (CT), positron emission tomography (PET) and magnetic resonance imaging (MRI) (Maas et al., 2007; Sidaros et al., 2008; Neuro Imaging for Coma Emergence and Recovery Consortium et al., 2012; Luyt et al., 2012; Stender et al., 2014; Wu et al., 2015), and physiological and biochemical disturbances at both the brain and body levels (Kaneko et al., 2009; Rundgren et al., 2009). Despite many efforts, however, identifying efficient biomarkers for the early prediction of outcome is still challenging and requires additional research. One of the reasons for this is that the DOC could have many different causes and could be associated with several neuropathological processes and different severities, such that any method when used alone carries the risk of false prediction (Bernat, 2016; Rossetti et al., 2016).

Recently, resting state functional MRI (fMRI) has been widely used to investigate the brain functions of DOC patients. Research suggests that these patients demonstrate multiple changes in brain functional networks, including the default mode (Vanhaudenhuyse et al., 2010; Silva et al., 2015), executive control (Demertzi et al., 2014; Wu et al., 2015), salience (Qin et al., 2015; Fischer et al., 2016), and sensorimotor (Yao et al., 2015), auditory (Demertzi et al., 2015), visual (Demertzi et al., 2014a) and subcortical networks (He et al., 2015). The within-network and between-network functional connectivity appeared to be useful indicators of functional brain damage and the likelihood of consciousness recovery (Silva et al., 2015; Di Perri et al., 2016). Taken together, these studies suggest that the brain networks and functional connectivity detected with resting state fMRI could be valuable biomarkers that can be used to trace the level of consciousness and predict the possibility of recovery.

With advances in medicine, prognostication of a DOC patient has moved toward a multidomain paradigm that combines clinical examination with the application of novel technologies (Gosseries et al., 2014). Multidomain assessment has the potential to improve prediction accuracy. More importantly, it can provide reassurances about the importance of each predictor for prognostication by offering concordant evidence (Stevens and Sutter, 2013; Rossetti et al., 2016). More than 20 years ago, the Multi-Society Task Force on persistent vegetative state (PVS) suggested that the etiology, incidence age and duration of DOC could help to predict the outcome (Multi-Society Task Force on PVS, 1994), and numerous studies have subsequently validated the clinical utility of these features (Jennett, 2005; Bruno et al., 2012; Estraneo et al., 2013; Celesia, 2016). Therefore, it is possible that a multidomain model that combines these clinical characteristics and resting state fMRI data could improve prognostic predictions at an individual level and could lead to the early identification of patients who could recover consciousness.

The present work had two major objectives. The first aim was to develop an approach to predict the prognosis of an individual DOC patient using clinical characteristics and resting state fMRI. The second aim, building on the first, was to further explore the different prognostic effects of these clinical and brain imaging features.

The study paradigm is illustrated in Figure 1. Resting state fMRI and clinical data from DOC patients were collected at the so-called T₀ time point when the patients' vital signs and consciousness level had stabilized and a diagnosis had been made. Outcomes were assessed at least 12 months after this T₀ time point; at a time referred to as the T₁ time point. The principal scales included the Coma Recovery Scale Revised (CRS-R) and the Glasgow Outcome Scale (GOS). Instead of predicting diagnosis, this study used the outcome as a target for regression and classification. Using the resting state fMRI and clinical data from the T₀ time point in a training dataset, a regression model was first developed to fit each patient's CRS-R score at the T₁ time point, after which the optimal cut-off value for classifying individual patients on the basis of consciousness recovery was calculated. In this way, we set up the prognostic regression and classification model. Two independent testing datasets were then used to validate the model.

Figure 1

Download asset Open asset

Data analysis

Request a detailed protocol

The data analysis pipeline is illustrated in Figure 2.

Figure 2

Download asset Open asset

In this paper, we describe the development of a prognostic model that combines resting state fMRI with three clinical characteristics to predict one-year outcomes at the single-subject level. The model discriminated between patients who would later recover consciousness and those who would not with an accuracy of around 88% in three datasets from two medical centers. It was also able to identify the prognostic importance of different predictors, including brain functions and clinical characteristics. To our knowledge, this is the first reported implementation of a multidomain prognostic model that is based on resting state fMRI and clinical characteristics in chronic DOC. We therefore suggest that this novel prognostic model is accurate, robust, and interpretable. For research only, we share the prognostic model and its Matlab code at a public download resource (https://github.com/realmsong504/pDOC; copy archived at https://github.com/elifesciences-publications/pDOC).

Brain functions are mediated by the interactions between neurons within different neural circuits and brain regions. Functional imaging can detect the local activity of different brain regions and explore the interactions between them, and has demonstrated potential for informing diagnosis and prognosis in DOC. On the one hand, many studies have focused on one modality of brain functional imaging, such as PET (Phillips et al., 2011), SPECT (Nayak and Mahapatra, 2011), task fMRI (Owen et al., 2006; Coyle et al., 2015), or resting state fMRI (Demertzi et al., 2015; Qin et al., 2015; Wu et al., 2015; Roquet et al., 2016). On the other hand, some cross-modality studies have compared the diagnostic precision between imaging modalities, for example, comparing PET imaging with task fMRI (Stender et al., 2014) or comparing PET, EEG and resting state fMRI (Golkowski et al., 2017). In our study, by combining brain functional networks detected from resting state fMRI with three clinical characteristics, we built a computational model that allowed us to make predictions regarding the prognosis of DOC patients at an individual level. We compared the models separately using only the imaging features or only the clinical characteristics and found that the combination of these predictors achieved greater accuracy. This validated the need for the use of accumulative evidence stemming from multiple assessments, each of which has different sensitivity and specificity in detecting the capacity for recovery of consciousness (Demertzi et al., 2017). Validations in additional and unseen datasets were also undertaken to evaluate the feasibility of the predictive model. Our results showed about 88% average accuracy across the two testing datasets, and good sensitivity and specificity in both the ‘subacute’ patients (i.e. 1 months ≤ duration of unconsciousness ≤ 3 months) and those in the prolonged phase (i.e. duration of unconsciousness >3 months), which suggested good robustness and the generalizability of our model.

Further, the sensitivity of 83% and 100% obtained across the two testing datasets demonstrated a low false-negative rate, which would avoid predicting non-recovery in a patient who can actually recover. Our method successfully identified 16 out of the total 18 patients who later recovered consciousness in the two testing datasets. In parallel, the specificity across the two testing datasets was up to 92% and 83%, respectively. Taken together, these results suggest that our method can precisely identify those patients with a high-potential for recovery consciousness and concurrently reduce false positives in predicting low-potential patients. In addition, although it has been widely considered that the prospect of a clinically meaningful recovery is unrealistic for prolonged DOC patients (Wijdicks and Cranford, 2005), our model correctly predicted 9 out of 10 DOC patients with longer than or equal to three months duration of DOC who subsequently recovered consciousness, including three patients with DOC longer or equal to 6 months duration, suggesting that it can also be applied to prolonged DOC patients.

According to the surviving consciousness level, DOC patients can be classified into distinct diagnostic entities, including VS/UWS and MCS. As MCS is often viewed as a state with a potentially more favorable outcome (Luauté et al., 2010), a misdiagnosis of VS/UWS could heavily bias the judgment of the prognosis, the medical treatment options and even the associated ethical decisions. Some influential studies have found that a few VS/UWS patients exhibit near-normal high-level cortical activation in response to certain stimuli or commands (Owen et al., 2006; Monti et al., 2010), and that late recovery of responsiveness and consciousness is not exceptional in patients with VS/UWS (Estraneo et al., 2010). Instead of predicting diagnosis, this study used one-year outcome as a target for regression and classification. Our method, which is based on the combined use of clinical and neuroimaging data, successfully identified seven out of the eight VS/UWS patients in the testing dataset who were initially diagnosed as VS/UWS but subsequently achieved a promising outcome.

By analyzing the sMC F-value for each predictor in the regression model, we investigated the prognostic effects of these predictors. In particular, the sMC F-value of the incidence age was greater than that of the other clinical characteristics, suggesting that incidence age might be the most important predictor among the clinical characteristics. Notably, the sMC F-value for the imaging features as a whole seemed to be greater than that for the clinical features, as shown in Figure 4B. We therefore speculate that the patient's residual consciousness capacity, indicated by brain networks and functional connectivity detected from resting state fMRI, might have a stronger prognostic effect than their clinical characteristics.

Some previous studies have shown that the resting state functional connectivity within the default mode network is decreased in severely brain-damaged patients, in proportion to their degree of consciousness impairment, from locked-in syndrome to minimally conscious, vegetative and coma patients (Vanhaudenhuyse et al., 2010). Moreover, the reduced functional connectivity within the default mode network, specifically between the MPFC and the PCC, may predict the outcome for DOC patients (Silva et al., 2015). Our model also validates that the aMPFC and PCC in the default mode network play important roles in predicting prognosis.

Above all, we found that the functional connectivity between the aMPFC and the DMPFC had the maximum sMC F-value in the prognostic regression model. The aMPFC is one of the core brain areas in the default mode network, whereas the DMPFC is located in the executive control network. Previous studies have demonstrated that this functional connectivity is negative connectivity in normal healthy populations, with the anti-correlation being proposed as one reflection of the intrinsic functional organization of the human brain (Fox et al., 2005). The default mode network directly contributes to internally generated stimulus-independent thoughts, self-monitoring, and baseline monitoring of the external world, while the executive control network supports active exploration of the external world. Correct communication and coordination between these two intrinsic anti-correlated networks is thought to be very important for optimal information integration and cognitive functioning (Buckner et al., 2008). A recent study reported that negative functional connectivities between the default mode network and the task-positive network were only observed in patients who recovered consciousness and in healthy controls, whereas positive values were obtained in patients with impaired consciousness (Di Perri et al., 2016). Further, our regression model suggests that the anti-correlations between these two diametrically opposed networks (i.e. the default mode network and the executive control network) should be the most crucial imaging feature for predicting the outcomes of the DOC patients. We therefore conclude that our prognostic model has good interpretability, and that it not only verifies the findings of previous studies but also provides a window to assess the relative significance of various predictors for the prognosis of DOC patients.

This study involved two testing datasets, which were found to be quite different, as shown in Table 1. First, the distributions of the etiology of the patients were remarkably different in the two datasets. The numbers of patients suffering from trauma/stroke/anoxia were 12/6/7 and 8/0/16 in the ‘Beijing HDxt’ and ‘Guangzhou HDxt’ datasets, respectively. The outcomes were also different. In the ‘Beijing HDxt’ dataset, 12 out of the total 25 patients recovered consciousness, compared with six out of the total 24 patients in the ‘Guangzhou HDxt’ dataset. Given that the characteristics of the two medical centers and their roles in the local health care system are different, we speculate that this could be one of the main reasons why the enrolled patient populations were heterogeneous. As described in the Introduction, DOC can have many causes and can be associated with several neuropathological processes and different severities, leading to the suggestion that information from different domains should be integrated to improve diagnosis and prognostication (Bernat, 2016). Our study demonstrates that the combination of imaging and clinical features can achieve a better performance than the use of single domain features.

However, some caution is warranted. First, although this study involved 112 DOC patients, the number of patients who would later recover consciousness was relatively low (i.e. 31/112). So, a much larger cohort will be needed for further validation. Second, the PPVs for the two testing datasets were remarkably different, with that for the ‘Guangzhou HDxt’ dataset being relatively low (67% versus 91%). Although our method predicted that nine patients in this dataset would recover consciousness, only six of them did (i.e. GOS ≥3), with the other three remaining unconscious at the end of the follow-up (i.e. GOS ≤2). Given that many additional factors are associated with the outcome of DOC patients, including medical complications, nutrition and so on, future work will need to integrate more information in order to provide more precise predictions. Third, the signal-to-noise ratio of resting state fMRI can influence functional connectivity analysis, so calibrations will be necessary when applying the predictive model across different sites, including standardizing the MRI acquisition protocols (e.g. scanner hardware, imaging protocols and acquisition sequences) and the patients' management strategies (see Appendix 10 for more information). Finally, a DOC patient’s prognosis can be considered according to at least three dimensions: survival/mortality, recovery of consciousness, and functional recovery. This study focused on predicting recovery of consciousness, and the patients who died during the follow-up were excluded. In the future, we will consider more outcome assessments, including survival/mortality and functional recovery.

In summary, our prognostic model, which combines resting state fMRI with clinical characteristics, is proposed to predict the one-year outcome of DOC patients at an individual level. The average accuracy of classifying a patient as ‘consciousness recovery’ or not was around 88% in the training dataset and two unseen testing datasets. Our model also has good interpretability, thereby providing a window to reassure physicians and scientists about the significance of different predictors, including brain networks, functional connectivities and clinical characteristics. Together, these advantages could offer an objective prognosis for DOC patients that will optimize their management and deepen our understanding of brain function during unconsciousness.

We have provided anonymised demographic and clinical characteristics of the DOC patients in Appendix 1. We have made the analysis pipeline, including fMRI preprocessing, feature calculation and extraction, regression and classification, and the results visualization publicly available. Also, we have uploaded the fMRI signals in each of region of interest for every DOC patient and healthy control involved in this study. Anyone is welcome to download them from GitHub (https://github.com/realmsong504/pDOC; copy archived at https://github.com/elifesciences-publications/pDOC).

1. 1. Bernat JL

   (2006) Chronic disorders of consciousness

   The Lancet 367:1181–1192.

   https://doi.org/10.1016/S0140-6736(06)68508-5

   • Google Scholar
2. 1. Bernat JL

   (2016) Prognostic limitations of syndromic diagnosis in disorders of consciousness

   AJOB Neuroscience 7:46–48.

   https://doi.org/10.1080/21507740.2016.1146367

   • Google Scholar
3. 1. Booth CM
   2. Boone RH
   3. Tomlinson G
   4. Detsky AS

   (2004) Is this patient dead, vegetative, or severely neurologically impaired? assessing outcome for comatose survivors of cardiac arrest

   Jama-Journal of the American Medical Association 291:870–879.

   https://doi.org/10.1001/jama.291.7.870

   • Google Scholar
4. 1. Bruno M-A
   2. Ledoux D
   3. Vanhaudenhuyse A
   4. Gosseries O
   5. Thibaut A
   6. Laureys S

   (2012)

   Coma and Disorders of Consciousness. London: Springer London

   11–23, Prognosis of Patients with Altered State of Consciousness, Coma and Disorders of Consciousness. London: Springer London, 10.1007/978-1-4471-2440-5_2.

   • Google Scholar
5. 1. Buckner RL
   2. Andrews-Hanna JR
   3. Schacter DL

   (2008) The brain's Default Network

   Annals of the New York Academy of Sciences 1124:1–38.

   https://doi.org/10.1196/annals.1440.011

   • Google Scholar
6. 1. Celesia GG

   (2016)

   Vegetative state two decades after the Multi-Society task force (MSTF) Report

   Brain Function and Responsiveness in Disorders of Consciousness: Springer pp. 171–184.

   • Google Scholar
7. 1. Chennu S
   2. Annen J
   3. Wannez S
   4. Thibaut A
   5. Chatelle C
   6. Cassol H
   7. Martens G
   8. Schnakers C
   9. Gosseries O
   10. Menon D
   11. Laureys S

   (2017) Brain networks predict metabolism, diagnosis and prognosis at the bedside in disorders of consciousness

   Brain 140:2120–2132.

   https://doi.org/10.1093/brain/awx163

   • PubMed
   • Google Scholar
8. 1. Corazzol M
   2. Lio G
   3. Lefevre A
   4. Deiana G
   5. Tell L
   6. André-Obadia N
   7. Bourdillon P
   8. Guenot M
   9. Desmurget M
   10. Luauté J
   11. Sirigu A

   (2017) Restoring consciousness with vagus nerve stimulation

   Current Biology 27:R994–R996.

   https://doi.org/10.1016/j.cub.2017.07.060

   • PubMed
   • Google Scholar
9. 1. Coyle D
   2. Stow J
   3. McCreadie K
   4. McElligott J
   5. Carroll Á

   (2015) Sensorimotor modulation assessment and brain-computer interface training in disorders of consciousness

   Archives of Physical Medicine and Rehabilitation 96:S62–S70.

   https://doi.org/10.1016/j.apmr.2014.08.024

   • PubMed
   • Google Scholar
10. 1. Damoiseaux JS
    2. Rombouts SA
    3. Barkhof F
    4. Scheltens P
    5. Stam CJ
    6. Smith SM
    7. Beckmann CF

    (2006) Consistent resting-state networks across healthy subjects

    PNAS 103:13848–13853.

    https://doi.org/10.1073/pnas.0601417103

    • PubMed
    • Google Scholar
11. 1. Demertzi A
    2. Antonopoulos G
    3. Heine L
    4. Voss HU
    5. Crone JS
    6. de Los Angeles C
    7. Bahri MA
    8. Di Perri C
    9. Vanhaudenhuyse A
    10. Charland-Verville V
    11. Kronbichler M
    12. Trinka E
    13. Phillips C
    14. Gomez F
    15. Tshibanda L
    16. Soddu A
    17. Schiff ND
    18. Whitfield-Gabrieli S
    19. Laureys S

    (2015) Intrinsic functional connectivity differentiates minimally conscious from unresponsive patients

    Brain 138:2619–2631.

    https://doi.org/10.1093/brain/awv169

    • PubMed
    • Google Scholar
12. 1. Demertzi A
    2. Antonopoulos G
    3. Voss HU
    4. Crone JS
    5. Schiff ND
    6. Kronbichler M

    (2014a)

    Audio-visual crossmodal fMRI connectivity differentiates single patients with disorders of consciousness. 

    Frontiers in Human Neuroscience: Belgian Brain Council 2014 MODULATING the BRAIN: FACTS, FICTION,FUTURE.

    • Google Scholar
13. 1. Demertzi A
    2. Gómez F
    3. Crone JS
    4. Vanhaudenhuyse A
    5. Tshibanda L
    6. Noirhomme Q
    7. Thonnard M
    8. Charland-Verville V
    9. Kirsch M
    10. Laureys S
    11. Soddu A

    (2014) Multiple fMRI system-level baseline connectivity is disrupted in patients with consciousness alterations

    Cortex 52:35–46.

    https://doi.org/10.1016/j.cortex.2013.11.005

    • PubMed
    • Google Scholar
14. 1. Demertzi A
    2. Sitt JD
    3. Sarasso S
    4. Pinxten W

    (2017) Measuring states of pathological (un)consciousness: research dimensions, clinical applications, and ethics

    Neuroscience of Consciousness 2017:.

    https://doi.org/10.1093/nc/nix010

    • PubMed
    • Google Scholar
15. 1. Di Perri C
    2. Bahri MA
    3. Amico E
    4. Thibaut A
    5. Heine L
    6. Antonopoulos G
    7. Charland-Verville V
    8. Wannez S
    9. Gomez F
    10. Hustinx R
    11. Tshibanda L
    12. Demertzi A
    13. Soddu A
    14. Laureys S

    (2016) Neural correlates of consciousness in patients who have emerged from a minimally conscious state: a cross-sectional multimodal imaging study

    The Lancet Neurology 15:830–842.

    https://doi.org/10.1016/S1474-4422(16)00111-3

    • PubMed
    • Google Scholar
16. 1. Dolce G
    2. Quintieri M
    3. Serra S
    4. Lagani V
    5. Pignolo L

    (2008) Clinical signs and early prognosis in vegetative state: a decisional tree, data-mining study

    Brain Injury 22:617–623.

    https://doi.org/10.1080/02699050802132503

    • PubMed
    • Google Scholar
17. 1. Dosenbach NU
    2. Nardos B
    3. Cohen AL
    4. Fair DA
    5. Power JD
    6. Church JA
    7. Nelson SM
    8. Wig GS
    9. Vogel AC
    10. Lessov-Schlaggar CN
    11. Barnes KA
    12. Dubis JW
    13. Feczko E
    14. Coalson RS
    15. Pruett JR
    16. Barch DM
    17. Petersen SE
    18. Schlaggar BL

    (2010) Prediction of individual brain maturity using fMRI

    Science 329:1358–1361.

    https://doi.org/10.1126/science.1194144

    • PubMed
    • Google Scholar
18. 1. Drysdale AT
    2. Grosenick L
    3. Downar J
    4. Dunlop K
    5. Mansouri F
    6. Meng Y
    7. Fetcho RN
    8. Zebley B
    9. Oathes DJ
    10. Etkin A
    11. Schatzberg AF
    12. Sudheimer K
    13. Keller J
    14. Mayberg HS
    15. Gunning FM
    16. Alexopoulos GS
    17. Fox MD
    18. Pascual-Leone A
    19. Voss HU
    20. Casey BJ
    21. Dubin MJ
    22. Liston C

    (2017) Resting-state connectivity biomarkers define neurophysiological subtypes of depression

    Nature Medicine 23:.

    https://doi.org/10.1038/nm.4246

    • PubMed
    • Google Scholar
19. 1. Estraneo A
    2. Moretta P
    3. Loreto V
    4. Lanzillo B
    5. Cozzolino A
    6. Saltalamacchia A
    7. Lullo F
    8. Santoro L
    9. Trojano L

    (2013) Predictors of recovery of responsiveness in prolonged anoxic vegetative state

    Neurology 80:464–470.

    https://doi.org/10.1212/WNL.0b013e31827f0f31

    • PubMed
    • Google Scholar
20. 1. Estraneo A
    2. Moretta P
    3. Loreto V
    4. Lanzillo B
    5. Santoro L
    6. Trojano L

    (2010) Late recovery after traumatic, anoxic, or hemorrhagic long-lasting vegetative state

    Neurology 75:239–245.

    https://doi.org/10.1212/WNL.0b013e3181e8e8cc

    • PubMed
    • Google Scholar
21. 1. Fan W
    2. Li H
    3. Shan Y
    4. Lv H
    5. Zhang H
    6. Liang Y

    (2011) Classification of vinegar samples based on near infrared spectroscopy combined with wavelength selection

    Analytical Methods 3:1872–1876.

    https://doi.org/10.1039/c1ay05101f

    • Google Scholar
22. 1. Fan Y
    2. Shen D
    3. Gur RC
    4. Gur RE
    5. Davatzikos C

    (2007) COMPARE: classification of morphological patterns using adaptive regional elements

    IEEE Transactions on Medical Imaging 26:93–105.

    https://doi.org/10.1109/TMI.2006.886812

    • PubMed
    • Google Scholar
23. 1. Fischer DB
    2. Boes AD
    3. Demertzi A
    4. Evrard HC
    5. Laureys S
    6. Edlow BL
    7. Liu H
    8. Saper CB
    9. Pascual-Leone A
    10. Fox MD
    11. Geerling JC

    (2016) A human brain network derived from coma-causing brainstem lesions

    Neurology 87:2427–2434.

    https://doi.org/10.1212/WNL.0000000000003404

    • PubMed
    • Google Scholar
24. 1. Fox MD
    2. Snyder AZ
    3. Vincent JL
    4. Corbetta M
    5. Van Essen DC
    6. Raichle ME

    (2005) The human brain is intrinsically organized into dynamic, anticorrelated functional networks

    PNAS 102:9673–9678.

    https://doi.org/10.1073/pnas.0504136102

    • PubMed
    • Google Scholar
25. 1. Galanaud D
    2. Perlbarg V
    3. Gupta R
    4. Stevens RD
    5. Sanchez P
    6. Tollard E
    7. de Champfleur NM
    8. Dinkel J
    9. Faivre S
    10. Soto-Ares G
    11. Veber B
    12. Cottenceau V
    13. Masson F
    14. Tourdias T
    15. André E
    16. Audibert G
    17. Schmitt E
    18. Ibarrola D
    19. Dailler F
    20. Vanhaudenhuyse A
    21. Tshibanda L
    22. Payen JF
    23. Le Bas JF
    24. Krainik A
    25. Bruder N
    26. Girard N
    27. Laureys S
    28. Benali H
    29. Puybasset L

    (2012) Assessment of white matter injury and outcome in severe brain trauma: a prospective multicenter cohort

    Anesthesiology 117:1300–1310.

    https://doi.org/10.1097/ALN.0b013e3182755558

    • PubMed
    • Google Scholar
26. 1. Gardumi A
    2. Ivanov D
    3. Hausfeld L
    4. Valente G
    5. Formisano E
    6. Uludağ K

    (2016) The effect of spatial resolution on decoding accuracy in fMRI multivariate pattern analysis

    NeuroImage 132:32–42.

    https://doi.org/10.1016/j.neuroimage.2016.02.033

    • PubMed
    • Google Scholar
27. Website

    1. Giacino J
    2. Kalmar K

    (2006) Coma recovery Scale-Revised

    Accessed June 2018.

    http://www.tbims.org/combi/crs
28. Book

    1. Giacino JT
    2. Schnakers C
    3. Rodriguez-Moreno D
    4. Kalmar K
    5. Schiff N
    6. Hirsch J

    (2009) Behavioral assessment in patients with disorders of consciousness: gold standard or fool's gold?

    In: Laureys S, Schiff N. D, Owen A. M, editors. Coma Science: Clinical and Ethical Implications.. Amsterdam: Elsevier Science Bv. pp. 33–48.

    https://doi.org/10.1016/S0079-6123(09)17704-X

    • Google Scholar
29. 1. Golkowski D
    2. Merz K
    3. Mlynarcik C
    4. Kiel T
    5. Schorr B
    6. Lopez-Rolon A
    7. Lukas M
    8. Jordan D
    9. Bender A
    10. Ilg R

    (2017) Simultaneous EEG-PET-fMRI measurements in disorders of consciousness: an exploratory study on diagnosis and prognosis

    Journal of Neurology 264:1986–1995.

    https://doi.org/10.1007/s00415-017-8591-z

    • PubMed
    • Google Scholar
30. 1. Gosseries O
    2. Zasler ND
    3. Laureys S

    (2014) Recent advances in disorders of consciousness: focus on the diagnosis

    Brain Injury 28:1141–1150.

    https://doi.org/10.3109/02699052.2014.920522

    • PubMed
    • Google Scholar
31. 1. Greicius MD
    2. Srivastava G
    3. Reiss AL
    4. Menon V

    (2004) Default-mode network activity distinguishes Alzheimer's disease from healthy aging: evidence from functional MRI

    PNAS 101:4637–4642.

    https://doi.org/10.1073/pnas.0308627101

    • PubMed
    • Google Scholar
32. 1. Guyon I
    2. Elisseeff A

    (2003)

    An introduction to variable and feature selection

    Journal of Machine Learning Research 3:1157–1182.

    • Google Scholar
33. 1. Hay E
    2. Ritter P
    3. Lobaugh NJ
    4. McIntosh AR

    (2017) Multiregional integration in the brain during resting-state fMRI activity

    PLoS Computational Biology 13:e1005410.

    https://doi.org/10.1371/journal.pcbi.1005410

    • PubMed
    • Google Scholar
34. 1. He JH
    2. Cui Y
    3. Song M
    4. Yang Y
    5. Dang YY
    6. Jiang TZ
    7. Xu RX

    (2015) Decreased functional connectivity between the mediodorsal thalamus and default mode network in patients with disorders of consciousness

    Acta Neurologica Scandinavica 131:145–151.

    https://doi.org/10.1111/ane.12299

    • PubMed
    • Google Scholar
35. 1. Hofmeijer J
    2. van Putten MJ

    (2016) EEG in postanoxic coma: Prognostic and diagnostic value

    Clinical Neurophysiology 127:2047–2055.

    https://doi.org/10.1016/j.clinph.2016.02.002

    • PubMed
    • Google Scholar
36. 1. Huettel SA
    2. Song AW
    3. McCarthy G

    (2009)

    Functional Magnetic Resonance Imaging

    Sinauer Associates, Functional Magnetic Resonance Imaging, 2 Edn.

    • Google Scholar
37. Book

    1. James G
    2. Witten D
    3. Hastie T
    4. Tibshirani R

    (2013)

    An Introduction to Statistical Learning

    Springer.

    • Google Scholar
38. 1. Jennett B

    (2005)

    Boundaries of Consciousness: Neurobiology and Neuropathology

    537–543, Thirty years of the vegetative state: clinical, ethical and legal problems, Boundaries of Consciousness: Neurobiology and Neuropathology, 10.1016/S0079-6123(05)50037-2.

    • Google Scholar
39. 1. Kaneko T
    2. Kasaoka S
    3. Miyauchi T
    4. Fujita M
    5. Oda Y
    6. Tsuruta R
    7. Maekawa T

    (2009) Serum glial fibrillary acidic protein as a predictive biomarker of neurological outcome after cardiac arrest

    Resuscitation 80:790–794.

    https://doi.org/10.1016/j.resuscitation.2009.04.003

    • PubMed
    • Google Scholar
40. 1. Kang XG
    2. Li L
    3. Wei D
    4. Xu XX
    5. Zhao R
    6. Jing YY
    7. Su YY
    8. Xiong LZ
    9. Zhao G
    10. Jiang W

    (2014) Development of a simple score to predict outcome for unresponsive wakefulness syndrome

    Critical Care 18:R37.

    https://doi.org/10.1186/cc13745

    • PubMed
    • Google Scholar
41. 1. Krishnan A
    2. Williams LJ
    3. McIntosh AR
    4. Abdi H

    (2011) Partial least squares (PLS) methods for neuroimaging: a tutorial and review

    NeuroImage 56:455–475.

    https://doi.org/10.1016/j.neuroimage.2010.07.034

    • PubMed
    • Google Scholar
42. 1. Kvalheim OM
    2. Arneberg R
    3. Bleie O
    4. Rajalahti T
    5. Smilde AK
    6. Westerhuis JA

    (2014) Variable importance in latent variable regression models

    Journal of Chemometrics 28:615–622.

    https://doi.org/10.1002/cem.2626

    • Google Scholar
43. 1. Kvalheim OM
    2. Karstang TV

    (1989) Interpretation of latent-variable regression models

    Chemometrics and Intelligent Laboratory Systems 7:39–51.

    https://doi.org/10.1016/0169-7439(89)80110-8

    • Google Scholar
44. 1. Li H
    2. Liang Y
    3. Xu Q
    4. Cao D

    (2009) Key wavelengths screening using competitive adaptive reweighted sampling method for multivariate calibration

    Analytica Chimica Acta 648:77–84.

    https://doi.org/10.1016/j.aca.2009.06.046

    • PubMed
    • Google Scholar
45. 1. Li H
    2. Xu Q
    3. Liang Y

    (2014)

    libPLS: an integrated library for partial least squares regression and discriminant analysis

    PeerJ Preprints 2:e190.

    • Google Scholar
46. 1. Luauté J
    2. Maucort-Boulch D
    3. Tell L
    4. Quelard F
    5. Sarraf T
    6. Iwaz J
    7. Boisson D
    8. Fischer C

    (2010) Long-term outcomes of chronic minimally conscious and vegetative states

    Neurology 75:246–252.

    https://doi.org/10.1212/WNL.0b013e3181e8e8df

    • PubMed
    • Google Scholar
47. 1. Luyt CE
    2. Galanaud D
    3. Perlbarg V
    4. Vanhaudenhuyse A
    5. Stevens RD
    6. Gupta R
    7. Besancenot H
    8. Krainik A
    9. Audibert G
    10. Combes A
    11. Chastre J
    12. Benali H
    13. Laureys S
    14. Puybasset L
    15. Neuro Imaging for Coma Emergence and Recovery Consortium

    (2012) Diffusion tensor imaging to predict long-term outcome after cardiac arrest: a bicentric pilot study

    Anesthesiology 117:1311–1321.

    https://doi.org/10.1097/ALN.0b013e318275148c

    • PubMed
    • Google Scholar
48. 1. Maas AI
    2. Steyerberg EW
    3. Butcher I
    4. Dammers R
    5. Lu J
    6. Marmarou A
    7. Mushkudiani NA
    8. McHugh GS
    9. Murray GD

    (2007) Prognostic value of computerized tomography scan characteristics in traumatic brain injury: results from the IMPACT study

    Journal of Neurotrauma 24:303–314.

    https://doi.org/10.1089/neu.2006.0033

    • PubMed
    • Google Scholar
49. 1. Magrassi L
    2. Maggioni G
    3. Pistarini C
    4. Di Perri C
    5. Bastianello S
    6. Zippo AG
    7. Iotti GA
    8. Biella GE
    9. Imberti R

    (2016) Results of a prospective study (CATS) on the effects of thalamic stimulation in minimally conscious and vegetative state patients

    Journal of Neurosurgery 125:972–981.

    https://doi.org/10.3171/2015.7.JNS15700

    • PubMed
    • Google Scholar
50. 1. Mehmood T
    2. Liland KH
    3. Snipen L
    4. Sæbø S

    (2012) A review of variable selection methods in partial least squares regression

    Chemometrics and Intelligent Laboratory Systems 118:62–69.

    https://doi.org/10.1016/j.chemolab.2012.07.010

    • Google Scholar
51. 1. Monti MM
    2. Vanhaudenhuyse A
    3. Coleman MR
    4. Boly M
    5. Pickard JD
    6. Tshibanda L
    7. Owen AM
    8. Laureys S

    (2010) Willful modulation of brain activity in disorders of consciousness

    New England Journal of Medicine 362:579–589.

    https://doi.org/10.1056/NEJMoa0905370

    • PubMed
    • Google Scholar
52. 1. Multi-Society Task Force on PVS

    (1994) Medical aspects of the persistent vegetative state (1)

    New England Journal of Medicine 330:1499–1508.

    https://doi.org/10.1056/NEJM199405263302107

    • PubMed
    • Google Scholar
53. 1. Nayak P
    2. Mahapatra AK

    (2011) Single photon emission computed tomography scanning: a predictor of outcome in vegetative state of head injury

    Journal of Neurosciences in Rural Practice 2:12–16.

    https://doi.org/10.4103/0976-3147.80079

    • PubMed
    • Google Scholar
54. 1. Noirhomme Q
    2. Brecheisen R
    3. Lesenfants D
    4. Antonopoulos G
    5. Laureys S

    (2017) "Look at my classifier's result": Disentangling unresponsive from (minimally) conscious patients

    NeuroImage 145:288–303.

    https://doi.org/10.1016/j.neuroimage.2015.12.006

    • PubMed
    • Google Scholar
55. 1. Owen AM
    2. Coleman MR
    3. Boly M
    4. Davis MH
    5. Laureys S
    6. Pickard JD

    (2006) Detecting awareness in the vegetative state

    Science 313:1402.

    https://doi.org/10.1126/science.1130197

    • PubMed
    • Google Scholar
56. 1. Phillips CL
    2. Bruno MA
    3. Maquet P
    4. Boly M
    5. Noirhomme Q
    6. Schnakers C
    7. Vanhaudenhuyse A
    8. Bonjean M
    9. Hustinx R
    10. Moonen G
    11. Luxen A
    12. Laureys S

    (2011) "Relevance vector machine" consciousness classifier applied to cerebral metabolism of vegetative and locked-in patients

    NeuroImage 56:797–808.

    https://doi.org/10.1016/j.neuroimage.2010.05.083

    • PubMed
    • Google Scholar
57. 1. Pignat JM
    2. Mauron E
    3. Jöhr J
    4. Gilart de Keranflec'h C
    5. Van De Ville D
    6. Preti MG
    7. Meskaldji DE
    8. Hömberg V
    9. Laureys S
    10. Draganski B
    11. Frackowiak R
    12. Diserens K

    (2016) Outcome prediction of consciousness disorders in the acute stage based on a complementary motor behavioural tool

    PloS One 11:e0156882.

    https://doi.org/10.1371/journal.pone.0156882

    • PubMed
    • Google Scholar
58. 1. Power JD
    2. Schlaggar BL
    3. Petersen SE

    (2015) Recent progress and outstanding issues in motion correction in resting state fMRI

    NeuroImage 105:536–551.

    https://doi.org/10.1016/j.neuroimage.2014.10.044

    • PubMed
    • Google Scholar
59. 1. Qin P
    2. Wu X
    3. Huang Z
    4. Duncan NW
    5. Tang W
    6. Wolff A
    7. Hu J
    8. Gao L
    9. Jin Y
    10. Wu X
    11. Zhang J
    12. Lu L
    13. Wu C
    14. Qu X
    15. Mao Y
    16. Weng X
    17. Zhang J
    18. Northoff G

    (2015) How are different neural networks related to consciousness?

    Annals of Neurology 78:594–605.

    https://doi.org/10.1002/ana.24479

    • PubMed
    • Google Scholar
60. 1. Raichle ME

    (2011) The restless brain

    Brain Connectivity 1:3–12.

    https://doi.org/10.1089/brain.2011.0019

    • PubMed
    • Google Scholar
61. 1. Roquet D
    2. Foucher JR
    3. Froehlig P
    4. Renard F
    5. Pottecher J
    6. Besancenot H
    7. Schneider F
    8. Schenck M
    9. Kremer S

    (2016) Resting-state networks distinguish locked-in from vegetative state patients

    NeuroImage: Clinical 12:16–22.

    https://doi.org/10.1016/j.nicl.2016.06.003

    • PubMed
    • Google Scholar
62. 1. Rossetti AO
    2. Rabinstein AA
    3. Oddo M

    (2016) Neurological prognostication of outcome in patients in coma after cardiac arrest

    The Lancet Neurology 15:597–609.

    https://doi.org/10.1016/S1474-4422(16)00015-6

    • PubMed
    • Google Scholar
63. 1. Rundgren M
    2. Karlsson T
    3. Nielsen N
    4. Cronberg T
    5. Johnsson P
    6. Friberg H

    (2009) Neuron specific enolase and S-100B as predictors of outcome after cardiac arrest and induced hypothermia

    Resuscitation 80:784–789.

    https://doi.org/10.1016/j.resuscitation.2009.03.025

    • PubMed
    • Google Scholar
64. 1. Schiff ND
    2. Giacino JT
    3. Kalmar K
    4. Victor JD
    5. Baker K
    6. Gerber M
    7. Fritz B
    8. Eisenberg B
    9. Biondi T
    10. O'Connor J
    11. Kobylarz EJ
    12. Farris S
    13. Machado A
    14. McCagg C
    15. Plum F
    16. Fins JJ
    17. Rezai AR

    (2007) Behavioural improvements with thalamic stimulation after severe traumatic brain injury

    Nature 448:600–603.

    https://doi.org/10.1038/nature06041

    • PubMed
    • Google Scholar
65. 1. Seeley WW
    2. Menon V
    3. Schatzberg AF
    4. Keller J
    5. Glover GH
    6. Kenna H
    7. Reiss AL
    8. Greicius MD

    (2007) Dissociable intrinsic connectivity networks for salience processing and executive control

    Journal of Neuroscience 27:2349–2356.

    https://doi.org/10.1523/JNEUROSCI.5587-06.2007

    • PubMed
    • Google Scholar
66. 1. Sidaros A
    2. Engberg AW
    3. Sidaros K
    4. Liptrot MG
    5. Herning M
    6. Petersen P
    7. Paulson OB
    8. Jernigan TL
    9. Rostrup E

    (2008) Diffusion tensor imaging during recovery from severe traumatic brain injury and relation to clinical outcome: a longitudinal study

    Brain 131:559–572.

    https://doi.org/10.1093/brain/awm294

    • PubMed
    • Google Scholar
67. 1. Silva S
    2. de Pasquale F
    3. Vuillaume C
    4. Riu B
    5. Loubinoux I
    6. Geeraerts T
    7. Seguin T
    8. Bounes V
    9. Fourcade O
    10. Demonet JF
    11. Péran P

    (2015) Disruption of posteromedial large-scale neural communication predicts recovery from coma

    Neurology 85:2036–2044.

    https://doi.org/10.1212/WNL.0000000000002196

    • PubMed
    • Google Scholar
68. 1. Stender J
    2. Gosseries O
    3. Bruno MA
    4. Charland-Verville V
    5. Vanhaudenhuyse A
    6. Demertzi A
    7. Chatelle C
    8. Thonnard M
    9. Thibaut A
    10. Heine L
    11. Soddu A
    12. Boly M
    13. Schnakers C
    14. Gjedde A
    15. Laureys S

    (2014) Diagnostic precision of PET imaging and functional MRI in disorders of consciousness: a clinical validation study

    The Lancet 384:514–522.

    https://doi.org/10.1016/S0140-6736(14)60042-8

    • PubMed
    • Google Scholar
69. 1. Steppacher I
    2. Eickhoff S
    3. Jordanov T
    4. Kaps M
    5. Witzke W
    6. Kissler J

    (2013) N400 predicts recovery from disorders of consciousness

    Annals of Neurology 73:594–602.

    https://doi.org/10.1002/ana.23835

    • PubMed
    • Google Scholar
70. 1. Stevens RD
    2. Sutter R

    (2013) Prognosis in severe brain injury

    Critical Care Medicine 41:1104–1123.

    https://doi.org/10.1097/CCM.0b013e318287ee79

    • PubMed
    • Google Scholar
71. 1. Steyerberg E

    (2008)

    Clinical Prediction Models

    A practical approach to development, validation, and updating, Clinical Prediction Models, Springer Science & Business Media.

    • Google Scholar
72. 1. Tan B
    2. Liang Y
    3. Yi L
    4. Li H
    5. Zhou Z
    6. Ji X
    7. Deng J

    (2010) Identification of free fatty acids profiling of type 2 diabetes mellitus and exploring possible biomarkers by GC–MS coupled with chemometrics

    Metabolomics 6:219–228.

    https://doi.org/10.1007/s11306-009-0189-8

    • Google Scholar
73. 1. Tran TN
    2. Afanador NL
    3. Buydens LMC
    4. Blanchet L

    (2014) Interpretation of variable importance in partial least squares with significance multivariate correlation (sMC)

    Chemometrics and Intelligent Laboratory Systems 138:153–160.

    https://doi.org/10.1016/j.chemolab.2014.08.005

    • Google Scholar
74. 1. Vanhaudenhuyse A
    2. Noirhomme Q
    3. Tshibanda LJ
    4. Bruno MA
    5. Boveroux P
    6. Schnakers C
    7. Soddu A
    8. Perlbarg V
    9. Ledoux D
    10. Brichant JF
    11. Moonen G
    12. Maquet P
    13. Greicius MD
    14. Laureys S
    15. Boly M

    (2010) Default network connectivity reflects the level of consciousness in non-communicative brain-damaged patients

    Brain 133:161–171.

    https://doi.org/10.1093/brain/awp313

    • PubMed
    • Google Scholar
75. 1. Wannez S
    2. Heine L
    3. Thonnard M
    4. Gosseries O
    5. Laureys S
    6. Coma Science Group collaborators

    (2017) The repetition of behavioral assessments in diagnosis of disorders of consciousness

    Annals of Neurology 81:883–889.

    https://doi.org/10.1002/ana.24962

    • PubMed
    • Google Scholar
76. 1. Welvaert M
    2. Rosseel Y

    (2013) On the definition of signal-to-noise ratio and contrast-to-noise ratio for FMRI data

    PLoS One 8:e77089.

    https://doi.org/10.1371/journal.pone.0077089

    • PubMed
    • Google Scholar
77. 1. Wijdicks EF
    2. Cranford RE

    (2005) Clinical diagnosis of prolonged states of impaired consciousness in adults

    Mayo Clinic Proceedings 80:1037–1046.

    https://doi.org/10.4065/80.8.1037

    • PubMed
    • Google Scholar
78. 1. Wold S
    2. Sjöström M
    3. Eriksson L

    (2001) PLS-regression: a basic tool of chemometrics

    Chemometrics and Intelligent Laboratory Systems 58:109–130.

    https://doi.org/10.1016/S0169-7439(01)00155-1

    • Google Scholar
79. 1. Wu X
    2. Zou Q
    3. Hu J
    4. Tang W
    5. Mao Y
    6. Gao L
    7. Zhu J
    8. Jin Y
    9. Wu X
    10. Lu L
    11. Zhang Y
    12. Zhang Y
    13. Dai Z
    14. Gao JH
    15. Weng X
    16. Zhou L
    17. Northoff G
    18. Giacino JT
    19. He Y
    20. Yang Y

    (2015) Intrinsic functional connectivity patterns predict consciousness level and recovery outcome in acquired brain injury

    Journal of Neuroscience 35:12932–12946.

    https://doi.org/10.1523/JNEUROSCI.0415-15.2015

    • PubMed
    • Google Scholar
80. 1. Yao S
    2. Song J
    3. Gao L
    4. Yan Y
    5. Huang C
    6. Ding H
    7. Huang H
    8. He Y
    9. Sun R
    10. Xu G

    (2015) Thalamocortical sensorimotor circuit damage associated with disorders of consciousness for diffuse axonal injury patients

    Journal of the Neurological Sciences 356:168–174.

    https://doi.org/10.1016/j.jns.2015.06.044

    • PubMed
    • Google Scholar
81. 1. Yu YT
    2. Yang Y
    3. Wang LB
    4. Fang JL
    5. Chen YY
    6. He JH
    7. Rong PJ

    (2017) Transcutaneous auricular vagus nerve stimulation in disorders of consciousness monitored by fMRI: the first case report

    Brain Stimulation 10:328–330.

    https://doi.org/10.1016/j.brs.2016.12.004

    • PubMed
    • Google Scholar
82. 1. Zandbergen EG
    2. de Haan RJ
    3. Stoutenbeek CP
    4. Koelman JH
    5. Hijdra A

    (1998) Systematic review of early prediction of poor outcome in anoxic-ischaemic coma

    The Lancet 352:1808–1812.

    https://doi.org/10.1016/S0140-6736(98)04076-8

    • PubMed
    • Google Scholar

Author details

1. Ming Song

   1. National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China
   2. Brainnetome Center, Institute of Automation, Chinese Academy of Sciences, Beijing, China

   Contribution

   Conceptualization, Software, Formal analysis, Validation, Visualization, Methodology, Writing—original draft, Writing—review and editing

   Contributed equally with

   Yi Yang

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-8846-5707

2. Yi Yang

   Department of Neurosurgery, PLA Army General Hospital, Beijing, China

   Contribution

   Conceptualization, Data curation, Validation, Investigation, Writing—original draft, Writing—review and editing

   Contributed equally with

   Ming Song

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-3096-0312

3. Jianghong He

   Department of Neurosurgery, PLA Army General Hospital, Beijing, China

   Contribution

   Resources, Data curation, Investigation, Writing—original draft, Writing—review and editing

   Competing interests

   No competing interests declared

4. Zhengyi Yang

   1. National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China
   2. Brainnetome Center, Institute of Automation, Chinese Academy of Sciences, Beijing, China

   Contribution

   Formal analysis, Methodology, Writing—original draft, Writing—review and editing

   Competing interests

   No competing interests declared

5. Shan Yu

   1. National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China
   2. Brainnetome Center, Institute of Automation, Chinese Academy of Sciences, Beijing, China

   Contribution

   Methodology, Writing—original draft, Writing—review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-4385-6306

6. Qiuyou Xie

   Centre for Hyperbaric Oxygen and Neurorehabilitation, Guangzhou General Hospital of Guangzhou Military Command, Guangzhou, China

   Contribution

   Data curation, Validation

   Competing interests

   No competing interests declared

7. Xiaoyu Xia

   Department of Neurosurgery, PLA Army General Hospital, Beijing, China

   Contribution

   Data curation

   Competing interests

   No competing interests declared

8. Yuanyuan Dang

   Department of Neurosurgery, PLA Army General Hospital, Beijing, China

   Contribution

   Data curation

   Competing interests

   No competing interests declared

9. Qiang Zhang

   Department of Neurosurgery, PLA Army General Hospital, Beijing, China

   Contribution

   Data curation

   Competing interests

   No competing interests declared

10. Xinhuai Wu

    Department of Radiology, PLA Army General Hospital, Beijing, China

    Contribution

    Data curation

    Competing interests

    No competing interests declared

11. Yue Cui

    1. National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China
    2. Brainnetome Center, Institute of Automation, Chinese Academy of Sciences, Beijing, China

    Contribution

    Methodology, Writing—original draft

    Competing interests

    No competing interests declared

12. Bing Hou

    1. National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China
    2. Brainnetome Center, Institute of Automation, Chinese Academy of Sciences, Beijing, China

    Contribution

    Writing—original draft, Writing—review and editing

    Competing interests

    No competing interests declared

13. Ronghao Yu

    Centre for Hyperbaric Oxygen and Neurorehabilitation, Guangzhou General Hospital of Guangzhou Military Command, Guangzhou, China

    Contribution

    Project administration

    Competing interests

    No competing interests declared

14. Ruxiang Xu

    Department of Neurosurgery, PLA Army General Hospital, Beijing, China

    Contribution

    Project administration

    For correspondence

    zjxuruxiang@163.com

    Competing interests

    No competing interests declared

15. Tianzi Jiang

    1. National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing, China
    2. Brainnetome Center, Institute of Automation, Chinese Academy of Sciences, Beijing, China
    3. CAS Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, Beijing, China
    4. Key Laboratory for Neuroinformation of the Ministry of Education, School of Life Science and Technology, University of Electronic Science and Technology of China, Chengdu, China
    5. Queensland Brain Institute, University of Queensland, Brisbane, Australia

    Contribution

    Funding acquisition, Writing—original draft, Project administration, Writing—review and editing

    For correspondence

    jiangtz@nlpr.ia.ac.cn

    Competing interests

    No competing interests declared

    "This ORCID iD identifies the author of this article:" 0000-0001-9531-291X

• 4,439

  views

• 715

  downloads

• 66

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.