Brain-computer interface in disorders of consciousness: answering simple questions with a P3 speller

Objective: In the recovery from coma, the acquisition of command following represents an important milestone, indicating emergence from the vegetative state (Schnakers et al., 2009). In some patients, recovery of consciousness may precede motor recovery. Brain-computer interfaces (BCI) might permit these patients to show non-motor dependent signs of awareness and in a next step might enable communication. This study aimed at testing to what extent an EEG-based BCI could help detecting signs of awareness and communication in disorders of consciousness. We employed a P300 based BCI where healthy volunteers and patients with locked-in syndrome and in a minimally conscious state were asked to answer yes or no to simple questions by paying attention to one out of four auditorily presented stimuli (‘yes’, ‘no’, ‘stop’, ‘go’).
Methods: We studied 13 patients with a minimally conscious state (MCS, 5 TBI – 8 anoxic, mean time post injury 70±109 months; mean age 42 ± 21) and 2 in pseudo-coma or locked in syndrome (LIS; brainstem stroke, time post injury 26 and 46 months; aged 63 and 29)) and 16 healthy controls (aged 45±19). Patients were evaluated using the Coma Recovery Scale Revised (CRS-R). An auditory P300 four choice speller paradigm (Furdea et al., 2009) based on the BCI2000 system (Schalk et al., 2004) was used. 16-Channel EEG was recorded using a g.tec USBAmp amplifier. A trial constituted of 15 presentation of four sounds the order of presentation being pseudo-randomized (sound duration: ~400ms; inter-stimulus interval: ~600ms). After a training session of 4 trials, patients and healthy subjects were required to answer 10 or 12 questions, respectively. Questions were of the following kind: “Is your name Quentin?”, “Is your mother’s name Dorothée?”. A stepwise linear discriminant analysis based on the training session was used to classify the data and to provide online feedback. Offline, all training and testing sequences were pooled. Sequences with artifacts were discarded and a leave-one-out approach was used to classify the data.

Results: Healthy subjects presented a mean correct response rate of 73% online and 93% offline. Note that online classification failed for one control subject due to a presumed change in cognitive strategy between training and testing sessions. LIS patients showed a correct response rate of 30 and 60% (online) and 36 and 79% (offline). Three MCS patients had a correct response rate of ≥50% offline (10, 18, 0% online and 50, 53, 57% offline). Two of these three patients did not show any command following at the bedside. The 10 remaining MCS cases showed online and offline correct answers <50% (mean 33±9% online and 25±13% offline).

Conclusion: Our auditory P300-based BCI permitted functional interactive communication in 15/16 controls (online) and in all offline. Our data obtained in patients with locked-in syndrome and disorders of consciousness demonstrate the potential clinical usefulness of the technique following coma but also show lower accuracy in patients as compared to healthy volunteers. This might be due to fluctuating attentional levels and exhaustibility in the MCS and to the suboptimal EEG recording quality due to movement, ocular and respiration artifacts in these challenging patients. Further algorithmic developments should include automatic artifact detection and single trial classification. Despite the need for further improvement in BCI devices adapted to post-coma patients, our results already indicate that MCS patients without any clinical sign of command-following can employ a yes-no speller offering the hope of functional interactive communication and a possibility for decision making and autonomy.
Bibliography
Furdea A, Halder S, Krusienski DJ, Bross D, Nijboer F, Birbaumer N, Kübler A, 2009, An auditory oddball (P300) spelling system for brain-computer interfaces, Psychophysiology. May; 46(3):617-25.
Schalk G., McFarland D.J., Hinterberger T., Birbaumer N., and Wolpaw J.R. 2004, BCI2000: A General-Purpose Brain-Computer Interface (BCI) System, IEEE Trans Biomed Eng, 51(6).
Schnakers C, Vanhaudenhuyse A, Giacino J, Ventura; Boly M, Majerus S, Moonen G, Laureys S, 2009, Diagnostic accuracy of the vegetative and minimally conscious state: Clinical consensus versus standardized neurobehavioral assessment, BMC Neurology, 9 (35).