Damien Lesenfants

Biased binomial assessment of cross-validated estimation of classification accuracies illustrated in diagnosis predictions

Multivariate classification is used in neuroimaging studies to infer brain activation or in medical applications to infer diagnosis. Their results are often assessed through either a binomial or a permutation test. Here, we simulated classification results of generated random data to assess the influence of the cross-validation scheme on the significance of results. Distributions built from classification of random data with cross-validation did not follow the binomial distribution. The binomial test is therefore not adapted. On the contrary, the permutation test was unaffected by the cross-validation scheme. The influence of the cross-validation was further illustrated on real-data from a brain–computer interface experiment in patients with disorders of consciousness and from an fMRI study on patients with Parkinson disease. Three out of 16 patients with disorders of consciousness had significant accuracy on binomial testing, but only one showed significant accuracy using permutation testing. In the fMRI experiment, the mental imagery of gait could discriminate significantly between idiopathic Parkinsons disease patients and healthy subjects according to the permutation test but not according to the binomial test. Hence, binomial testing could lead to biased estimation of significance and false positive or negative results. In our view, permutation testing is thus recommended for clinical application of classification with cross-validation.

The auditory P300-based single-switch brain-computer interface

OBJECTIVE Within this work an auditory P300 brain-computer interface based on tone stream segregation, which allows for binary decisions, was developed and evaluated. METHODS AND MATERIALS Two tone streams consisting of short beep tones with infrequently appearing deviant tones at random positions were used as stimuli. This paradigm was evaluated in 10 healthy subjects and applied to 12 patients in a minimally conscious state (MCS) at clinics in Graz, Würzburg, Rome, and Liège. A stepwise linear discriminant analysis classifier with 10×10 cross-validation was used to detect the presence of any P300 and to investigate attentional modulation of the P300 amplitude. RESULTS The results for healthy subjects were promising and most classification results were better than random. In 8 of the 10 subjects, focused attention on at least one of the tone streams could be detected on a single-trial basis. By averaging 10 data segments, classification accuracies up to 90.6% could be reached. However, for MCS patients only a small number of classification results were above chance level and none of the results were sufficient for communication purposes. Nevertheless, signs of consciousness were detected in 9 of the 12 patients, not on a single-trial basis, but after averaging of all corresponding data segments and computing significant differences. These significant results, however, strongly varied across sessions and conditions. CONCLUSION This work shows the transition of a paradigm from healthy subjects to MCS patients. Promising results with healthy subjects are, however, no guarantee of good results with patients. Therefore, more investigations are required before any definite conclusions about the usability of this paradigm for MCS patients can be drawn. Nevertheless, this paradigm might offer an opportunity to support bedside clinical assessment of MCS patients and eventually, to provide them with a means of communication.

Automated Analysis of Background EEG and Reactivity During Therapeutic Hypothermia in Comatose Patients After Cardiac Arrest

Visual analysis of electroencephalography (EEG) background and reactivity during therapeutic hypothermia provides important outcome information, but is time-consuming and not always consistent between reviewers. Automated EEG analysis may help quantify the brain damage. Forty-six comatose patients in therapeutic hypothermia, after cardiac arrest, were included in the study. EEG background was quantified with burst-suppression ratio (BSR) and approximate entropy, both used to monitor anesthesia. Reactivity was detected through change in the power spectrum of signal before and after stimulation. Automatic results obtained almost perfect agreement (discontinuity) to substantial agreement (background reactivity) with a visual score from EEG-certified neurologists. Burst-suppression ratio was more suited to distinguish continuous EEG background from burst-suppression than approximate entropy in this specific population. Automatic EEG background and reactivity measures were significantly related to good and poor outcome. We conclude that quantitative EEG measurements can provide promising information regarding current state of the patient and clinical outcome, but further work is needed before routine application in a clinical setting.

Posterior Cingulate Cortex-Related Co-Activation Patterns: A Resting State fMRI Study in Propofol-Induced Loss of Consciousness

Background Recent studies have been shown that functional connectivity of cerebral areas is not a static phenomenon, but exhibits spontaneous fluctuations over time. There is evidence that fluctuating connectivity is an intrinsic phenomenon of brain dynamics that persists during anesthesia. Lately, point process analysis applied on functional data has revealed that much of the information regarding brain connectivity is contained in a fraction of critical time points of a resting state dataset. In the present study we want to extend this methodology for the investigation of resting state fMRI spatial pattern changes during propofol-induced modulation of consciousness, with the aim of extracting new insights on brain networks consciousness-dependent fluctuations. Methods Resting-state fMRI volumes on 18 healthy subjects were acquired in four clinical states during propofol injection: wakefulness, sedation, unconsciousness, and recovery. The dataset was reduced to a spatio-temporal point process by selecting time points in the Posterior Cingulate Cortex (PCC) at which the signal is higher than a given threshold (i.e., BOLD intensity above 1 standard deviation). Spatial clustering on the PCC time frames extracted was then performed (number of clusters = 8), to obtain 8 different PCC co-activation patterns (CAPs) for each level of consciousness. Results The current analysis shows that the core of the PCC-CAPs throughout consciousness modulation seems to be preserved. Nonetheless, this methodology enables to differentiate region-specific propofol-induced reductions in PCC-CAPs, some of them already present in the functional connectivity literature (e.g., disconnections of the prefrontal cortex, thalamus, auditory cortex), some others new (e.g., reduced co-activation in motor cortex and visual area). Conclusion In conclusion, our results indicate that the employed methodology can help in improving and refining the characterization of local functional changes in the brain associated to propofol-induced modulation of consciousness.

Volitional electromyographic responses in disorders of consciousness

Abstract Objective: The aim of the study was to validate the use of electromyography (EMG) for detecting responses to command in patients in vegetative state/unresponsive wakefulness syndrome (VS/UWS) or in minimally conscious state (MCS). Methods: Thirty-eight patients were included in the study (23 traumatic, 25 patients >1 year post-onset), 10 diagnosed as being in VS/UWS, eight in MCS− (no response to command) and 20 in MCS+ (response to command). Eighteen age-matched controls participated in the experiment. The paradigm consisted of three commands (i.e. ‘Move your hands’, ‘Move your legs’ and ‘Clench your teeth’) and one control sentence (i.e. ‘It is a sunny day’) presented in random order. Each auditory stimulus was repeated 4 times within one block with a stimulus-onset asynchrony of 30 seconds. Results: Post-hoc analyses with Bonferroni correction revealed that EMG activity was higher solely for the target command in one patient in permanent VS/UWS and in three patients in MCS+. Conclusion: The use of EMG could help clinicians to detect conscious patients who do not show any volitional response during standard behavioural assessments. However, further investigations should determine the sensitivity of EMG as compared to neuroimaging and electrophysiological assessments.

Detection of response to command using voluntary control of breathing in disorders of consciousness

Background: Detecting signs of consciousness in patients in a vegetative state/unresponsive wakefulness syndrome (UWS/VS) or minimally conscious state (MCS) is known to be very challenging. Plotkin et al. (2010) recently showed the possibility of using a breathing-controlled communication device in patients with locked in syndrome. We here aim to test a breathing-based “sniff controller” that could be used as an alternative diagnostic tool to evaluate response to command in severely brain damaged patients with chronic disorders of consciousness (DOC). Methods: Twenty-five DOC patients were included. Patients’ resting breathing-amplitude was measured during a 5 min resting condition. Next, they were instructed to end the presentation of a music sequence by sniffing vigorously. An automated detection of changes in breathing amplitude (i.e., >1.5 SD of resting) ended the music and hence provided positive feedback to the patient. Results: None of the 11 UWS/VS patients showed a sniff-based response to command. One out of 14 patients with MCS was able to willfully modulate his breathing pattern to answer the command on 16/19 trials (accuracy 84%). Interestingly, this patient failed to show any other motor response to command. Discussion: We here illustrate the possible interest of using breathing-dependent response to command in the detection of residual cognition in patients with DOC after severe brain injury.

Electromyographic decoding of response to command in disorders of consciousness

Objective: To propose a new methodology based on single-trial analysis for detecting residual response to command with EMG in patients with disorders of consciousness (DOC), overcoming the issue of trial dependency and decreasing the influence of a patients fluctuation of vigilance or arousal over time on diagnostic accuracy. Methods: Forty-five patients with DOC (18 with vegetative/unresponsive wakefulness syndrome [VS/UWS], 22 in a minimally conscious state [MCS], 3 who emerged from MCS [EMCS], and 2 with locked-in syndrome [LIS]) and 20 healthy controls were included in the study. Patients were randomly instructed to either move their left or right hand or listen to a control command (“It is a sunny day”) while EMG activity was recorded on both arms. Results: Differential EMG activity was detected in all MCS cases displaying reproducible response to command at bedside on multiple assessments, even though only 6 of the 14 individuals presented a behavioral response to command on the day of the EMG assessment. An EMG response was also detected in all EMCS and LIS patients, and 2 MCS patients showing nonreflexive movements without command following at the bedside. None of the VS/UWS presented a response to command with this method. Conclusions: This method allowed us to reliably distinguish between different levels of consciousness and could potentially help decrease diagnostic errors in patients with motor impairment but presenting residual motor activity.

Detecting Consciousness with a Brain-Computer Interface

Recent electrophysiological and neuroimaging studies showed command-specific changes in EEG or fMRI signals of unresponsive patients providing motor-independent evidence of conscious thoughts. These promising results have paved the way for a new application for Brain-computer Interface (BCI): detecting consciousness in patients with disorders of consciousness (DOC). In the present abstract, we review the first results obtained by BCI-like applications in patients with DOC and discuss the challenges facing BCI research. We believe that patients with DOC may benefit from BCI based diagnosis. BCIs may detect changes in the signal in response to command and, in some cases, may permit communication.

Cognitive processing in non-communicative patients: What can event-related potentials tell us?

Event-related potentials (ERP) have been proposed to improve the differential diagnosis of non-responsive patients. We investigated the potential of the P300 as a reliable marker of conscious processing in patients with locked-in syndrome (LIS). Eleven chronic LIS patients and 10 healthy subjects (HS) listened to a complex-tone auditory oddball paradigm, first in a passive condition (listen to the sounds) and then in an active condition (counting the deviant tones). Seven out of nine HS displayed a P300 waveform in the passive condition and all in the active condition. HS showed statistically significant changes in peak and area amplitude between conditions. Three out of seven LIS patients showed the P3 waveform in the passive condition and five of seven in the active condition. No changes in peak amplitude and only a significant difference at one electrode in area amplitude were observed in this group between conditions. We conclude that, in spite of keeping full consciousness and intact or nearly intact cortical functions, compared to HS, LIS patients present less reliable results when testing with ERP, specifically in the passive condition. We thus strongly recommend applying ERP paradigms in an active condition when evaluating consciousness in non-responsive patients.