The Berlin Brain-Computer Interface – Single trial classifications of phantom finger movements of arm amputees

Introduction: The Berlin Brain-Computer interface (BBCI) is based on single-trial classifications of multichannel EEG signals, such as lateralized readiness potentials (LRP), discriminating right vs. left limb movements with accuracies up to 95% in healthy subjects. The current study used 128-channel EEG recordings in patients with traumatic amputations of one arm or hand to compare EEG correlates of real vs. phantom finger/hand movements. Objective: We studied LRPs and event-related desynchronizations (ERDs) associated with phantom movements of amputees over the contralateral primary motor cortex, and analyzed the single-trial classification with the BBCI standard EEG-classifier to clarify if classification rates are as accurate as shown earlier in comparable studies with healthy subjects. Methods: We studied eight patients (1 female, 7 male; 37–74 years) with amputations between 16 and 54 years ago. A digital metronome played alternately two distinct sounds in a steady rhythm, with intersound intervals between 1 and 1.4s, as chosen by each patient for convenient performance. Concomitant with the higher sound, the patient had to perform either a finger tap on a keyboard using the healthy hand or a phantom movement with a phantom finger/hand. The lower sound called for rest. Results: 1) In 5 of 8 patients we found a typically shaped contralateral LRP associated with the phantom movement. The younger patients, with more recent amputations, tended to show better defined LRP topographies. 2) While real movements were preceded by standard ERDs, phantom finger/hand movements showed a less distinct ERD. 3)The 'right vs. left' single-trial classifications performed with the standard BBCI-classifier yielded hit rates between 60–78%. 4) The subgroup with LRPs associated with phantom movements showed hit rates in the single-trial classification not better than the group without lateralization of RP. Conclusions: Patients with traumatic amputations show lateralized RPs associated with phantom limb movements. LRP topographies appeared better defined in cases with more recent amputations. Interestingly, even with no lateralized RP, classifications well above chance level were still possible, but the underlying physiology needs further study. Together, these findings indicate a possibility to exploit lateralized RPs for BCIs in amputees, for example, to control a prosthesis. At present, the hit rates are still too low for such BCI to be useful in the everyday life of amputees.