Paralysis caused by spinal cord injury, amyotrophic lateral sclerosis (LAS), brainstem stroke, cerebral palsy, and other disorders of the nervous system, has a severe impact on the lives of millions of individuals world-wide. While many assistive technologies have been developed for this patient population, the emerging class of technologies known as brain-machine interfaces (BMI) has a seemingly unprecedented potential to improve their quality of life.
In the domain of non-invasive BMI control, classification of imagined finger movement offers the greatest promise for drastically increase degrees of freedom of current systems. However, no classifier has yet been invented that could perform this task with significantly greater than chance probabilities. Currently known solutions can only provide accuracy in distinguishing between brain states driven by large regions of activation. Moreover, the research literature in this area is populated exclusively by reports on brain activity associated with actual finger movements, in contrast to the BMI aim of decoding brain activity associated with imagined movements.
Herewith a new method of spatiotemporal and spectral EEG feature extraction was developed, the “Spiral CovaWave” method, followed by the employment of a multi-class support vector machine (SVM) classifier that generates predictions and probabilities that serve as input into a new voting scheme which outputs a system decision. The method, motivated by domain-specific knowledge of the physiological encoding of imagined finger.
- Specific and highly accurate classifier
Applications and Opportunities
- Brain-Machine Interface (BMI) applications