Combining brain state classification and classical conditioning for basic BCI communication
(2012) Organization for Human Brain Mapping — Location: Beijing, China (10.June.2015)
Files
No attached file found for this publication.
Details
- Authors
- Abstract
- Introduction: Brain-computer interfaces (BCI) provide alternative methods for communicating and acting on the world, by conveying messages and commands without using the normal output pathways of peripheral nerves and muscles (Birbaumer et al., 1999; Pasqualotto et al., 2011; Wolpaw et al., 2002). Patients with Alzheimer's disease (AD) would benefit from a BCI that is able to convey information about their basic thoughts and emotions. The possibility to discriminate between emotional states by pattern classification of BOLD signals has been demonstrated in both offline (Lee et al., 2010) and online (Sitaram et al., 2011) situations. A possible way to develop a BCI that can be used by AD patients is through the modulation of cerebral responses with a semantic classical conditioning paradigm (Furdea et al., 2011), moving away from the more common operant conditioning paradigm, which requires subjects to actively self-regulate their brain activation (Birbaumer, 2006). The aim of our study is to condition subjects to associate emotionally negative and positive stimuli as unconditioned stimuli (US) with incongruent and congruent word pairs (eliciting "negative" and "affirmative" thinking) as conditioned stimuli (CS), respectively. We investigated whether brain signals related to congruent and incongruent word pairs could be classified with more than chance accuracy, in view of an application for basic yes/no communication in AD patients. Methods: The paradigm consisted of six blocks comprising the different phases of conditioning (habituation, acquisition, extinction) (Fig.1). The US, drawn from the International Affective Digitized Sounds (Bradley & Lang, 1999), consisted of a scream and a baby laugh, representing a negative and a positive emotional sound respectively. The CS, presented aurally, were congruent (e.g. 'animal-dog') and incongruent (e.g. 'animal-chair) word pairs. The unconditioned and conditioned responses (UR and CR) were the changes in the BOLD signal related to the CS and US. In the first block, 50 US and 50 CS were presented randomly. In the second and third blocks, 25 congruent word pairs, followed by the baby laugh, and 25 incongruent word pairs, followed by the scream, were presented randomly. In the fourth and fifth blocks, respectively 40% and 20% of the CS were paired with the US. In the sixth block, only the CS was presented. Functional imaging was performed on 11 healthy subjects (6 females, 5 males, age 21-28) on a 3T scanner. To classify the signals corresponding to congruent and incongruent word pairs, both univariate (General Linear Model) and multivariate (Support Vector Machine, SVM) analyses were performed. Results: The data that emerges from the univariate analysis shows that the classical conditioning allows a differentiation of "yes" and "no" responses. Interestingly, the differential activations took place in brain areas such as insula, superior temporal gyrus and the superior frontal gyrus, which are recognized to be involved in emotional processing (Sitaram et al., 2011). Results from SVM analyses, performed after feature selection from the third, fourth and fifth image after each word pair (temporal-spatial voxel selection performed with a threshold of 7.5% for the percentage of univariately separable samples), indicate that it is possible, after classical conditioning, to discriminate between affirmative and negative responses with more than chance accuracy. Conclusions: The present results are encouraging and show that basic yes/no discrimination may be possible within an fMRI-BCI setting. This discrimination may be obtained using a "passive" procedure, such as classical conditioning, which does not require subjects to be actively involved in a task, and can be therefore used with patients with dementia. A further step, which is already in progress, is the testing of the paradigm with AD patients.
- Affiliations
Citations
Liberati, G., Van der Heiden, L., Veit, R., Dalboni da Rocha, J., Kim, S., Raffone, A., Birbaumer, N., Olivetti Belardinelli, M., & Sitaram, R. (2012). Combining brain state classification and classical conditioning for basic BCI communication. Organization for Human Brain Mapping. Published. Organization for Human Brain Mapping, Beijing, China. https://hdl.handle.net/2078.5/194720 (Original work published 2012)