Improving EEG reading of brain states for clinical applications using a data-driven joint model of FMRI and EEG

In recent years researchers have learned a great deal about the function of the human brain through neuroimaging techniques. Different techniques have their own strengths and weaknesses and each offers a window on brain function with a different perspective. Two of the most common methods for measuring the amount and location of brain activity associated with different sensations, thoughts and feelings are electroencephalography (EEG) and functional magnetic resonance imaging (fMRI). EEG records from electrodes attached to the scalp the electrical signals from co-ordinated activity of large numbers of nerve cells. FMRI, however, records using an MRI scanner, the local changes in blood oxygenation associated with alterations in neural activity. EEG has the advantage of being able to detect rapid changes in neural activity (millisecond temporal resolution) but suffers from a poor ability to pinpoint the location of brain activity (spatial resolution). FMRI, however, has good spatial resolution (a few millimetres) but poor temporal resolution (a few seconds) because the signal relies on changes in blood flow: the plumbing of the brain. FMRI and EEG can therefore be regarded as complementary with EEG giving the 'when' and fMRI giving the 'where' of brain activity.However, while very useful in research, doctors and scientists want also to develop these neuroimaging techniques for practical uses which rely on reading the state of the brain or measuring the activity of the brain. These include, but are not limited to, brain-computer interfaces (BCI, disabled patients using brain signals to control a device), assessment of the effects of new medicines targeted at the brain and the diagnosis of epilepsy (the type and source of seizures from within the brain). EEG has been used, for many years in some cases, in these applications and has the considerable advantage of being portable and comparatively cheap and therefore appropriate for a routine lab or clinical setting. Why is the usefulness of EEG limited? As we have seen, its spatial resolution is comparatively poor but it can also be insensitive because of many signals from the brain being present and mixing together. FMRI is a more recent technique able to discriminate very well different patterns of brain activity but requires an MRI scanner: clearly not portable and comparatively expensive. In research labs such as ours at Cardiff University Brain Research Imaging Centre (CUBRIC), it has become possible to perform EEG and fMRI simultaneously. Our research proposal aims to improve the ability of EEG to discriminate different brain states or responses to specific types of stimulation, such as pain, drugs or for control of BCIs. To exploit the day-to-day practical advantages of EEG we wish to improve its stand alone capabilities. We will use fMRI in this project to help us do this. How can fMRI help us to improve EEG? We will use EEG and fMRI measurements acquired simultaneously on healthy volunteers. We will relate these two types of measurements together in what it known as a statistical model derived from the data. This procedure will discover associations or correlations between the EEG and fMRI data. Subtle features of the EEG signal, which are not normally easily identified but which are associated with the spatial location of the source of neural activity, will be highlighted by their association with the fMRI data, which is good at pinpointing locations in space. Having established and codified the relationship between the EEG and fMRI data in mathematical terms, EEG data alone will be used to simulate fMRI scans. These simulated fMRI scans will be used, applying what we know about the representation of brain activity by fMRI, to interpret the EEG signal effectively improving its spatial resolution. This will improve the ability of EEG on its own to tell the difference between brain states for the uses in BCI, development of medicines and clinical conditions.

There is potential for developing new mobile imaging techniques and equipment for brain analysis based on EEG and joint FMRI/EEG models, which would bring some of the advantages of the FMRI technology closer to the patient without the need for FMRI equipment. There is potential for a spin-off company taking advantage of the developed techniques. Cardiff University has a dedicated Research and Commercial Division with all the necessary systems in place for IP exploitation and a strong history of doing this successfully. The project has the long term potential to benefit the commercial private sector, the public healthcare sector and also charities concerned with the severely disabled, for whom brain computer interfaces may have benefits. Any improvement in EEG signal inversion algorithms that would permit categorisation of more than simplest tasks/classes would be of huge benefit to the brain computer interface (BCI) industry. Improved BCIs may have a role in helping impaired patients communicate. Such improved algorithms would be disseminated through presentation at well targeted industry meetings and would also be protected for potential commercial exploitation. There are additional healthcare benefits of successful EEG algorithms. EEG assists in the clinical diagnosis of different forms of epilepsy. The development of improved EEG models for identifying subtypes of epilepsy and their spatial distribution within the brain by the use of FMRI may permit these improved EEG models to be applied in the clinic. So far the use of combined EEG-FMRI has been largely to pinpoint epileptic foci. However, the use of FMRI to train a greater discriminative EEG procedure may have benefit at a lower cost for a greater number of patients. A long term impact could be the improved diagnosis, localisation and assessment of drug treatment (and side effects) in epilepsy. This potential benefit would initially be realised through collaboration with our clinical colleagues in the NHS (Dr Hamandi and his group and the Welsh Epilepsy Research Network) and through charitable funding agencies with which CUBRIC has an ongoing relationship (e.g. Waterloo Foundation). The UK pharmaceutical industry, in common with that in the rest of the world is experiencing ever increasing costs of bringing new therapeutic compounds to market. This threatens their long term success. Effective measurement technologies are needed to quantify drug effects, especially in the brain for neurological and psychiatric indications. Wise has ongoing collaborations with the pharmaceutical industry e.g. Pfizer UK Ltd, and previously Merck and GSK, to develop EEG and FMRI for assessing the effects of novel drugs in the human brain. These technologies are therefore beginning to assist in drug discovery and development. While FMRI is proving useful in this regard, EEG has the advantage of being more practical and widely available and measures neuronal activity more directly. Within Wise's group EEG and FMRI is being used to assess drug action especially in the field of pain, analgesia and sedation where we have experience. Wise meets regularly with those concerned with assessing technology in the UK pharmaceutical industry and therefore has a route to disseminating this research. Clinical monitoring of awareness and depth of sedation: EEG-based monitoring systems exist for this (e.g. bi-spectral monitoring). However, the development of such EEG devices would benefit from a stronger neuroanatomical and systems-neuroscience foundation which could be provided by the proposed contribution to EEG analysis from FMRI data. CUBRIC works closely with anaesthetists in researching the brain in sedation using EEG-FMRI. We therefore have the opportunity to test algorithms directly and develop our approaches with clinical practitioners as well as in collaboration with a commercial partner with whom we have had early discussions on depth of anaesthetic monitoring (ANT, Berlin).