Tuned in: tracking epilepsy with melody analysis

A new European project is turning brain pattern data into music to help spot the anomalies that could lead to an epileptic seizure. The project's researchers explain how they are tapping into the universal language of melody to improve our understanding of one of the world's most common neurological disorders.


The project is investigating whether tying EEG data sets to musical melodies could help researchers

Of all the organs that make up the human body, the brain is undoubtedly the most complex and the least well-understood. The relative mystery surrounding the human brain underscores so many unanswered questions about its influence over every aspect of human function and behaviour.

This mystery makes the treatment of neurological and neuropsychiatric disorders a much harder proposition than the challenges faced by those in many other medical fields. With researchers still struggling to get to grips with the brain's fundamental inner workings, the treatment or management of illnesses such as dementia, autism and Parkinson's disease remains an uphill struggle for medical institutions around the world.

Epilepsy is one such disorder. Around three percent of people around the world will be diagnosed with epilepsy in their lifetimes, and people in resource-poor countries are particularly at risk - nearly 90% of the approximately 50 million people who have epilepsy live in the developing world.

But even among highly-industrialised nations with the benefit of advanced medical facilities, epilepsy is a serious problem - a 2002 clinical audit carried out by the UK's National Institute for Health and Clinical Excellence (NICE) found that epilepsy causes around 1,000 deaths every year in the UK, concluding that the majority of these deaths were associated with seizures, and that 42% of deaths were potentially avoidable.

Data sonification: predicting epileptic seizures

"Lots of data is required to build a reliable database of melodies and a melodic baseline."

Given that most epilepsy-related deaths are caused by seizures that often strike out of the blue, researchers around the world have been working for years to develop methods of predicting imminent epileptic events.

In the US, NeuroVista is working on an implantable device that constantly monitors electroencephalogram (EEG) data to detect oncoming seizures, and other major seizure-predicting studies are ongoing in Australia, France and at US universities. While some of these studies are producing promising results, the extreme difficulty of accurately reading and interpreting EEG data, which records the brain's electrical activity, is a recurring problem.

A just-launched European research project, spearheaded by the Italian Association for the Research on Brain and Spinal Cord Diseases (ARCEM) and the Mario Negri Institute for Pharmacological Research, is taking an outside-the-box approach to the problem. The project has pulled together neuroscientists, IT specialists, musicians and music analysts in an attempt to predict impending seizures using a method called data sonification.

In simple terms, data sonification is the process of expressing visual data, like EEG read-outs, in audio form. The project is investigating whether tying EEG data sets to musical melodies could help researchers, and eventually doctors, spot the abnormal brain activity that prefaces a possible seizure.

One of the project's neuroscientists, Dr Massimo Rizzi, explains: "In the context of epilepsy research, the change of hidden temporal patterns associated to the brain's electrical activity, as represented by the EEG, may characterise the transition from apparently normal brain function to a state which may precede the occurrence of a seizure. The sonification technique may help highlighting and characterising the temporal patterns embedded in the EEGs of epileptic sufferers, providing a powerful tool for seizure prevention, ameliorating the daily life of patients and promoting research for new therapeutic interventions."

Spotting the bum note

But what advantages does turning EEG data into streams of melody bring over the current methods of studying visual brain pattern data to predict epileptic seizures? According to the project team, turning visual data into audio melodies could help researchers sequence and conceptualise the brain's activity over time, increasing the possibility of catching the hidden signs that could signal a seizure. These signs would be expressed as an abnormal or jarring sound in the melody, essentially turning seizure prediction into a clinical search for a bum note.

"The human ear is naturally trained to analyze series of data, detect anomalies and spot irregularities."

For Dr Domenico Vicinanza, a theoretical physicist and composer who is co-ordinating the project's sonification team, the ability to study the changes in brain signals over a period of time is a major advantage over existing methods of tracking epilepsy, and could potentially lead to the prediction of seizures further in advance than has previously been thought possible.

"The musical marker approach is expected to be more accurate than traditional visual techniques because the changes occurring in the electrical activity of the epileptic brain can be better described in terms of temporal patterns," he says. "Those patterns can hardly be expressed by analytical techniques like the ones based on the Fourier transform. In fact, tools based on Fourier transform are designed to describe any signal as a summing up of different independent waveforms, thus not taking in consideration any temporal information embedded in the signal."

Data sonification also brings a more human advantage - our intrinsic ability to spot the bum note in a melody. "The human ear is, to some extent, naturally trained to analyze series of data, detect anomalies and spot irregularities," says Vicinanza. "We can say that the ear constantly performs auditory data analysis, for example, every time we recognise a person on the phone. The ear naturally recognises patterns, structures and sequences. If we are searching for a particular value which, for some reason, is distinguished from a series of data, identifying it on a graph can be difficult, while finding it through a melody could result in a much easier task. Everybody can spot a wrong note in a song, even without any special music training."

The melody analysis project timeline

"The project has pulled together neuroscientists, IT specialists, musicians and music analysts."

Although the theories behind data sonification and melody analysis for predicting seizures sound promising, the project team is in the very early stages of its work, which involves the collection, sonification and analysis of huge volumes of data, combining sonification techniques with recurrence quantification analysis - all of which takes a great deal of time.

Vicinanza says the team will have more concrete information on the validity of the technique in the coming months, and hopes to publish the first set of preliminary results by the end of 2013. "One of the biggest challenges is the fact that this is an almost green field," he says. "We are working hard designing and testing sonification algorithms, effective mappings between EEG data and notes. Lots of data is required to build a reliable database of melodies and hopefully build a sort of melodic baseline that can lead to the identification of music markers."

One technological innovation that is speeding up the arduous process is the introduction of high-speed research networks. The project is receiving support from the pan-European GÉANT data network, which allows the instant transfer of massive data streams between researchers and analysts throughout Italy and Europe.

"Only a few years ago many scientific researchers relied on physically transporting data cassette tapes for analysis," Vicinanza says. "Today, thanks to the high-speed research and education networks, such as GÉANT and its NREN [National Research and Education Network] partners, this data can be transferred in real time between scientists and researchers located at different institutes and in different countries."

While Rizzi is confident in the data sonification technique's utility for clinical epileptologists, he says it's still hard to foresee when the data sonification technique could be ready for clinical trials and introduction to clinics and hospitals. For now, the team is gathering as much data as possible to confirm that its technique is worth pursuing.

For Vicinanza, the ultimate goal would be to bring the benefits of data sonification to a wide range of hospitals and research organisations. "I feel this could be a good candidate for a hospital-based system and a research tool. I expect data sonification to be a technique supporting physicians, doctors and researchers, providing complementary information and an alternative perspective to the data they analyse every day."

Music, like mathematics, is one of the world's universal languages. Rhythms and melodies transcend linguistic barriers and speak to a shared space in the human psyche. Although it's still too early to prove the effectiveness of the data sonification technique for predicting seizures, the beauty of ARCEM's project is in making use of this collective musical mother tongue to bring complex clinical data down to a more human level.

Related content

The tangled web: Münchausen by Internet and the threat to online health forums

Münchausen by Internet, which involves online posters who deceive health forums with fictitious stories of illness and bereavement, is an emerging phenomenon that sheds light on the spaces where medicine and social media collide.

Kinect evolved: stroke recovery with Microsoft's motion sensor

Could Microsoft's Kinect drive a revolution in home-based stroke recovery through the use of affordable motion tracking?

Follow Chris Lo on Google+