Predict Pain
Research type
Research Study
Full title
Electroencephalographic predictors of central neuropathic pain in subacute spinal cord injury
IRAS ID
280703
Contact name
Aleksandra Vuckovic
Contact email
Sponsor organisation
Greater Glasgow & Clyde NHS Board
Clinicaltrials.gov Identifier
Duration of Study in the UK
2 years, 6 months, 0 days
Research summary
Summary of Research
Spinal cord injury (SCI), induced by damage to the spinal cord, can cause life-altering levels of disability including the development of chronic pain. Central Neuropathic Pain (CNP) typically develops within months after injury in 40-50% of SCI patients, affecting everyday activity, sleep and mood. There is no cure for CNP, it can be very difficult to treat and is often refractory to any pharmacological treatments.In a previous study (study no. 14/WS/1029) the principle investigators showed that the likelihood of CNP developing can be predicted by defining characteristics of brain waves that are related to pain. We will use electroencephalograph (EEG) to measure brain activity in people early after SCI, before they develop pain, knowing that about half will develop pain within a year. We aim to recruit 80 participants, aged 18-80; 40 with subacute spinal injury (level C3-T12) and no symptoms of CNP; 20 with symptoms of CNP and 20 able-bodied participants. Completeness of injury is irrelevant. Patients will be recruited by clinical consultants within national spinal units in Glasgow and Stoke Mandeville. Patients will undergo two EEG recording sessions in which they will imagine movements while we record EEG. Sessions will also involve basic sensory testing and completion of questionnaires. Able-bodied participants will be recruited by the PhD candidate at the University of Glasgow and undergo only one EEG session (identical to SCI patients).
The primary aim of this study is to use early EEG markers of CNP to optimise and validate an existing computer program based on machine learning to enable more accurate prediction of pain in newly injured patients with the hope of aiding future treatments. Secondary aims include characterising EEG features which might describe different phases in patients’ development of CNP and exploring possible differences between pain at/below the level of SCI based on EEG markers.
Summary of Results
1. Confirmation of EEG Pain Markers in Spinal Cord Injury An important part of this project has been to use the electroencephalograph (EEG) data collected in this study to confirm known markers of central neuropathic pain (CNP) and to identify any new ones. We looked at both oscillatory features, based on frequency power spectrum, and non-linear features of pain, using Higuchi’s fractal dimension (HFD). HFD is a common feature within the EEG community that’s used to characterise various neurological disorders, including pain, which we’ve discovered during this project.
We were able to demonstrate in our new dataset, that people with SCI that go on to develop CNP within 6 months of injury, have a shifted dominant peak in their brain frequency responses during, when compared to able-bodied people and people with SCI who do not develop pain.
With respect to HFD features, people with SCI and CNP tend to have increased complexity of their EEG signals compared to people that either don’t develop pain or able-bodied controls. This increased signal complexity is consistent across both our resting state EEG and EEG recorded during movement imaginations.
Given that all recordings are taken from participants shortly after injury, these markers are present in the people that develop pain before they develop any physical symptoms. This suggests that brain markers could be used for both prognosis and diagnosis of CNP.2. Predicting Pain After Spinal Cord Injury Using EEG Markers We wanted to see if certain brain activity patterns, measured using EEG, could reliably predict whether people with recent SCI would develop chronic neuropathic pain in the future. To do this, we analysed two separate groups of patients: one that had already been collected with 20 people the new dataset collected in this study (35 people). None of the participants had pain when their brain activity was recorded, but some went on to develop pain within six months, while others did not.
We focused on specific features of the brain waves, such as the strength of certain frequency bands and a measure called the Higuchi fractal dimension (HFD), which captures the complexity of brain activity.
We tested how well these features could distinguish between people who developed pain (PDP) and those who didn’t (PNP) in three scenarios: 1. Separately analysing each group; 2. Combining both groups; 3. Using one group for training the machine learning model and the other for testing it.
We also tried a new way to adjust the HFD features to make the data more comparable across participants.
When we trained and tested the model on the same group, we could correctly identify who would develop pain more than 80% of the time. Combining the two groups improved accuracy slightly, with the best results (86.4%) achieved using the HFD features and a specific type of model called a support vector machine (SVM).
When testing how well the model worked on new data, the accuracy dropped to 66.6%, however this is still much better than random guessing. The HFD features, especially when adjusted with our new normalisation method, seemed to give the most consistent results.
We found promising signs that brain activity patterns can predict future pain after SCI however more data would help to confirm or improve these results. One type of machine learning model, the SVM, worked particularly well with HFD features, particularly when normalising the data.REC name
West of Scotland REC 1
REC reference
20/WS/0135
Date of REC Opinion
11 Nov 2020
REC opinion
Further Information Favourable Opinion