Meeting Banner
Abstract #1109

Multivariate Discrimination of Changes in Regional Cerebral Blood Flow Representing Ongoing Post-Surgical Pain Using Gaussian Process Classification.

Matthew A. Howard1, Jonathan O'Muircheartaigh2, Kristina Krause1, Nathalie Massat3, Nadine Khawaja1, John P. Huggins4, William Vennart4, Tara F. Renton5, Andre Marquand1, Steven C R Williams1

1Neuroimaging, King's College London, Institute of Psychiatry, Camberwell, London, United Kingdom; 2Clinical Neuroscience, King's College London, Institute of Psychiatry, United Kingdom; 3Centre for Cancer Prevention, Queen Mary University Of London; 4Pfizer Global Research and Development; 5Dental Institute, King's College London, United Kingdom

Recent reports have described the application of arterial spin labelling (ASL), to interrogate perfusion changes associated with the central representation of ongoing pain. We used Gaussian Process Classification, a supervised 'machine learning' multivariate analysis technique, to provide probabilistic classification of 'No Pain' from 'Ongoing Pain' states, as experienced following wisdom tooth extraction, using only ASL-derived indices of regional cerebral blood flow (rCBF) in each state. GPC classified between states with accuracy above 90%; 80% accuracy could be maintained using only two rCBF maps per state. This methodology has potential to impact on efficient economic assessment of novel treatments for pain.

Keywords

accuracies accuracy amount analgesics applied arterial assay ates best bilateral blood brain cancer cerebral classification classified clinically college computed conjunction correctly counting cross decision dental detected development domain early efficient examinations exams exciting exhaustively extraction flow free handed individual individuals institute john king kingdom labels learning leave linear making males maps methodologies model multivariate nation novel ongoing owing pain pendent period permutation permuted persistent phys post potential prevention probabilistic pulsed queen randomly recovery recruited reduced regional relatedness requiring resting robust sensitive separate session side sided sing spin subject surgeries surgery surgical take therapeutics third tooth trained underlying variate whole