PREDICTION OF PARKINSON'S DISEASE TREMOR ONSET USING ARTIFICIAL NEURAL NETWORKS S. Pan , K. Warwick , J. Stein , M.N. Gasson , S.Y. Wang , T.Z. Aziz , J. Burgess a a c a b,c c a a Department of Cybernetics, University of Reading, UK b University Laboratory of Physiology, University of Oxford, UK c The Functional Neurosurgery Group, Department of Surgery, Radcliffe Infirmary, Oxford, UK Corresponding author: e-mail: s.pan@reading.ac.uk Abstract In this paper we present the initial results using an artificial neural network to predict the onset of Parkinson’s Disease tremors in a human subject. Data for the network was obtained from implanted deep brain electrodes. A tuned artificial neural network was shown to be able to identify the pattern of the onset tremor from these real time recordings. Keywords Parkinson’s Disease, DBS, and Artificial Neural Networks 1. Introduction Signal transmission in the brain is electrochemical. This means that electrical activity which corresponds to normal or abnormal brain processes can potentially be recorded. Influencing neural function electrically is therefore a reality, indeed for movement disorders and pain, implanting deep brain electrodes to stop tremors or relieve pain has been conceptually possible since the 1960’s. However the difficulty of accurately targeting structures deep in the brain, lack of safe electrodes, and problems of miniaturising the necessary electronics were not overcome until the mid-1980’s. Now, implantation of deep brain stimulating electrodes is quite routine. The most common use of such technology is for the treatment of movement disorders such as Parkinson’s disease. This is partly because of the disabling long term side effects of the normal treatment with L-DOPA and because many movement disorders such as multiple system atrophy or dystonia do not respond to dopaminergic treatment at all. Even in those patients who do respond, after 5-10 years about half develop serious long term side effects such as dyskinesias, unpredictable ‘on-off’ switching and freezing [1] which are resistant to any other drug therapy [2]. Other side effects can even occur more rapidly [3]. Although functional neurosurgeons have found that lesioning the thalamus, sub-thalamus or globus pallidus can dramatically alleviate Parkinsonian symptoms such as tremor and dystonia, most have now moved towards implanting deep brain stimulating electrodes into these targets. [4]. Deep Brain Stimulation (DBS) can have similar effects to lesioning, yet, because it is usually largely reversible, it is less dangerous, especially when targeting small structures such as the sub-thalamic nucleus [5]. It is however a relatively expensive solution. Also, commercial stimulators utilise continuous stimulation at high frequencies (typically 100-180Hz for movement disorders and 5-50Hz for pain), with amplitudes in the region of 5mA. The net result is the need for battery replacement every 18 months or so, along with associated surgery depending, upon the disease treated. The accumulated cost and time of such therapies means that the number of patients that can benefit is limited. One solution is to pursue complex battery technology to prolong the life of the unit. However, an alternative is to design an ‘intelligent’ stimulator which can respond to indicative changes in neural activity prior to an event by delivering pulses to suppress it at the onset rather than continuously stimulating the brain. Such a device will both prolong the life of the battery by reducing the drain associated with constant stimulation and potentially produce a clinically superior outcome. At the Radcliffe Infirmary, Oxford, UK, deep brain electrodes are routinely implanted into the thalamus, pallidum or sub-thalamic nucleus to alleviate the symptoms of Parkinson’s disease, multiple sclerosis and dystonia. In pain patients, electrodes are implanted into the sensory thalamus or periventricular / periaqueductal grey area. The depth electrodes are then externalised for a week to ascertain effect prior to internalisation. A control unit and battery is then implanted in the chest cavity and the electrode connections internalised if good symptom relief is achievable. During the period prior to internalisation, an opportunity exists to record the local field potentials (LFPs) from the