Prediction of Parkinson’s Disease Tremor Onset Using Artificial Neural Networks Kevin Warwick
Road Map Parkinson’s Disease (PD) and Treatments Deep Brain Stimulation (DBS) Treatment on PD Tremor PD Brain Data Analysis Artificial Neural Network Prediction Future Research
Parkinson’s Disease What is Parkinson's Disease (PD)? Symptoms of Parkinson’s Disease Tremor, Slowness of movement, Rigidity, Poor balance How is PD caused in theory Substantia nigra Dopamine Sir William Richard Gowers (1886)
Current PD Treatment Surgical Drugs – Various forms of Levodopa (L-dopa) Advantage – effective for early PD patients, cheap Disadvantage – kyskinesias, unpredictable on-off switching and freezing, chemical side effects. Surgical Radio frequency Lesioning – non-reversible Deep Brain Stimulation (DBS) – blocks nerve traffic Levodopa
PD Patient’s Brain Data Gathering
Road Map Parkinson’s Disease (PD) and Treatments Deep Brain Stimulation (DBS) Treatment on PD Tremor PD Brain Data Analysis Artificial Neural Networks Prediction Future Research
DBS Treatment on PD What is Deep Brain Stimulation? Electrode, a connecting cable and Neurostimulator Video (Dana Centre)
Research Project What is wrong with current DBS? Large amount of battery usage Continuous stimulation – overrides brain Open-Loop Side Effects? Create an intelligent DBS system which can predict PD tremor onset using Artificial Neural Networks (ANNs)
Road Map Parkinson’s Disease (PD) and Treatments Deep Brain Stimulation (DBS) Treatment on PD Tremor PD Brain Data Analysis Artificial Neural Networks Prediction Future Research
PD Patients Brain Data Plot
Frequency Domain Analysis
Road Map Parkinson’s Disease (PD) and Treatments Deep Brain Stimulation (DBS) Treatment on PD Tremor PD Brain Data Analysis Artificial Neural Networks Prediction Future Research
ANN Design Principles
MLP Results
Road Map Parkinson’s Disease (PD) and Treatments Deep Brain Stimulation (DBS) Treatment on PD Tremor PD Brain Data Analysis Artificial Neural Networks Prediction Comments & Future Research
Comments Aim is to accurately predict tremors and then apply stimulation in order to stop them occurring Only stimulate when prediction indicates Needs research into nature of stimulation – frequency/magnitude/time prior for effect/generality with different people Doesn’t matter so much if false positive predictions Does matter if false negatives
More Comments Every PD case is a little bit different Not everyone can benefit from stimulation Probably not everyone will benefit from our prediction + stimulation (subset) Doesn’t have to be neural network Hybrid methods probably best Neural net good for nonlinearities though
Future Research More Tests on resting data – no false calls Test on a broader range of patients to assess general results with our current MLP (male, female, age difference) Test on other problems (Epilepsy/Tourette’s/Emotional) Apply feedback loop (MRC project)
Surprise Biggest shock has been the prediction time 15/20 seconds seems typical But this means we know for sure what will happen in the human brain 15/20 seconds before it happens How long could this time be? What else could we predict this way?
Other Points Developing here an AI brain within the human brain Use for other diseases – not just PD Could be used for emotional changes/control Ethical concerns: What should predictor/stimulator be used for in humans? What about responsibility/murders etc?