Post on 14-Dec-2015
Dr. Athanasios Tsanas (‘Thanasis’), EPSRC post-doctoral research fellow, Oxford Centre for Industrial and Applied Mathematics &Institute of Biomedical Engineering, University of Oxford
Current issues in healthcare2. Probabilistic Parkinson’s disease detection using speech signals
Centre for Doctoral Training in Healthcare Innovation
Death of dopaminergic cells in basal ganglia Dopaminergic cells facilitate motor control 60-80% dopaminergic cells have died before clinical diagnosis too late to intercept Parkinson’s disease symptom progression
No known biomarker of Parkinson’s disease (PD) Difficult diagnosis: 100% only post-mortem Speech may be one of the earliest indicators of PD onset
Early diagnosis can improve the quality of life of PD subjects Optimized treatment (hitherto no known cure available) Possibly surgery at later stages
Lungs: inefficient control of expiration
Vocal folds: incomplete collision, vortices,
aperiodicity
Vocal tract: loss of muscle co-ordination
Speech signal
Extract feature vector
Feature selection
Report results
Statistical machine learning (mapping features PD)
Sustained vowel “ahh…”
What sort of speech characteristics should we be aiming to extract from the signals?
Which are the state of the art speech signal processing algorithms, and can you suggest some ways to extend/improve them?
Who are the leading researchers/research groups on the topic of mining speech signals for medical information?
Hint: background information, and check some useful references: http://people.maths.ox.ac.uk/tsanas/Preprints/DPhil%20thesis.pdf