Data science and AI for health
Researchers at Surrey are combining machine learning and computational technology with mathematical algorithms to increase understanding of health data for the efficient and timely diagnosis and treatment of disease.
Symmetric Projection Attractor Reconstruction (SPAR)
A team led by Professor Philip Aston (Department of Mathematics) has developed the Symmetric Projection Attractor Reconstruction (SPAR) technique which provides a visual representation of changes in the shape and variability of a waveform using an ‘approximately periodic’ signal. This method has been applied to cardiovascular signals such as ECG, continuous blood pressure and PPG (a measure of blood volume change). The aim is to identify novel signals that facilitate early detection of disease.
Features extracted from the SPAR attractors have also been used for signal classification using machine learning, and the attractor images themselves have been used as input for deep learning classification.
Electroencephalograms (EEG) project
Research by Dr Daniel Abasolo (Centre for Biomedical Engineering) focuses on the analysis of electromagnetic brain signals (electroencephalograms) of patients with Alzheimer’s Disease using non-linear signal processing with the aim of being able to diagnose dementia earlier. This technique can also be applied to patients with epilepsy, in sleep studies, and for the characterisation of healthy ageing.
Other recent projects in this area include:
- Development of new analysis techniques for extracting information when considering pairs of signals (bivariate analysis of electroencephalogram recordings) or high-density electroencephalogram data (multivariate analysis).
- Use of advanced connectivity measures and graph theory for the characterisation of healthy ageing magnetoencephalogram recordings.
- Application of deep learning to electroencephalogram analysis for the diagnosis of Alzheimer’s disease and mild cognitive impairment.
Researchers led by Philip Evans and Kevin Wells in Surrey’s Centre for Vision, Speech and Signal Processing (CVSSP) are applying state-of-the-art computer vision techniques to solve medical imaging problems. Working in collaboration with the Royal Surrey NHS Foundation Trust, National Physical Laboratory and industry partners, they are developing methods to make radiomics (the method of extracting features from medical images) robust and repeatable across different hospitals scanners. CVSSP’s methods have been applied to scans of lung, head and neck cancer.
Other areas of work include:
- Using artificial intelligence to determine when lung cancer patients need to see a dietician using a few simple questions, avoiding the need for patients to fill in lengthy questionnaires.
- Development of an end-to-end imaging test object (phantoms) for Magnetic Resonance (MR) guided radiotherapy which combines image guided radiotherapy (IGRT) for adaptation with dose verification.