Personal tools

You are here: Home / Training / CDT in Healthcare Innovation / Student Research Areas / 2014 Abstracts

2014 Abstracts

Facial recognition of developmental disorders

Mohsan Alvi

The aim of this project is to use facial recognition software as a tool for automatic identification of dysmorphic  disorders such as Downs syndrome and Williams syndrome.  The long term goal is to develop high throughput data mining tools to aid identification, and genetic diagnosis, of ultra-rare developmental disorders.  This approach is aimed at leveraging the next generation sequencing data being produced in clinical genetics with digital facial phenotyping to aid in identification of the causative genetic abnormalities.
The project will involve using machine learning methods (such as metric learning) to automatically cluster faces with similar developmental disorders. The methods will use visual descriptors that are computed automatically from face images.
The project is in collaboration with Dr Christoffer Nellaker and Prof Chris Ponting of the Department of Physiology, Anatomy and Genetics.

Machine Learning Methods for Identifying Physiological Deterioration in Acutely-Ill Patients

Glen Colopy

The Biomedical Signal Processing cluster at the IBME has shown that integrating physiological data into a single decision support system can allow early warning of patient deterioration, and that clinicians acting on the outputs of such a system can improve patient outcomes.  Existing work has typically made strong assumptions about the nature of the physiological data (treating them in a population-generic manner).  Recent work has investigated the use of Bayesian non-parametric regression using multi-task Gaussian processes (GPs) for modelling physiological time-series in a patient-specific manner.  The goal of this project will be to treat these processes in a “strongly Bayesian” manner, with distributions over all hyperparameters of the GP – to date, either maximum a posteriori estimates of the hyperparameters, or simple approximations to the joint distribution of the hyperparameters, have been used.
This project will link with the Machine Learning Research Group (MLRG) in the IEB, where Prof. Steve Roberts and Prof. Michael Osborne will be key collaborators.  Their work has shown that Bayesian quadrature can provide an efficient means of estimating the joint distribution over the hyperparameters of a GP.  Further extensions have shown that change-point detection may be performed straightforwardly using the GP regression framework in a coherent probabilistic manner.

Signal processing and image analysis in cardiovascular disease: linking ECG and MRI to assess risk of sudden death in hypertrophic cardiomyopathy

Johanna Ernst

Hypertrophic Cardiomyopathy (HCM) is a genetic disorder that affects 1 in 500 individuals and is the most common cause of sudden cardiac death in children and young adults. Reliable methods to identify those subjects in risk of sudden cardiac death are lacking. HCM is known to modify the structure of the heart, which in turn can affect electrical patterns.
The aim of this project is the combined analysis of cardiac structure, using Cardiac MRI, and electrical conduction, using the electrocardiogram, in order to improve the assessment of risk in HCM patients.

Automated measurements of MRI spine images

Amir Jamaludin

Back pain is a major public health and economic problem. It is a very common debilitating condition that affects all ages but particularly those of working age and the elderly. Approximately 60% of adults have a significant episode of back pain at some point over their life.
The overall goal of this project is to improve and standardize MRI lumbar spine analysis. The objective is to employ modern computer vision and machine learning methods to automate diagnosis procedures by learning new measures on large image databases.  We have access to a large database of spinal MR images (~2500 patients) with ground-truth radiologist measurements and associated clinical meta-data.
The short project will investigate automating the prediction of the radiological measurements using segmentation and machine learning.

Quantifying Impaired Metabolism Using Chemical Exchange Saturation MRI

Yunus Msayib

In a condition like stroke it is important to be able to understand what is happening to cells that are being starved of nutrients because their blood supply has been reduced. This can be difficult to achieve in critically ill patients. However, Magnetic Resonance Imaging methods now permit such information to be acquired with a high level of detail and Oxford has a facility that permits such images to be obtained in stroke patients soon after hospital admission.

This project will exploit a relatively new MRI technique that can measure pH in the brain tissue of patients, which can be used as an indicator of cell health – potentially providing a prediction of which tissue can be rescued if treatment is applied swiftly. The project will improve the accuracy with which we can make these pH measurements making this technique more useful in clinical research and ultimately in clinical practice.

New imaging biomarkers for monitoring the maternal-fetal system cardiovascular system during pregnancy

Elina Naydenova

In the developing world poor maternal nutrition can lead to sub-optimal growth of the placenta which is known to be a key factor in pre-disposal to premature birth (1 in 3 babies in India are premature) and to place the mother at risk (pre-eclampsia). Little is understood about how the cardiovascular system development of a intrauterine growth retarded (IUGR) fetus differs from a healthy fetus. It has been suggested that this might be assessed from fetal brain scans.
The short project will involve working with a PhD student in the laboratory on assessment of fetal brain scans on an existing database of data using a developed machine-learning method to establish whether vascular delays can be observed in ultrasound scans of IUGR fetuses versus those of a healthy population. The project will involve designing a machine-learning method to do this, running it on the data, and visualisation of interpretation of the results.

Analysis of the developing heart using machine-learning image analysis

Hasmila Omar

The EPOCH study led by the Department of Cardiovascular Medicine is acquiring ultrasound data on fetuses and neonates around birth to gain better understanding of the effect of prematurity and hypertensive disorders of pregnancy on offspring cardiovascular health. This short project will look at how image analysis can assist in the automated analysis and interpretation of the data.

Assessing myocardial infarction with state-of-the-art Magnetic Resonance Imaging

Ben Villard

After myocardial infarction, early clinical decisions are key to reduce the risk of sudden death and maximise functional recovery. Cardiac Magnetic Resonance imaging is becoming a crucial tool to assess myocardial infarction patients and inform best treatment. This project aims at developing image analysis tools to quantify risks and predict the development of the heart in the weeks to months that follow, facilitating the choice of an optimal treatment to eventually reduce the burden of the disease.