Personal tools

You are here: Home / Training / CDT in Healthcare Innovation / Student Research Areas / 2011 Abstracts

2011 Abstracts

Machine Learning Methods for the Automation of Cell Image Analysis

Carlos Arteta, Pembroke College

The vast amount of microscopy data that needs to be analysed in cell-based experiments demand the use of automatic methods, which must be accurate and reliable while providing quantitative information. It has been shown that, in combination with the appropriate imaging platforms and labelling techniques, image analysis methods can automatically provide the quantitative information that is required to better understand complex biological systems. The areas of application are numerous; for example, drug discovery, tissue engineering, genomics and proteomics are some of the sciences that greatly benefit, and to some extent, depend on accurate, fast and intelligent cellular image analysis algorithms. In many cases, cellular image analysis is based on a series of common tasks that have been studied in the computer vision and machine learning community, and that are still actively researched. Those tasks include object detection, segmentation, tracking, counting and classification. My Current research is in the exploration of such methods to develop intelligent systems for the automatic analysis of experiments based on cellular imaging.

Ultrasound-mediated epidermal delivery of vaccines

Sunali Bhatnagar, St Anne's College

Background: The human skin comprises three tissue layers: the stratified, avascular, cellular epidermis, the underlying dermis of connective tissue, and the subcutaneous fat. It performs many varied function, but one of the main ones is to act as a protective barrier against microorganisms, chemicals and radiation. This actual barrier function of the skin is mainly due to the almost impermeable nature of the stratum corneum, the multi-layered top part of the epidermis, with its dead, dense cell layers. It is also crucially important in controlling the percutaneous absorption of drug and other chemical and as a consequence limits the use of the skin as route of drug administration to only a few small, hydrophobic therapeutic molecules. Newer more complex drugs such as peptides, proteins or vaccines cannot penetrate through the intact stratum corneum.

The human skin has another important property that can be used for medical purposes: it has a highly effective immunological surveillance and effector system. The epidermis of the skin contains a large number of antigen-presenting cells, so-called Langerhans cells, which can be used for transcutaneous immunization. In the past this has been addressed using needle-free injection devices or microneedle projections. However, these devices are complex to manufacture and require extensive formulation development of the often very sensitive vaccine substances.

Project: Ultrasound has been shown in the past to alter the permeability of the stratum corneum for large, hydrophilic molecules and thus increase their penetration into the skin without sophisticated pharmaceutical formulation or device development. The effect is attributed to a disturbance of the stratum corneum structure due to cavitation caused by ultrasound waves. This could be of particular interest in the delivery of vaccines directly to the antigen-presenting cells within the epidermis. Additionally, it is believed that ultrasound can “activate” exactly these antigen-presenting cells and thus provide an increased immune response, something very important in the vaccination of children and the elderly as well as in reducing the doses of very expensive vaccines.

Distinguishing normal and abnormal outcomes in labour from fetal heart rate monitoring

Alex Clibbon, St Catherine's College

Trying to decide whether or not to intervene during labour is a very difficult decision for a doctor: intervene too infrequently and there is the danger of brain damage; intervene too frequently and too many woman have to undergo Caesarean sections, with their attendant risk and need for prolonged recovery. What is needed is a more targeted approach, identifying which fetuses are most at risk. We will do this byanalysing tens of thousands of records that we have acquired and the related outcomes to find what can most reliably predict clinical outcome. This will be very valuable in helping doctors, particularly inexperienced ones, to target interventions more precisely.

Automated Analysis of 3D+T Echocardiography Imaging: Towards an Improved Clinical Diagnosis

Joao Domingos, Wolfson College

Real-time 3D echocardiography (RT3DE) is an imaging modality that is increasingly used in clinical practice to assess cardiac function. RT3DE promises dynamic 3D image acquisition with the potential of a more objective and complete cardiac functional analysis than the conventional 2D echocardiography, the backbone of echocardiography imaging in current clinical practice. Some of the drawbacks of RT3DE is the time it takes the clinician to navigate the 3D volumes to the anatomy of interest, to obtain standardized views that are similar to the 2D acquisitions or to perform volumetric cardiac function quantification. RT3DE uptake has also been slow due to imaging challenges such as missing anatomical information, speckle noise and limited field-of-view. To automate the clinical workflow and facilitate the subsequent processing tasks, this doctoral research focusses on providing cardiologists (or sonographers) with novel tools for utilizing 3D echocardiography data in clinical practice. It evaluates image quality and information content of 3D volumes to design and test computer vision and machine learning algorithms, aiming for the automated real-time detection of standardized echocardiographic imaging planes, left and right ventricle volume estimation and wall motion analysis.

Detection of abnormal activity for early warning of patient deterioration using mobile networks

Elnaz Gederi, Wolfson College

Changes in physical activity and sleep structure has been shown to correlate with a range of short and long term medical conditions from cardiac problems to mental health issues. Historically monitoring has taken the form of infrequent and relatively brief recordings (such as ECG stress tests of overnight polysomnographic studies). Potentially important fluctuations in physiology between such tests are therefore not captured. Moreover, the time of recording may be anomalous due to either random fluctuations or causal factors (such as ‘white coat syndrome’).

Although continuous monitoring is preferable, it is expensive and requires the analysis of large and often noisy tracts of data. Traditional continuous or long term monitoring methods are based upon actigraphy, generally using a wrist-worn accelerometer. However, accelerometry is of limited use and provides a two-dimensional view of physical activity and levels of ambient light.

This project has two parallel aims. The first aim is to analyze existing baseline accelerometry data from 100 patients recorded over several weeks. Signal processing algorithms will be used to filter data and extract features and classifiers will be used to map these features onto clinical labels of patient deterioration specific to schizophrenia, obsessive compulsive disorder and depression.

The second aim of the research project will be the augmentation and analysis of the monitoring systems. Currently the data are recorded on a wrist-worn device which must be downloaded to a computer whenever the patient comes in for further clinical evaluation (usually monthly). A GPRS-based quasi-real time data upload system is required to allow continuous review of the patient and the ability to intervene when early warning signs are encountered.  The student will build on existing systems.

Recording modalities will include accelerometry, audio, GPS, and video, with potential extensions to signals like drug dosing, temperature, pulse oximetry, respiration and electrocardiography.

Whole organ ultrasound for targeted drug delivery

Susan Graham, Wolfson College

Conventional chemotherapeutic approaches to cancer treatment have seen unacceptable levels of side effects for cancer patients.  To reduce these side effects, treatments must have increased selectivity for tumour targeting and mechanism of action against the cells.  Nanotechnology has shown great promise in the area of tumour selectivity and uptake.  To further enhance drug delivery, this project focuses on the combination of drug delivery via a nano-vehicle, and drug release using diagnostic levels of ultrasound.  The enhancement that nanoparticles can bring both in their tumour localisation and ultrasound effect enhancement (inertial cavitation), will be utilised to initiate targeted drug release and local tissue damage.  With the nanoparticles targeted to tumours, an ultrasound field is  applied to the whole organ (for example the liver) to initiate targeted drug release.  This treatment would then be able to be applied to existing drugs, using existing diagnostic ultrasound equipment to reduce the side effects, and increase the efficacy of today’s chemotherapeutics.

Measurement of lung function through modelling

Chris Harrison, Wadham College

Measuring how well a person’s lungs are functioning is not easy. Doctors do various tests but trying to calculate things such as lung volume and its response time require a mathematical model that can be used with clinical data. In this project we will start to develop such a model, one that can be applied in both healthy subjects and patients with lung problems. This will have potential to be applied clinically to provide more accurate and reliable measurements for use in ongoing diagnosis and treatment of people with lung disease.

Development of novel particles of cancer therapy

Rachel Morrison, St Hilda's College

Globally, cancer is one of the leading causes of morbidity and mortality in both the developed and developing world.  This project will seek to develop novel particles for the more effective treatment of cancer.  The project will involve synthesising and characterising the particles to optimise their therapeutic potential.  Various characterisation techniques will be used including: scanning electron microscopy (SEM), transmission electron microscopy (TEM), X-ray diffraction (XRD), disc centrifugation (particle size) and zeta potential.  Cell based models will also be used to determine toxicity of the particles and their therapeutic benefit.

Active targeting of sonosensitive nanoparticles for cancer therapy

Rachel Myers, Wadham College

Recent research at the IBME has enabled the development of a novel generation of sonosensitive nanoparticles, which can encapsulate a drug have been found to enable lowering of the local cavitation threshold in tissues by up to a factor of 10. If adequately conjugated, these particles could be targeted to specific sites of pathology, thus removing the need to target the ultrasound field used to excite them. The objective of the project is to investigate the feasibility of this principle, by depositing sonosentive nanoparticles at known locations within a tissue-mimicking material and by exciting it with an unfocussed, low-power ultrasound field. Sites of cavitation activity can be mapped in real time using a novel technique known as passive acoustic mapping, which enables localization of sources of broadband and acoustic emissions associated with cavitation.  A key aim will be to optimize the ultrasound field (amplitude, frequency, duty cycle) to achieve the greatest therapeutic differential between treated and untreated sites.

Detection of abnormal activity for early warning of patient deterioration using mobile networks

Maxim Osipov, Wolfson College

Changes in physical activity and sleep structure has been shown to correlate with a range of short and long term medical conditions from cardiac problems to mental health issues. Historically monitoring has taken the form of infrequent and relatively brief recordings (such as ECG stress tests of overnight polysomnographic studies). Potentially important fluctuations in physiology between such tests are therefore not captured. Moreover, the time of recording may be anomalous due to either random fluctuations or causal factors (such as ‘white coat syndrome’).

Although continuous monitoring is preferable, it is expensive and requires the analysis of large and often noisy tracts of data. Traditional continuous or long term monitoring methods are based upon actigraphy, generally using a wrist-worn accelerometer. However, accelerometry is of limited use and provides a two-dimensional view of physical activity and levels of ambient light.

This project has two parallel aims. The first aim is to analyze existing baseline accelerometry data from 100 patients recorded over several weeks. Signal processing algorithms will be used to filter data and extract features and classifiers will be used to map these features onto clinical labels of patient deterioration specific to schizophrenia, obsessive compulsive disorder and depression.

The second aim of the research project will be the augmentation and analysis of the monitoring systems. Currently the data are recorded on a wrist-worn device which must be downloaded to a computer whenever the patient comes in for further clinical evaluation (usually monthly). A GPRS-based quasi-real time data upload system is required to allow continuous review of the patient and the ability to intervene when early warning signs are encountered.  The student will build on existing systems.

Recording modalities will include accelerometry, audio, GPS, and video, with potential extensions to signals like drug dosing, temperature, pulse oximetry, respiration and electrocardiography.

“Intelligent” quality assurance of fetal ultrasound scans for improved perinatal care

Khair Othman, Wolfson College

Machine learning is a powerful paradigm for learning patterns in data including images. In the Biomedical Image Analysis laboratory we have recently started investigating this paradigm as a way to select “good” and “bad” ultrasound images which is essential for good fetal biometric measurement.

In this short project we propose to investigate the application of the techniques to recognise a good fetal brain scan plane and to assess the accuracy of such an approach on the Intergrowth-21st database (currently approximately 4000 scans).

Flow Diverters for Cerebral Aneurysm Treatment: Design, Optimisation, Assessment and Clinical Deployment

Thomas Peach, Magdalen College

Cerebral aneurysms are relatively common and affect 2% to 5% of the adult population. They appear as sac-like out-pouchings (like a small berry) of the arterial wall inflated by the pressure of the blood. Most remain asymptomatic; however, there is a small but inherent risk of rupture:  0.1% to 1% of detected aneurysms rupture every year. Rupture of an aneurysm is associated with a 30% to 50% mortality rate and is the cause of around 5% of strokes.

Relatively recent developments in imaging technology have led to new healthcare policies in terms of the application of medical imaging modalities: nowadays, many admissions that involve head-related symptoms include Magnetic Resonance or Computer Tomography scanning. This trend has led to a dramatic increase in coincidentally detected asymptomatic aneurysms. This increase has resulted in substantial pressure in the healthcare sector for better, safer and cheaper interventional protocols for this disease. The research proposed in this project addresses exactly this unmet clinical need, looking to develop Information Technology-based tools and techniques that improve the design and deployment of the most promising treatment implant for aneurysms (flow diverters) and to translate these tools so that they function effectively in a clinical setting.

In summary, we intend to develop and integrate computational tools that allow for the representation of diseased vasculature; that permit the virtual, in silico, deployment of existing and new design stents and that evaluate their performance regarding flow reduction and thrombosis. This platform will be tested on a representative set of clinical cases and will be translated within the interventional planning workflow of one of our partner hospitals, where the clinicians responsible for the interventions will assess the software’s efficacy and utility.

Dynamic modelling of vital sign data from post-operative patients

Marco Pimentel, St Cross College

Patients who undergo upper-gastro-intestinal (GI) surgery have a high incidence of post-operative complications, with up to 30% requiring readmission to the Intensive Care Unit (ICU) several days after surgery.  Failure to identify such deteriorations in a timely manner has led to the design of a clinical trial at the Oxford Cancer Hospital, in which ambulatory post-operative cancer patients are monitored using wearable sensors. The vital signs recorded with these sensors are transmitted in real time, via the hospital wireless network, to a central server for analysis and display. Vital sign measurements are also made periodically by the nurses on the ward.

Our approach to monitoring in-hospital patients has so far relied on constructing static models of normality, based on the vital signs acquired from a large population of acutely ill patients.  We now propose to improve upon existing techniques by investigating dynamic models tuned to post-operative cancer patients. These patients are recovering from surgery, and the project will adopt a new approach in which the aim will be to learn the vital sign trajectories associated with “normal recovery”, allowing “abnormal” trajectories to be identified.

The research project will firstly investigate the changes in vital sign distributions between admission to the Upper GI ward and subsequent discharge, when the patient is well enough to go home. Dynamic models based on windowed sections of data will be learnt to describe these trajectories, using for example Gaussian processes. These models will then be tested on “abnormal data” from patients who deteriorate sufficiently after surgery to be re-admitted to ICU.

Multimodality imaging of chemo-embolic agents

Sheetal Sanak, St Cross College

The aim of this project is to develop a new type of delivery system for chemotherapy based on injectable capsules which both localise the release of the drug whilst simultaneously blocking the blood supply to the tumour. To enable successful implantation and assess the effectiveness of the treatment, the capsules need to be visible under different types of imaging such as ultrasound, X-ray or Magnetic Resonance Imaging (MRI). The work will involve developing a system for testing the capsules which will enable their formulation to be optimised.

Vital sign data fusion for detecting patient deterioration in Emergency Department

Mauro Santos, St Hilda's College

We have previously shown that the integration of continuously-monitored parameters from bedside monitors in high-dependency or step-down units can provide early warning of adverse events. We are now extending this approach to the Emergency Department, in which there is a mixture of continuous monitoring and periodic recording of vital signs by nurses. We have just completed a 500-patient observational study in the Emergency Department at the John Radcliffe Hospital, during which we recorded both the vital sign data from the bedside monitors and the nurse observations (which were then scanned and converted to electronic data).

The research project will firstly investigate the differences in vital sign distributions between continuous and periodic data, and then use machine learning techniques (Gaussian Mixture models and Support Vector Machines) to construct data fusion models of normality for Emergency Department patients. These models will then be tuned to optimise the detection of patient deterioration in the Emergency Department.

A surface body coil for Electromagnetic Acoustic medical imaging

Ning Zhang, Wadham College

This project involves the electromagnetic design and building of a surface radio frequency coil working at 434 MHz for medical imaging. The output of the project will be utilised in a new medical imaging technique currently under development within the group.