MSc Projects 2010-11

Here is a list of projects being offered for MSc students in the Department. For details of deadlines and for registering submissions, see the Moodle page. Full instructions are available here.

Understanding the effect of surface roughness on electrode impedance

Supervisors: Prof. Nick Donaldson and Dr. Anne Vanhoest

PROJECT TAKEN BY UNDERGRADUATE STUDENT

A polarisable electrode like platinum has an impedance that has both capacitance and resistance but these are not combined in a simple way. The interface itself may be perfectly capacitive but the roughness means that some of the surface capacitance is accessed through the resistance in tissue within the crevices. In this project, the effect of different types of roughness will be explored using models made from resistors and capacitors in 2D arrays. The impedance of these models will be measured on an Impedance Analyser. An enjoyment of model-making and mathematics will be helpful in this project. Topics: electrode impedance, impedance of networks.

Wind Turbines in the Jet Stream: a feasibility study

Supervisors: Prof. Nick Donaldson and Dr. Anne Vanhoest

Student: (Project available)

The power from wind turbines rises as the cube of the wind speed. The speed of the jet stream is sometimes 200 miles per hour. Obviously the power that might be extracted could be enormous. However the course of the jet stream moves over the earth's surface and is high in the atmosphere. Might turbines be placed in the jet stream to extract some of this power? One idea is to fly a large kite, which lifts the turbine, from a ship that moves about the ocean under the jet stream. Electricity from the generator could electrolyse sea water and the hydrogen could be stored in the ship. This is a project for two students who should come up with one concept and then write two reports about its feasibility. One should report on the engineering aspects; the other on the meteorological and economic aspects of this concept. Good reports should use stated principles of engineering; use data from references; and be generally quantitative rather than merely descriptive. Of course it is essential that none of the possible difficulties (for example: the weight of the wire going to the kite) is neglected. Topics: aerodynamics, turbines, energy generation, meteorology, investment.

Is there a vegetal analog to the Action Potential?

Supervisors: Dr. Anne Vanhoest and Prof. Nick Donaldson

PROJECT TAKEN BY UNDERGRADUATE STUDENT

The electrical activity of selected plants will be studied in response to various stimuli (touch, light, heat, electricity). The student is asked to establish whether this vegetal activity is analog to an Action Potential and could be used in demonstrations where ex-vivo animal nerves would otherwise be required. This is a lab-based project, the student will be taught how to use stimulation and recording equipment and is expected to be "hands on" and practical. There will also be a preliminary literature review based on Fromm and Lautner (Plant, Cell and Environment, 2007). Topics: basic nerve physiology, electrical stimulation, lab practice, literature review

Thick-film humidity sensors - a feasibility study

Supervisors: Dr. Anne Vanhoest and Prof. Nick Donaldson

Student: (Project available).

To study the feasibility of using standard thick-film methods to create an implantable humidity sensor. This project will take place in the lab, characterizing existing sensor candidates and researching other materials, suitable in terms of their bio-compatibility and sensing properties. If the progress are satisfactory and the student shows good lab behaviour a second stage will involve the production of new sensors in the cleanroom. Topics: humidity sensor, electronics, thick-film printing, material science

History of Electrotherapy

Supervisors: Prof. Nick Donaldson and Dr. Anne Vanhoest

PROJECT TAKEN BY UNDERGRADUATE STUDENT

Electrotherapy (medical application of electricity to the body) started in the 1740s and for 2 centuries it was offered as a treatment of many illnesses. However, by the 1970s, the methods had been abandoned yet, nowadays, the treatment of motor disorders by exploiting neural plasticity with electrical or magnetic stimulation is a hot research topic: we may wonder whether some methods that were in use for some 200 years will be rediscovered but now with some scientific understanding of the therapeutic mechanisms. Although this is a literature based project (using the Wellcome Library of the History of Medicine as well as UCL's own resources). The student should aim to critically appraise the treatments used, at least concerning the physics involved (but not necessarily the neurophysiology). Topics: biophysics, history of physics and physiology, nerve stimulation, plasticity of CNS.

Portable Impedance Analyser

Supervisors: Prof. Nick Donaldson and Dr. Anne Vanhoest

PROJECT TAKEN BY UNDERGRADUATE STUDENT

A device that measures the impedance between two terminals across a range of frequency is useful in many fields. We are particularly interested in electrodes and therefore the impedance between pairs of electrodes in electrolyte or in the body. Analog Devices have developed an integrated circuit for this function which can interface to a PC via an USB port. The project is to develop a portable impedance analyzer for use in the lab and during operations. The hardware will be developed and a program must be written for the PC to give useful displays. Topics: impedance, impedance spectra, PC programming, USB.

Electrolytic Pump

Supervisors: Prof. Nick Donaldson and Dr. Anne Vanhoest

PROJECT TAKEN BY UNDERGRADUATE STUDENT

Heart pacemakers were the first successful neuroprosthesis and the way their electronics is protected from body fluid is orthodox for implanted devices. They use a titanium case that has a metal-in-glass feed-through for the connection to the electrode, and the case is closed by welding. This construction is expensive and simpler methods might reduce the costs of implanted devices. One possibility is to use a plastic enclosure, through which water will diffuse, but pump the water out by converting the liquid water, condensed in a salt-filled sponge, into gas by electrolysis. The gas would escape through a pressure-relief valve. The project will be to test this idea by experimenting with chambers which include a humidity sensor and tests are done with various salts, electrodes and drive voltages. The aim will be to show that the relative humidity can be maintained at a level significantly below 100%. Topics: electrolysis, relative humidity, properties of salts.

Time-resolved optical detection of auditory activation

Supervisors: Dr Adam Gibson and Dr Nick Everdell

Student: MSc.

The adult auditory cortex is a challenging target for optical imaging and spectroscopy as it is deep in the brain and covered with a thick layer of bone. We will use our most sensitive optical system first to detect whether any change in activation can be detected during auditory stimuli and second to try and determine the depth and localisation of the activation.

Maximising sensitivity and specificity of Terahertz images

Supervisors: Dr. Adam Gibson and Dr. Caroline Reid

Student: MSc. (probably MIC)

We have Terahertz images from cancer and healthy tissue. The data can be processed to give many different datatypes. In this project, we will examine these datatypes and determine which combinations of datatypes maximise the difference between healthy and cancerous tissue. One method which will be examined will be support vector machines.

Maximising sensitivity and specificity of optical mammography.

Supervisors: Dr. Adam Gibson and Dr. Danny Alexander

Student: MSc. (probably MIC)

We have imaged about 50 women using optical mammography. For each woman, we record images of blood volume, blood oxygenation and optical scatter. These gave us a sensitivity of 85.8% and a specificity of 66.8%. In this project, you will determine the combination of these images which maximises the sensitivity and specificity with which the system can distinguish between healthy tissue and tumour. One method which will be examined will be support vector machines.

Demonstrating pulse plethysmography and oximetry

Supervisors: Dr Nick Everdell and Dr Adam Gibson

Student: Margus Must

This project is aimed at producing medical physics demonstrations for schools. The idea is to use low cost materials and equipment that are readily available for GCSE and A level students.

The first part of the project would be to produce a demonstration of pulse plethysmography – which is basically measuring and displaying the pulse, obtained at the finger, or another suitable location. For instance, if the measurement were made using light, a laser pen might be a suitable source, a simple photodiode a suitable detector, and the measurement could be displayed on an oscilloscope. Other measurement techniques could be experimented with.

The project could be extended to include pulse oximetry, which is the measurement of the oxygen saturation of arterial blood. This is an extension of the light measurement technique, where 2 different wavelengths are used, to measure the ‘colour’ of the blood, and hence its oxygen content.

Identification and Quantification of Brain Tissue Haemodynamic Signals and Systemic Interference in Functional Near-Infrared Spectroscopy

Supervisors: Dr Ilias Tachtsidis and Dr Heidrun Wabnitz

Student: Nuttanont Panitchob

Brain tissue functional near-infrared spectroscopy (or fNIRS) is a technique that uses non-invasive optical head reflection measurements to monitor brain tissue haemodynamics by resolving the concentrations of oxygenated haemoglobin (HbO2) and deoxygenated haemoglobin (HHb). For several years we have been using the technique to monitor the brain haemodynamic changes secondary to brain neuronal activation during frontal lobe cognitive tasks (such as anagram solving). From these studies we demonstrated the ability of fNIRS to monitor brain function. However, we also identified certain limitations regarding the contamination of the brain fNIRS haemodynamic response from systemic originated changes such as blood pressure increases (http://www.ucl.ac.uk/medphys/research/borl/nirs/current_projects/funct_adult). As part of recent research collaboration with the Biomedical Optics department of Physikalisch-Technische Bundesanstalt (PTB) in Berlin, Germany (http://www.ptb.de/cms/en/fachabteilungen/abt8/fb-83.html) we have performed a study in 14 adults using simultaneously a state of the art time-domain fNIRS system and multimodal systemic measurements (blood pressure, heart rate, breathing rate etc) during a word performing task. Preliminary results have identified systemic changes affecting the fNIRS measurements; however questions remain about the degree of interference. Further investigation is necessary and likely to include coherence analysis and/or independent component analysis and/or wavelet cross-correlation methods. This project is mainly computational and will be suitable for a student with a general interest in monitoring brain physiology, fair knowledge of MatLab and signal processing methods.

Computational investigation of optical spectroscopy algorithms to decouple optical measurements of brain oxygenation and metabolism

Supervisors: Dr Ilias Tachtsidis and Prof. Chris Cooper

Student: (project still available)

Brain tissue near-infrared spectroscopy (or NIRS) is a technique that uses non-invasive optical head reflection measurements to monitor oxygenation by resolving the concentrations of oxygenated haemoglobin (HbO2) and deoxygenated haemoglobin (HHb); NIRS can also monitor changes in energy metabolism by resolving the redox state of cytochrome-c-oxidase (oxCCO). In order to resolve the three chromophores (HbO2, HHb and oxCCO) we employ a linear algorithm that relates the optical attenuation measurements with the concentrations of the chromophores scaled with their optical absorption characteristics and the total pathlength of light. This spectroscopic approach has the potential to cause crosstalk between the concentration measurements of the three chromophores – by crosstalk we mean a genuine change in a chromophore concentration inducing a spurious measured concentration change in another. The presence of this crosstalk artefact has been investigated by our group and others, both experimentally and using computational simulations. It has been suggested that “spectroscopic crosstalk” can be minimised by utilising a large number of optical wavelengths. However questions remain about: (1) wavelength combination; (2) wavelength resolution and bandwidth; (3) pathlength selection. This project requires the student to implement an optical spectroscopy algorithm based on previous work in our group and investigate the above questions using computational simulations. To enhance the simulations experimental data from healthy volunteers and head injured patients will also be available for analysis. This project is mainly computational and will be suitable for a student with a general interest in monitoring brain physiology and fair knowledge of MatLab.

Electrical impedance tomography of epileptic seizures

Supervisors: Prof David Holder and Dr. Dominic Heaney

Student: A similar project has been assigned to a student. Prof. Holder has reached his full student quota for this Academic Year.

EIT is a novel medical imaging method, with which images of the electrical impedance of the head can be produced with a box about the size of a paperback book, laptop and EEG electrodes on the head. It is portable, safe, fast and inexpensive. The supervisor’s research has been to develop its use in imaging functional activity in the brain. One exciting application lies in its use to image changes in the brain due to epileptic activity. In epilepsy, abnormal activity may occur in the form of seizures in which there is continuous abnormal activity lasting a minute or so. There is also “interictal” activity, in which there is non-seizure epileptic activity which lasts for a second or so and is not noticed by the patient. This is associated with a small local increase in blood flow which can be detected with functional FMRI sin some subjects. It could also be detected with EIT and this is the topic of this project.
Using software already developed, the purpose of the work will be to record EIT and EEG at the same time in patients brought into hospital for evaluation of their epilepsy. The recordings will be over several hours. Using the developed software, all examples of interictal epileptic activity will be marked on the EEG. All the current sections of EIT will be averaged together and used to reconstruct an image of the associated change in blood flow. The signal to noise ratio in such recordings is low; the hope is that averaging of many examples will lead to accurate images. If successful, this would provide a new method for diagnosis in epilepsy.
Students will work together to collect EIT data during repeated evoked activity in about 10 volunteers, and then will help produce images using Matlab code written for this purpose. Digital photos will be taken around the head, and then photogrammetric software will be used to localise their positions. Images will be reconstructed using an MRI of the patient’s head, which needs to be converted to a Finite Element model with software for segmenting medical images and meshing them.
Skills to be acquired: Students will spend time in the lab in Medical Physics at UCL learning relevant methods and analysing the data, and some time in Prof Holder’s department at UCH, learning how to collect evoked responses using scalp electrodes. Skills to be acquired will include one or more of: medical image reconstruction; photogrammetric software use; medical image segmentation and meshing software; EEG electrode placement and use; experimental design and data analysis. The project would be suitable for a single student or more than one working in a team, with a background in medical physics, engineering, computing, or medicine.

A new method for diagnosing muscle disease with bioimpedance measurement

Supervisors: Prof David Holder and Dr. Ori Gilad

Student: A similar project has been assigned to a student. Prof. Holder has reached his full student quota for this Academic Year.

At present, electrophysiological diagnosis of different muscle diseases in undertaken with needle recording of the electrophysiological signals – this is termed “Electromyography” (EMG). The idea behind this project is to make recordings of electrical impedance to aid in this diagnosis.
The work will be to review relevant literature concerning muscle impedance studies, and test electrical impedance measuring equipment in the laboratory for this purpose. If successful, this will then be tested in a small number of patients with muscle disease.
Skills to be acquired: Students will spend time in the lab in Medical Physics at UCL learning relevant methods and analysing the data, and some time in Prof Holder’s department at UCH, acquiring muscle impedance data. Skills to be acquired will include one or more of: biomedical instrument use and assessment, EMG, biophysical modelling. experimental design and data analysis. The project would be suitable for a single student or more than one working in a team, with a background in medical physics, engineering, or medicine.

Electrical Impedance Tomography of evoked physiological activity

Supervisors: Prof David Holder

Student: A similar project has been assigned to a student. Prof. Holder has reached his full student quota for this Academic Year.

EIT is a novel medical imaging method, with which images of the electrical impedance of the head can be produced with a box about the size of a paperback book, laptop and EEG electrodes on the head. It is portable, safe, fast and inexpensive. The supervisor’s research has been to develop its use in imaging functional activity in the brain. One possible use could be to image increases in blood volume which occur over some tens of seconds during normal brain activity, such as during the standard clinical techniques of stimulation of the visual system by flashing lights or the somatosensory system by mild electrical stimulation at the wrist. Such imaging can already be performed by fMRI (functional MRI); the advantages of EIT are that similar images could be acquired with portable much less expensive technology which would increase its availability.
EIT data has been collected in these situations before and led to a landmark publication in which reliable single channel data were observed but, unfortunately, the data was too noisy to form into reliable images. Since then, the electronics and imaging software have been improved – for example, we can now collect images at multiple frequencies whereas before they were only collected at one. This gives greater opportunities to reduce noise.
Students will work together to collect EIT data during repeated evoked activity in about 10 healthy volunteers, and then will help produce images using Matlab code written for this purpose. Digital photos will be taken around the head, and then photogrammetric software will be used to localise their positions. Images will be reconstructed using an MRI of the patient’s head, which needs to be converted to a Finite Element model with software for segmenting medical images and meshing them. The accuracy of these images will be compared with similar studies using fMRI.

Skills to be acquired: Students will spend time in the lab in Medical Physics at UCL learning relevant methods and analysing the data, and some time in Prof Holder’s department at UCH, learning how to collect evoked responses using scalp electrodes. Skills to be acquired will include one or more of:
Medical image reconstruction;
Photogrammetric software use;
Medical image segmentation and meshing software;
EEG electrode placement and use;
Experimental design and data analysis.
The project would be suitable for a single student or more than one working in a team. Would suit students with a background in Medical Physics, Engineering, Computing or Medicine.

Assessment of a novel method for automated EEG analysis

Supervisors: Prof David Holder and Dr. T. Tidswell

Student: Sivachelvan Jeyaseelan

In Electroencephalography (EEG), about 20 voltages are recorded from the scalp, over about 20 minutes. These are subsequently analyzed, to see if there are abnormalities in conditions such as epilepsy. The analysis is currently undertaken by a Clinical Physiologist and medically qualified Consultant. Methods have been developed to analyse records by computer, in order to quantify the information and save time. These employ the power of modern computers and are intended to speed up the laborious process of visual analysis of many pages of EEG.
In the supervisor’s group, a novel method for automated analysis of the EEG has been developed, which employs independent component analysis and a quadratic classifier. It is written in Matlab. The purpose of this project is to assess the accuracy of this new method on a group of about 30 EEG recordings in normal subjects. The outcome will be a comparison of its accuracy compared to visual identification by an expert and should lead to a peer reviewed publication in a scientific journal.

Skills to be acquired: students will spend some at UCH in the supervisor’s department learning the basic of EEG collection and analysis, and learn how to use the newly developed Matlab based EEG analysis software. They will then run it on the EEG records and use statistics to compare the results with the “gold standard” of visual inspection. Skills to be acquired will include one or more of:
Matlab programming;
Simple medical statistical analysis;
EEG use;
Experimental design and data analysis.
Suitable backgrounds: would suit students with a background in Medical Physics, Engineering, Computing or Medicine. Some experience in programming, ideally with Matlab, would be desirable.

Porosimetry of absorbent materials

Supervisors: Prof Alan Cottenden and Mihaela Soric

Student: (project still available)

Porosimetry involves measuring the capillary pressure of porous materials as a function of their saturation and, theoretically, data should correlate with the wicking and fluid retention properties of the materials. This project will test this theoretical prediction by conducting careful porosimetry, wicking and retention measurements on some absorbent fabrics used in medical devices. The results will help us with our ongoing efforts to understand the fluid handling properties of existing materials and develop better ones. The project will be primarily experimental and based at the UCL Archway campus (by Archway tube station).

How does the fibre footprint of a fabric on a surface depend on the stiffness of the surface?

Supervisors: Prof Alan Cottenden and David Cottenden

Student: (project still available)

As part of our work to reduce the impact of incontinence pads on the skin of their wearers, we have recently developed an optical microscopy technique for visualising and quantifying the fibre footprint of nonwoven fabrics of the kind used to face incontinence pads on skin surrogate materials. An associated mathematical model predicts that – up to a point - the footprint should be the same for compliant materials (like real skin) and stiffer materials (like glass). If proved to be true, this would be very helpful as it is much easier to make measurements on glass. The purpose of this project will be to test this prediction by measuring the footprint of a sample fabric against glass and against various silicone rubber skin surrogates under a range of loads and comparing the data. The project will be primarily experimental and based at the UCL Archway campus (by Archway tube station).

Using infrared light to investigation the absorption properties of nonwoven felts used in medical applications

Supervisors: Prof Alan Cottenden and Prof. Jem Hebden

Student: (project still available)

Absorbent nonwoven felts are used in a number of medical applications, notably incontinence pads and wound dressings. It is a major objective of the Continence and Skin Technology Group to establish a better understanding of how fluids interact with absorbent materials and build mathematical models which will enable the development of more effective medical products. This project will use an infrared device to investigate the absorption properties of nonwoven felts by mapping the distribution of fluid in them under a number of equilibrium (e.g. retention under gravity) and dynamic (e.g. horizontal and vertical wicking) experimental configurations. Data from the new device will be compared with both experimental data using other techniques and predictions based on existing mathematical models. The project will be primarily experimental and based at the UCL Archway campus (by Archway tube station).

Development of a new laboratory method for characterising the leakage performance of experimental absorbent materials

Supervisors: Prof Alan Cottenden and Mihaela Soric

Student: (project still available)

The Continence and Skin Technology Group is currently collaborating with two other universities and five companies to develop a new washable absorbent pant for lightly incontinent women. As part of the project, we have developed a test rig for measuring how much fluid a pad will hold under standard conditions before it leaks. Currently, we simply apply fluid at 2 ml/s until leakage occurs. The proposed project will involve investigating the impact on pad performance of using different flow rates and intermittent dosing patterns which may better reflect real use. The methodology developed in the project will be used to evaluate experimental fabrics provided by Leeds University and the data will be fed into our ongoing product development work. The project will be primarily experimental and based at the UCL Archway campus (by Archway tube station).

MRI of Human Disease

Supervisors: Prof Roger Ordidge; the second supervisor will be chosen according to the disease or organ which is the subject of the project.

Student: Zhe Toh

This project requires the student to study the application of MRI (or MR
Spectroscopy) to a specific human disease of their choice. This would
involve an extensive literature search followed by choice of the ideal
examination protocol and simulation of the ideal contrast for the study
using researched relaxation parameters and a MatLab simulation study of
contrast opimisation.

Skills: Interest in developing MRI knowledge and in becoming competent in
Matlab programming. Literature based research of existing problems and
methodology.

Implementation of de-blurring routines in Digital Tomosynthesis

Supervisors: Prof Robert Speller and Dr. Caroline Reid

Student: Vladimir Petelca

Conventional projection radiography is the most widely implemented method of x-ray screening methods, be it mammography, security screening of passenger luggage or quality control of industrial products. This method can be limited, firstly, through effective ‘stacking’ of objects in a projection image resulting in flattened images from which it is difficult to discriminate objects and, secondly, distortion of image information due to variations in x-ray absorption properties of imaged structures, for example, difficulty in distinguish between a thin sheet of strong absorber and a thick slab of weak absorber. These effects can lead to important information from the screening process being overlooked. It is proposed that this may be overcome through the use of digital x-ray tomosynthesis, an imaging technique currently attracting much attention in the medical imaging field. Tomosynthesis is a refinement of conventional geometric tomography methods where a finite number of projection images are acquired at varying orientations of the x-ray tube around the imaged object, from which a 3D image of the object is created. From these 3D images retrospective reconstruction is used to create focussed 2D slice images of an arbitrary number of planes through the object. It is proposed this method will be performed on objects moving on a conveyor system, leading to the name ‘On-belt Tomosynthesis’ (ObT).

Typically in tomosynthesis, a set of slice images are generated from the summation of a set of shifted projection images acquired at different orientations of the tube. This is referred to as the Shift-and-add (SAA) reconstruction. This SAA reconstruction takes into consideration the fact that the projection of objects at different heights above the detector will be dependent on the relative heights of the objects above the imaging plane. While one benefit of the SAA image reconstruction method is the small computing power required to run the algorithms, the resulting images are heavily affected by blurring as they contain images from every plane of the imaged object; one plane in focus and all the others smeared on top. Code for the SAA image reconstruction method has been developed. This project aims to implement a number of de-blurring routines, implemented on top of the SAA image reconstruction method, have been developed to remove artefacts and improve the reconstructed image quality. Test procedures will then be developed to assess the relative image quality of the reconstructed images. This project would be suitable for a student with an interest in computer programming and will involve mathematics.

Intelligent CT

Supervisors: Prof Robert Speller and Dr. Peter Munro

Student: Yi Zheng

X-ray and gamma ray imaging are still the most frequently carried out examinations despite the concern for radiation burden. Recently the UCL Radiation Physics Group developed a technique capable of reducing dose without a loss of image quality – the technique is called I-ImaS. This technique adjust the imaging conditions on-the-fly to suit each local region in the object being imaged. The technique was designed for planar imaging but now we wish to extend this to CT.

A project to investigate the I-ImaS_CT concept. Last year a student demonstrated that the concept is viable for dose reduction but did not really study the steering algorithm that should be used to control exposure. Furthermore only one type of imaging task was studied. The project requires phantoms to be built, software to be written and many experiments carried out using the X-Tek CT system. This is an u/g merging into an MSc project.

"RadiCal" - a new concept in detector development

Supervisors: Prof Robert Speller and Dr. Konstantin Ignatyev

Students: Evangelos Inglesis and George Randall

Many monitoring devices of radioactivity exist but none can identify the direction in which the radiation source exists without the use of a collimator. Recently a new device has been suggested to overcome this problem – the RadICal detector.

The RadiCal concept could be tested both experimentally and by the use of modelling techniques. The route that will be taken will depend upon the student’s interests. Experimental work will be undertaken with existing equipment and modelling can be adapted from existing codes. The project will to test the feasibility of the concept and attempt to optimise the design for different applications.

Deconvolution of experimental phase contrast patterns to simulate acquisitions with higher spatial resolution devices

Supervisors: Dr Sandro Olivo and Dr. Konstantin Ignatyev

Student: (project still available)

X-ray phase contrast imaging is a new, exciting imaging modality with the potential of revolutionizing the world of diagnostic radiology over the next years. It generates image contrast from the phase changes of x-rays instead of x-ray absorption, and as a consequence it solves the main problem of diagnostic radiology i.e. poor image contrast due to low absorption differences. Fields like mammography were demonstrated to benefit enormously from this approach.
Deconvolution is a classic method of increasing image quality, especially in terms of resolution.
The student would be provided with a dataset of phase contrast images, open-source deconvolution and phase contrast modeling software, and would have the accomplish the following tasks:
- process phase contrast images acquired with a given spatial resolution;
- simulate phase contrast images with a finer resolution;
- match processed experimental and simulated images.
This project would allow the student to acquire fundamental skills in data analysis and simulation methods. Some basic degree of computing skills is required. For students at MSc level, some degree of interaction/modification/improvement of the open source code provided would be required.

Simulation/analysis of Talbot self-images through methods based on Fresnel/Kirchoff integration

Supervisors: Dr Sandro Olivo and Dr. Peter Munro

Student: (project still available)

One of the basic problems in optics is near field diffraction, which can be described by means of Fresnel/Kirchoff diffraction integrals. This aspect has recently acquired more and more relevance in medical imaging due to an increasing number of fields which strongly benefit from the introduction of x-ray phase contrast techniques (see http://www.medphys.ucl.ac.uk/research/acadradphys/researchactivities/pci.htm).
At given distances, called the Talbot distances, near field diffraction can create self-images of an object. This property is explicitly exploited in some x-ray phase contrast imaging approaches, whereas in other methods it is considered a disturbance that has to be minimized.
The student would be provided with software capable of solving Fresnel/Kirchoff diffraction integrals for a given number of objects, and would be expected to:
- modify it to account for a broader range of objects;
- study near-field diffraction and self-imaging aspects of such objects;
- analyze the influence of these effects on x-ray phase contrast imaging approaches.
This project would allow the student to acquire fundamental skills in simulation methods and data analysis, and to familiarize with theoretical models. Computing skills are required.

Optimal techniques for modeling x-ray phase contrast imaging systems

Supervisors: Dr Peter Munro and Dr. Sandro Olivo

Student: (project still available)

The bulk of current day x-ray imaging systems image an object's x-ray absorption profile. Most objects of interest in medicine and biology are, however, predominantly phase objects which delay incident x-rays in accordance their composition. Our group is currently building a novel system for performing phase contrast imaging and need to model a variety of phenomena to aid in the design of the system.

Models of our and related systems employ scalar diffraction theory in the Fresnel regime. The application of such theory is still in its infancy in x-ray imaging. We currently use a reasonably direct application of scalar diffraction to model our system however we are proposing a project in which a student would consider a number of techniques which may be employed to efficiently model phase contrast imaging systems. In particular, the student would be encouraged to consider how the chirp z transform, Wigner distributions and the method of stationary phase can be used to complement or replace our existing modeling techniques. The student may then apply the developed techniques in the study of important phenomena which arise in x-ray phase contrast imaging systems such as Talbot self imaging.

The student will need to have good mathematical skills and be experienced in computer programming, preferably in Matlab or C. Skills in developing and testing models will be acquired as well as a good understanding of x-ray phase contrast imaging systems.

Cardiac functional analysis using image registration techniques

Supervisors: Dr Xiahai Zhuang

Student: (project still available)

This project is to investigate the algorithms
used to compute cardiac functional indexes. These indexes are measured from
the volume size of local regions (image segmentation techniques) and
displacements of the myocardium at different time points (image
registration techniques). This project will work with DICOM data and
extract useful information from DICOM header using provided libraries. The
majority of algorithms in the project may come from the existing tools or
softwares, but investigators are expected to design and implement a
workable pipe-line for the functional analysis. Therefore, the project may
also involve some new ideas/methods to improve the performance of the
designed pipe-line.

skills required: Computing, c++ program, user interface using visual studio (e.g c#).

Calculating Image Quality Metrics in a Sparse Domain

Supervisors: Dr David Atkinson and Dr Andrew Melbourne

Student: (project still available)

Magetic Resonance Imaging (MRI) can suffer from patient motion during the acquisition of scan data. This can lead to a ghosting and blurring of the image. If the motion and its timing during the acquisition can be estimated, then corrections can be applied to the acquired k-space data and higher quality images achieved. New techniques such as compressed sensing are being investigated to reduce the amount of data required and shorten acquisition times.

One mechanism to determine motion is to apply a correction for a trial motion
and assess the image quality, accepting corrections that improve the image. Image quality metrics such as entropy are usually computed directly on images. Entropy works best when there are motion ghosts overlying image regions that would otherwise be dark, for example the air outside the head.
For subtle changes within the brain, the entropy changes may be less distinct and less sensitive to motion.
Recently there has been an interest in the use of domains where the data is relatively
sparse, e.g. wavelet domain or gradient domain. The interest stems from the application of compressed sensing to reduce acquisition times. In this project, the student will investigate the effect of motion on the signal in the sparse domain. It may be the case
that entropy, or other measures, are more sensitive to motion in this domain and we can provide a more effective correction. Furthermore, the study could indicate how motion may interact with a compressed sensing reconstruction.

The project will require skills in MATLAB, complex numbers, MRI understanding and optimisation.

Faster Motion Correction of MR Images

Supervisors: Dr David Atkinson and Dr Andrew Melbourne

Student: (project still available)

Magetic Resonance Imaging (MRI) can suffer from patient motion during the acquisition of scan data. This can lead to a ghosting and blurring of the image. If the motion and its
timing during the acquisition can be estimated, then corrections can be applied to the acquired k-space data and higher quality images achieved.

"Autofocus" is a technique for iteratively finding this motion but because of the
large number of unknowns, it can be a slow algorithm and subject to getting stuck in local minima. Autofocus usually relies on computing an image quality measure such as
entropy on a complete image. To form this image, traditionally all the acquired k-space data from throughout the scan is Fourier Transformed.
In this project, the student will investigate finding the motion parameters for
successively acquired portions of k-space in turn. By finding the motion for each portion of k-space in turn, the number of possible combinations of motion and timings is reduced, hopefully leading to a faster algorithm that does not stick in local minima. The intermediate images will be wrapped or aliased so it is unclear how well entropy or other focus criteria will perform and this will be investigated.

The project will require skills in MATLAB, complex numbers, MRI understanding and
optimisation.

Image registration and contrast enhanced MRI

Supervisors: Dr Andrew Melbourne and Dr David Atkinson

Student: (project still available)

Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) is a widely used technique for quantitative assessment of the vascular properties of tissue and is now widely used in oncology for screening, diagnosis and staging. Observation of the rate of change of image intensity and the pattern of enhancement yields quantitative information that allows the assessment of the underlying vascular status. Unfortunately, motion artefacts can have a strong influence on the subsequent analysis of pharmacokinetic parameters estimated by model-fitting.

Correction of motion artefacts may be carried out using automated 'image registration'
algorithms to align spatial features. These algorithms often need modification to improve
robustness to local contrast change. A number of recent registration methods operate using the results of model fitting procedures (produced from data with sub-optimal motion correction), to subsequently guide an image registration algorithm prior to a second round of improved model-fitting. Model-fitting may be explicit, motivated by physiological considerations or implicit, substituting physiological models for data-driven techniques such as principal or independent component analysis.

In this project, the student will investigate the use of a principal components analysis
based registration algorithm in combination with a highly optimised, open-source image
registration algorithm. Of particular importance is an investigation of the influence of
registration algorithms on the accuracy of pharmacokinetic parameters applied to both real and simulated MRI data. Furthermore, the study could investigate some of the bottlenecks present in DCE-MRI registration and analysis that could be parallelised using a high-performance GPU.

This project will require skills in Matlab, C++, and some basic knowledge of MRI acquisition.

TB: Learning from 3D CT to aid diagnosis from 2D X-rays

Supervisors: Dr John Hipwell and Dr. Jamie McClelland

Student: Argyrios Christodoulidis

TB has reemerged as a major cause of death in the developing world due to the global pandemic of the human immunodeficiency virus (HIV) and drug resistant strains of M. tuberculosis (Mtb). Successful control of tuberculosis is dependent on early detection and diagnosis of active disease. Current strategies in resource poor settings depend on clinical examination and sputum smear microscopy. Facilities for X-ray imaging and the skills for accurate interpretation are rarely available.
The continued reduction in cost of digital X-ray detectors, makes affordable, low maintenance X-ray equipment widely available, substantially reducing the imaging cost per patient. In addition, the increasing coverage of cell phone networks in the developing world, coupled with advances in the power and capacity of hand-held cell-phone devices, means portable equipment can be developed to image, screen and analyse chest X-rays for signs of TB. Low cost X-ray, together with reliable CAD will facilitate accessible and rapid assessment of advanced TB, allowing early intervention which will limit the dissemination of tuberculosis and allow early access to treatment programmes for the patient, reducing the risk of development of severe disease and drug resistance.
This project will build on our experience in aligning 3D medical imaging modalities, such as CT and magnetic resonance imaging, with 2D X-rays. You will adapt these techniques, previously developed on the head, pelvis, knee and breast, to enable abdominal CT and chest X-rays to be aligned according to the differences in position between X-ray and CT. Having established the transformation to relate locations in the CT image to points in the chest X-ray, we will investigate the appearance of TB pathology in CT and develop 3D models of this pathology which can be projected onto the 2D chest X-ray to detect the presence or absence of the disease in the chest X-ray.

There will be a high computing component to this project. The student will be encouraged to use the Insight Toolkit C++ library.

Mammographic Image Analysis

Supervisors: Dr John Hipwell and Prof. David Hawkes

Student: Amanda TURNER

The lifetime risk of being diagnosed with breast cancer in women is 1 in 9 and breast cancer is now the most common cancer in the UK with more than 44,000 women being diagnosed with breast cancer each year.
In the literature various factors have been reported that influence the probability of developing cancer and the actual phenotype of the tumour. These include (1) Hormonal factors such as age of menarche and menopause, and age at which children are born, (2) environmental factors such as smoking, (3) family history and (4) known gene mutations and polymorphisms such as those in the BRCA1 and BRCA2 genes conferring a high cancer risk to a small proportion of the population, and those in FGFR2 conferring a modest increase in risk to a much larger population.
There is also an established link between the crude, manual measurement of “breast density” from X-ray mammograms and breast cancer risk. This project will focus on developing a more sophisticated, textural analysis of mammograms and investigating whether they can be used to distinguish two cohorts of data: a set of normals and a set of women with a known high risk of breast cancer. These images are currently being collected by geneticists in Dundee as part of the EU project HAMAM.

There will be a high computing component to this project. The student will be encouraged to use the Insight Toolkit C++ library.

Imaging the Microstructure of the Brain PART 1

Supervisors: Dr Ivana Drobnjak

Student: (project still available)

Measuring microstructure parameters of brain tissue, such as axon radius, in vivo is a challenge in diffusion MRI. Recent studies showed that using pulse sequences with oscillating or chirped diffusion-gradient pulses may provide better sensitivity to microstructure features than the typically used pulsed-gradient spin-echo (PGSE) or stimulated-echo (STEAM) sequences, both of which use a rectangular diffusion-gradient pulse.

However, diffusion-gradients are usually very strong (can go as high as is |G|=0.08T/m on clinical scanners) and when allowed to oscillate on high frequencies they could potentially create significant gradient-related artifacts such as eddy-currents, gradient-nonlinearities or gradient-inhomogeneities. In order to achieve reliable and robust diffusion–weighted images it is necessary to investigate the extent to which these gradient-related artefacts can affect MR images.

To this end, a computational model of the MR image acquisition process was built, POSSUM, which uses a geometric definition of the object (brain), Maxwell’s equations (to model the magnetic field in the scanner), Bloch equations (to model the behaviour of the nuclear magnetisation). POSSUM forms a part of the FMRIB Software Library (FSL) that is a very widely used analysis package for FMRI (used in over 600 laboratories worldwide).

This project aims to extend the current POSSUM software to make simulations of gradient-related artefacts, and then apply it in order to look into the potential effect these artefacts can have on MR images. The project would involve finding fast, accurate and efficient numerical/analytical solutions to a set of ODE’s and then implementing those in C++ language into the already existing software environment. The project would also involve learning about Magnetic Resonance Imaging.

Imaging the Microstructure of the Brain PART 2

Supervisors: Dr Ivana Drobnjak

Student: Jeanne Gaspard

In many brain diseases such as Alzheimer’s or multiple sclerosis, the microstructure of the brain changes and neurons change their size, density or organization. In order to be able to understand what is happening in the brain a way of “looking into” the brain is needed. However, measuring, non-invasively, the microstructure (e.g. the size of neurons) in a living brain is a big challenge.

In order to do so, we are using diffusion MRI, which is a magnetic resonance imaging (MRI) method that produces in-vivo images of biological tissues weighted with the local microstructural characteristics of water diffusion. Recent research showed that imaging strategy can be optimised to be neuron-size specific.. For example, if we would like to image very big neurons (with big radiuses) we would use very low frequency diffusion gradients (magnetic fields in the scanner), while if those were very small neurons we would use high frequency ones.

This project focuses on understanding the relationship between the imaging sequence and the size of the neurons. Expressing the sequence in a parameterised form using, for example, frequency and magnitude of diffusion gradients, can do this. Finding analytical expression for the imaging protocols would significantly improve the way we image brain microstructure and could potentially benefit in diagnosing a wide range of diseases, in particular cancers and dementias.

Requirements for this project are: programming in MATLAB, some knowledge of MRI, and mathematics.

Reconstruction and comparison of whole-brain structural connectivity networks

Supervisors: Dr Jonathan Clayden and Dr Chris Clark

Student: Christopher Parker

A complex network of region-to-region connectivity provides the physical substrate for information processing in the brain. Diffusion magnetic resonance imaging, combined with computational methods, can be used to reconstruct this pattern of connectivity; but it is not yet clear how robust this reconstruction is, what the natural variability is between different people, or how such networks should be compared.

The aim of this project will be to reconstruct whole brain connectivity networks from MRI data using computational methods, and to investigate the variability in connectivity patterns across individuals. The aim is to move towards a robust "connectivity fingerprinting" approach that might be useful for clinical diagnosis or prognosis.

Some programming experience will be required for the project. Any previous experience with MRI, Unix/Linux, image processing or vectorised programming (using R, Matlab or similar) would be useful, but none of these are strictly required.

A phantom for proton radiography

Supervisors: Dr. Gary Royle and Edgar Gelover Reyes

Student: Maria Panayiotou

Proton therapy is a highly advanced form of radiotherapy treatment which can deliver a conformal radiation dose to the tumour site much more accurately than conventional radiotherapy. It has proven to be very successful in treating difficult tumours close to critical organs and paediatric cancers. The better conformality of the treatment site means that it is even more important to be able to accurately localize the tumour volume within the patient. A system for proton radiography is currently being developed which aims to image the patient with the proton beam immediately prior to treatment to ensure they are correctly positioned. This project will involve learning about proton therapy and proton radiography, performing some computer simulations of a proton treatment and developing test objects that we can use to evaluate the performance of the proton radiography technique.

A 3D optical scanner for radiotherapy dose mapping

Supervisors: Dr. Gary Royle and Dr.Jenny Griffiths

Student: Nya Boayue

Modern radiotherapy equipment is capable of delivering very complex 3D radiation dose volumes to patients to match the shape of the region to be treated. It is important to be able to check that the equipment is delivering the dose exactly in accordance with the planned treatment. 3D imaging detectors exist on a small scale but current technology prevents patient size systems. Dosimetry gels exist that can change its chemical properties according to the amount of radiation dose it receives. Certain gels can change colour or opacity. The aim of this project is to develop a simple tomographic optical scanner that can image the gel and produce a 3D map of the opaque region, thereby producing a 3D dose map. This project will involve learning about modern radiotherapy treatments and computerized tomography techniques, optimizing optical scanning equipment, and producing 3D images of test objects.

Image correction procedures for diffraction CT

Supervisors: Dr. Gary Royle and Kate Pepper

Student: (project still available)

X-ray diffraction is a technique which has long been used to identify the internal composition of an object. The group at UCL has applied x-ray diffraction analysis methods to investigate disease states of biological samples. A system has been built which can produce high resolution 2D and 3D images of such samples. The system is ultimately to be used to locate and identify cancerous cells within a tissue sample. This project will involve learning about x-ray imaging and x-ray diffraction analysis, obtaining some images of test objects using the system and implementing algorithms to improve the quality of the images produced in order to correct for the various interactions in the sample.

Registration of images (in medicine), improving robustness and testing for accuracy

Supervisors: Dr. Tryphon Lambrou and Prof. Andrew Todd-Pokropek

Student: (project still available)

Several algorithms exist (and have been developed) for registering similar images in medicine, for example variance minimisation and mutual information maximization, for rigid registration, optic flow and thin plate spline for non-rigid registration. The aim of this project is to pursue this work with the view of evaluating the performance of these types of algorithms, in particular when working with relatively different types of image data, for example images from MRI and Nuclear Medicine, and for different applications, for example images of the heart or liver as opposed to the more conventional images of the head. In general these algorithms are slow (if they are robust) and in any case perform more poorly as the difference between the types of image increases. Part of this project will be analytical, that is studying the performance of the algorithms for example for different parameters of the optimization. For non-rigid registration, the process of evaluation requires the assessment of the accuracy of registration at every point in the field of view, although not necessarily with equal weighting. The process can be tested by deforming an object in a know manner, and then re-registering it back, both of which can be represented as a vector field. The main part of this method will be to devise a method for evaluating the difference between these two vector fields (for example by fitting an active shape model) and investigating parameters to intercompare different non-rigid registration methods. An extension of this which is novel is to look at the vector deformation field, to decompose it into divergence and curl components and to use these to look at the local field (for example to decide if the local field can be expressed as a rigid transformation.

Deformation of images (in medicine) based on constrained models

Supervisors: Dr. Tryphon Lambrou and Prof. Andrew Todd-Pokropek

Student: (project still available)

Deformable images are required for matching to atlases and for other special purposes such as image subtraction in angiography. Several models for deformation exist, polynomial, thin plate spline and diffusion based. For the example of brain images, it is clear that the constraints should be position (and possibly shape) dependent, for example much greater weight being applied at the centre than at edges. This project will investigate and test the accuracy of several model based deformable models with the aim of find the most appropriate for use in generating a brain atlas including various moments (mean, variance) about the expected values and their positional reliability. Again a critical parameter is that of validation and overall measures of the vector displacement field provide tools for assessing local and overall accuracy. Part of the project, since simulated data would be required to provide a suitable testbed and gold standard is to investigate methods for creating images of a different type from real acquired images for example PET like imaged from MR raw data.
An associate task is to develop a Finite Element soft tissue deformation model (cf Schnabel IPMI 2001) for an organ such as the liver or the heart.

Texture analysis of radiographic data

Supervisors: Dr. Tryphon Lambrou and Prof. Andrew Todd-Pokropek

Student: (project still available)

Images such as mammograms and liver CT and MR images contain features which it appears may associate texture as an indication of the clinical type and stage of different tumours. The aim would be to extend (and devise new parameters) being local texture estimates, such as local spatial frequency and fractal order, as an aim to provide a good texture estimate. An alternative target would be to investigate the classification of MR bone image by the use of texture to investigate its accuracy with respect to identification of Osteoporosis. Considerable amounts of data are available and a suitable programming environment could be either PC or Sun. A particular aim would be to compare several different texture analysis methods on one of the available datasets. A sub-aim would be to investigate optimal window size over which texture should be evaluated in such images, a good example of which is the use of wavelet for classification.
An alternative approach would be to implement and test the Pattern Fed Objects (PAFO) algorithm as an alternative method for texture analysis and segmentation.

Lesion detection and classification in lung CT scans

Supervisors: Dr. Tryphon Lambrou and Prof. Andrew Todd-Pokropek

Student: Huu K Nguyen

It is proposed to use lung CT scans for screening for lung cancer. Early indications of lung cancer are the presents of small lesions (2-8mm diameter) in the lung field, but the problem is complicated by the present of considerable numbers of other structures such as blood vessels etc. The first task is to develop a method for removing confounding information for example filtering out linear structures. The second is to develop a sensitive detector for the presence of small lesions (which are not always circular/ spherical) One method which could be applied is the use of a neural net after suitable preprocessing. The final task (if there is time) is to look at classification methods of which the most promising is likely to be a fuzzy texture evaluation (although the neural net might be extended to perform the same process). The difference between objects detected which are normal and those corresponding to possible tumours is typically the variation of shape of the variation of pixel values within the object. The use of a fuzzy method is suggested since the exact localisation of the lesions is difficult. Several methods exist (for example radial profiles, matched filters, neural nets etc) which should be intercompared.
An interesting extensionn of this is the extraction of arterial (venous) and bronchial trees with the aim of improving detection of nodules.
[This project could use liver scan images as an alternative]

Shape description and segmentation of bones in a projection image

Supervisors: Dr. Tryphon Lambrou and Prof. Andrew Todd-Pokropek

Student: (project still available)

Conventional x-rays are projections where groups of objects are superimposed. This project which has important implications in the automated analysis of image of infants suffering from different bone dysplasia syndromes (in collaborations with Great Ormond St. Hosp. for Sick Children) is concerned with taking the image of a skeleton, identifying different bones (e.g. vertebrae, ribs, tibia, fibula, femur etc, and segmenting them in such as way that individual shape outline of the different bones can be identified (with the possibility of overlapping structures being the major problem) and then characterizing the shapes into short, long, straight, bent, deformed and various other more specialized categories with the aim of eventually linking into an expert system being developed independently. Data would be acquired on a laser digitized and be (initially) of high resolution - 2000x2500 pixels.
A previous project used snakes to segment such images which proved to be not very robust. An alternative techniques which could be employed is the use of active appearance models where the image is decomposed into images of different modes of variation and placement of nodes can be optimised. An alternative problem of considerable interest other than that of segmentation is that of the robust extraction of symbolic (or textual) descriptions. An active shape model is a useful extension.
[This project needs extending into analysis methods]

Classification of microcalcifications in mammograms

Supervisors: Dr. Tryphon Lambrou and Prof. Andrew Todd-Pokropek

Student: Divesh Tanwani

Work has already been undertaken in the detection of individual and groups of microcalcifications on mammograms. Simple parameters such as size, roundness, linear arrangement etc can assist in the classification of the mammogram into malignant or benign. However, so far, no information from other structures in the breast, for example presence of a lesion, segmental distribution etc has been implement in order to use other than purely local parameters. This project would extend the classification and description of clusters of microcalcificaion by including other descriptive parameters associated with other detected structures in the mammographic images.
[An interesting extension would be into defining and ‘detecting’ asymmetry]

Segmentation of White and Grey matter to classify abnormal lobes in brain images

Supervisors: Dr. Tryphon Lambrou and Prof. Andrew Todd-Pokropek

Student: (project still available)

MR Brain image segmentation for example into white and grey matter is complication by several factors: non-uniform signal response, partial volume effects, difficulty in identification of individual lobes etc. Good simulated data is available form Montreal Inst. Of Neurology which can be used as a test bed. So far quantitative assessment of the ratio of grey to white matter has not be validated in the no gold standard data is available. This project would use existing tools to investigate the accuracy of such classification measure and to investigate their sensitivity with respect to the several sources of error. The use of a 3D clustering algorithm such as proposed by Udupa could be of considerable interest.
An associate project is that of the detection, classification of tracking of the sizes, shapes and ‘intensities’ of multiple sclerosis lesions over time to investigate if such temporal information can be of prognostic value; e.g. is the condition essential stable or is it deteriorating rapidly. The distribution of plaques is different regions of the brain seem to be of considerable significance.
An associated problem with respect to MS lesions is to develop a robust method to define spatial localisation (above below left right) between such different rather fuzzy objects, which requires notions of fuzzy logic to be employed.

Pre-processing of CT images for cranioplasty planning

Supervisors: Dr. Clifford Ruff and Dr. Liu Ming

Student: (project still available)

The cranioplasty unit of the medical physics department in UCLH provides the NHS with over 150 custom built cranioplasty plates per year. A number of the patients who require plates to be made have existing repairs that need to be removed from the CT images before the new plate can be designed to custom fit the patients skull. The repairs of generally made of a denser material than the bone they replace.

At present the pre-existing repairs are removed by hand, the aim of this project is to develop software that automatically removes the pre-existing repairs from the CT images, as such some programming experience, preferably in C++, is required.

Statistical Shape Modelling of the skull

Student: Lumbani Munthali

Supervisors: Dr. Clifford Ruff and Dr. Liu Ming

The cranioplasty unit of the medical physics department in UCLH provides the NHS with over 150 custom built cranioplasty plates per year. Patients normally present with a skull defect that needs covering with a custom built titanium plate. The current method of designing the plate relies on mirroring the patients anatomy to provide an initial estimate of the repair shape. This method is not available however when the defect crosses the patients midline. Statistical Shape analysis should be able to provide a method of finding a repair shape that matches the patients anatomy in these cases, however in order for this to function a training set of skulls needs to be landmarked for modelling purposes.

We have available to us shape modelling software (in the form of Point Distribution Models) and a number of complete CT skulls images that could form the training set for a PDM. The aim of this project is to develop and test possible labelling schemes to capture the part of the skull being modelled.

Ventilation imaging of the lungs using deformable registration of CT images

Student: Laura Shell

Supervisors: Dr. Jamie McClelland and Prof. David Hawkes

Images of the local ventilation in the lungs, that is the amount of air entering a particular area of the lungs during respiration, have a number of potential uses. They can help to avoid the most functional and healthy regions of the lung when planning Radiotherapy or other more invasive procedures. They can be used to help diagnose and further understand various respiratory diseases such as COPD and can help to assess the response of these diseases to different treatments. There are a number of established methods of forming such images including using Xeon CT, Hyper-polarised MR imaging, and SPECT imaging, but each of these has a number of associated problems or drawbacks. Another relatively recently proposed method uses deformable registration of CT images, either 4DCT images acquired while the patient is freely breathing, or high quality CT images acquired at breath-hold to form the ventilation images. The CT values in the aligned datasets can be used to calculate the change in density of the lung tissue, and the registration results themselves can be analysed to calculate the local volume change of the lung tissue, and these values can be combined to calculate the local change in air volume (ventilation) and also the local change in tissue volume (amount of mass). However, this method is potentially very sensitive to the accuracy of the deformable registration results as well as any artefacts in the CT images (which are often present in 4DCT images).

In this project the student will investigate the effect of using different CT images (breath-holld vs 4DCT) and the effect of using different deformable registration results on the ventilation images that are generated. This project will require good computational abilities, in particular with Matlab and/or C++, and a basic understanding of (or interest to learn about) CT acquisition, lung physiology, and deformable image registration. This project will hopefully lead to a conference presentation and/or a journal paper.

Computer-Aided Diagnosis of Prostate Cancer from MRI

Student: Ellen Thomma

Supervisors: Dr. Tim Carter, Dr. Shonit Punwani and Dr. Dean Barratt

Current evidence suggests that multi-functional (i.e. combined T2-weighted, diffusion weighted and contrast enhanced) magnetic resonance imaging has both a sensitivity and specificity of 70-90% for prostate cancer. These large and complex images require very specialist knowledge and are time-consuming for radiologists to report. A computer-aided detection algorithm would speed up this process, and reduce the chance of a cancer being overlooked.

We have a large and probably unique dataset of MR images and aligned pathology. You will use this dataset to train a simple machine learning algorithm to identify prostate cancer. The project can then be developed in two directions, according to your interests. Either further computer analysis of the MR images can be performed (to improve the current identification of features indicative of cancer or to develop new ones) or alternative machine learning algorithms can be investigated. This project will involve working closely with our clinical partners at UCLH, as well as programming in Matlab.

Development of a 3D-ultrasound-guided system for prostate cancer biopsy

Student: (project still available)

Supervisors: Dr. Dean Barratt

PROJECT AIMS: The aim of this project is to develop one or more components of a new needle biopsy device that supports 3D ultrasound imaging of the prostate gland and provides realtime feedback of the current ultrasound probe/needle position relative to the gland using 3D graphics. In order to achieve this, the probe will be tracked by a 3D position sensing device, which continuously measures its 3D position and orientation. In this project, the aim will be to build and test a basic prototype system, with the opportunity to focus on developing one particular software or hardware component. An important task will be to investigate the accuracy and robustness of electromagnetic versus optical probe tracking in a clinical environment. Electomagnetic tracking devices can be prone to error due to ferro-magnetic materials or electromagnetic fields, whereas optical devices require a line-of-sight between the (infra-red) camera system and the passive (reflective) or active (light-emitting) markers attached to the tracked object. A further important feature of the system will be the ability to register MR images, obtained prior to biopsy, to the 3D ultrasound images to enable tumour targeting. A commercial research ultrasound scanner (the Ultrasonix RP) will be used as the main development platform, and a number of commercial tracking devices are available for experiments. The Ultrasonix scanner allows full access to image data and has a programmable interface with which custom-designed software (written in C/C++ or Matlab) can be integrated.

SKILLS: Good practical skills and familiarity with C/C++ programming and Matlab.

BACKGROUND: Prostate cancer is now the most common cancer among men in the UK and many other countries in Europe, and is a leading cause of cancer-related death. Following an initial blood test and/or digital rectal examination, the standard method for diagnosing the disease is transrectal needle biopsy in which tissue samples are obtained by inserting a needle into the prostate gland through the rectal wall. These samples are subsequently analysed histologically to determine the presence of cancer cells and the grade (i.e. aggressiveness) of the cancer. The procedure is routinely guided by ultrasound imaging, but current clinical ultrasound scanners have two major limitations: Firstly, they present only two-dimensional images representing sections through the prostate. This makes accurate three-dimensional navigation and needle tip placement practically difficult, especially for the inexperienced clinician. As a result, predetermined locations within the prostate gland are often very difficult to sample accurately, leading to substantial variation in the results obtained by different operators and between different hospitals. It is also difficult to determine with any accuracy precisely where samples have come from. Secondly, it is, in general, impossible to reliably distinguish cancerous tissue from normal or benign tissue on ultrasound images. The clinician is therefore not able to target suspected tumours during biopsy. This means that clinicians resort to systematic sampling of the whole gland, typically taking 10-15 samples using modern protocols. Although this method is effective for detecting the majority of cancers, around 1 in 10 patients who have clinically-significant prostate cancer – i.e. cancer that requires treatment – require one or more repeat biopsies before their cancer is detected. Moreover, because standard prostate biopsy is subject to sampling error, the grade of disease is not determined accurately in 20-40% of patients. This is a particular problem because cancer grade is a key factor in determining the prognosis and therefore significant grading inaccuracy has profound implications for the clinical management of patients with prostate cancer.

CLINICAL IMPACT: A key motivation for the proposed system is to improve the accuracy of the conventional prostate cancer biopsy technique for detecting and characterising the disease in terms of histopathological grade and tumour volume. We hope that this will in turn lead to more accurate diagnosis and greater confidence in the assignment of appropriate treatment options. Furthermore, the ability to perform targeted biopsy using MR images opens up the possibility of highly selective sampling and sampling schemes tailored to the individual patient. Targeted sampling in particular is likely to require far fewer biopsy samples than conventional biopsy to arrive at an accurate diagnosis, minimising patient discomfort and risk of infection.

Vessel-based image registration for guiding bile-duct interventions

Student: (project still available)

Supervisors: Dr. Dean Barratt and Dr. Erik Rijkhorst

PROJECT AIMS: The aim of this project will be to investigate the application of a vessel-based image registration technique to the problem of registering (i.e. spatially aligning) 3D MRCP images of the bile-ducts with 2D ERCP (x-ray) images acquired during a minimally-invasive surgical procedures, such as biopsy, stenting, and photodynamic therapy, to aid surgical navigation. The project will involve image processing (in Matlab or C/C++/ITK) to enhance the ducts using filtering techniques normally used to enhance blood vessels and other tubular structures. It will also involve adapting a new deformable vessel-based registration algorithm to solve the 2D-3D registration problem. Simulated and patient data will be used to test the registration accuracy by comparing the relative locations of image-visible anatomical landmarks before and after registration.

SKILLS: Good mathematical ability and familiarity with C/C++ and ITK programming or Matlab.

BACKGROUND: Bile duct carcinoma is a rare but serious condition, which often presents in the advanced stages and is difficult to treat surgically; most patients diagnosed with the disease are not eligible for surgery and the life expectancy of patients treated with the chemo- or radiotherapy alone is severely limited (typically 12-18 months). Therefore, in many cases, treatment is largely palliative. One common intervention is the placement of stents in the affected bile duct to open up the lumen and relieve symptoms. This is generally done endoscopically using ERCP imaging - an x-ray technique in which the ductal tree is visualised by injecting radiographic dye into a common duct. More recently, the efficacy of novel, endoscopically-delivered therapies, such as photodynamic therapy (PDT) are being investigated as future treatments. However, all of these techniques are technically demanding and would potentially benefit from 3D guidance using 3D MR images, fused with interventional ERCP images, to provide a “road-map” to guide the surgeon through the complex 3D anatomy.

CLINICAL IMPACT: The major anticipated benefits of MR/ERCP-guided bile-duct cancer interventions are the potential for technically easier, more reliable, and shorter procedures through improved 3D navigation.