Edinburgh Data-Intensive Research - PhD student project

References:

1. Dickie, D.A., et al. (2012). Do brain image databanks support understanding of normal ageing brain structure? A systematic review. Eur Radiol. 22, 1385-1394. 2. Farrell, C., et al. (2009). Development and initial testing of normal reference MR images for the brain at ages 65–70 and 75–80 years. Eur Radiol. 19, 177–183. 3. Mazziotta, J.C., et al. (2009). The myth of the normal, average human brain - the ICBM experience: (1) Subject screening and eligibility. Neuroimage. 44, 914-922. 4. Freedman, D. (2010). Statistical Models and Causal Inference: A Dialogue with the Social Sciences. Cambridge University Press, Cambridge. 5. Breteler, M., et al. (1994). Cerebral white matter lesions, vascular risk factors, and cognitive function in a population-based study. Neurology. 44,1246-1252.

Distributed multi-modal image collection and analysis

Fan.Zhu — Fri, 16 Nov 2012 15:12:56 +0000

Multimodal Image data banks, of normal [3] and pathological subjects, are of great utility for improving collaboration and performing research with greater statistical power. The acquisition of images is expensive and time consuming; therefore it is important to reuse them. We are currently developing human brain image data banks, one of which is likely to be the largest bank of normal aging brains in the world.

To maximise the benefits of such data banks it is necessary to investigate ways to organise and aggregate multimodal images and their associated clinical, cognitive and metadata. For derived data, provenance data should be automatically captured. Issues include: spatial registration between differently images, differentiation of the factors contributing to image variation, reliability and performance of analysis methods; and satisfaction of ethics, security and privacy policies. The knowledge representation must be open to diversity and innovation, yet support inference over the accumulated data. To gain commitment from data and method contributors, data must be identified and citable with proper attribution.

The student would pioneer data-intensive methods of automated analysis to produce metadata and to support statistical analysis, data mining and information presentation. The student would work closely with existing teams of specialists within CIVIS.

Attachment	Size
PhD_CIVIS_BrainBank_dj2_mpa.doc	37.5 KB

Supervisors @ NeSC:

Project status:

Still available

Degree level:

PhD

Other supervisors:

Dominic E. Job

Background:

Computer Science, mathematical sciences or engineering with strong foundations in computing

Subject areas:

Student project type:

References:

1. Slomka P, Baum, R. Multimodality image registration with software: state-of-the-art. EJNMMI. 36 2. Gorgolewski K, Burns CD, Madison C, Clark D, Halchenko YO, Waskom ML, Ghosh SS. (2011). Nipype: a flexible, lightweight and extensible neuroimaging data processing framework in Python. Front. Neuroimform. 5:13 3. Dickie, D.A., Job, D.E., Poole, I., Ahearn, T.S., Staff, R.T., Murray, A.D., Wardlaw, J.M., 2012. Do brain image databanks support understanding of normal ageing brain structure? A systematic review. Eur. Radiol. 22, 1385-1394.

Early detection of infarcts by improving brain perfusion imaging analysis

Fan.Zhu — Fri, 16 Nov 2012 14:54:27 +0000

Outcome after severe ischemic stroke may improve with thrombolysis. Some studies have shown that parametric perfusion maps and other information might be useful in selecting patients for this potentially hazardous treatment. Traditionally, perfusion source images are deconvolved in order to create these parametric maps; such as cerebral blood flow and volume [1]. The time-domain information latent in the perfusion source images has great potential to discriminate abnormal tissues and we have recently made advances in applying denoising and signal processing methods [2] [3], to detecting stroke lesions using parallelised algorithms [4].

The research would investigate machine learning combined with image and signal processing to identify lesion extent as will as predict response to treatment in stroke patients [5]. The aim would be to produce new methods and validated models that are also suitable for reuse in other brain image interpretation tasks. The goal would be to develop reliable methods that could be used by technical staff in daily clinical practice. (This would require validation and commercial collaboration that is not part of the PhD.)

The input data come from clinical measurements (NIH Stroke Score etc) in combination with perfusion weighted and other imaging types (angiography, structural, etc). The research would use machine-learning techniques, such as pattern recognition, clustering (tissue type segmentation, lesion classification) and association rule learning, and develop methods of improving evidential power. The research would apply to both CT and MRI images with the objective of developing a clinical tool and exploring whether the additional contrast mechanisms available in MR (diffusion, FLAIR etc.) have a significant impact upon treatment decisions.

This project may be funded by CIVIS.

Attachment	Size
CIVIS_acute_stroke_lesion_recognition-1.doc	39 KB

Supervisors @ NeSC:

Project status:

Still available

Degree level:

PhD

Other supervisors:

Joanna.Wardlaw

Background:

Informatics, Neuroscience, statistics or Mathematics.

Subject areas:

Student project type:

References:

1. L. Ostergaard, R. Weisskoff, D. Chesler, C. Gyldensted, and B. Rosen, “High resolution measurement of cerebral blood flow using intravascular tracer bolus passages. part i: Mathematical approach and statistical analysis,” Magn Reson Med, vol. 36, no. 5, pp. 715–25, 1996. 2. F. Zhu, T. Carpenter, D. R. Gonzalez, M. Atkinson, and J. Wardlaw, “Computed tomography perfusion imaging denoising using Gaussian process regression,” Physics in Medicine and Biology, vol. 57, no. 12, pp. N183–198, 2012. 3. F. Zhu, D. R. Gonzalez, T. Carpenter, M. Atkinson, and J. Wardlaw, “Automatic Lesion Area Detection Using Source Image Correlation Coefficient for CT Perfusion Imaging”, IEEE Transactions on Information Technology in Biomedicine, Under Revision. 4. F. Zhu, D. Gonzalez, T. Carpenter, M. Atkinson, and J. Wardlaw, “Parallel per- fusion imaging processing using GPGPU,” Computer Methods and Programs in Biomedicine, 2012. 5. I. Kane, T. Carpenter, F. Chappell, C. Rivers, P. Armitage, P. Sandercock and J. Wardlaw, “Comparison of 10 different magnetic resonance perfusion imaging processing methods in acute ischemic stroke effect on lesion size, proportion of patients with diffusion/perfusion mismatch, clinical scores, and radiologic outcomes”, Am Heart Assoc, vol . 38, no. 12, pp 3158-64, 2007.

Ad hoc Cloud Computing

Gary.McGilvary — Mon, 27 Sep 2010 14:07:05 +0000

Student:

Gary McGilvary

Commercial cloud providers offer computational services via co-located machines within data centres, whereas private clouds typically offer services via a set of dedicated servers. While both cloud models appeal to the mass market, there exists a long tail of potential cloud users that are unable to take advantage of either public or private cloud computing. Organizations with policy, monetary or application constraints may not be able to migrate to the commercial cloud and combined with the inability to procure and manage an internal private cloud, organizations cannot take advantage of cloud computing which may stifle their ability to grow and operate more efficiently.

To address such problems we aim to create an ad hoc cloud computing framework which deploys ad hoc clouds on top of existing organizational infrastructures. Ad hoc clouds harvest resources from non-exclusive organizational hosts — machines used for everyday business, e.g. employees’ workstations — and allows them to be used by others. In particular, within an ad hoc cloud, we create a set of cloudlets; we define a cloudlet as set of connected ad hoc guests that provide a particular service or execution environment.

Many cloudlets may exist and can be deployed over multiple organizational hosts to provide a unique service to the organization's employees. In the diagram above, cloudlet `1' and `2' may hold the necessary environments for Matlab and BLAST applications to execute respectively. These applications execute on ad hoc guests; virtual machines that contain the application environment necessary and execute on organizational hosts. Note that cloudlet members are likely to be distributed across an organization and not co-located as depicted above. However, an ad hoc guest may either be dedicated to a single cloudlet or attached to many cloudlets to steal unused CPU cycles or other resources from the underlying organizational host.

This computational model of stealing CPU cycles is employed by volunteer computing systems such as the Berkeley Open Infrastructure for Network Computing (BOINC) and Condor however an ad hoc framework has different aims. Firstly, the ad hoc framework aims to target more diverse applications as opposed to typical CPU intensive applications commonly executed by volunteer computing frameworks. Secondly, it aims to be perceived as a cloud computing framework and offer the advantages associated with cloud computing to organizations. While being different from volunteer computing systems, the requirements of ad hoc cloud computing also differ when compared to commercial and private cloud models where machines are dedicated for a single purpose.

Hence the objective of this research is to determine the feasibility, cost effectiveness and performance of ad hoc cloud computing infrastructures deployed over organizational infrastructures. We specifically investigate:

* how to introduce reliability upon a volunteer network to ensure the continuity of cloud jobs over a set of ad hoc hosts. We solve this by employing a P2P virtualization snapshot mechanism where snapshots are periodically transferred to select hosts within the network. A snapshot can be restored when a particular host terminates to limit the downtime of a single cloud job.

* how to minimize the interference from cloud jobs on organizational processes executing upon ad hoc hosts. We solve this by employing virtualization and dynamically adjusting the amount of resources the virtual machine is able to consume.

* how to schedule cloud jobs to optimal ad hoc hosts. We perform this based on a job’s characteristics, host availability, capability, load and reliability.

* how to dynamically monitor and control ad hoc hosts. This is performed using our C2MS tool; a dynamic monitoring and control mechanism called the Cloudlet Control and Management System (C2MS) to allow continued monitoring and control for dynamic nodes which may enter and leave defined groups, or cloudlets.

* the difference in cost and performance of ad hoc cloud computing compared to private and public cloud infrastructures.

The foundation of our ad hoc cloud computing approach our V-BOINC solution.

Supervisors @ NeSC:

Project status:

In progress

Degree level:

PhD

Other supervisors:

Dr. Adam Barker Dr. Ashley Lloyd

Student project type:

Optimising Distributed Data Integration and Data Mining Service through Transformation of Data Workﬂow into Parallel Stream

Jano.van.Hemert — Tue, 01 Sep 2009 18:41:38 +0000

Student:

Chee Sun Liew

Over the past decades, running large-scale experiments using computational tools has become popular in modern science. The data processing steps involved in such experiments are usually complex and compute intensive. A challenge arises when the demand comes from large collaboration projects that involve running computations across institutions and continents, where the data and machines are located on distributed sites. The common solution to make the experiments more manageable is executing the processing steps as a workflow, using domain-specific or generic workflow management systems. The workflow management systems map a scientific workflow onto available resources for execution. The resources can be heterogeneous and scattered in geographically distributed locations. Thus, the key factors in determining the success of scientific workflow execution will rely on the data integration, resource mapping and the process execution itself. In my thesis, I propose to improve the overall performance of scientific workflows by optimising the resource integration and mapping mechanism, and transforming the data workflows into parallel streams to speed up process execution.

Supervisors @ NeSC:

Project status:

Finished

Degree level:

PhD

Student project type:

Improving quality and reliability of results in gene expression studies by accounting for systematic artefacts

Jano.van.Hemert — Fri, 14 Aug 2009 15:39:11 +0000

Student:

Rob Kitchen

With the growing complexity and procurement costs of these high-throughput platforms, it is becoming increasingly common for the experiments to be deployed in central ‘core facilities’. This service-oriented paradigm is a recent development and one that is generally welcomed by lab-researchers and data-analysts as it encourages the standardisation of experimental protocols and reduces costs of hardware maintenance.

Despite the advances in manufacturing-quality, experimental efficiency, and overall reliability seen in these hardware platforms, certain issues regarding the accuracy of results gleamed from such experiments remain unresolved amongst those in the scientiﬁc community. Among those in question are high-throughput genome sequencing, gene microarrays, and quantitative polymerase chain reaction (qPCR) assays.

The main hypothesis is that the quality, and therefore also the reliability, of results gathered through gene-expression studies can be improved through careful management of the entire experimental workﬂow, rather than simply through optimisation of its individual steps.

Supervisors @ NeSC:

Project status:

Finished

Degree level:

PhD

Other supervisors:

Prof Peter Clarke (School of Physics); Dr Varrie Ogilvie (Molecular Medicine Centre, University of Edinburgh)

Subject areas:

e-Science

Bioinformatics

Student project type:

Improving knowledge curation in structured wiki-like collaborative environments

Jano.van.Hemert — Tue, 10 Feb 2009 15:22:34 +0000

Student:

Luna De Ferrari

This work aims at deﬁning, modelling and evaluating the integrated use of collaborative software and machine learning for building high quality knowledge resources. A possible scenario is Molecular Biology, where high-throughput data production is overwhelming the traditional centralised data annotation by paid experts. Many biological resources have moved to collaborative software platforms, predominantly wikis, in an effort to involve the wider community and replicate the success story of Wikipedia. However, it has been shown that even a widespread effort will ﬁnd it diffcult, if not impossible, to scale the annotation up to the current rate of data growth in Biology [1].

Supervisors @ NeSC:

Project status:

Finished

Degree level:

PhD

Other supervisors:

Igor Goryanin, School of Informatics; Stuart Aitken, AIAI

Student project type:

References:

[1] William A Baumgartner, K. Bretonnel Cohen, Lynne M Fox, George Acquaah-Mensah, and Lawrence Hunter. Manual curation is not suffcient for annotation of genomic databases. Bioinformatics, 23(13):i41–i48, Jul 2007.

Planning Emergency Movement for the Built Environment

Jano.van.Hemert — Tue, 10 Feb 2009 15:17:46 +0000

Student:

Thomas French

The goal of this project is to investigate methods for ﬁnding emergency movement plans in dynamic and uncertain environments, specifically buildings. Current techniques used to solve these problems, like (Opasanon, 2004), make unrealistic assumptions about human behaviour during emergency movement. For example, they assume that occupants travelling through a building do not directly interact, and, therefore, provide instructions that presume people who arrive at a decision point at the same time will split up if told to do so. In practice, it is not clear how this could be implemented and how it would be possible to distribute risk exposure fairly. Furthermore, empirical research on human behaviour during emergency evacuation (for example, (SFPE, 2002)) indicates that people tend to have a social response to evacuation; people act as a group and try to evacuate with people to whom they have some attachment. We explore approaches that model a limited social response in the route planning process.

Supervisors @ NeSC:

Project status:

Finished

Degree level:

PhD

Other supervisors:

Austin Tate, AIAI; Stephen Potter, AIAI; Gerhard Wickler, AIAI; Jose Torero, School of Engineering

Subject areas:

e-Science

Algorithm Design

Genetic Algorithms/Evolutionary Computing

Projects:

FireGrid

Student project type:

Final version of the thesis submitted.

References:

(Opasanon, 2004) S. Opasanon. On Finding Paths and Flows in Multicriteria, Stochastic and Time-Varying Networks. PhD thesis, University of Maryland, 2004. (SFPE, 2002) SFPE. SFPE Handbook for Fire Protection Engineering. National Fire Pro- tection Association, 3rd edition, January 2002.

Gaussian Process deconvolution for perfusion imaging: evaluation of the usage of distributed and parallel computing

Jano.van.Hemert — Mon, 17 Nov 2008 09:31:17 +0000

Student:

Fan Zhu

Original project description:

Deconvolution is used in perfusion imaging to obtain the impulse residue function (IRF) that is then used to create parametric maps of relevant haemodynamic quantities such as CBF, CBV and MTT [1]. A popular method to achieve this is Singular Value Decomposition (SVD), but it has been shown that for MRI Gaussian Process Deconvolution (GPD) [2] is comparable to SVD when determining the maximum of the IRF, and superior estimating the full IRF. Furthermore, it clearly outperforms SVD when the signal-to-noise ratio improves. Gaussian Process regression [3] arises from a Bayesian approach to the regression problem, and as in the case of other kernel-based methods the scalability with data size is very poor. This constitutes the main drawback of this technique to compute deconvolution when compared with SVD.

The currently running Wyeth-TMRC multicenter project on acute stroke brings the opportunity to test this technique with data from several SINAPSE centres and different modalities. This PhD project will benefit from the expertise in these centres and would seek to collaborate with them through the centres’ contacts: M.J. McLeod (Aberdeen), J. Wardlaw (Edinburgh) and K. Muir (Glasgow).
The project will research the possibilities that distributed (and parallel) computing brings to make this method usable in practice. There have been some previous works like the dataparallel approach proposed in [4]. The project will study the consequences on the final results of the local learning used in it. As a by product, the project will produce a data processing framework prototype reusable for other types of image processing.

Supervisors @ NeSC:

David.Rodriguez

Project status:

Finished

Degree level:

PhD

Other supervisors:

Prof Joanna Wardlaw (SF Brain Imaging Research Centre, University of Edinburgh) Dr Trevor Carpenter (BRIC, University of Edinburgh)

Background:

MSc in Computer Science essential. Strong background in imaging and distributed computing important.

Subject areas:

Projects:

SINAPSE

Student project type:

References:

1. Ostergaard L, Weisskoff RM, Chesler DA, Gyldensted C and Rosen BR “High resolution measurement of cerebral blood flow using intravascular tracer bolus passages: I.Mathematical approach and statistical analysis” Magn. Reson. Med. 36 715–25 2. Andersen IK et al; “Perfusion Quantification Using Gaussian Process Deconvolution”. Magnetic Resonance in Medicine 48:351-361 (2002). 3. Williams CKI and Rasmussen CE; “Gaussian processes for regression”. Advances in neural information processing systems, (1996), 514-520. 4. Choudhury A, Nair PB and Keane A; “A Data Parallel Approach for Large-Scale Gaussian Process Modeling”. Proc. the Second SIAM International Conference on Data Mining (2002).

A Framework for Metadata Driven e-Science Implementations

Jano.van.Hemert — Wed, 26 Sep 2007 16:12:17 +0000

Student:

Yin Chen

Many e-Science applications are data intensive. Metadata is at the heart to serve the semantic interpretation, discovery, and integration of large-scale heterogeneous scientific data. The project explores observations of how metadata are used in a variety of e-Science disciplines, such as annotation, workflow, information integration, provenance, and curation. It aims to analyse the requirements for metadata central to e-Science applications, and examines state-of-the-art approaches. Its goal is to provide a flexible framework that comprises of the five basic elements necessary for delivery of successful metadata-driven projects.

Supervisors @ NeSC:

Project status:

Finished

Degree level:

PhD

Other supervisors:

Dr Stuart Aitken (School of Informatics, University of Edinburgh)

Subject areas:

Databases

Distributed Systems

Software Engineering

Student project type: