If you are thinking of doing your BSc or MSc/MRes project at the Computational Neuroimaging Lab, here there are a few ideas. Any of the suggested topics may be considered for a BSc or MSc/MRes thesis with the main difference being the depth at which you are expected to carry out the research. There might be some students already working on these topics, but more candidates are sought.
- Topological data analysis (highly challenging and strongly mathematically oriented. Look for these only if you are looking for a first class degree.)
- Probabilistic and statistical analysis (Mid level of complexity and usually enjoyable, but do not get confused, you will be seriously challenged!)
- Knowledge Representation and Interpretation (Feeling adventurous? Follow an unusual path with unforeseeable open ends. You need to be able to deliver with little guidance.)
- Data science and data analytics (Work with real datasets. Beware! These projects are only available under very serious commitment.)
- Programming (Low conceptual challenge, but extreme riguor in the forms. Not your usual free-style programming.)
Topological data analysis
Topology is the branch of mathematics concerned with shapes. Topological spaces, such as manifolds are very expressive mathematical objects that represent surfaces with some persistent local feature, e.g. being locally flat. Manifolds are usually employed to discover the implicit structure of cloud of points and analysed extremely complex datasets. We apply manifold based analysis to optical neuroimaging data to decode brain activity under a variety of circumstances. Some projects ideas in this area include:
- Modelling and understanding different projections of tensors to ambient spaces. Most often, you depart your data analysis from a set of observations. By definition these observations are points in a given space, but the explicit space in which your data is expressed is not necessarily the most convenient for analysis. Depending on your data analysis goals you may want to explore projections to other space which more clearly highlight that particular aspect of your data that can help you to answer your data analysis question. But the question is which projection is more adequate given a certain data analysis goal? Often this is done by trial and error. Here we intend to understand some common projections to afford guidelines on their implications in order to guide future uses.
- Segregational analysis using seeded manifolds: Segregational analysis looks at studying brain regions activity on isolation. Segregational analysis addresses one of the most important questions in neuroscience; which areas of the brain activate following some stimuli? There are already excellent tools for segregational analysis from classical statistics, but there is always room for improvement. We hypothesize that the topological point of view can give us an edge on certain inferences over the standard inferences one makes with statistics, but this remains to be proven!
- Use of causal manifolds for establishing effective connectivity in optical neuroimaging: Causality is a top deep question in science in general. It is so difficult toresolve that we have some necessary but still no sufficient conditions, opening the door to multiple mathematical definitions of the concept. Yet, this elusive concept is critical to undertand how different regions of the brain collaborate with each other to produce the complex behaviours that allow you to interact with your environment. This project complements our current efforts to model effective connectivity in the brain using causal manifolds.
Probabilistic and statistical analysis
Statistics is the branch of mathematics that studies data in order to measure and control uncertainty and it is essential for science. Statistics is by far the most important tool in data analysis and experimental science. The application of statistics is based on mathematical models that associate concepts, but building and parameterizing such models is sometimes not straight forward. Neuroscience continues making questions at the edge of what our statistical models can answer always demanding for finer, better models that express more and more variance (e.g. allows less and less unexplained uncertainty). Some projects ideas in this area are:
- Parameterization of the temporal window selected for detection of brain activity: A common experimental design paradigm in neuroscience will expose the volunteer repeatedly to a certain stimulus. The choice of the optimal temporal window for analysis of the brain response is not known. This project intends to investigate the implications of different choices of such window.
- Event related analysis: When studying repeated stimulation of the brain (whether with the same of different stimuli), sometimes it is not possible to separate the stimulations periods sufficiently to give time to the brain to go back to a baseline state. Hence, it becomes the goal of the statistical model to dissociate the effects of all contributing stimuli; this is called event related analysis. Literature has some efforts presenting models for this situation but they are cumbersome hence deterring neuroscientists from using this valuable experimental approach. This project seeks to implement an existing event related model to incorporate it into an optical neuroimaging data analysis tool to facilitate this type of analysis and consequently giving support to neuroscientists wanting to use the aformentioned experimental approach.
- Information theory models for connectivity analysis: First proposed by Claude Shannon for telecommunications, information theory has become an important branch of probabilities to quantify information and entropy, both central concepts across many domains. Brain connectivity refers to the interplay of brain regions at functional level and it has been suggested that by measuring how much information is shared between the observations made at multiple brain regions, it is possible to establish their degree of connectivity. But the work so far has concentrated on associative relations. This projects explores the feasibility of employing information theory for the analysis of effective (causal) connectivity.
Knowledge Representation and Interpretation
Knowledge representation is a branch of artificial intelligence that affords the generation of knowledge by means of logical reasoning over carefully designed representations of knowledge; e.g. in the form of ontologies. While this is an extremely powerful branch of artificial intelligence, it remains severely underused due to the difficulties in capturing existing knowledge in the first place. We are making efforts to capture neuroscientific knowledge so that it can be later exploited to generate further knowledge. Some projects ideas in this area are:
- Ontological representation of fNIRS experimental design for automatization of analysis: OntoNIRS is an ontology to represent optical neuroimaging knowledge but it is still in early development. We week motivated students to work on different aspects of the development of OntoNIRS including; the automated labelling of scientific papers to populate the ontology, the enrichment of the glossary underpinning the onotology, the automated extraction of neuroscientific facts from literature, etc.
- Automating neuroscience: A branch of AI known as automation of science has proven that the generation of scientific knowledge can be automated. This paradigm has ben partially fulfilled in several domains, but never attempted on a grand scale. In this project we seek to develop a minimal reasoning proof of concept over neuroscientific knowledge.
Data science and data analytics
Through our national and international collaborations, we are in possession of a large number of neuroimaging datasets waiting to be analysed; here are a few examples;
- Surgical neuroergonomics data: Does neurostimulation improves surgeons performance? And if it does, does it manisfest in the brain activity? In collaboration with surgeons from Imperial College London we have fNIRS data ready to be analysed in order to answer this question. We hypothesized that the prefrontal cortex activity attenuates with successive sessions perhaps due to an habituation effect. Can you show evidence of this effect (or perhaps refute the hypothesis)?
- Carotid Endarterectomy: Recovery from thrombectomy usually has bad prognosis due to impaired perfusion. In order to help to improve the surgical procedure, patients were monitored using two different neuroimaging modalities. In collaboration with the MGH-HST Martinos A. Center for Biomedical Imaging, we seek to carry out a preliminary analysis on the electroencaphalography part of the dataset.
- Time-resolved hemodynamics frequencies: (Dataset expected for early June) In collaboration with Politechnico di Milano, we seek to study the haemodynamics frequencies as studied with time resolved optical neuroimaging data.
|The projects related to data science and data analytics vary very quickly as our research progresses. The above are available at the time of writing this, but please make sure that you speak to the lab leader to confirm which datasets are available for analysis at the time of your project. Also please beware that undertaking one of these projects demands a very high degree of commitment to the project as our works affects the research made by our colleagues at other research groups.|
We also offer a range of programming-based projects. To get involved in these projects you are expected to be a proficient programmer, but importantly, also being a group-based developer as you will be using (and perhaps modifying) code developed by others and in turn your code will be used and perhaps modified by others; thus, you will have to comply with programming standards (you will not be allowed to program “freely” but to strictly abide to certain rules), you will be demanded enormous documenting efforts following standards, and required to provide reasoned evidence of your software architecture solution.
- Imperial College Near Infrared Spectroscopy Neuroimaging Analysis (ICNNA) complements: ICNNA is a Matlab-based analysis tool for fNIRS neuroimaging data that we develop in collaboration with Imperial College London. The software is already well established in the fNIRS community, but as the field of fNIRS advances so does ICNNA has to adapt to the new requirements of the community. Many subprojects can be chosen here including;
- Updating of the internal architecture.
- Separation of the SciMeth library.
- Compatibility with .snirf standards
- Compatibility with .BIDS standards
- Compatibility with Homer 3, NIRS ToolBox and MNE-NIRS.
- Incorporation of new files for reading new devices file formats.
- Improved packaging for version upgrades.
- Updating the user documentation, commenting the code on its specific functions.
- Updating the technical documentation, commenting the code on its specific functions.
- Enrichment of the signal processing capabilities.
- Development of a pipelines system.
- Development of a plug-ins architecture.
- Enrichment of the statistical analysis of the signals.
- NIRFAST: NIRFAST is an open source software for multi-modal optical imaging to predict light transport in tissue. It is the world leading application of its type and we recently have joined the team. We are planning to enrich NIRFAST with analytical capabilities based on manifolds.
- OCTant (Optical Coherence Tomography analysis tools): OCTant is a suite of tools for OCT analysis in Python developed by our group. We seek to expand the functionalities of OCTant to provide automatic and semi-automatic segmentation of of OCT images.