Skip to main content Skip to navigation

Details on the Talks in the Afternoon Session

Tom Nichols (Dept of Statistics, University of Warwick) and Nicola Filippini (Dept of Psychiatry, University of Oxford)


Big Data Squared: Neuroimaging Genetics
 

Genetics and neuroimaging are two scientific disciplines where massive data plays an essential role. Over 1 million genetic markers are needed for a detailed view of a human individual, and over 1 million volume elements are needed to show the brain with high resolution Magnetic Resonance Imaging (MRI). "Neuroimaging Genetics" is an emerging field that seeks to find genetic variants that explain why different people have different brain structure and brain function, and then link these effects to debilitating illnesses like Alzheimer's disease, depression and schizophrenia.

In this 2-part talk, neuroscientist Nicola Filippini will talk about exciting developments driven by combining genetics and brain imaging, and the promise Neuroimaging Genetics has for transforming psychiatric research. Then statistician Thomas Nichols will give a flavour of the tools used to jointly model genetic and brain MRI data, and show how advanced computational and statistical tools are allowing analyses thought simply impractical just 5 years ago.

John Aston (Dept of Statistics, University of Warwick) and Nigel Coleman (Phonetics Laboratory, University of Oxford)


Statistics and Speech

There is considerable interest in comparing and contrasting languages, both as a tool to understand communication and as a reference for cultural and historical differences. However, while it is relatively easy to compare two written words (for example, how many letters need to be changed to go from one to another, do the words have the same common root), it is more difficult to compare speech, as every word is pronounced slightly differently each time it is said, and people's voices and pronunciations vary considerably.
A relatively new area of statistics, called functional data analysis, can help in this regard. It is possible to compare and contrast languages by treating sound *waves* as curves and surface representations. We will explain how speech is converted into statistical objects, and then how these objects can be analysed to give distances between languages, as illustrated by the sound of the digits "one" to "ten" in various Romance languages.
The ideas also facilitate reconstruction of intermediate and historical sounds, allowing us to hear back in time.

Nigel Stallard and Gavin Perkins (Warwick Medical School)


Data monitoring in a randomised controlled clinical trial for adrenaline in out-of-hospital cardiac arrest

Cardiac arrest is a major cause of mortality, with survival rates for those having an out-of-hospital arrest about 6%. Adrenaline has been a standard treatment for many years, backed up by evidence of a resulting increase in the number of patients surviving long enough to reach hospital. Longer-term, however, the impact of adrenaline is less clear, and recent analysis of observational data has suggested that its use may actually decrease the number of patients surviving to hospital discharge. A team at Warwick Medical School are planning a randomised clinical trial to compare adrenaline with placebo in patients with out-of-hospital cardiac arrests. The fact that patients with a life-threatening condition will be randomised to a group in which a currently standard therapy is withheld means that it is important to closely monitor the ongoing trial to be able to stop quickly if it becomes apparent that adrenaline does indeed improve long-term survival. The exact rule that will be used to stop the trial will be decided by a Data and Safety Monitoring Committee, but statistical calculations and simulations will be used to illustrate the properties of different rules to assist them in their choice. The data monitoring will also give the opportunity to stop the trial if there is a strong indication that adrenaline is harmful. As this could increase the risk of reaching this conclusion erroneously, an appropriate statistical analysis method will be used to allow for this.