Data analysis competition
Data analysis competition Biomag 2010:
Connectivity and multivariate classification approaches
Organizers: Ole Jensen and Jan Mathijs Schoffelen
Sponsored by:
SEE BELOW FOR THE WINNERS!!!
Deadline for submission: March 14, 2010.
While there are a number of advanced data analysis techniques that allow us to embrace distributed electrophysiological activity measured by MEG, these tools are somewhat underexploited. This includes measures of functional connectivity and multivariate classification approaches. The aim of the competition is to
- promote the development and application of new analysis techniques for distributed brain activity.
- make the audience aware of novel approaches.
- elucidate the pros and cons of the different techniques.
- which assumptions are behind a given approach?
- what are the limitations?
- attract signal-processing experts from outside the MEG field.
- encourage a discussion on the cognitive insight which the techniques can bring about.
The competition has three parts. Partcipant can do one or more parts. They will be evaluated separately. The winners will be announced at the Biomag 2010 meeting (time and day will be announced later). The winners will then shortly present their findings and pros and cons will be discussed.
A board will evaluate the outcome of the competition. Hard criteria are difficult to define for competition 1 and 2 and the outcomes will be evaluated by how compelling a case has been made.
Please send an e-mail to Ole Jensen (ole.jensen AT donders.ru.nl) if you plan to attend.
Competition 1. Data set on multivariate classification
We will provide four data sets from the study:
In this study subjects were asked to covertly attend left, right, up or down. We demonstrated that these four directions of attention could be classified reliably based on posterior alpha activity.
Question: Apply a classifier to predict in which direction the subjects were covertly attending. In this competition we will only focus on attention to the left and right visual fields.
Evaluation: Contestants should report the classification rate (proportion of correctly classified trials) for each subject as computed using leave-one-out cross-validation and report the employed classification procedure.
You can find the data sets here: competition 1 - dataset. Please consult the Readme file for more details. For questions on data please contact Marcel van Gerven - marcelge AT cs.ru.nl
Winners:
Team 1: Roberto Santana, Concha Bielza, Pedro Larra˜naga
Team 2: Jinyin Zhang, Xin Li, Gustavo Sudre
Competition 2. Real data set on functional connectivity
We will provide MEG data set and the MRI from one subject from the study
This study has already revealed task-dependent power-power correlations between central beta and posterior gamma activity during mental rotation.
Question: In the study task dependent power-power correlations were identified at the sensor level. Identify functional connectivity at source level (could be power-power correlations but not necessarily).
Evaluation: Is a compelling case made for functional connectivity in sources space? Do the results make sense from a cognitive neuroscience perspective?
You can find the data set here including detail on the triggers etc: competition 2 - dataset. You can also download the preprocessed data as a .zip-file (approx. 1.3 GB): competition 2 - preprocessed. For questions on data please contact Floris de Lange - floris.delange AT fcdonders.ru.nl
Winner:
Bernadette van Wijk, Vladimir Litvak, Karl Friston
Competition 3: Simulated data set on functional connectivity.
A realistic data set has been simulated in which two regions are functionally connected. A realistic forward model has been used.
Question: The data set contains 2x175 trials assigned to condition A and B. Several sources are simultaneously active. There is condition dependent change in functional connectivity. Can you identify the sources (sensors and/or sources space) and quantify the functional connectivity?
Evaluation: Identification of the sources. Is the actual task-dependent functional connectivity identified? Description of the type of connectivity.
You can find the data files here, including additional information and some additional files: competition 3 - dataset. You can also download it as a .zip-file (approx. 520 MB) competition 3 - dataset.zip. For questions on data please contact Jan-Mathijs Schoffelen - j.schoffelen AT fcdonders.ru.nl
No winner
What to submit
A power point presentation of about 5-8 slides. It should contain
- Key components of method ; focus on whats new/innovative
- Key results
- Suggestions for future approaches
Please make the slides understandable for a non-expert audience. If you win the competition we will ask you to present the results at the Biomag workshop. If nobody from your team can attend Biomag2010 we will appoint somebody from the evaluation board to do the presentation.
Evaluation and prizes
Board that will evaluate results: Ole Jensen, Jan-Mathijs Schoffelen, Joachim Gross, Cheryl Aine, Sylvain Baillet
Prizes: 500 Euros per competition (to be shared among team members). The awards are sponsored by MEG International Services Ltd. and Elektra Oy.
Winners will be announced at the workshop
"Data analysis competition: Connectivity and multivariate classification approaches"
at Biomag2010 meeting March 30, 9:00-10:30. The winners are expected to present and discuss their findings at the workshop. If the winners cannot participate in Biomag2010 a member of the board will present.