Completed 9/18/14

Winners: Congratulations to Andrew Triedman and Jonathan Nicholas!

Neural Decoding Competition, 2014 version

This dataset was collected to measure the electrophysiological correlate of rapid visual categorization. In this paradigm, a visual stimulus is presented very briefly (e.g, 50 ms) and the participant is expected to report the presence or absence of a target stimulus (in this case, animal vs. non-animal images) as quickly as possible. Here electrocorticography (ECoG) gives us a unique opportunity to study visual recognition at both a high temporal and spatial resolution.

The data were recorded from 4 human patients who underwent brain surgery for the treatment of epilepsy. In the experiment, the patients were told to fixate on the ‘+’ sign at the center of the screen. The stimuli were shown for 20 ms, after which the patient was asked to press either one of two keys depending on whether or not they saw an animal. For half of the trials an eraser mask was presented shortly after the stimulus to test whether it is possible to interrupt visual processing and force the visual system to operate in a strictly feedforward manner by suppressing re-entrant signals (see figure below).

Competition Objectives

Contestants should submit a short report describing their approach, as well as the predicted values for the test sets provided at cgm-info. The report should be detailed enough to allow replication of the method (including the code used is welcome, but not mandatory).

Winners will be decided by a panel of judges, based on several criteria, including overall decoding success in the test sets (in particular, more points will be awarded for better performance across multiple sets), but also on other aspects such as innovation and creativity.

Winners will be invited to present their work at a BIBS lunch. There will also be prizes, including an iPad as first prize.

Note that people of all levels are welcome to participate. There will be separate prizes for students and all categories. Team entries are also allowed.

Data Formats

Data can be downloaded at https://dropbox.brown.edu/download.php?hash=71265ee7. You will need a Brown ID to access the data. The data are quite large so please be patient!

The stimuli used in the experiment can be downloaded here: https://dropbox.brown.edu/download.php?hash=e2678ad4

The neural data (located in ./baseSets/) is from 4 human patients: 004, 006, 007, 008. Each .mat file contains a Matlab structure that contains epoch data, epoched ECoG data, sampling rate, and event data.

Here is an explanation of the most important fields.

EEG.times

The time points in milliseconds at which the neural data is sampled during each trial. For example, in Subject 004 each trial has 217 time points at which neural data is sampled, ranging from ~200 ms before trial onset to ~645 ms after trial onset. Note that the sampling rate varies between subjects (256 – 2000 Hz). If you want to resample the data, you can use the MATLAB function resample.

EEG.data

The epoched raw neural data, pre-processed to remove 60 Hz line noise using a multitaper filter. The first dimension corresponds to electrodes (64 in the case of Subject 004). The second dimension is the number of time points at which the neural data is sampled during each trial (see EEG.times). The third dimension is the number of trials (1,200 trials in the case of Subject 004). Dimensions of the data are: [N_electrodes X N_sample_points X N_trials]

EEG.chanlocs

A structure which contains cell arrays containing the locations of each electrode (64 in the case of Subject 004) in terms of hemispheres, lobes, gyri, etc. Explained in greater detail below.

EEG.event

A structure containing trial information. Explained in greater detail below.

Here is an explanation of the structure contained in EEG.event, which contains trial information

EEG.event.type

The type of trial: if the image presented contained an animal (type = 1) or no animal (type = 0)

EEG.event.imageName

The filename of the image presented during the trial

EEG.event.rt

The reaction time of the participant in seconds

EEG.event.correct

If the participant responded with the correct response (e.g. participant correctly indicated the presence of an animal in the image). 1 corresponds to a correct response, and 0 to an incorrect one. Also, if participant took too long to respond (i.e. EEG.event.rt was longer than 1.5 seconds) it is an incorrect response.

EEG.event.mask

If an eraser mask was presented after the appearance of the stimulus. 1 if a mask was used and 0 if a mask was not used.

EEG.event.badTri

The inverse of EEG.event.correct

Here is an explanation of the structure contained in EEG.chanlocs, which contains trial information

EEG.chanlocs.hemis

Electrode location by hemisphere. Calculated by Talairach daemon

EEG.chanlocs.lobes

Electrode location by lobe. Calculated by Talairach daemon

EEG.chanlocs.gyrus

Electrode location by gyrus. Calculated by Talairach daemon

EEG.chanlocs.matter

Electrode location by grey/white matter. Calculated by Talairach daemon

EEG.chanlocs.ba

Electrode location by Brodmann area. Calculated by Talairach daemon

EEG.chanlocs.coors

Electrode location in Talairach coordinates

EEG.chanlocs.regions

Electrode location by region. Calculated by FreeSurfer parcellation.

EEG.chanlocs.regionCodes

Electrode location by region using unique numeric codes. Calculated by FreeSurfer parcellation.

You can find the 2013 neural decoding challenge at http://compneuro.clps.brown.edu/neural-decoding-competition/