IMPAC

A big data challenge

Brain images from more than 2000 individuals

Patient vs Control distribution

Gender distribution

Age distribution

Multimodal imaging data

Structural MRI

Preprocessed with FreeSurfer and FSL
Gray matter volume, area, and thickness
Average for each Desikan cortical parcel.

Functional MRI

Resting state fMRI
Time series extracted on different atlases

Coding framework, for competition and collaboration

The challenge will be carried out on the RAMP platform. It enables competition and collaboration on data-science problems, using the Python language. To start "hacking", a starting kit is available. It provides a simple working example which can be expanded to more advanced solutions.

Starting kit

What is this challenge?

Autism spectrum disorder (ASD) is a severe psychiatric disorder that affects 1 in 166 children.

There is evidence that ASD is reflected in individuals brain networks and anatomy. Yet, it remains unclear how systematic these effects are, and how large is their predictive remain unclear. The large cohort assembled here can bring some anwsers. Predicting autism from brain imaging will provide biomarkers and shed some light on the mechanisms of the pathology.

Joining the competition

The goal of the competition is to predict the diagnostic status from brain imaging data in the hidden test set. The data of the test set are on a server, hidden from participants. Prediction is done by submitting Python code that will be first trained by the server on training images, and then applied to predict on the hidden test set.

Prizes

The best performers will be awarded prizes at the end of the competitive period, based on the metrics used during the challenge: 1^st - 3000 €, 2^nd - 2000 €, 3^rd - 1000 €, from 4^th to 10^th- 500 €.

The timeline

This challenge is made of 2 phases:

a competitive phase up to July 1 in which participants can submit their solutions to the RAMP platform. A Q&A session regarding the challenge and technical platform is organized May 14;
a collaborative phase from July 1 to July 7. This last day will coincide with the awards ceremony.

2 events in Paris

Anyone can participate remotely. In addition, two events will be organized as follows:

May 14, 2018: a Q&A with the team to help get started, La Paillasse, 226 rue Saint-Denis, 75002 Paris
July 6 & 7, 2018: collaboration after the competition, Institut Pasteur, 25-28 Rue du Dr Roux, 75015 Paris

More about the data

Structural MRI (sMRI)

For each subject, a set of structural features have been extracted and stored in a CSV file. Each row corresponds to a subject and each column to a specific features. The features extracted are:

Normalized brain volume computed using subcortical segmentation of FreeSurfer;
Cortical thickness and area for right and left hemisphere of FreeSurfer.

Functional MRI (fMRI)

Each subject also comes with fMRI signals extracted on different brain parcellations and atlases, and a set of confound signals. The brain atlases and parcellations are:

BASC parcellations with 64, 122, and 197 regions (Bellec 2010);
Ncuts parcellations (Craddock 2012);
Harvard-Oxford anatomical parcellations;
MSDL functional atlas (Varoquaux 2011);
Power atlas (Power 2011).

The Nilearn documentation gives details on the parcellations, and how they were extracted.

The time series for each subject are stored in a CSV file. Each row represent a scan while each column a region defined by the brain parcellation or atlas used.

Multimodal: predicting from sMRI and fMRI data

The data are openly downloadable and a Jupyter notebook provides a starting kit. It explain: (i) how to read data, (ii) how design, train, and test a model, (iii) how to submit it on the competition platform.

Refer to the starting kit for more information.

Questions

Organizers

Institutions

The group of Neuroanatomy of the Unit of Human Genetics and Cognitive Functions is interested in the development and evolution of brain anatomy. We are particularly interested in Autism Spectrum Disorders.

More

The Robert Debre Hospital investigates different fields such as genetic, brain imaging, neurocognitive science to identify new biomarkers as well as cell models for new therapeutical targets.

More

The Paris-Saclay CDS aims at developing methods and tools so as to be capable of analyzing large amounts of data and extracting useful information from them for multidisciplinary sciences.

More

Ingenieurs et Scientifiques de France (IESF) is the unique representative of the engineers and scientists social corps. The IESF is a major actor of the scientific, technological, and economic debates.