The Neuro Bureau Preprocessing Initiative: open sharing of preprocessed neuroimaging data and derivatives

• ¹ Child Mind Institute, Center for the Developing Brain, United States
• ² Nathan Kline Institute for Psychiatric Research, United States
• ³ The Neuro Bureau Research Institute, United States
• ⁴ Université de Montréal, Département d’anthropologie, Canada
• ⁵ Centre de recherche de l’institut de gériatrie de Montréal, Canada
• ⁶ McGill University, Montreal Neurological Insitute, Canada
• ⁷ University of Debrecen Medical and Health Sciences Centre, Hungary
• ⁸ Université de Montréal, Département d’informatique et de recherche opérationnelle, Canada

Introduction
Grass-roots initiatives such as the 1000 Functional Connectomes Project (FCP) and International Neuroimaging Data- sharing Initiative (INDI) [1] are successfully amassing and sharing large-scale brain imaging datasets, with the goal of recruiting the broader scientific community into the fold of neuroimaging research. Unfortunately, despite the increasing breadth and scale of openly available data, the vast domain-specific knowledge and computational resources necessary to derive scientifically meaningful information from unprocessed neuroimaging data has limited their accessibility. The Neuro Bureau Preprocessing Initiative [2] has taken on this challenge, generating and openly sharing preprocessed data and common derivatives for the large-scale ADHD-200 dataset [3]. This initiative has grown to include preprocessed DTI data and derivatives for 180 healthy individuals from INDI’s Beijing Enhanced Sample [4]. The next planned release will include resting state and structural data from the 1,112 subject Autism Brain Imaging Data Exchange (ABIDE) dataset [5].

Methods
Four teams are currently participating in the preprocessing initiative, each one using different toolsets and preprocessing strategies (fig. 1). Preprocessed data, derivatives, and quality control metrics are made openly available for download through the Neuroimaging Informatics Tools and Resources Clearinghouse (NITRC) [6]. The ADHD-200 release included two fMRI preprocessing pipelines as well as maps of grey matter density for voxel-based morphometry (fig. 1). The Beijing diffusion imaging release includes DTI scalars along with voxel specific diffusion distributions for performing probabilistic tractography. Figure 2 illustrates various derivatives generated through these initiatives. The future ABIDE preprocessing initiative will incorporate three functional preprocessing piplines and cortical measures (fig. 3). The analytical procedures employed in the preprocessing are extensively documented on the NITRC website [2].
The Neuro Bureau preprocessing initiative also includes an on-going working group to release derivatives, which can be readily compared across different preprocessing strategies, so that investigators can directly test the impact of the method- ological choices on the scientific outcome of a study. Most of the ong-going work consists of improving and harmonizing the quality control procedures and the derivatives generated by different processing pipelines. Interested teams are welcome to join the effort and contribute new analytical pipelines for future release.

Results
Intended to buttress the ADHD-200 Global Competition [7] and accelerate ADHD imaging research, the ADHD-200 preprocessing effort has yielded more than 6,500 downloads from 780 unique IP address globally (see fig. 4), inspired a team of biostatisticians to win the competition and resulted in eight peer-reviewed publications - with many more in preparation or submission. The DTI preprocessing initiative has resulted in 572 downloads from 134 unique IP addresses. Based on the success of the previous preprocessing efforts four teams have agreed to continue this effort by preprocessing the recently released ABIDE dataset (fig 3).

Conclusion
By openly sharing a wide range of preprocessed data and derivatives, the Neuro Bureau Preprocessing Initiative seeks to make neuroimaging research accessible to a wider audience of researchers. It has already enabled computer scientists, mathematicians, and statisticians who lack neuroimaging expertise to develop and test novel data analysis strategies. We see several important benefits to our initiative: (1) facilitate the generation and test of novel hypotheses about brain function, (2) provide a resource to train future generations of neuroimaging researchers and, (3) facilitate the replication of published results by providing a benchmark set of test images. By providing a breadth of derivatives and preprocessing strategies, we also hope to establish a platform for comparing their relative merits, as well as testing the robustness of neuroscientific findings. This already broad resource will soon be enhanced by the inclusion of the phenotypically rich ABIDE dataset which consists of data from an important clinical population.