case study
Published: 01 March 2021

Using Automatic Analysis to improve neuroimaging

Dr Tibor Auer, a Research Fellow at our School of Psychology, explains why using Automatic Analysis will help the drive for more efficient and reproducible neuroimaging pipelines. 

Tibor Auer
Dr Tibor Auer

The issue 

Neuroimaging research relies upon several open source tools, such as SPMFSLFreeSurferEEGLAB and FieldTrip. This offers methodologies for a wide range of measurements of brain structure and function.  

However, there is currently no well-established framework that integrates them into flexible, reproducible and scalable pipelines, which current studies increasingly demand.  

The solution 

Automatic Analysis (aa) is a unique MATLAB-based integration framework designed for these packages, which reduces the barrier for those with limited computational background.  

aa addresses key issues in neuroimaging and provides an executable, reproducible and publishable description for complex neuroimaging workflows. It not only offers the tight integration of tools and improved usability, but it promotes reliability and transparency by providing detailed diagnostics and reporting facilities. 

In 2020, MathWorks, the developer of MATLAB, a leading mathematical computing software for engineers and scientists, identified aa as the vital link connecting open source projects. 

The outcome 

The initial development of aa was inspired by various research projects at the MRC Cognition and Brain Sciences Unit in Cambridge.  

In 2012, the project was moved to GitHub to facilitate a transparent and efficient collaboration of more than ten researchers at four institutes in four countries. As a result, the project also greatly benefitted from the integration of revision and communication both within the developer community and between the userbase and the developers. 

By working on openly available datasets following the clear specification of data sharing standard BIDS, we were able to conduct testing and, in addition, provide example workflows and workshops processing real data.  

aa addresses the challenges of processing multimodal datasets. For example, it can combine anatomy, functional MRI, diffusion and EEG information. It also supports parallelised execution, and provides a detailed report of interim and final results. 

The capabilities of aa have been shown in challenging conditions involving large-cohort, complex, multimodal datasets [1,2]; and its usability has been shown for diverse pipelines [3]. Knowledge of aa has spread by word of mouth to more than 100 researchers to date. 

Although aa itself is fully open, MATLAB is not. For Python-using neuroimagers, NiPype is a great pipeline tool relying predominantly on FSL, FreeSurfer, and other pre-compiled and Python-based tools.  

However, it is important to mention that MATLAB is still the dominant environment in neuroimaging as reflected by the number of studies based on it. 

Share what you've read?