1665
Educational Course - Half Day (4 hours)
As the number of neuroimaging studies continues to grow, so does the need for robust and reliable methods to analyze and interpret complex brain data. However, neuroimaging pipelines' inherent flexibility and customizability have resulted in a proliferation of disparate approaches, leading to inconsistent and often unreproducible results. This variability not only hinders progress in understanding brain function and behavior but also jeopardizes the validity and reliability of neuroimaging research findings. Understanding and addressing this challenge is essential to advancing scientific progress and ensuring its credibility.
This proposed educational course will cover the knowledge and tools to design, evaluate, and improve robust neuroimaging pipelines. A key feature of the course is the inclusion of multiple tools—such as BABS, BrainLife, NeuroDesk, and Nipoppy—each selected for its unique strengths and suitability for different user needs. By exploring these diverse tools, participants will gain a comprehensive understanding of how to address analytical variability in various contexts, whether working with large datasets, leveraging cloud-based infrastructures for parallel processing, shifting analysis between different computing environments, or implementing flexible workflows.
By the end of the course, participants will acquire an understanding of the consequences of analytical variability on research outcomes, practical skills for assessing and mitigating variability using cutting-edge tools and methodologies, and the confidence to implement these principles in their research. Ultimately, this course will equip neuroimaging researchers with the needed practices to drive progress in the field while ensuring the reliability and reproducibility of their findings.
1. Assess and mitigate analytical variability: Participants will be able to identify and assess the sources of analytical variability in neuroimaging pipelines, develop strategies to mitigate these variations, and implement evidence-based methods to minimize their impact on research outcomes.
2. Explore open-source tools addressing analytical variability: With the engaging demonstrations, participants will learn to use BABS, BrainLife, NeuroDesk, and Nipoppy and how to choose the most suitable tool based on context, such as large dataset analysis, parallel processing in cloud-based infrastructure, platform accessibility, or container flexibility.
3. Apply classical statistical and machine learning methods: Participants will be able to apply classical statistical and machine learning methods to analyze and interpret neuroimaging data, including techniques for handling variability and bias in pipeline outputs.
These learning objectives align with the ACCME's requirements for continuing medical education (CME) and reflect the core skills and knowledge researchers and students will gain from this course.
The target audience for this educational course is researchers and trainees working in neuroimaging, including early-career scientists, postdoctoral fellows, and graduate students conducting research in neuroscience, psychiatry, psychology, or related fields. This audience is expected to have a basic understanding of neuroimaging techniques and methods, but may not possess advanced expertise in pipeline design and analysis, making them well-suited for this comprehensive training program.
Presentations
In this introduction, we will review the evidence of analytical variability in the three main neuroimaging modalities, structural MRI, functional MRI (task, resting state and dynamic functional connectivity), and diffusion MRI. We will review the impact of the analytical flexibility on scientific conclusions and show that in many instances, conclusions will differ as a function of a chosen analytical pipeline. We will consider the rationales for choosing one pipeline over others, and lay out the motivation for this course.
We will demonstrate a workflow for large datasets and how its full audit trail is generated to validate the reproducibility of an analysis by BABS.
Presenter
Satra Ghosh, Massachusetts Institute of Technology Cambridge, MA
United States
We will introduce the cloud-based infrastructure, parallel processing, show validation experiments and the reproducibility analysis and demonstrate a pipeline reproducibility and variability.
Presenter
Anibal Heinsfeld, The University of Texas at Austin Austin, TX
United States
We will introduce its accessibility to run on any platform and infrastructure, explain the assessment of its reproducibility, and demonstrate a pipeline.
Presenter
Steffen Bollmann, The University of Queensland Brisbane, Queensland
Australia
We will introduce its flexibility to use existing containers and demonstrate its workflow.
Presenter
Michelle Wang, McGill University Montreal, Quebec
Canada
We will introduce and explain statistical approaches to deal with analytical variability.
Presenter
Thomas Nichols, PhD, University of Oxford
University of Oxford
Oxford, Oxfordshire
United Kingdom