Machine Learning with Big Data in Neuroimaging

With tremendous advances in imaging techniques, computing infrastructure, and data storage, human neuroscience has access to vast amounts of structured and unstructured data from a variety of measurement modalities, such as fMRI, diffusion MRI, PET, SPECT, and PET-MRI, at an unprecedented pace. These imaging data are likely to increase in quantity and quality in the near future, thus posing both challenges and opportunities to investigate the functions and dysfunctions of the human brain. Turning such massive data into a rich source of information is reliant on the ability to mine value from vast and complex data—finding a needle in a haystack—through data analytics; machine learning (ML), network analysis (NA), and artificial intelligence (AI) are at its core, due to their ability to learn from data and to offer data-driven insights, decisions, and predictions. However, a swath of challenges—and opportunities—arise in every stage of an analytics workflow, ranging from data wrangling, data selection, constraint learning, and predictive modeling to model deployment. Some examples of challenges include dimension reduction or representation learning from high-dimensional complex and time-series data; structural and functional connectivity estimation; graph networks; kinetic modeling; multimodal data integration; resolution enhancement; missing values and confounding bias; model interpretability; uncertainty estimation; and distribution shifts between source and target domains. In addition, there is a large gap between general-purpose ML/NA/AI techniques and their application to neuroimaging; as such, there are great challenges in integrating knowledge of neuroanatomy and physiology into models.

This Topic focuses on studies and results that address the aforementioned challenges and investigate the potential of ML and AI techniques that bridges the gap between AI and neuroimaging in both healthy and diseased populations.

We welcome contributions that address, but are not limited to, the following themes:
- Dimension reduction approaches to study the relationship between brain structure and function in healthy and diseased populations.
- Analytics tools based on ML and AI techniques for structural or functional imaging data in healthy and diseased populations.
- Novel tools for analysis and visualization of brain connectivity in healthy and diseased populations.