You are here: Machine Learning for Precision Health»Computational Neuroimaging, Structural and Functional Connectomics

Computational Neuroimaging, Structural and Functional Connectomics

Predictive analytics from brain connectivity measures of diffusion and functional MRI holds great potential to become a useful clinical tool to quantify disease severity, evaluate therapy efficacy, and predict individualized outcomes.The value of such an analytical tool is heightened by the many diseases and disorders with characteristic structural or functional connectivity deficits including: neurodegenerative diseases, stroke, traumatic brain injury (TBI), epilepsy, developmental disorders (such as autism), and neuropsychiatric disorders (such as PTSD and Schizophrenia).

Brain images

Magnetoencephalography (MEG) characterizes task specific brain function. Our lab precisely quantifies spatiotemporal electrical activation patterns (signatures) of specific tasks using functional modalities such as MEG. This modality provides exquisite temporal resolution of the electrical activation pattern within spatially distributed brain regions involved in listening to and processing of speech. Shown here are 3 snapshots after a word is spoken within a 200ms time window post stimulus presentation.

Connectivity diagram

MEG enables the characterization of the human brain at rest. Though the resting brain is not actively engaged in a targeted task, the spatiotemporal electrical activation pattern at rest contains a wealth of information about brain health. This information can be transformed into maps of functional connectivity and represented succinctly in 2D maps called connectograms. Shown here is the connectogram for 3 frequency bands including alpha and low and high beta bands for a human subject at rest.

Capitalizing on the potential of connectivity analysis requires end-to-end optimization of image acquisition, feature extraction, and processing, a function not provided by the available software pipelines. Through my research we develop approaches to optimize image acquisition, image reconstruction, and brain connectivity network extraction.

We also determine the optimum spatiotemporal scale for connectivity analysis and find the best machine learning approach to make individualized outcome predictions whose reliability we verify through rigorous statistical methods.