Rapid geodesic mapping of brain functional connectivity: Implementation of a dedicated co-processor in a field-programmable gate array (FPGA) and application to resting state functional MRI

Ludovico Minati, Mara Cercignani, Dennis Chan

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

Graph theory-based analyses of brain network topology can be used to model the spatiotemporal correlations in neural activity detected through fMRI, and such approaches have wide-ranging potential, from detection of alterations in preclinical Alzheimer's disease through to command identification in brain-machine interfaces. However, due to prohibitive computational costs, graph-based analyses to date have principally focused on measuring connection density rather than mapping the topological architecture in full by exhaustive shortest-path determination. This paper outlines a solution to this problem through parallel implementation of Dijkstra's algorithm in programmable logic. The processor design is optimized for large, sparse graphs and provided in full as synthesizable VHDL code. An acceleration factor between 15 and 18 is obtained on a representative resting-state fMRI dataset, and maps of Euclidean path length reveal the anticipated heterogeneous cortical involvement in long-range integrative processing. These results enable high-resolution geodesic connectivity mapping for resting-state fMRI in patient populations and real-time geodesic mapping to support identification of imagined actions for fMRI-based brain-machine interfaces.

Original languageEnglish
Pages (from-to)1532-1539
Number of pages8
JournalMedical Engineering and Physics
Volume35
Issue number10
DOIs
Publication statusPublished - Oct 2013

Fingerprint

Brain Mapping
Field programmable gate arrays (FPGA)
Brain
Magnetic Resonance Imaging
Brain-Computer Interfaces
Computer hardware description languages
Graph theory
Alzheimer Disease
Topology
Costs and Cost Analysis
Coprocessor
Processing
Population
Costs

Keywords

  • Dijkstra's algorithm
  • Field-programmable gate array (FPGA)
  • Functional connectivity
  • Graph theory
  • Network topology
  • Resting-state functional MRI (rs-fMRI)

ASJC Scopus subject areas

  • Biomedical Engineering
  • Biophysics

Cite this

@article{ea4adebe869b48eb8543709676857ba9,
title = "Rapid geodesic mapping of brain functional connectivity: Implementation of a dedicated co-processor in a field-programmable gate array (FPGA) and application to resting state functional MRI",
abstract = "Graph theory-based analyses of brain network topology can be used to model the spatiotemporal correlations in neural activity detected through fMRI, and such approaches have wide-ranging potential, from detection of alterations in preclinical Alzheimer's disease through to command identification in brain-machine interfaces. However, due to prohibitive computational costs, graph-based analyses to date have principally focused on measuring connection density rather than mapping the topological architecture in full by exhaustive shortest-path determination. This paper outlines a solution to this problem through parallel implementation of Dijkstra's algorithm in programmable logic. The processor design is optimized for large, sparse graphs and provided in full as synthesizable VHDL code. An acceleration factor between 15 and 18 is obtained on a representative resting-state fMRI dataset, and maps of Euclidean path length reveal the anticipated heterogeneous cortical involvement in long-range integrative processing. These results enable high-resolution geodesic connectivity mapping for resting-state fMRI in patient populations and real-time geodesic mapping to support identification of imagined actions for fMRI-based brain-machine interfaces.",
keywords = "Dijkstra's algorithm, Field-programmable gate array (FPGA), Functional connectivity, Graph theory, Network topology, Resting-state functional MRI (rs-fMRI)",
author = "Ludovico Minati and Mara Cercignani and Dennis Chan",
year = "2013",
month = "10",
doi = "10.1016/j.medengphy.2013.04.014",
language = "English",
volume = "35",
pages = "1532--1539",
journal = "Medical Engineering and Physics",
issn = "1350-4533",
publisher = "Elsevier BV",
number = "10",

}

TY - JOUR

T1 - Rapid geodesic mapping of brain functional connectivity

T2 - Implementation of a dedicated co-processor in a field-programmable gate array (FPGA) and application to resting state functional MRI

AU - Minati, Ludovico

AU - Cercignani, Mara

AU - Chan, Dennis

PY - 2013/10

Y1 - 2013/10

N2 - Graph theory-based analyses of brain network topology can be used to model the spatiotemporal correlations in neural activity detected through fMRI, and such approaches have wide-ranging potential, from detection of alterations in preclinical Alzheimer's disease through to command identification in brain-machine interfaces. However, due to prohibitive computational costs, graph-based analyses to date have principally focused on measuring connection density rather than mapping the topological architecture in full by exhaustive shortest-path determination. This paper outlines a solution to this problem through parallel implementation of Dijkstra's algorithm in programmable logic. The processor design is optimized for large, sparse graphs and provided in full as synthesizable VHDL code. An acceleration factor between 15 and 18 is obtained on a representative resting-state fMRI dataset, and maps of Euclidean path length reveal the anticipated heterogeneous cortical involvement in long-range integrative processing. These results enable high-resolution geodesic connectivity mapping for resting-state fMRI in patient populations and real-time geodesic mapping to support identification of imagined actions for fMRI-based brain-machine interfaces.

AB - Graph theory-based analyses of brain network topology can be used to model the spatiotemporal correlations in neural activity detected through fMRI, and such approaches have wide-ranging potential, from detection of alterations in preclinical Alzheimer's disease through to command identification in brain-machine interfaces. However, due to prohibitive computational costs, graph-based analyses to date have principally focused on measuring connection density rather than mapping the topological architecture in full by exhaustive shortest-path determination. This paper outlines a solution to this problem through parallel implementation of Dijkstra's algorithm in programmable logic. The processor design is optimized for large, sparse graphs and provided in full as synthesizable VHDL code. An acceleration factor between 15 and 18 is obtained on a representative resting-state fMRI dataset, and maps of Euclidean path length reveal the anticipated heterogeneous cortical involvement in long-range integrative processing. These results enable high-resolution geodesic connectivity mapping for resting-state fMRI in patient populations and real-time geodesic mapping to support identification of imagined actions for fMRI-based brain-machine interfaces.

KW - Dijkstra's algorithm

KW - Field-programmable gate array (FPGA)

KW - Functional connectivity

KW - Graph theory

KW - Network topology

KW - Resting-state functional MRI (rs-fMRI)

UR - http://www.scopus.com/inward/record.url?scp=84883140389&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84883140389&partnerID=8YFLogxK

U2 - 10.1016/j.medengphy.2013.04.014

DO - 10.1016/j.medengphy.2013.04.014

M3 - Article

C2 - 23746911

AN - SCOPUS:84883140389

VL - 35

SP - 1532

EP - 1539

JO - Medical Engineering and Physics

JF - Medical Engineering and Physics

SN - 1350-4533

IS - 10

ER -