A heterogeneous accelerator platform for multi-subject voxel-based brain network analysis

Wang, Yu; Xu, Jun; Ren$^{, }$; Zhang, Bo; Wu, Kan; He, Chaobin

doi:10.1109/iccad.2011.6105352

Cited by 4 publications

(6 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our FPGA implementation has 32 GPEs (128 GPEs in total) that utilise about 69% of LUTs and 86% of BRAMs, with an operational clock frequency of 150MHz. We compare our performance results to CPU and GPU results reported in [18] and [6] respectively. An AMD Phenom II X4 965 quad-core CPU running at 3.4 GHz is used in [18] for APSP-BFS, while an AMD Cypress GPU (Radeon HD 5870) is used in [6] for APSP-FW.…”

Section: Resultsmentioning

confidence: 87%

“…We compare our performance results to CPU and GPU results reported in [18] and [6] respectively. An AMD Phenom II X4 965 quad-core CPU running at 3.4 GHz is used in [18] for APSP-BFS, while an AMD Cypress GPU (Radeon HD 5870) is used in [6] for APSP-FW. Figure 4 shows the scalability of our FPGA-based APSP design for graphs with different sparsity values.…”

Section: Resultsmentioning

confidence: 87%

“…The resulting graphs have 38368 vertices with varying sparsity values. This is the same dataset used in [18]. For the high performance reconfigurable computing system, we use the Convey HC-1 server [19] which has four Virtex-5 LX330 FPGAs which are connected to a shared memory subsystem.…”

Section: Resultsmentioning

confidence: 99%

“…Because of the small-world phenomenon the number of vertices in each BFS level grows very rapidly, leading to most edges being examined in one or two BFS levels, the critical BFS levels. Algorithm 4 describes the bottom-up algorithm that replaces lines (6)(7)(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24) in Algorithm 3, i.e. steps 1-4.…”

Section: Hybrid Bfs Kernelmentioning

confidence: 99%

See 3 more Smart Citations

Parallel FPGA-based all pairs shortest paths for sparse networks: A human brain connectome case study

Betkaoui

Wang

Thomas

et al. 2012

22nd International Conference on Field Programmable Logic and Applications (FPL)

Self Cite

View full text Add to dashboard Cite

This paper proposes a highly parallel and scalable reconfigurable design for the All-Pairs Shortest-Paths (APSP) algorithm for very sparse networks. Our work is motivated by a computationally intensive bioinformatics application that employs this memory-latency bound algorithm. The proposed design methodology takes advantage of distributed on-chip memory resources of modern FPGAs to reduce accesses to high-latency off-chip memories. We develop design optimisations that yield different FPGA configurations which are selected at run time based on the input graph data. Using human brain network data, we are able to achieve performance results superior to those from multi-core CPU and GPU, while attaining linear scaling over the number of processors introduced. Our FPGA-based APSP design is over 10 times faster than a quad-core CPU implementation and 2-5 times faster than an AMD Cypress GPU implementation.

show abstract

Section: Resultsmentioning

confidence: 87%

Section: Resultsmentioning

confidence: 87%

Section: Resultsmentioning

confidence: 99%

Section: Hybrid Bfs Kernelmentioning

confidence: 99%

See 2 more Smart Citations

Parallel FPGA-based all pairs shortest paths for sparse networks: A human brain connectome case study

Betkaoui

Wang

Thomas

et al. 2012

22nd International Conference on Field Programmable Logic and Applications (FPL)

Self Cite

View full text Add to dashboard Cite

show abstract

“…However, the field has recently being gaining more attention due to the rapid evolvement of modern neuroimaging techniques, giving more rooms for network information. Structural description of connectivity of the human brain, the connectome, currently exists as a viable radiological approach to investigating, detecting and analyzing vascular diseases [8]. With the complex nature of the brain, its investigation is usually limited to what the eye could grasp.…”

Section: Introductionmentioning

confidence: 99%

ConnectViz: Accelerated Approach for Brain Structural Connectivity Using Delaunay Triangulation

Adeshina

Hashim

2015

Interdiscip Sci Comput Life Sci

View full text Add to dashboard Cite

Stroke is a cardiovascular disease with high mortality and long-term disability in the world. Normal functioning of the brain is dependent on the adequate supply of oxygen and nutrients to the brain complex network through the blood vessels. Stroke, occasionally a hemorrhagic stroke, ischemia or other blood vessel dysfunctions can affect patients during a cerebrovascular incident. Structurally, the left and the right carotid arteries, and the right and the left vertebral arteries are responsible for supplying blood to the brain, scalp and the face. However, a number of impairment in the function of the frontal lobes may occur as a result of any decrease in the flow of the blood through one of the internal carotid arteries. Such impairment commonly results in numbness, weakness or paralysis. Recently, the concepts of brain's wiring representation, the connectome, was introduced. However, construction and visualization of such brain network requires tremendous computation. Consequently, previously proposed approaches have been identified with common problems of high memory consumption and slow execution. Furthermore, interactivity in the previously proposed frameworks for brain network is also an outstanding issue. This study proposes an accelerated approach for brain connectomic visualization based on graph theory paradigm using compute unified device architecture, extending the previously proposed SurLens Visualization and computer aided hepatocellular carcinoma frameworks. The accelerated brain structural connectivity framework was evaluated with stripped brain datasets from the Department of Surgery, University of North Carolina, Chapel Hill, USA. Significantly, our proposed framework is able to generate and extract points and edges of datasets, displays nodes and edges in the datasets in form of a network and clearly maps data volume to the corresponding brain surface. Moreover, with the framework, surfaces of the dataset were simultaneously displayed with the nodes and the edges. The framework is very efficient in providing greater interactivity as a way of representing the nodes and the edges intuitively, all achieved at a considerably interactive speed for instantaneous mapping of the datasets' features. Uniquely, the connectomic algorithm performed remarkably fast with normal hardware requirement specifications.

show abstract

ConnectViz: Accelerated approach for brain structural connectivity using Delaunay triangulation

Adeshina

Hashim

2015

Interdiscip Sci Comput Life Sci

View full text Add to dashboard Cite

Stroke is a cardiovascular disease with high mortality and long-term disability in the world. Normal functioning of the brain is dependent on the adequate supply of oxygen and nutrients to the brain complex network through the blood vessels. Stroke, occasionally a hemorrhagic stroke, ischemia or other blood vessel dysfunctions can affect patients during a cerebrovascular incident. Structurally, the left and the right carotid arteries, and the right and the left vertebral arteries are responsible for supplying blood to the brain, scalp and the face. However, a number of impairment in the function of the frontal lobes may occur as a result of any decrease in the flow of the blood through one of the internal carotid arteries. Such impairment commonly results in numbness, weakness or paralysis. Recently, the concepts of brain's wiring representation, the connectome, was introduced. However, construction and visualization of such brain network requires tremendous computation. Consequently, previously proposed approaches have been identified with common problems of high memory consumption and slow execution. Furthermore, interactivity in the previously proposed frameworks for brain network is also an outstanding issue. This study proposes an accelerated approach for brain connectomic visualization based on graph theory paradigm using Compute Unified Device Architecture (CUDA), extending the previously proposed SurLens Visualization and Computer Aided Hepatocellular Carcinoma (CAHECA) frameworks. The accelerated brain structural connectivity framework was evaluated with stripped brain datasets from the Department of Surgery, University of North Carolina, Chapel Hill, United States. Significantly, our proposed framework is able to generates and extracts points and edges of datasets, displays nodes and edges in the datasets in form of a network and clearly maps data volume to the corresponding brain surface. Moreover, with the framework, surfaces of the dataset were simultaneously displayed with the nodes and the edges. The framework is very efficient in providing greater interactivity as a way of representing the nodes and the edges intuitively, all achieved at a considerably interactive speed for instantaneous mapping of the datasets' features. Uniquely, the connectomic algorithm performed remarkably fast with normal hardware requirement specifications.

show abstract

A heterogeneous accelerator platform for multi-subject voxel-based brain network analysis

Cited by 4 publications

References 22 publications

Parallel FPGA-based all pairs shortest paths for sparse networks: A human brain connectome case study

Parallel FPGA-based all pairs shortest paths for sparse networks: A human brain connectome case study

ConnectViz: Accelerated Approach for Brain Structural Connectivity Using Delaunay Triangulation

ConnectViz: Accelerated approach for brain structural connectivity using Delaunay triangulation

Contact Info

Product

Resources

About