High-performance Visual Computing for Biomedical Sciences

  • Abstract: Determining the detailed connections in brain circuits is a fundamental unsolved problem in neuroscience. Understanding this circuitry will enable brain scientists to confirm or refute existing models, develop new ones, and come closer to an understanding of how the brain works. High-resolution, large-scale medical images play a central role in the brain analysis, and also pose very challenging computational problems for 3D segmentation and visualization in terms of developing suitable algorithms, coping with the ever-increasing data sizes, and maintaining interactive performance. My research focuses on convergence of different research fields and various computing technologies to address scientific questions. Specifically, my main research goal is to develop interactive visualization and computation methods for neuroscience and medicine applications. In this talk, I will introduce some of my past and recent research results in GPU-accelerated biomedical image analysis. First, I will talk about the Fast Iterative Method, a parallel algorithm to solve a class of Hamilton-Jacobi equations for weighted distance computation and its application in DT-MRI white matter connectivity analysis. Second, I will introduce our GPU-accelerated semi-automatic segmentation and interactive visualization system for processing terabytes of electron microscopy image data, a first step towards the complete reconstruction of neuronal connections in the mammalian brain. Third, I will discuss our recent development in the visual computing framework for large-scale collaborative histopathology workflows

  • Speaker: Dr. Won-Ki Jeong
    • Research scientist at the Center for Brain Science at Harvard University
    • Ph.D. in Computer Science, University of Utah, 2008

Introduction

  • The Connectome Project: How is the brain wired?
    • Synapse: where 2 neurons touch each other
    • Challenges:
      • Data size
      • Complexity: 100 billion neurons
  • Visual Computing
    • Applications: Hydrodynamics Project, Quantum Chemistry
    • Employ GPU:
      • Massively parallel SIMD processor
      • Programmable, high precision
      • High throughput for data parallel problems
    • Data Structure: on-demand Data structure
  • Research Goal
    • Interactive visualization computation methods for biomedical sciences
    • Convergence of research fields & technologies
    • Neuroscience & medicine

Part I: Trace Brain Tracks

  • Used in MRI
  • Problems
    • Error accumulation
    • Ambiguity near multiple directions overlap
    • Difficult to find region-to-region paths
    • Stopping criteria
  • Proposed Method
    • Region-to-region volumetric pathway
      • Minimum Volumetric Pathway: Hamilton-Jacobi
        • Find Probability of R1 connected to R2
      • Hamilton-Jacobi Solver???
        • Old: Fast Marching (Dijsktra)
        • propose Fast Iterative Method (employ Parallelism)

Part II: Neutral Process Segmentation & Visualization in Electron Micrographs

  • How is the Brain Wired?
  • Neural Process Reconstruction

Neural Process Segmentation

  • 3D neural process as a collection of 2d rings
    • 2D membrane segmentation
    • 3D tracking
  • GPU Muliphase Level set Solver
  • Visualization
    • Out-of-core EM data visualization
    • Data reduction

Part III: Real-time Microscope for Visualizing

  • Proposed Method
    • Local, adaptive data structure
      • Demand driven, adaptive image hierarchy
    • On-the-fly processing
      • Interactive image registration
    • Texture compression
      • Image stack, GPU acceleration
  • Data Structure:
    • Process each image stack independently
  • On-the-fly Image Composition
    • Mapping images to reference space
    • Resampling within visible ranges
  • Compression: Vector Quantization
    • Clustering multiple pixels into 1
    • Decompression: table-lookup using GPU & Interpolation
  • * Collaborative Pathology System*
    • Multiple users can individual access the Visualization data
    • Client-server model
    • Multi-touch user interface

Future Research Plan

  • Biomedical Image Analysis
    • Multimodal imaging
    • Full brain atlas connectivity
    • Neural activity simulation
    • Information Visualization
  • High-performance Visual Computing
    • GPGPU
    • Heterogeneous computing
    • Parallel visualization
    • Large-data processing
Topic revision: r2 - 02 May 2011, ToanMai
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback