c31 lecture 16

Clinical assessment eg for cancer using improved

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ng compressive sensing • Better clinical assessment (e.g., for cancer) using improved registration and segmentation algorithms Magnetic resonance (MR) angiograph of an aneurysm Hemodynamic simulation Very useful for surgical procedures involving blood flow and vasculature Both may take hours to days to construct Intracranial aneurysm reconstruction with hemodynamics 41 Overview of CDSC Research Program Customizable Heterogeneous Platform (CHP) $ $ $ $ $ $ $ $ DRAM DRAM I/O I/O CHP CHP Fixed Fixed Core Core Fixed Fixed Core Core Fixed Fixed Core Core Fixed Fixed Core Core DRAM DRAM CHP CHP CHP CHP Custom Custom Core Core Custom Custom Core Core Custom Custom Core Core Custom Custom Core Core Prog Prog Fabric Fabric Prog Prog Fabric Fabric Domain-specific-modeling (healthcare applications) Ap p accelerator accelerator accelerator accelerator Reconfigurable RF-I bus Reconfigurable optical bus Transceiver/receiver Optical interface n on ai ati m iz Do ter ac ar Architecture ch lic at io n m od el in g modeling CHP creation Customizable computing engines Customizable interconnects Design once CHP mapping Source-to-source CHP mapper Reconfiguring & optimizing backend Adaptive runtime Invoke many times 42 Application Thrust: Highlights ♦ Compressive sensing Applied to CT (computed tomography) and MR (magnetic resonance) imaging For CT, quality is maintained or improved with ~20% of conventional sampling Currently testing with real-world phantom data and physical acquisition constraints for MR Exploring for additional reconstruction modalities (e.g., MR spectroscopy, MRS), angiography ♦ Medical image processing pipeline Library of algorithms across all stages of the pipeline have been implemented • Denoising/deblurring, registration, segmentation • Threaded, CnC, GPU, and FPGA 43 Domain-Specific Modeling Overview Create executable models (DSCG + step code) Producer-Consumer Producer-Consumer edges edges (data dependence) (data dependence) (step 1) (step 1) [item] [item] Static Analysis Static Analysis Parent-Child edges<t2 Parent-Child edges<t2 (control dependence) > (control dependence) > (step 2) (step 2) Abstract execution Abstract execution (step 1) (step 1) (step 2) (step 2) Code Code generation generation Static Dynamic characteristics characteristics data types, custom instruction Mapping tool Auto chain generation patterns, accelerator opportunities, vector parallelism, task parallelism, data access patterns, … of stub code for...
View Full Document

This note was uploaded on 04/03/2014 for the course CS 31 taught by Professor Melkanoff during the Fall '00 term at UCLA.

Ask a homework question - tutors are online