Ultrasound Scan Conversion Using OpenCL

This is the course project of Parallel Computing, Spring 2018 at SCU. It uses OpenCL to accelerate the scan conversion of ultrasound frames.

Scan conversion is necessary to convert from the polar coordinates associated with the ultrasound data to the Cartesian coordinates used by standard video monitors, as is shown below.

Table of Contents

Features
Benchmarks
- Comparison of different interpolation methods (CPU version, compared to bicubic)
- Speedup (cpu_time / kernel_time)
Prerequisites
Building
Usage
Evaluation and Benchmarking
Built with
License

Features

Cross platform:
- Windows
- Linux
- macOS (not tested)
OpenCL acceleration (Optional)
Multiple interpolation algorithms:
- Nearest
- Bilinear
- Cubic for row and linear for column
- Catmul-Rom Spline
- Bicubic

Benchmarks

Comparison of different interpolation methods (CPU version, compared to bicubic)

Speedup (cpu_time / kernel_time)

Prerequisites

OpenCL 2.0 (optional)
CMake (for building)
A C compiler supporting the timespec_get function of C11
Python 3 with the following packages (for benchmarking):
- matplotlib
- numpy
- scikit-image

Building

git clone this repository, and cd to the directory to run build commands.

Build Options

The following options are defined in config.h.in, and can be turned on/off with cmake -D OPTION=ON|OFF. CMake will generate config.h according to command line options, which will be included by the source files as a configuration file.

Options	Default	Description
`NDEBUG`	ON	Disable `DBG_PRINT` and `assert`
`TIMING`	ON	Enable timing
`USE_OPENCL`	ON	Use OpenCL

Build Types

By default, the Release version will be built. To build the Debug version, pass -D CMAKE_BUILD_TYPE=Debug to CMake when generating a build system.

On Windows, when generating a Visual Studio solution, the build type is not controlled by CMake. Instead, it is set in build options of MSBuild or Visual Studio. For command line users, please use CMake in build tool mode as is shown below.

On Unix-like systems

To build with the default configurations, run the following commands:

mkdir -p build/linux/release
cd build/linux/release
cmake ../../..
make

The executable scanconv will be generated under "build/linux/release".

On Windows

For example, to build the Debug version, run the following commands:

mkdir build\windows\debug
cd build\windows\debug
cmake -D CMAKE_BUILD_TYPE=Debug ..\..\..
cmake --build . --config Debug

The executable scanconv will be generated under "build\windows\debug\Debug".

Usage

Synopsis:

./scanconv <dat_file>

For example:

$ ./scanconv image.dat
$

Then raw.png and interpolated.png will be generated under the working directory. raw.png is the raw image decoded from image.dat, and interpolated.png is the interpolated version.

Evaluation and Benchmarking

cd to the building output directory, and

run python3 evaluate_cpu.py to evaluate different interpolation methods (CPU version, compared to bicubic);
run python3 profile.py to see the speedup of OpenCL over CPU.

Built with

stb_image_write.h: image writing to disk (png)

License

Licensed under the MIT License.

uncleshushu / scanconv