F-RWP3: Using high-performance graphics cards to improve medical image resolution

Graphics cards provide a very cost-effective way to achieve high computational performance for intrinsically parallel problems. This project will employ nVidia GeForce graphics cards to enable our image processing problem to be solved up to 40 times faster than a serial programme running on the main PC processor.

Deconvolution algorithms provide a way of improving the resolution of ultrasound images, but their high computational requirements have prevented them being commonly used. These algorithms are however highly parallel in nature and very well-suited to efficient implementation on a graphics card.

The aim of the project is thus to develop software that will enable enhanced ultrasound images to be produced in real-time using a simple deconvolution algorithm running on a graphics card in a conventional PC. Ultrasound image acquisition will be performed using our Stradwin system, and the graphics card will be programmed using the nVidia CUDA environment.

This project will involve programming in C++ and C.

The ultrasound data on the left is of lower resolution than the data on the right. The goal of the project is to take low resolution data, like that on the left and process it using a deconvolution algorithm to produce an image with resolution as close as possible to that shown on the right. The edges of the voids are much clearer in the right-hand image.

Click here for other medical imaging projects offered by Richard Prager or Graham Treece.