Medical image processing

High performance hardware is required for the demanding workloads associated with medical image processing. Such systems represent significant investments and they need to be supported throughout long lifecycles. Implying lifecycle management (LCM) costs are a major component in total cost of ownership (TCO).

Today’s typical hardware architectures are x86 CPUs combined with GPUs for offloading specific tasks such as image reconstruction. While meeting performance requirements and being easy-to-use, these architectures suffer from frequent component obsolescence and incompatibility issues, escalating LCM costs.


FPGAs as medical image processing accelerators

FPGAs are a good fit with the requirements for medical image processing, because of their long availability and inherent flexibility. Additionally, the traditional bottleneck of FPGAs (programmability) has strongly improved due to the maturing support of framework languages such as OpenCL.

The long availability of FPGAs is rooted in their widespread use in the embedded computing market. Like in the medical market, embedded systems typically have long lifecycles. FPGA vendors support these lifecycles by offering their products lasting 15 years or longer, eliminating the costs of frequent redesigns and last-time buys and reducing cost of ownership.

As an array of programmable gates, FPGAs are very flexible by nature. Both compute resources and I/O functions can be optimally handled in an FPGA. This make FPGAs a very future-proof solution, simplifying platform management and reducing time-to-market.

Finally, the ease-of-use has improved significantly in recent years. FGPA programming is done in hardware description language (HDL), which is typically seen as cumbersome. Yet, in recent years frameworks such as OpenCL, have matured and offer C-like environments that enable easy programming.


Defining the future medical image processing architecture

Prodrive Technologies is exploring Intel® IoT technologies and Intel® FPGAs as hardware accelerators in the medical image processing architecture of the future. Prodrive Technologies will define a benchmarking setup that will combine Intel® Xeon® based CPUs with various types of hardware accelerators, including Intel®’s Arria 10 FPGA series.

The different architectures will be benchmarked using RabbitCT, an open source CT back projection image reconstruction algorithm. This algorithm is then optimized for each of the architectures to determine to what extent performance can be maximized, but also to determine the ease-of-use of each platform.

The results will show in the future medical image processing platform, which optimizes performance, minimizes cost of ownership, and is future-proof.

We will be present at the RSNA – Radiological Society of North America from November 26 to December 1 2017. Be sure to visit out booth #6663, North, Hall B.