More processing power and advanced configurability let designers run the algorithms needed for cutting edge platforms.
As baby boomers and Gen Xers reach the age where they more regularly encounter health issues, they’re seeking ways to effectively detect health issues early and treat those conditions with less invasive therapies and procedures that let them recover faster and live longer. To this end, medical equipment companies are turning to advanced electronics to create fantastically innovative and truly life-saving and enriching technologies, such as diagnostic imaging equipment and robotic-assisted surgical equipment.
At the heart of many of these medical technologies are advanced mathematical algorithms that companies traditionally implement in IC-based processors—specifically digital signal processors (DSPs). Now there’s a trend toward more flexible devices known as field-programmable gate arrays (FPGAs) to improve image-processing quality and performance and equipment responsiveness.
Physicians today use imaging technology, MRI and CT scanners, and very small arthroscopic cameras to diagnose health problems early, monitor those problems, and even assist in surgical procedures. The need to have higher image resolution requires the use of precise micro-array detector geometry and complex software/hardware systems for analyzing photonic and electronic signals. These systems generate images that require high-precision and extremely fast data processing. What’s more, this medical equipment must be highly reliable and customers expect it to last for a long time as the equipment can be very expensive. Thus, medical equipment vendors are constantly upgrading existing equipment, as well as developing new generations of equipment to sell to physicians. Each technology must be faster, smaller, and have better or new features. Much of this is accomplished with algorithms and the processors they run on.
To develop efficient and flexible medical imaging equipment, it’s important to consider the following:
Minimally-invasive surgical market stats
BCC Research predicts the global market for minimally-invasive surgical (MIS) systems will reach $23.0 billion in 2014, with a CAGR of 7.8%. They also claim that the U.S. has the largest market share, with the segment expected to reach $13.6 billion in 2014. In particular, cardiothoracic surgery is the largest application segment for MIS systems, representing 70% of the total market, followed by orthopedic surgery (12%), gastrointestinal surgery (10%), and gynecology (3%) (see Figure 1).
1. Global market for MIS devices and equipment (source: BCC Research).
As their name suggests, Intuitive Surgical has standardized on FPGA-based design blocks to innovate 3D vision and precise control for robotic-assisted, minimally invasive surgery, which means significantly less trauma and faster recovery for patients than conventional open surgery. By combining the benefits of MIS with the precision, dexterity, and control of robotics and the accuracy of breakthrough 3D visualization, the da Vinci surgical system has brought MIS to a broader range of specialties (see Figure 2). Starting with cardiac and general surgery, the platform has enabled methodologies for urology, pediatric, gynecology, colorectal, and head and neck surgeries.
2. The da Vinci robotic surgical system (photo courtesy of Intuitive Surgical)
The da Vinci system puts the surgeon in a comfortable, seated position, with an adjustable-zoom, 3D view of the operating field. Controls translate large motions into very tiny, controlled movements of surgical instruments attached to robotic arms. The extreme precision speeds up many complex procedures and greatly improves the patient experience.
The engineering challenges
Designing MIS systems requires overcoming numerous engineering challenges. Today, video and image processing is more important and a "must have" in MIS systems, as models have evolved from 3D standard-definition stereo vision to the dual-console, multi-window 3D high-definition (HD) systems.
Updating a video processing subsystem from previous-generation machines to multi-windowed video sources lets the surgeon monitor vital patient data during surgeries. An increase in video processing bandwidth allows data to be displayed from auxiliary video sources, along with the view of the operating field. The surgeon has instant feedback from an ultrasound or heart-lung machine without taking his eyes off of the procedure in progress. Besides the technical challenges of giving surgeons an expanded, immersive view that could shorten procedures and improve outcomes, video applications utilizing an FPGA can be easily aligned with the other overall system objectives including stringent safety and reliability requirements.
These display enhancements are suited for a unified programmable logic architecture that medical system designers are familiar with. When combined in a total processing system that includes an industry-standard ARM dual-core Cortex-A9 MPCore, this extensible processing platform (EPP) enables high-performance video processing through the DSP elements for streaming video, while also promoting a reduction in the real estate for video processing. This, in turn, increases design re-use within and across design teams, thus replicable, modular designs can be easily realized that enable many functions to be standardized and building blocks quickly placed into new designs.
Some of the key building block functions required for these sophisticated imaging algorithms include CT image reconstruction, which requires interpolation, fast Fourier transform (FFT) and convolution functions. In MIS and ultrasound applications, important processing methods include color flow processing, convolution, compounding, and elasticity estimation. General imaging algorithms include functions such as color space conversion, graphic overlays, 3D/median/temporal filtering, scaling, frame/field conversions, de-interlacing, and sharpening.
The modular design methodology has revolutionized manufacturability, testability, reliability, and serviceability. A flexible, customizable design block focuses on modules and cards, not systems. Thus, system servicing can occur in the field. Plus the power of FPGA programmability simplifies updates. Instead of replacing modules or subsystems, a new capability can often be introduced or enhanced with an in-field firmware upgrade. Service teams can also quickly query for consistency across all the processors in the system for improved process control while ensuring that systems are optimally configured for surgery.
Embracing an open, scalable design flow takes advantage of a reusable processor design. FPGA-based building blocks shorten design time and cost-effectively leverage the expertise of the in-house architectural experts. The reusable blocks and scalable design flow can go beyond just video to benefit general communications and servo-control designs. Re-use is especially important in medical devices as designs can be hardened by multiple teams, for varied applications, while gaining a high degree of safety and ultimately deliver a more reliable product.
Another example of FPGA use in advanced medical applications is electrosurgery, which is now widely used because of the need to control bleeding in operative procedures. The electrosurgical tools available to surgeons meet the challenge of improved patient care.
Covidien, of Dublin, Ireland, is a leading global healthcare products company that creates innovative FPGA-based medical solutions. Its ForceTriad energy platform is a full-featured radiofrequency energy system that provides an operating room with state-of-the-art technology (see Figure 3). This device lets surgeons precisely manage energy and the desired tissue effect with consistent power delivery. The energy platform delivers both enhanced monopolar and bipolar energy with the next generation of LigaSure tissue fusion technology in a self-contained unit.
3. The Covidien ForceTriad generator (photo courtesy of Covidien)
FPGAs, with their inherent benefits of integration, flexibility, and time-to-market advantages, provide upgradeable, high-speed control of the RF energy delivery for multiple Covidien applications. This includes monopolar and bipolar electrosurgery, a range of LigaSure tissue fusion devices as well as video image display, safety monitoring, and control functions. From the human standpoint, surgeons can potentially shorten their operating room procedure time by having control of power and mode adjustments from the sterile field, while reducing the demands of staff during procedures.
Some other medical imaging applications in which FPGAs are frequently used include:
FPGAs offer a combination of performance and flexibility that suits them for medical video imaging and processing applications. The flexible FPGA interconnect, integrated DSP blocks, distributed block memories, and external memory controllers enable development of fully customized, high-resolution imaging and video designs. The large number of programmable high-performance I/Os provide wide I/O bandwidth to the memory. FPGAs with large numbers of user configurable I/O pins are configurable to support multiple independent memory buses where each memory bus can be connected to an independent memory bank and multiple data accesses are supported. The wide I/O bandwidth and the inherent parallelism of FPGAs make these devices a perfect choice for processing vast amounts of streaming data in parallel (see Figure 4). These scalable devices are easily configured to meet system architecture requirements and allow platform-based development strategies that enable a common architecture to be scaled across product families.
4. Block diagram of an FPGA-based in a medical imaging application
A key advantage of an FPGA is that it allows design teams to implement algorithms that would typically require multiple DSPs on one FPGA chip. The FPGA’s hard processor blocks and high-speed DSP slices provide system-on-a-chip capabilities that support eight channels of full 1080p HD video (up to 28.05 Gbits/s) across the simplified interconnect. To help accelerate the implementation of sophisticated imaging algorithms onto these platforms, design teams use high-level development tools and IP cores.
Design teams use FPGAs to perform beam forming, color flow processing, convolution, hybrid/flexible estimation and human-machine interface (HMI) functions They simultaneously provide more processing power and enable the flexibility to adapt to the changing algorithms. Ordinary imaging algorithms include functions such as color space conversion, image overlay, 2D/median/time filtering, scaling, frame/field conversion, contrast enhancement, sharpening, edge detection, limiting, translation, Polar/Cartesian conversion, uniformity correction and pixel replacement.
Various imaging algorithms are often implemented in an FPGA, including image enhancement, stabilization, distributed wavelet analysis, vector processing, and image compression. Image enhancement is commonly performed with convolution (linear) filtering. High-pass filtering enhances the image’s detail, but also makes noise more visible. Low-pass filtering suppresses the noise at the expense of blurring the detail. Most images contain some areas with fine detail and others with broad detail. Linear convolution filtering is a technique that enhances detail in the former and reduces noise in the latter by producing both high- and low-pass filtered images and combining them according to a mask.
Design teams use image stabilization technology to normalize the average noise in successive frames when a video data sequence is being rotated or scaled. Wavelet transform is an analysis of the algorithm. It overcomes some limitations of FFT analysis as the signal comes from the time to the frequency domain and measures lost time information. To gather signal event information, a designer will use a wavelet analysis technique to receive a variable time window analysis of a small portion of the signal by accurately analyzing the low-frequency information using a longer time interval and the high-frequency information using a shorter interval.
S-transform (ST) combined with FFT and wavelet transform reveals the frequency of changes in space and time. However, STs require a large amount of calculation. Therefore, using a traditional CPU is too slow to implement. Design teams can implement distributed vector processing in parallel processing in the FPGA to accelerate compute-intensive algorithm calculations. This lets it solve a problem as much as 25X faster than a normal DPS implementation. Xilinx’s Virtex-6 FPGA embedded kit, along with additional IP and partner reference designs can accelerate integration of these algorithms onto FPGAs, including those systems with the highest performance and smallest footprints.
Kamran Khan is a technical marketing engineer at Xilinx. He has more than six years of technical marketing and application experience in the semiconductor industry focusing on FPGAs. Prior to joining Xilinx, he worked on developing FPGA-based applications and solutions for customers in an array of end markets, including medical.