- This article is filed under:
- » View All
Medical Imaging is Never Just 'Good Enough'
by Winn Hardin, Contributing Editor - AIA Posted 12/20/2005
The progression from Willhelm Röntgen's x-ray to Robert Ledley's computed tomography scan (CAT-scan) system and Texture Analysis Computer (TEXAC) continues today, pushing medical imaging and analysis to ever more complex systems with greater capabilities.
From Röntgen's 2D radiograph to Ledley's 3D CAT files, physicians continue to seek diagnostic systems capable of seeing inside the human body with greater fidelity, speed and specificity than previously possible. The machine vision industry is helping in concrete ways by developing new hardware and software that targets the core needs of medical imaging systems: speed, resolution and image processing power – and then feeding those lessons learned back into industrial systems.
The dynamics of medical imaging
Although physicians resisted the first transitions from film-based x-ray systems to digital image collection systems, today the technology is widely accepted and accompanied by further needs for increased spatial resolution. A typical x-ray system places a 50 to 100 keV x-ray source, typically tube based, on one side of the subject and an image intensifier or other conversion mechanism on the other side of the subject to convert the transmitted x-rays into visible light. The visible radiation from the intensifier is then collected by a large-area silicon detector, or high-resolution CCD sensor.
Adimec (North American office located in Stoneham, Massachusetts, USA and headquarters in Eindhoven, Netherlands) introduced its first 12-bit, megapixel CCD camera for medical imaging in the early 1990s and has progressed to high-resolution – up to 4 megapixels – and higher-performance sensors in recent years. ‘‘The dynamic range of an x-ray image is very large,’‘ explains Adimec's technical marketing manager, Miriam van Baalen. ‘‘Our cameras must show small differences in grey scale because when you look at an x-ray image, small differences in grey scale can indicate a fracture to a bone. Thus, dynamic range is very important, but so is the image quality itself.’‘
The dynamic range of a camera is mainly defined by two parameters; Qmax and noise of the sensor. According to Adimec's product support engineer, Bill Cortright, the challenge in designing a camera in general and especially for medical X-ray imaging is to optimize these two parameters. ‘‘Adimec is able to maximize the camera’s dynamic range by optimizing the design of the camera such that the noise floor is lowered and the maximum Q per pixel is increased. This, in combination with low noise video processing and high performance A-to-D conversion, allow us to deliver maximum dynamic range from the camera.’‘
Many digital cameras offer 12-bit grey scale depth, but will not use the full capability of the digitizer – not all 12-bits are always used to display true differences in grey scale values. For example, with some industrial cameras, the operator may have a selectable 8- or 10-bit output from a 12-bit camera because all 12-bits are not completely comprised of useful video information. They may also contain noise information. With the Adimec MX12D medical cameras, explains Cortright, ‘‘we are able to provide full 12-bit dynamic range at the output of the camera. In fact, the internal processing accuracy is significantly higher than 12 bits.’‘
Speeding up image acquisition
Just as industrial applications require fast frame rates for maximum throughput, frame speed is also an important camera feature for medical imaging. Capturing of real time images(>30fps) is critical in medical imaging applications. Moving to higher resolution cameras, based on the requirement for image quality in combination with real time, results in a need for higher pixel read out rates and increased image processing further down the line.
Multi-tap or multi-channel sensors -- sensors with more than one read-out channel -- allow a sensor to offload pixel data faster, which is helpful as the number of pixels increases per sensor. However, answering one system need, dual-channel sensors pose a new challenge for medical imaging applications.
According to Adimec's van Baalen, ‘‘Sensors with two outputs require the data streams to be matched perfectly to each other, making sure there aren't any artifacts introduced in the image. With a dual tap operation, you process two halves of the image simultaneously, and then stitch them back together to get one image at full resolution. You must not be able to see a seam in the image. Doctors want to see a perfect image and the standard is rising all the time. Currently, the image quality standard for medical imaging is about 10x higher than the requirement in machine vision.’‘ Adimec has developed several proprietary methods to improve image quality for medical applications, says Cortright, and insure against artifacts.
After you get the image data out of the sensor, the information still has to be filtered, processed and displayed for the physician. This process is complicated by the additional needs of real-time x-ray imaging, such as x-ray fluoroscopy, used during orthoscopic and endoscopic procedures. These systems allow physicians to track where catheters and probes are inside the human body. Image processing specialists, such as DALSA, are optimizing image processing boards to meet the specific needs of real-time medical imaging systems.
According to Inder Kohli, product manger for medical imaging at DALSA, real-time x-ray systems not only require additional image processing power to handle increases in resolution and dynamic range, but also in frame rate. ‘‘With the XRI-1200, we've integrated all the critical processing tasks on a single piece of hardware. From getting frame averages to correct for noise and lens aberration, to rotating the image to maintain the visual perspective so that the doctor can look at the images regardless of how the patient is laid out or how the sensors are placed around the patient, the XRI-1200 handles all the processing so it doesn’t burden the PC.’‘
Image rotation and orientation are particularly helpful for endoscopic procedures, Kohli said, where the probe may take many turns inside the body but the doctor needs to maintain a constant visual perspective. ‘‘The doctor doesn't want to think: is the probe coming from the top, or the bottom of the body?’‘
In addition to standard noise reduction algorithms, DALSA implemented a motion sensing algorithm for real-time x-ray imaging in a field gate programmable array (FPGA) and backed with significant buffer memory – up to 2 GB per board. ‘‘If there's motion, there's a blurry effect in the image, and that's not acceptable. What we do is detect if there's motion as well as how much motion there is. If there's rapid movement, there's no sense averaging in that particular frame into the display. If the motion is small enough, we do adaptive averaging,’‘ similar to weighting the fidelity of each frame before averaging.
‘‘Adding a floating point morphing engine that corrects for visual perspective also lets people use less expensive lenses. We've used FPGA's for image processing before, but this is the first time we've used it specifically for medical imaging. It allows us to adapt to different OEM requirements quite easily. Unfortunately, there isn't a standard set of requirements for every OEM,’‘ Kohli explained.
As we can see, from the sensor to image processing engines, machine vision suppliers are developing both hardware architectures and algorithms that accommodate larger data streams in medical imaging systems caused by improvements in spatial resolution and dynamic range while improving throughput. While this is great news for the medical industry and patients alike, it also brings a silver lining for industrial machine vision systems. ‘‘By supplying cameras to the medical market we have had to deliver cameras with top image quality,’‘ explains Adimec's van Baalen. ‘‘The added good news is that we've been able to transfer technology improvements back to the industrial side as well, maximizing performance of our products across the board.’‘
There are currently no comments for this article.
Leave a Comment:
All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.