In this article, guest author Bill Hornof DVM DACVR discusses bit depth in veterinary digital radiography. This article is an excerpt from the new book from Animal Insides Press titled "Purchasing a Veterinary Digital Radiography System Without Getting Your Head Handed To You. This is one of those topics that gets alot of focus during the sales process but might not be where you want to focus your time when you search for a digital system. Nonetheless, Dr. Hornof is here to explain all you need to know...and then some. Take it away Dr. H....
Before starting on an explanation of image processing, we really need to begin to think like a computer, and that means to a computer digital radiographs are not images but arrays of pixel values. These values will have a range that will depend on how many bits were used to digitize the radiation absorbed by the pixels in the image. But how many bits are necessary?
We have already begun a discussion of noise, as it relates to dose. We previously discussed the random noise caused by the statistical uncertainty of detecting radiation. However, there are other causes of random noise that can add to the noise spectrum. Things like electronic noise in the system, thermal effects, and electromagnetic interference can all add to the baseline noise present in a system. The dynamic range of a system is defined as the minimum dose detectable above noise, divided into the maximum dose the system can record.
As stated previously dynamic range is greater than 1,000:1 in a digital system. As a reminder a 10 bit system can contain 1024 different values, a 12 bit system can contain 4096, a 14 bit system 16,384 and a 16 bit system 65,536. At first glance it would seem like a 10 bit system would not be able to completely capture information from digital radiographic system with a dynamic range of >1,000:1. However, in practical application that entre dynamic range is never used. The radiation detected beneath the patient (the clinically relevant diagnostic information) will represent a subset of the entire dynamic range.
With CR systems a prescan is conducted to determine where the patent information lies within the entire possible spectrum before the digitization occurs. This is done by creating a histogram. A histogram is a graph of the range of pixel values on the Y-axis and the number of pixels with that value on the X-axis. By analyzing the histogram the portion of the histogram that contains the patient information can be determined (Figure 1).
Figure 1. A histogram created from the raw data. In a histogram the pixel value (X-Axis) is plotted against the number of pixel with that value (Y-Axis). The pixel values span a range of 0-16,000 indicating this is a 14-bit system. Notice the pixels from the both the shoulder and the background are separated from the pixels of the body. With the LUT chosen for display, this is reflected in the radiograph as the shoulder being white and the background behind the elbow being black.
How many shades of gray are necessary to record a radiograph? The human eye can only register 50-100 simultaneous shades of gray. The answer then to how many shades are needed to display a radiograph is there must be enough such that the narrowest window (steepest LUT) still contains >50 pixel values that range from black to white.
Figure 2. Conventional radiograph digitized at different bit depths. In a bit depth of 4 bits (16 gray shades) and 5 bits (32 gray shades) was used. Individual gray shades (banding) are visible in the LUT used, while in B there is no apparent difference between the 6 bit (64 gray shades) image and the 12 bit (4096 gray shades) image. However, in C if the window is narrowed and recentered to look at the underexposed areas of the radiograph both within and outside the collimator, banding is clearly visible in the 6-bit image, but not the 12-bit image.
At this point it is probably worth pointing out that the difference in bit depth between color and grayscale images. A typical digital color camera has 24-32 bit color, while digital radiographic systems have only 10-16 bits of grayscale. What’s the difference? A color image is produced by combining red, green and blue. Thus a 24-bit color system will have one third of that for each color or 8 bits (256 shades) of each color, and they can be mixed and matched to produce millions of colors. However, to produce a shade of gray the intensity of each of the three colors must be exactly equal. Thus a 24-bit color camera can produce only 256 shades (8 bits) of gray.
As seen in Figure 2, a radiograph digitized at a bit depth of only 6 will have 64 shades of gray and banding will not be visible, unless the window is narrowed to evaluate the over or under exposed areas. Digital radiographic systems use a minimum of 12 bits to digitize the radiation exposure, but some use as many as 16 bits.
However, the For-Presentation radiograph, the one that has been processed and is ready for display, ranges from only 10 bits to 16 bits, depending on the system. The immediate question is, are more bits better? The answer is after a point, no. If the image processing used by a system capturing raw data at any bit depth of 12 bits or more results in a quality 10 bit For-Presentation radiograph, that is all that is needed, nothing is lost including dynamic range. However, if the image processing is flawed, information can be lost regardless of bit depth. If we remember a 10-bit radiograph has 1,024 shades of gray, and if we can only perceive 50 simultaneous shades of gray, it means we can look at the darkest or brightest 5% of the pixels in the image without banding. That’s plenty of gray shades provided the image has not been clipped. Clipping means valuable patient information has been lost in the image processing steps. We will cover this in more detail in the section on image processing.
A common misconception is that bit depth affects the dynamic range of a system, and it does not. Additionally computers think in terms of bytes. One byte is 8 bits (256 gray shades), and in order to store even one more bit it requires 2 bytes. That means the cost in terms of memory and speed to store 16 bits is the exactly the same as storing 10. So why not digitize everything at 16 bits? The answer is in the early days of digital, special software was necessary to display bit depths greater than 8, and some PACS systems could not accept 14-16 bit images.
Systems developed years ago needed to compress the bit depth of their for-presentation images for PACS systems to properly display them. This is not much of a factor today, but since the bit depth compression algorithms used by these vendors has already passed FDA approval, and because they do not cause any degradation in image quality, they have largely been left in place. Thus a system delivering a 10-12 bit for-presentation image is likely to be from a system with years of experience in the market, while 14-16 bit systems tend to be newer and have not had to develop bit compression algorithms. Although vendors tout greater bit depth as an advantage, image quality wise there is no difference.
About Dr. Hornof: Bill received his Bachelor of Science, Master of Science and Doctor of Veterinary Medicine degrees from The University of California, Davis. He was named a Diplomate of the American College of Veterinary Radiology in 1979 and joined the faculty of the University of California Davis that same year. He had a distinguished academic career with over 100 publications in imaging and informatics. In 2004 he retired as head of Radiology and Hospital Computing at UC Davis to become Chief Medical Officer for Eklin Medical Systems. He is now serving as Chief Medical Officer for SoundEklin, a VCA Company.