Digital Imaging

(For a printable version without graphics, please click here)

Learning Objective - To better understand the advantages and limitations of digital imaging. To gain some visual experience with the interpretation of medical images which are recorded in digital form and to become more skillful in making judgements about image quality.

Digital imaging is the future of medical images. Regardless of the information input, if it can be recorded digitally that is a much more useful format than traditional (analog) ways of making a permanent record. Some of the advantages of recording the image in a digital format include:

1. The possibility of changing density relationships (brightness and contrast) without re-examining the patient.
2. The possibility of transmitting to a remote location without any loss of image quality.
3. Eliminate the need for chemical processing.
4. The ability to produce multiple "original" images without re-examining the patient.
5. More reliable permanent storage because the image can be reproduced and stored in several locations without chemical deterioration.
6. The potential for rapid computer interpretation.

Several radiographic techniques have been inherently digital since the imaging method was developed. These would include computed tomography and magnetic resonance imaging. Other methods are gradually being converted to digital approaches even though there is a long history of analog recording. Many hospital radiographic departments are converting conventional radiography to digital methods as rapidly as possible even though they have operated film/screen imaging systems for many years. This conversion is done by changing the signal receptor from film to a photon sensitive detector which can give a numerical output. Other methods such as ultrasound and nuclear medicine imaging are also relatively easy to convert to digital systems. To better understand the advantages and limitations of digital imaging, computed tomography will be used as the example for this discussion.

The basic idea is to have the image recorded as a numerical matrix in which the density variations are symbolized by numbers. In computed tomography, this information is obtained by rotating an x-ray tube around the patient as the residual x-ray beam exiting the body of the patient is recorded by a set of photon sensitive detectors. The images below show a typical, clinical computed tomographic machine in the image on the left. The center image shows diagrammatically how the patient's body is located between the x-ray source and the detectors. The image on the right shows the computed tomography machine with the gantry cover removed so that you can see how the detectors are positioned around the aperture through which the patient is moved.

The number of photons reaching the detectors gives information about the density of the part of the body through which the beam of "x" photons passed. If the beam passed through a relatively thick bone, the number of photons received by the detectors recording the exit beam would be relatively low while the number would be high if the beam passed primarily through air-bearing lung. The problem is that the number of photons recorded by the detectors indicates the total tissue density across the entire path through the body. It is not specific for a single small spot within the body. However, if you record enough of these paths through the body each taken from a different angular perspective, it is mathematically possible to reconstruct the density at any point within the imaged area.

The diagrams below show how this can be done with a simple, four (4) pixel (picture element) matrix using an iterative method.

The diagram on the left shows the four (4) element picture with no density numbers within the imaged area but the ray sums are recorded around the edge, in other words, when the detector was positioned at the right side of the upper row it recorded a number three (3) when it was at the right side of the lower row recorded a number seven (7) when it was at the bottom of the right column it recorded a six (6) and when it was at the bottom of the left column it recorded a four (4). From those numbers, the machine can quickly determine what the correct set of numbers would be inside the scanned object (in this simple illustration, it is possible to have more than one solution). As you can see, the machine would first take the row total (ray sum) and divide it by the number of pixels in the row and put that number inside of each pixel. For instance, in the upper row the sum number was three (3) so each of two pixels would have one half of that number or 1.5. Once the row numbers are distributed proportionally across the line of pixels, the machine will quickly look at the column totals to see if they are correct.

If we look at the left column 1.5 and 3.5 do not total 4 so an adjustment has to be made. Since the excess is 1 and the number of pixels is 2, 0.5 should be subtracted from each pixel. If we do that, the resulting total would be 4 which is correct. The upper pixel in the left column would now be 1 and the lower pixel would be 3. If you use the same approach to correct the incorrect numbers in the right column, those pixel values will be changed to 2 in the superior pixel of the right column and 4 in the inferior pixel of the right column which total 6 corresponding to the ray sum. I think it is easy to see how this method could be used to calculate the pixel values for a much larger matrix of numbers. Actually, modern computed tomographic equipment uses a 512 row by a 512 column matrix which results in calculating the correct numerical density for 262,144 individual pixels.

This would be a very elegant system if humans were good at interpreting the matrix of numbers. Unfortunately, we continue to train physicians to be more familiar with anatomic relationships so we either must build a machine to do the image interpretation or convert the matrix of numbers into something that is recognizable to physicians as anatomy.

The list of numbers shown in the image below is the information contained in a small portion of a single computed tomographic image. This actual numerical information can be obtained from any part of the image in which some quantitation of tissue density is needed. The image on the right below shows the entire matrix of numbers for a body cross-section. It is interesting to see how anatomic this numerical array can be using only the density variations of the printed numbers. For example, the number 8 would give a much higher density (blackness on paper) than the number 1, so if you print the entire 262,144 number matrix the density variation gives some indication of the anatomy but it is not predictable because the density of each number might not vary directly with the tissue density it represents. So, we must devise a better system for reconstructing tissue density from the numerical array.

Window

Although there is a more detailed description of the use of the gray tone "window" in the computed tomography film package, a simplified discussion will be given here. Just remember that the only part of the information collected from a computed tomographic study which is available to you visually, on the films, is the relatively small part that is included within the gray tone window (this is likely to be less than 10% of the total information of the CT image). If you only examine the anatomic films printed with the "window" selected by the Radiology Department, there will be a lot of information stored electronically which you have never seen.

Physiologically, humans only perceive about sixteen (16) levels of gray. In other words, if you were given a continuous black to white stripe and asked to divide it as many times as possible so that no gray tone could be again sub-divided into a recognizably different shade of gray, most people would end up with about sixteen (16) divisions of the black to white stripe. Given this limitation, the machine will allow you to determine which matrix number should be printed as white and which should be printed as black. The numerical range between the white and black levels establishes the CT window "width". The center of that numerical range becomes the window "level". The numerical range (window width) is divided by sixteen (16) to determine the numbers which are included in each individual gray tone. For instance, if we established a CT window which was sixteen hundred (1600) CT units in width, each gray tone would cover a range of one hundred (100) CT units. Any structure which was within that one hundred (100) CT unit density step would have the same gray tone on the picture.

I think you can see that this approach requires a lot of judgement about where to set the image window in order to obtain the most clinically useful information from any computed tomographic study. It is possible to set the window inappropriately and completely miss the important diagnostic information from a particular study. Intuitively, you might think it would be a good idea to open the window as wide as possible. Maybe even include the entire two thousand (2000) CT number range. That way, it might seem that you would not miss any tissue density variation and would always be able to make the correct diagnosis. Unfortunately, if the window is set too wide, each gray tone will include such a wide range of tissue density that the lesion is likely to be indistinguishable from the surrounding normal tissue. In other words, the liver metastases would end up having the same density as the surrounding liver and therefore would not be detectable.

Computed tomography is frequently thought to produce images with better anatomic definition than conventional radiographs. From the previous discussion, you can understand how the image can be very accurate in displaying tissue density variation. However, many people erroneously believe that computed tomography has good spatial resolution. Spatial resolution is usually defined by how closely two (2) lines can be positioned and still be detectable as separate lines. Computed tomography is not very good for this. The problem is relatively easy to understand if you consider distributing the numerical matrix over an area the size of the cross-section of a human body. For that size image you will find that the area of a pixel is approximately one square millimeter. Since there can be no gray tonal variation within an individual pixel, it would take a minimum two (2) pixels to display the presence of a density change and probably at least three (3) pixels to understand the density change in relationship to an anatomical structure. While even the most basic conventional radiograph could easily record forty eight (48) line pairs in that same 3-millimeter space. Just remember that computed tomography does not have anywhere near the spatial resolution capability of conventional radiography even though the computed tomogram seems to give a more anatomically understandable image.

Pixel size is quite important in defining anatomic structure by computed tomography. The first computed tomography machine that was commercially available in the United States (about 1972) used an 80 by 80 pixel matrix. A cross-sectional image of the brain from that machine is shown below on the left. A similar cross section of the brain done on a more modern 512 by 512 matrix machine is shown below on the right. From this comparison, you might wonder why computed tomography does not use a matrix with even smaller pixel size such as a 1024 by 1024 matrix. If you would like a short discussion of that issue, click here.

Discussion #1

80 by 80 pixel matrix	512 by 512 pixel matrix

The "window" which is selected determines what anatomy can be seen on any images printed on film. The images below show the extreme difference in the appearance of images made from the same numerical matrix (the same patient study) just by changing the window. The image on the left is a very high contrast window in which almost every structure in the picture is either black or white. This results from using an extremely narrow window which produces a great change in the film density for very little change in the tissue density. The image in the middle is produced using a "bone" window. This window is centered at a fairly high CT number. The image on the right gives some anatomic detail within the air-bearing lung so the center of that window must have been very low actually less than zero (0). Before finishing this lesson be sure that you understand the concept of the display window for computed tomography. This is an important issue in determining the quality of the clinical images which you will be studying. If you feel uncertain about the way the window is selected or what effect it will have on the image, come to the Radiology Department and review more examples of computed tomographic images with varied window settings.

The two images below again illustrate the change in image appearance caused by nothing more than a different window setting. The image on the left shows no anatomic detail in the lungs but there is good anatomic detail of the chest wall. The image on the right is set to show anatomy within the lungs but the chest wall and mediastium are completely white. At this stage, you should understand how the window was varied in order to produce these two images from the same matrix of numbers.

Patient Study

The next set of images shows a patient who suddenly and unexpectedly developed paraplegia. The first radiographic image obtained was an image of the thoracic spine which is recorded below. To appreciate the pathology on this image, you will have to look directly at the radiographic film on a viewbox. The density variations are so subtle that it will not be possible to get all of the important information from the computer screen or a paper print.

The composite CT images are shown on the left below and an enlarged image is displayed on the right. If you look at the spine anatomy shown on these films, it is easy to understand why the patient is paraplegic. There is considerable tumor involvement of the mid-thoracic spine with extensive bone destruction at several vertebral levels. Although this is very obvious on the computed tomogram, the same information was available on the conventional radiograph of the thoracic spine and I would encourage you to carefully study the thoracic spine film on a viewbox. Particularly, look at the elements of the spine which are posterior to the vertebral body. You will see destruction of the spinous processes and pedicles in the same pattern that is shown on the computed tomographic images. A careful interpretation of the conventional radiograph in this case would have made the computed tomography unnecessary because no additional information was gained from the computed tomogram even though those tomographic images were much easier to interpret.

Enlarged Image

Issues

1. In what situation would computed tomography be superior to magnetic resonance imaging for digital imaging tomography?

2. Why is the gray tone display window used for computed tomography and how is the optimal window determined?

3. What are the most severe limitations of computed tomography in "missing" small lesions?

4. What is the risk of computed tomography (approximately what is the radiation exposure and how much risk is created by the frequently used contrast material)?

5. What is the approximate cost of this examination?

1. Larson, EB. Promoting Effective Use of New Imaging Techniques. AJR 138:788 - 789.

Send us comments: Dr. David Adcock, DAVID@MED.SC.EDU.

This page was last updated 26 August 2002.
This page copyright 1999, The Board of Trustees of the University of South Carolina
URL: 6digimg.html
You are visitor 01 July 1999.

Please report any problems or inaccuracies to obibb@med.sc.edu