British Journal of Radiology 74 (2001),69-72 © 2001 The British Institute of Radiology
A digital frame of reference for viewing digital images
D S Brettle, MSc, MIPEM
Department of Medical Physics & Engineering, Leeds Teaching Hospitals NHS Trust, Lincoln Wing, St James's University Hospital, Beckett Street, Leeds LS9 7TF, UK
 |
Abstract
|
|---|
Digital imaging is becoming widespread in diagnostic radiology. Most diagnostic digital images do not relate explicitly to the physical processes involved in their generation but are, in essence, a "pseudo" image generated from digital data using pre- and post-processing. Without knowledge of how the image was generated, there is a potential to misinterpret the image data. A new design of digitally generated graphic is presented that is intended to help maintain the frame of reference when viewing digitally processed images. The intention is that the digital frame of reference (DFOR) be included with all digitally processed images and be processed using the same factors as were used on the image. An unprocessed DFOR can then be displayed adjacent to the processed DFOR to re-introduce a frame of reference and to clearly illustrate the effect of any processes that have been applied to the image. This would allow the viewer to perceive any artefacts that may have been introduced into the image by the processing. This is particularly important where the image requires interpretation by the viewer, as in medical diagnosis. This paper presents a grey scale version of the DFOR that is suitable for applications such as medical imaging. The DFOR includes: grey scale from 0 to the maximum bit depth in 0%, 30%, 70% and 100% steps on a 50% background; the full frequency range from 0 to the Nyquist frequency; high, medium and low contrast boundaries; and linear/curvilinear features. The same method could be extended to any other digital image system and could be easily modified to include colour.
|
Introduction
|
|---|
Digital imaging systems allow the image acquisition and the display functions to be separated. This allows virtually unlimited post-processing of the image data. Therefore, when viewing a digital image there is a danger that the traditional frame of reference associated with film images will be used even when the image presented is in fact a "pseudo" image. For example, a traditional radiographic image has grey values ranging from black, representing totally irradiated areas, to white, representing areas of no irradiation. If, however, the image was inverted, and the viewer was unaware of the fact, they would still interpret the image using this frame of reference (or context), i.e. that black related to maximum radiation and white to minimum radiation. This is an extreme example, but illustrates that if an observer is not aware of the context of an image there is the potential for misinterpretation of that image. This is a particular concern in applications such as medical imaging were the interpretation or diagnosis is paramount.
In digital medical imaging, image processing is widely used to enhance image presentation. This can range from simple sharpening up to complex non-linear algorithms [1]. By definition, all image processing will introduce artefacts, some intentional, some not. An example of this is the commonly used "unsharp mask" algorithm [2] that may introduce a halo artefact at boundaries [3, 4]. If the viewer's frame of reference is not adjusted to allow for these artefacts, the artefacts may be wrongly interpreted as genuine clinical features. A misdiagnosis may become even more likely with increasing use of more complex algorithms where the artefacts are not as obvious.
It is therefore suggested that each digital image will require its own "reference point" to communicate to the clinician "what has happened to the image". There appear to be three options to achieve this; (1) always have the raw image presented; (2) state the algorithms used and train the staff to understand the potential artefacts; or (3) present each image along with graphical information that relates to the processes applied to that image.
The aim of this paper is to present a new graphical icon optimized for grey scale medical imaging. The icon is embedded in the digital image and processed in the same way as the image. This icon will graphically present information relating to the impact of any image processing on the original image data and can be readily compared with the unprocessed icon, thus re-introducing a frame of reference.
|
Method
|
|---|
The digital frame of reference (DFOR) was generated using commercial graphics packages provided by the Corel Graphics suite v8.0. (Corel Corporation, Canada). The presented DFOR includes the following elements:- Full grey scale range from 0 to maximum bit depth in 0%, 30%, 70% and 100% steps on a 50% background. This covers the full range of the grey scale and allows changes in the gamma function to be observed. The areas are also large enough to allow measurements of five points on the gamma curve, allowing quantitative assessment if required.
- Full frequency range from 0 to Nyquist frequency. The zero frequency is provided by the main bulk of the features. An ellipse intercepts the outer circle at four points. Where the ellipse intercepts the outer circle, a single point is generated that is, by definition, the Nyquist limiting frequency. Therefore, frequencies from 0 to the Nyquist limit are covered. This allows visualization of frequency-dependent effects.
- High, medium and low contrast boundaries. Contrast boundaries are areas where processing artefacts are often produced. Having grey scale boundaries ranging from 20% to 100% allows the magnitude of these effects to be visualized.
- Linear and curvilinear features. Processing effects are often directional. Having features that cover all directions should show any directional effects. Additionally, linear edges may overestimate the effect of processing on clinical images, where natural curvilinear features are often more common. Therefore, linear and curvilinear elements are included.
Suggested dimensions of the DFOR image are shown in Figure 1
. The circular detail composed of four quadrants is initially presented on a large area square background (XxX) to allow for later cropping to remove edge effects from the detail boundaries. This is not essential and is for aesthetic reasons only. All dimensions are related to this initial square, allowing easy reproduction and scaling. An ellipse is generated with a grey level of 50% and a ratio of 2:1. It is replicated four times and offset 0.0005X outward from the edge of the circle at positions 12, 3, 6 and 9 o'clock. The ellipse is rotated to the vertical for the 3 and 9 o'clock positions and to the horizontal for the 12 and 6 o'clock positions. It is intended that the value X would be approximately 20% of the image size. The final image can be cropped by a circle 0.4X in radius before or after processing. This is optional, but for clarity has been applied post processing to all the images presented.
View larger version (33K):
[in this window]
[in a new window]
|
Figure 1. Layout and dimensions for the digital frame of reference (DFOR). (a) The main features of the DFOR are the four quadrants of varying contrast covering the complete grey scale range. The ellipses intercept the outer diameter to produce a range of frequencies. They also introduce internal curvilinear features. (b) The cropping mask used to crop (a). This helps improve the aesthetic presentation of the DFOR by removing boundary processing artefacts.
|
|
|
Results
|
|---|
The DFOR has been processed with two simple algorithms to illustrate its sensitivity. The algorithms used include a simple sharpening filter comprised of a 3x3 convolution kernel and an unsharp mask filter using a blurring radius of 30. The results are shown in Figures 2a–c. It can be seen that presentation of the processed DFOR clearly illustrates the artefacts introduced by the processing. To indicate how the DFOR could be used clinically, an example of a chest radiograph is shown. Figure 3a shows a normal chest radiograph that, when processed, begins to exhibit the appearance of a chest with increased lung markings (Figure 3b), a potential sign of disease. The DFOR shown with the processed image shows the high level of applied processing.
View larger version (44K):
[in this window]
[in a new window]
|
Figure 2. Effects of simple image processing. (a) The unprocessed digital frame of reference (DFOR) for comparison with (b) and (c); (b) the DFOR after it has been sharpened with a medium level convolution kernel; (c) the DFOR after it has been processed with an unsharp mask filter. Radius=30.
|
|
View larger version (80K):
[in this window]
[in a new window]
|
Figure 3. Clinical example of use of the digital frame of reference (DFOR). (a) Digital chest radiograph before processing. (b) The chest radiograph has been processed with unsharp mask and sharpening algorithms. The image now has the appearance of increased lung markings. The DFOR is shown at 15% of the image size. (DFORs displaced for clarity.)
|
|
|
Conclusion and discussion
|
|---|
Digital imaging systems have many advantages over conventional film systems. One of these is the ability to process the images to enhance the presentation of certain features. This is routinely done on most digital acquisition systems to enhance image presentation. However, little is done to communicate to the viewer how the original data has been manipulated and in particular the artefacts that may have been introduced. The presented graphical icon has been designed to communicate to the viewer "what has happened to the image". The design is easy to generate digitally and presents signals that cover the full range of the system. It is intended that this icon would be imbedded into the image and processed with the same parameters as the image. This could then be presented side-by-side with an unprocessed DFOR, allowing the viewer to acquire a new frame of reference for the image.
This design can be used in any digital imaging modality and can be readily modified for specific applications, e.g. for colour and specific contrast/frequency ranges. A noise component has been excluded from this design for simplicity. An area of Gaussian noise could be added to the central area but it is unlikely that this would provide information that cannot be deduced from the existing features. The main advantages of this method of communicating the processes applied to an image over the other possible methods are that it is scalable, visual, unambiguous, easy to reproduce and takes up relatively little image space. This design is intended to be used in clinical images and, as such, its clinical application and validation need to be established. One method to achieve this could be to conduct a psychophysical experiment with unprocessed and processed images were the algorithm is known to introduce artefacts of a pseudo clinical nature, e.g. unsharp mask filtering. Multiple observers would then rank their confidence of diagnosis for a range of selected pathologies without the DFOR present. The experiment could then be repeated with the DFOR. The change in the confidence would give an indication of the impact the DFOR could have in clinical applications. Combining the DFOR with other technologies, such as image watermarking [5], would help to provide secure, easily interpretable and ultimately reliable images for all disciplines requiring a high level of image probity.
|
Footnotes
|
|---|
UK Patent applied for: application number GB0003157.5. 
Received for publication July 11, 2000.
Revision received October 18, 2000.
Accepted for publication October 20, 2000.
|
References
|
|---|
-
Sonka M, Hlavac V, Boyle R. Image processing, analysis and machine vision. London: International Thomson Computer Press, 1993.
-
Prokop M, Schaefer CM, Oestmann JW, Galanski M. Improved parameters for unsharp mask filtering of digital chest radiographs. Radiology 1993;187:521–6.[Abstract/Free Full Text]
-
Chang DC, Wu WR. Image contrast enhancement based on a histogram transformation of local standard deviation. IEEE Trans Med Imaging 1998;17:518–31.[Medline]
-
Prokop M, Schaefer-Prokop CM. Digital image processing. Eur Radiol 1997;7 (Suppl. 3):S73–82.
-
Zhao J. Watermarking by numbers. Nature 1996;384:514
Top
Abstract
Introduction
