Image analysis – Applications – Biomedical applications
Reexamination Certificate
1998-06-02
2002-03-05
Ahmed, Samir (Department: 2026)
Image analysis
Applications
Biomedical applications
C382S168000
Reexamination Certificate
active
06353674
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to digital image processing. More specifically, the invention relates to a method for segmenting a radiation image such as an x-ray image into diagnostic areas (or body parts) and direct exposure areas (background area).
BACKGROUND OF THE INVENTION
In radiation imaging systems such as computed radiography systems, it is now common practice to apply signal processing to the digital signal representation of an image so as to generate an enhanced image. In this way the radiologist is aided in better perceiving even the subtlest diagnostic details in the image.
This image enhancement is for example achieved by a multiscale contrast enhancement, suitable window/level setting and sensitometric mapping etc.
A great number of image enhancing processing methods that are applicable to radiation images have been described in the state of the art.
For example in European patent application 546 600 optimal window/level setting is described. In this patent application the automatic determination of two characteristic histogram points has been described. The lower and the upper boundary of the diagnostically useful signal range, are determined resulting in an appropriate window/level setting.
In European patent application 549 009 a method has been described for deriving a sensitometric mapping from these characteristic histogram points.
In European patent application 527 525 a method known as Multiscale Image Enhancement and Amplification (MUSICA) is disclosed to enhance details independent of their scale or extent so as to optimally map the input dynamic range on the available output dynamic range.
It has been recognized that the effectiveness of these techniques primarily depends on the optimal selection of parameters that are used. These parameters are not fixed, but are selected in function of the image content and the examination type. The techniques for determining the parameters commonly rely on a computation of the histogram of the image.
However, the histogram is often based on data originating from the image in its entirety. More specifically, the histogram takes into account data originating from foreground or collimated areas (even from different images in the case of split screen images), as well as data from direct exposure area (also called background area) in addition to data from diagnostically relevant area.
These background and foreground area may influence the selection of the parameters, which in its turn may give rise to undesired effects.
Techniques have been developed for delineating the exposed area in case of collimated images.
For example, in European patent application 610 605 a technique has been described that that is suitable for the purpose of delineating the exposed area in so-called collimated radiographic images.
In European patent application 742 536 this technique is further elaborated to cope with a multiply exposed image (split screen images), an imaging process known as partitioning.
However, the exposed area delineated by the above methods may still comprise direct exposure area in addition to the diagnostically relevant area. For certain examination types, these direct exposure area represent a relatively large fraction of the total exposed area. The presence of direct exposure areas obstructs the image processing in general and specifically the contrast enhancement.
It is thus required to identify the direct exposure area in a radiation image.
In U. S. Pat. No. 5,268,967 and corresponding EP 0 576 961 a method for automatic foreground and background detection in digital radiographic images is disclosed.
The method involves an edge detection step, a blockwise analysis of a tiled image by calculating edge strength and variance in each block to classify it as either homogeneous, non-homogeneous or mixed type, and a block refinement applied to blocks of mixed type. The method however does not produce a pixel precise delineation of the foreground, since the result of the block refinement may still be coarse and non-straight for collimation regions with straight edges. The method appears to be time consuming since it needs to operate on the full scale image.
U. S. Pat. No. 5,426,684 extends the idea of calculating statistics within each block by also considering a plurality of texture features. Thus, in addition to the one-dimensional feature such as gray level and variance, features derived from a two-dimensional co-occurrence matrix are computed. Each sample block is then classified with a previously trained neural network classifier to determine its class. Pixel values belonging to the same class are accumulated to form separate histograms for each class. Each of the histograms are then used to optimize tone scale reproduction. It is commonly known in texture analysis that the blocks need be sufficiently large to compute a reliable two-dimensional statistic. 39×39 pixel subregions are mentioned to this purpose. Hence, the problem of pixel precise delineation, especially for straight border foreground regions, remains unalleviated.
In U. S. Pat. No. 5,046,118 a tonescale transformation function is determined based on the histogram entropy of the input image. The method computes the background threshold from the histogram by (1) assuming the histogram is bimodal and (2) determining the gray level that separates the two modes by maximizing the a posteriori entropy of the image. However, many images show a multimodal histogram, in which the additional modes appear and either belong to the diagnostic area, due to the presence of substantially large tissue and bone portions in the body region, or belong to the background, which is split by the body region, each part having received a different scatter fraction (mostly extremities examinations). Less frequently, both the diagnostic area and the background show multiple modes.
In U. S. Pat. No. 5,124,913 a rule based technique automatically determines the final scan gain in a storage phosphor radiography system. The rules are based on the peaks detected by a method based on the cumulative histogram.
In U. S. Pat. No. 5,164,993, combined use is made of the histogram, the cumulative histogram and the entropy of subsections of the histogram to create the fine tonescale transformation. Using these three functions, the histogram can be divided into a region of interest, a low-signal foreground region and a high-signal background region. To delineate these histogram intervals, the positions of a so-called left-point (border between foreground and region of interest), right point (border between region of interest and remaining anatomy) and background point (border between remaining anatomy and true direct x-ray background) are determined. However, the method assumes that each of the intervals contains one mode only, which is not always true.
Therefore a method coping with multiple histogram modes is required.
OBJECT OF THE INVENTION
It is an object of the invention to provide a method for segmenting a radiation image into direct exposure areas and diagnostically relevant areas, that does not show the drawbacks of the prior art.
Further objects will become apparent from the description hereafter.
SUMMARY OF THE INVENTION
The objects of the present invention are achieved by a method of segmenting a radiation image represented by a digital signal representation into direct exposure area and diagnostically relevant area, comprising the steps of
calculating a histogram of said radiation image,
calculating local centroids of said histogram,
constructing a set of archetype histograms with associated location and strength of its local centroids and with an associated threshold value,
selecting from said set an archetype histogram that corresponds with said calculated histogram on the basis of the correspondence of location and strength of the calculated local centroids with location and strength of the local centroids associated with the archetype histogram,
selecting the threshold value associated with a selected archetype histogram,
applying said threshold value to the
Agfa-Gevaert
Ahmed Samir
Hoffman, Warnick & D'Alessandro LLC
Merecki John A.
LandOfFree
Method of segmenting a radiation image into direct exposure... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of segmenting a radiation image into direct exposure..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of segmenting a radiation image into direct exposure... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2821917