Image analysis – Applications – Biomedical applications
Reexamination Certificate
2000-11-17
2004-08-10
Boudreau, Leo (Department: 2621)
Image analysis
Applications
Biomedical applications
C382S173000
Reexamination Certificate
active
06775399
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to image processing systems generally. More specifically, the present invention relates to improved digital image processing through the removal of non-region of interest information.
BACKGROUND OF THE INVENTION
Computed Radiography (CR) has gained worldwide acceptance in radiology departments. CR not only provides digital radiographic image data for easy communication and storage, but also produces images with a wide dynamic range that is suitable for various exposure and diagnostic conditions. For a specific exam, however, a CR image needs to be processed digitally to show anatomical information and enhanced diagnostic details. Besides the anatomical portion of a CR image under exam, the CR image often contains collimators of different geometry and thickness such as patient monitoring devices, clothing, and/or image markers. Because the gray-level distributions of these non-anatomical objects often overlap with the gray-level distribution of the anatomy, an automatic image processing scheme may not produce the desired image quality. In addition, a CR image may also include multiple exams or views that are separated by collimators and each exam may have different gray-level distributions.
CR uses photostimulable phosphor imaging plates (IP) and associated hardware and software to acquire and display x-ray projection images. A CR system provides a significant advantage over conventional screen film system in terms of the exposure latitude (about 10,000:1) or wide dynamic range. However, the image contrast produced by a CR system is low and digital image processing is required to extract the diagnostic information from the image data by enhancing image contrast. Since the acquisition and display are separate processes in a CR system, different image processing techniques can be applied to correct for under- or over-exposures and to enhance image quality.
If an IP contains only anatomy, referred to as diagnostic regions of interest (ROI), standard image processing may be applied to produce desired image quality in an automated fashion. However, in computed radiography imaging, collimation is frequently employed to shield irrelevant body parts (i.e., not of interest) from radiation exposure as well as to present radiation scattering from x-ray opaque materials. Collimators are also applied to partition an IP plate into different views so that multiple exams can be exposed on the same IP plate. A view refers to a region on an IP plate that is not covered by the collimators. If an image contains only one exam, the view is the entire uncollimated region. If an IP plate contains more than one exam, the views are the regions that are partitioned by the collimators but not covered by any collimator. Besides collimated regions, a CR image may also contain direct exposure (DE) region, which is a region that has been directly exposed to the x-ray source without attenuation by, for example, collimators, anatomy, or markers, hardware devices, and so on. Therefore, a CR image may contain one or more collimated regions, one or more DE regions, and one or more ROIs.
In an ideal condition, each collimated region contains high intensity pixels with a uniform distribution and the DE regions contain low intensity pixels with a uniform distribution. Therefore, the collimated and the DE regions could be easily identified.
In a clinical setup, however, the ideal imaging setting is often not achievable and the three physically distinct regions (collimated, ROI and DE) can overlap in their distributions.
FIG. 1
shows a typical prior art single view CR image
100
having collimated regions
112
, ROI
114
, and DE regions
116
. Some pixels in collimated region
112
have lower intensity (i.e., are darker) than some pixels in the ROI
114
due to use of relatively thin collimators. Furthermore, some pixels in the DE region
116
may have higher intensity (i.e., are lighter) than the pixels in the ROI
114
due to the presence of objects
124
captured within the view, such as hardware devices for patient monitoring, cloth, air pockets, markers, hardware, and/or radiation scattering. Additionally, when a CR image has a view which is significantly skewed, processing of the ROI is extremely difficult. To compound matters, collimated and DE regions provide no useful diagnostic information and make it difficult to produce a resulting high quality, post processing ROI image, even with the use of sophisticated image processing algorithms.
In the case of an IP with multiple views, such as the two human foot views
202
,
204
of CR image
200
of prior art
FIG. 2
, a CR image will typically contain more collimated regions
206
, DE regions
208
and ROIs
210
than a single view CR image. That is, each view will contain ROI
210
and DE regions
208
bounded by collimated regions
206
. Because the exposure conditions and body parts for the views may vary from one exam (i.e., view) to another, the image enhancement of the CR image containing multiple views is complex. In such cases, all views of different exams need to be identified so that special image processing algorithms can be applied to each view to achieve reasonably good image quality. This processing can require human intervention and take a relatively long amount of time (e.g., several minutes or more).
SUMMARY OF THE INVENTION
The present invention is a region of interest (ROI) segmentation system and method that facilitate the isolation of ROI from other data within a digital image. The digital image may be any known type of digital image, such as a computed radiography (CR), digital radiology (DR), digital fluoroscopy (DF), nuclear medicine (NM), computer topography (CT), ultrasound, magnetic resonance (MR), or some other form of digital image. The ROI segmentation system accepts as an input a digital image that contains one or more views, wherein each view within a digital image corresponds to a different exposure. An input digital image of about 8 megabytes (and about 2000×2000 pixels) is typically processed in about 3-5 seconds, while larger images may take longer. Preferably, the input digital image includes at least about 10 percent of ROI. As an output, the ROI segmentation system produces a mask that allows generation of an image substantially void of all ancillary (i.e., non-ROI) information from each view of the original input digital image, wherein direct exposure (DE) regions such as markers and hardware devices and any collimated regions are clearly distinguished.
The ROI segmentation system includes a collimation subsystem configured to detect and remove collimated regions from an input digital image using, for the most part, boundary recognition algorithms. A collimation pre-processor module quickly detects sharp edges of collimated regions and well-defined hardware and marker images. The input image is sub-sampled using bilinear interpolation to produce a sampled digital image of about 2 MB and 1000×1000 pixels.
For edges that are less well-defined, a primary processor module accomplishes a more rigorous analysis. The collimation subsystem primary processor divides the sub-sampled digital image and averages a number of rows and columns to produce 1-dimensional averaged data. An array of the accumulated edge strength data is computed from each of the averaged row or column data. The primary processor processes each of the averaged row and column data and the edge strength data top-to-bottom and bottom-to-top and right-to-left and left-to-right to, ultimately, generate change in pixel intensity information. Using this information, the collimation subsystem primary processor determines most remaining collimated regions.
If collimated regions remain, a secondary processor which is configured to perform even greater analysis using a Hough transform-based process is implemented. Such a process may be necessary, for example, when a collimated edge is at an angle greater than 7 degrees with respect to its corresponding IP plate boundary or when the intensity distribution o
Analogic Corporation
Boudreau Leo
McDermott, Will & Emery LL
Miller Ryan J.
LandOfFree
ROI segmentation image processing system does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with ROI segmentation image processing system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and ROI segmentation image processing system will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3326980