Image analysis – Applications – Biomedical applications
Reexamination Certificate
1998-05-29
2001-06-19
Johns, Andrew W. (Department: 2621)
Image analysis
Applications
Biomedical applications
Reexamination Certificate
active
06249594
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to systems and methods for automatically inferring boundary contours of organs, tumors, prostheses, or other objects of medical interest from two-dimensional and three-dimensional images of the physical patient anatomy from computed tomography imaging, magnetic resonance imaging, or the like.
BACKGROUND OF THE INVENTION
Clinical Imperatives and Issues
Often in the course of clinical diagnosis, the patient's internal anatomy is imaged to determine the extent to which disease has progressed. The diseased tissue may be evidenced by some variance from normal anatomy or function. Several imaging modalities are commonly used to generate pictures (or images) of a patient's anatomy and function suitable for diagnostic and radiotherapy treatment purposes, or for surgical planning. These include conventional X-ray plane film radiography; computed tomography (“CT”) imaging, which also uses X-rays, magnetic resonance imaging (“MRI”), which produces images of internal anatomy and information about physiological function; and nuclear medicine imaging techniques, such as positron emission tomography (“PET”) and single photon emission computed tomography (“SPECT”), which produce images with combined anatomic and physiologic or biochemical information.
A common property shared by all the imaging modalities just mentioned is that the images are digital. That is, the images are represented as regular arrays of numerical values, which represents a physical measurement produced by a scanner. If these images are two-dimensional (“2-D”), the discrete picture elements are termed pixels. However, if the images are three-dimensional (“3-D”), the discrete volume elements are termed voxels. For 3-D imaging modalities, single slices or sections are composed of pixels, but those same picture elements are equivalently termed voxels when considering a set of stacked images as a volume of data.
The digital images from 2-D or 3-D imaging modalities are substantially exact maps of the pictured anatomy, so that each pixel value represents a sample of a property at a location in the scanner's, and, therefore, patient's, coordinate system. Thus, the distances between pixel/voxel centers are proportional and have meaning in the sense of real physical spacing in the patient anatomy. Moreover, the relative positioning of the numerical array represents the proportional spacing of objects in the patient anatomy.
The numeric value of each pixel represents a sample of a property at that location. In CT images, for example, the numbers are a measure of relative X-ray absorbing power, so that spaces inside the lungs are usually pictured as dark (low CT number) while bone is generally bright (high CT number).
CT imaging and MRI are two of the most frequently used imaging modalities because both provide detailed pictures of the internal anatomy of a patient. The instruments that employ these imaging techniques provide data that in appearance is 2-D or 3-D. However, the 3-D images, as stated, are a collection of 2-D samples, in this form of slices or sections, of the anatomy that have been combined to create a 3-D images. More specifically, to recreate the 3-D images from the 2-D image samples, the physician, scientist, or other skilled professional must recombine the 2-D image samples (slices or sections) of the anatomic elements (organs, tumors, surgically-implanted prostheses, etc.) A common way to recombine 2-D image samples to form 3-D images is to manually draw individual contours on a contiguous set of 2-D image slices or sections using computer graphics. Once these manually drawn contours are made, they are assembled to accurately construct 3-D representations of organs, tumors, and the like. The resulting 3-D reconstructions convey to the viewer the relative sizes, shapes, and mutual spatial relationships among the anatomic elements in the same anatomical scale as the original.
In the 2-D context of a slice or section, the individual anatomic elements may be represented by contours coinciding with each object's boundaries. Alternatively, in the 2-D context of a slice or section, anatomy elements may be represented by 2-D templates identical in size and shape to the object 2-D templates are patterns of pixels all having the same value which represent a single region in an image. A representation by 2-D region-templates or by 2-D edge-contours are equivalent, since either representation can be readily computed from the other.
As stated, 3-D reconstructions of patient anatomy are most often prepared using computer graphics by manually drawing the individual contours on a contiguous set 2-D image slices or sections and then combining them. This method is referred to as contouring. Contouring is very time-consuming and labor intensive. The time and labor necessary to use this method increases significantly with the number of image slices, and the number and sizes of the organs, tumors, etc. in the anatomical area of interest. The quality of the contouring and the later produced 3-D images, depend on the resolution and contrast of the 2-D images, and on the knowledge and judgment of the physician, scientist, or skilled professional performing the reconstruction.
Three-dimensional radiation therapy treatment planning (“RTTP”) is a medical procedure that currently makes the greatest use of 3-D reconstructions. This is even despite the labor and time required to contour the organs and tumors to generate a useful plan. In fact, the largest fraction of the plan preparation time involves contouring.
An example of a manually contoured CT image slice or section is shown in
FIG. 1
generally at
100
. In
FIG. 1
, the manually contoured organs are liver
102
, spleen
104
, left kidney
106
, right kidney
108
, and spinal cord
110
.
FIG. 2
, at
200
, shows an example of a 3-D reconstruction that uses as an element the 2-D slice or section shown in
FIG. 1
at
100
. The reconstruction in
FIG. 2
is composed of contours from a contiguous set of slices or sections. In
FIG. 2
, the 3-D reconstruction of the liver is at
202
, the spleen is at
204
, the right kidney is at
206
, the left kidney is at
208
, and the spinal cord is at
210
.
Another method that may be used for forming representations of organs, tumors, and the like is the segmentation method. Segmentation is the identification of image objects as distinct regions or segments of an image. This method also may be used to generate 3-D reconstructions of a patient's anatomy.
According to the segmentation method, a decision is made with regard to the image contents as to whether a given pixel of a 2-D slice or section belongs to a specific set of organs, tumors, lesions, or other objects known to exist in that slice or section. Therefore, given that both contouring and segmentation may equally be used to generate 3-D reconstructions, contouring and segmentation are taken to have the same meaning for description purposes herein.
Pixel/voxel values, which exist as intensities or gray levels, and their distributions across an image form a useful set of properties for segmentation. In a typical slice or section, the edges of objects that are shown usually are associated with large value differences with nearby pixel values. Further, the interiors of discrete objects tend to have relatively constant values. As such, discrete objects exhibit distinct gray level textures in such a manner that adjoining objects or regions with different textures appear to have visible boundaries between them. Each of these qualities of edgeness and texture are associated with one or more computational methods that may be used to generate a numerical value for that property. As quantified, these properties can be used to make decisions about the segment identity of individual pixels.
Prior Art Segmentation
A number of autosegmentation methods have been proposed in the prior art. These prior art methods may be separated into two principal types: (1) semi-automated segmentation methods in which physicians, technic
Computerized Medical Systems, Inc.
Hale and Dorr LLP
Johns Andrew W.
Nakhjavan Shervin
LandOfFree
Autosegmentation/autocontouring system and method does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Autosegmentation/autocontouring system and method, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Autosegmentation/autocontouring system and method will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2528098