Image analysis – Image segmentation
Reexamination Certificate
2003-02-04
2004-04-13
Patel, Jayanti K. (Department: 2623)
Image analysis
Image segmentation
C382S199000
Reexamination Certificate
active
06721450
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates generally to the field of imaging, and image processing and analysis techniques, such as those used in medical diagnostics and other fields. More particularly, the invention relates to a technique for rapidly segmenting and identifying features of interest in discrete pixel images and processing data extracted from the images following identification of the features.
BACKGROUND OF THE INVENTION
A great number of applications have arisen in recent years employing discrete pixel images of various types. In general, such images are comprised of an array of picture elements organized in rows and columns of an image matrix. Each element in the matrix may be characterized by various parameters, such as intensity, color, shade, and so forth. These characteristics are most often associated with or defined by information expressed in digital terms and stored in data files defining the image. By organizing and displaying the matrix of picture elements with reference to the parameters, a recognizable and useful reconstructed image may be obtained. The resulting image is typically displayed on a computer monitor or printed on a tangible medium such as paper or photographic film.
A number of modalities are known for acquiring discrete pixel images, particularly in the medical diagnostics field. For example, magnetic resonance imaging (MRI) systems have been developed which permit high-quality images of anatomies of interest in a subject. Such systems make use of controllable magnetic fields which affect precession of gyromagnetic material within the subject. By exciting the gyromagnetic material in the presence of specific fields, the materials can be made to emit signals which are captured and processed. The signals can be transformed to identify the particular intensity of the emissions at specific locations within the subject. By further processing the digitized signals corresponding to these intensities, image data is obtained which can be analyzed, stored, transmitted, and further processed to create a reconstructed image. Other modalities in the medical diagnostics field include computed tomography (CT) systems, digital x-ray systems, positron emission tomography (PET) systems, ultrasound systems, and so forth. In modem systems of all of these types, data sets may be obtained directly, or indirectly, which define features of interest within a subject patient for image reconstruction and processing.
Many applications exist for analysis of identifiable structures within discrete pixel images. In conventional images, the features are typically recognizable by contrasts, outlines, expected shapes, and other somewhat intuitive boundaries or characteristics recognizable to the experienced viewer. Depending upon the image quality, a qualified viewer may identify these features and even take measurements of the structures. However, conventional techniques for manually doing so are quite cumbersome, time-consuming, and imprecise. In the medical diagnostics field, for example, measurements of internal anatomies such as structures of the heart, brain, and other organs and tissues, have conventionally been made by hand or through software requiring substantial user interaction. Due to the drawbacks of these techniques, however, attempts have been made to automate the process.
Techniques which have been advanced for automating feature identification and measurement are not entirely satisfactory. A key aspect of this is the “segmentation problem” which refers to the identification of homogeneous regions in a given image, defined by a set of edges or boundaries. Processes have been developed that use mathematical constructs to deform candidate boundaries to correspond to the limits of a feature of interest. Such deformable boundaries are sometimes referred to as “snakes.” However, the methods employed for segmentation based upon these techniques are particularly susceptible to error due to noise present in the image data. Such noise, which may be visible as bright or dark spots or regions on the reconstructed image, may cause the candidate boundary to stop its expansion or contraction before reaching the limits of the feature, or may cause other anomalies in the result. While such noise may be eliminated or reduced by smoothing and other filtering techniques, these also tend to reduce the resolution of the feature of interest, thereby reducing the likelihood that the snake will converge accurately on the feature.
Another serious drawback of heretofore known techniques for automatic segmentation via snakes is the extremely demanding computational requirements involved. In particular, known techniques adopt algorithms which require sequential calculation of a very large number of parameters as the boundary expands or contracts towards the feature of interest. As a result, to be carried out in a reasonable amount of time, very sophisticated and powerful computers are required, particularly for noisy images, and those including a large number of picture elements.
Further drawbacks in existing techniques for automatic segmentation are rooted in the particular algorithms used to generate the candidate boundary and to deform the boundary toward the feature of interest. In particular, the algorithms typically include mechanisms for converging and stopping the evolution of the candidate boundary. Depending upon the techniques employed, the boundary may not fully conform to the feature of interest, particularly where the feature includes concavities and similar contours. Also, where adjacent features are present, the boundaries may not recognize distinctions between the adjacent features and, again, fail to conform accurately to the feature of interest. In either case, the resulting analysis is unreliable due to the failure to accurately recognize the boundary (and hence the surface area, volume or other parameter) of the feature of interest.
There is a need, therefore, for an improved technique for performing segmentation on discrete pixel images. There is, at present, an increasing need for a technique which will permit such segmentation on a series of images, such as images acquired over a three-dimensional space or over time, or both. Such techniques would be particularly useful in analyzing moving tissues, such as those of the heart. The technique would also find useful applications in still structures such as the brain, and so forth.
SUMMARY OF THE INVENTION
The present invention provides a technique for segmenting features of interest in discrete pixel images designed to respond to these needs. The technique makes use of a computationally efficient algorithm for initiating and expanding or contracting a candidate boundary for the feature. In accordance with certain aspects of the technique, a candidate boundary or snake is initialized and evolved by reference to mean curvature-weighted normals. Gradients of intensities within the image may be calculated before this evolution to greatly simplify the computational demands during the evolution. The candidate boundary thus converges extremely rapidly toward the feature and stops upon reaching the feature. The technique is particularly robust and tolerant of noise within the image. However, a diffused or smooth gradients may be used during the evolution to improve the tolerance to noise.
The technique may be used to segment both static and dynamic (or changing) structures. In both cases, the approach may be applied to a series of data sets representing individual images or slices of a subject. Over space, the technique may thus be used to reconstruct a three-dimensional representation of the feature. Over time, the technique may be used to analyze changes in the feature.
In a presently preferred implementation, the segmentation technique is used to measure and analyze features of the heart, such as the endocardium and the myocardium. By time analysis of this information, important characteristics of the heart may be evaluated, such as the ejection fraction. In other structures, both two-dimensional and
Tannenbaum Allen Robert
Yezzi, Jr. Anthony Joseph
Fletcher Yoder
Patel Jayanti K.
LandOfFree
Curvature based system for the segmentation and analysis of... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Curvature based system for the segmentation and analysis of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Curvature based system for the segmentation and analysis of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3259721