Image analysis – Image segmentation
Reexamination Certificate
2000-09-20
2002-12-24
Boudreau, Leo (Department: 2621)
Image analysis
Image segmentation
C382S133000, C382S280000, C382S288000
Reexamination Certificate
active
06498863
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates generally to analysis of digitized images of arrays.
2. Discussion of the Related Art
Array technology has dramatically expanded the ability of researchers to perform a variety of biotechnological operations. It has proven especially valuable in applications for DNA genotyping and gene expression. By supporting simultaneous examination of up to thousands of samples, the method enables researchers to gain a better overall picture than what may be obtained from a traditional “one gene in one experiment” approach. Array technology also reduces the time and the volume of materials needed for an experiment. Where previously a researcher might screen an array of unknowns with a defined probe, array technology facilitates a reverse approach. Defined DNA fragments can be affixed to known positions in an array and probed with an unknown sample.
Often, a distinction is made in array technology between microarrays and macroarrays. Objects on a microarray typically have diameters of less than 200 microns, while those on a macroarray are usually larger than 300 microns. Arrays may use glass slides, micromirrors, or semiconductor chips as substrate materials. The shape, alignment, and density of objects on an array may be controlled by a variety of methods. Robotic systems that precisely deposit small quantities of solution in an ordered configuration on the substrate material are common. Photolithographic techniques, similar to those used in the semiconductor electronics industry, are also used. Masks define activation sites where chemical coupling may occur. Repeating this process to build upon previously exposed sites allows for the formation of various gene sequences at different positions throughout the array. Alternatively, a solution of biomaterial may be spread across the substrate while specific sites on the chip are electrically activated to induce chemical bonding at those locations. Often a high density array is partitioned into sub-arrays to distinguish various aspects of an experiment, to separate different tests, or due to mechanical limitations in the means by which samples are applied to the substrate.
In a typical experiment, the array is exposed to a chemical sample to induce hybridization between array object and sample compounds. A variety of detection strategies may be used to locate the specific sites on the array where hybridization has occurred. These include use of fluorescent dyes, autoradiography, radiolabels, bioelectronic detection (observance of electron transfer reactions between samples and the substrate), and laser desorption mass spectrometry. Signals emitted from objects via detection media provide information about the reactions. Signals may be measured for absolute intensities or, in the case where different colored dyes are used to detect the degree of presence of different compounds, for ratios of intensities within specific frequencies.
Measurement and analysis of signals emitted from an array of objects usually involves examination of an image of the array. The labeled samples are excited and a detector system captures an image of the emitted energy. Accuracy of subsequent signal analysis is heavily dependent on the parameters of the detector system and its ability to reproduce faithfully the image of the array. Detector pixels need to be sufficiently small so that representation of each object includes enough pixels to support statistical analysis of the intensity of the object.
In addition to issues with image accuracy, signal measurement and analysis may be further complicated by the introduction of noise onto either the array or the image. Noise may arise from variations in sources used to excite labeled samples, fluorescence scattered by adjacent samples (“blooming” effect), dust and other impurities on the array, and other sources. Signal strength may be limited by the size of the objects, irregularities in preparing sites on the array, shortfalls of sample compounds deposited at certain locations, uneven application of detection media, and other causes. Fluctuations in the intensity of energy emitted by the background of the image can cloud the distinction between signal and noise objects and add to the difficulty of image analysis.
Image analysis is often aided by use of a grid overlay to assist the researcher in locating signal objects and identifying noise. However, alignment of the grid overlay often must be adjusted for the particular array, especially where a standard grid overlay must be modified to create sub-grids that correspond to underlying sub-arrays. When done manually, grid alignment can take days. Additionally, errors in human judgment in performing this process can substantially diminish the likelihood that other researchers will be able to reproduce the results of an experiment.
Digital image processing software has helped to mitigate some of the concerns with matching the grid overlay to the array. Orientation markers included on array substrates and captured in the image can be used to align grid overlays on the image. A typical algorithm to distinguish sub-grids detects signals from objects, determines the “center of mass” for each object, identifies corner coordinates for each sub-array, and aligns sub-grids accordingly.
Measurement of background intensity usually involves distinguishing background pixels from object pixels. Often it is assumed that those pixels nearest the edge of the image represent the background of the image. However, this is not always the case, particularly for high density arrays. Where an image consists of relatively low intensity objects against a high intensity background, a histogram of the image will likely be unimodal in character and provide useful information for distinguishing background pixels from object pixels. With an accurate characterization of background intensity, objects on the image may be identified. Often, objects are defined simply as those pixels with a value of intensity greater than a multiple of the standard deviation of the distribution of intensities for background pixels added to the mean value of this distribution.
Each of these methods for processing the digital image of an array has benefits and disadvantages. While the processes described in the current art have admirably employed statistical tools in the course of their analyses, they have not included other mathematical methods, such as evaluations of fourier transform spectrums, in their overall schemes. Also, current methods have taken a “background first” approach in which objects are defined in terms of “not being background”. On whole, development within the art has tended to focus on identifying and correcting individual impediments to accurate image analysis. What has been needed is a comprehensive scheme to integrate a variety approaches into a single optimal method that can be used with any array regardless of object size or density, corrects the orientation of the array on the image, flattens fluctuations in background intensity, removes noise, and adjusts its efficacy to variations in background and noise intensity.
SUMMARY OF THE INVENTION
The present invention provides a method, system, and product for analyzing a digitized image of an array to create an image of a grid overlay. It is designed to function on images of arrays of various sizes and densities and with different levels of noise intensity. In general, the invention locates the center of each object on the array, determines a standard shape and size for the objects, and creates a final image of the grid overlay. In a preferred embodiment, preliminary procedures normalize the image so that optimal results can be obtained, sub-grids are identified and objects are repositioned within their corresponding sub-grids, and noise is removed by filtering processes based on object size, intensity, and location. Unlike prior art approaches, the present invention takes an “objects first” approach by initially identifying objects on the image and then defining the backgr
Gaidoukevitch Yuri C.
Ryan Kevin W.
Sequera Dean E.
Boudreau Leo
Media Cybernetics Inc.
Sheral Ishrat
Sterne, Kessler, Goldstein&Fox P.L.L.C.
LandOfFree
Method, system, and product for analyzing a digitized image... 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, system, and product for analyzing a digitized image..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method, system, and product for analyzing a digitized image... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2994727