Image analysis – Applications – Biomedical applications
Reexamination Certificate
2000-09-27
2004-01-20
Mariam, Daniel G. (Department: 2621)
Image analysis
Applications
Biomedical applications
C382S305000, C382S173000
Reexamination Certificate
active
06681035
ABSTRACT:
This invention relates to an improved method of segmenting objects in an image. In particular, it relates to a method of active contours using a Viterbi search algorithm for unsupervised segmentation of cell nuclei.
BACKGROUND TO THE INVENTION
Various techniques exist for identifying and segmenting objects in an image. Typically an accuracy rate of greater than 90% is considered to be good. However, in biological systems, particularly screening situations, an accuracy rate approaching 100% is required.
One application in which a high level of screening confidence is required is the screening of pap smears to identify cervical cancer. Obviously, false positive and false negative indications cannot be tolerated. However, a high degree of accuracy in very large data sets has proven to be difficult to achieve. It has been established by Bengtsson [The measuring of cell features; Anal Quant Cytol; 9 (3) June 1987, 212-217] and others, that segmentation of cell nuclei is the key step in devising a computer-implemented method of unsupervised analysis.
The most common method of slide preparation for pap smear screening is the Papanicolaou process. Unfortunately, the stain used in the Papanicolaou slide preparation process does not produce a strong nucleus-cytoplasm contrast. Weak image gradients along the nuclear border and artefacts in the cytoplasm add to the difficulty of segmentation.
A segmentation success rate of 98.3% was reported on a data set of 4700 images by MacAuley [Development, implementation and evaluation of segmentation algorithms for the automatic classification of cervical cells; Ph.D Thesis; University of British Columbia; August 1989]. However, the images were obtained using a special sample preparation technique which included the use of Feulgin-Thionin and Orange(II) stains. These stains enhance the contrast of the cytoplasm to the background and the nucleus to cytoplasm. The improved contrast compared to the most common stain greatly simplifies the segmentation process. Even so, an error rate of 1.7% will be unacceptable in many cases.
Previous studies on segmentation of cervical cell nuclei from the cell cytoplasm have used global thresholding techniques, edge detection and post-processing without achieving acceptable results. An example that is typical of the prior art can be found in U.S. Pat. No. 4523278. This patent describes a method and system for automatic detection of cells and determination of cell features from cytological smear preparations. The method only uses cell nuclei for evaluation and classifies the nuclei according to a feature set consisting of topological parameters of the boundaries of the nuclei.
Another, more recent technique is active contours, such as the approach used by Gunn and Nixon [A robust snake implementation: a dual active contour; IEEE Trans Pattern Anal Mach Intell 19(1) (1997) 63-68]. Their approach is to use inner and outer contours which are required to lie within and outside the object of interest. A driving force pushes the contours towards each other which enables each contour to overcome local minima in an image. The process is halted when both contours meet. Although the Gunn and Nixon approach simplifies the initialisation, it still requires the setting of three parameters and cannot guarantee a globally minimum energy contour within the space bounded by the initial inner and outer contours. The method is not suitable where very high confidence levels are desired.
OBJECT OF THE INVENTION
It is an object of the invention to provide an improved method of image segmentation.
It is a further object of the invention to provide an improved method of unsupervised cell nucleus segmentation so that selected cell nuclei may be chosen for subsequent determination of nuclei characteristics. Other objects will be evident from the following discussion.
SUMMARY OF THE INVENTION
In one form, although it need not be the only, or indeed the broadest, form the invention resides in an improved method of cell nuclei segmentation including the steps of:
(i) locating cell nuclei in a sample;
(ii) initialising the segmentation method by determining a contour for a selected cell nucleus;
(iii) conducting a first segmentation step employing a Viterbi algorithm with a first predetermined value of &lgr; to obtain a first nucleus boundary;
(iv) conducting a second segmentation step employing a Viterbi algorithm with a second predetermined value of &lgr; to obtain a second nucleus boundary;
(v) comparing the first and second nucleus boundary; and
(vi) storing an image of the cell nucleus if the first boundary and the second boundary are substantially the same within a predetermined limit.
Whilst any suitable value of &lgr; may be used in the Viterbi algorithm for conducting the first and second segmentation steps, it is preferred that one value be low and the other value be high. Preferably the low value is about 0.1 and the high value is about 0.7.
The cell nucleus locating step preferably includes taking a low resolution image of a sample, for example cells on a pap smear slide. The scene in the low resolution image may be segmented to identify cells. This may be done with any suitable algorithm, such as a water immersion algorithm. A high resolution image may then be taken of each cell for further processing.
The initialisation step may include locating an approximate centre of a nucleus for each high resolution cell image, for example by using a converging squares algorithm, then setting up a search space as an initial step towards an approximate determination of the contour of a nucleus in each high resolution image.
In a further form the invention resides in an improved method of cell nucleus segmentation from an image containing a cell nucleus including the steps of:
(i) constructing a search space in the image;
(ii) searching the search space with a Viterbi algorithm using a first predetermined value of &lgr; to find a boundary of the nucleus;
(iii) determining a centroid within the boundary;
(iv) constructing an updated search space in the image centred on the centroid;
(v) searching the updated search space with the Viterbi algorithm using the first predetermined value of &lgr; to find an updated boundary of the nucleus;
(vi) comparing the update boundary with the boundary;
(viii) repeating steps (iii) to (vi) until the boundary and the update boundary are substantially the same; and
(ix) storing an image of the cell nucleus and the found boundary when the update boundary is substantially the same as the boundary.
Preferably a high value of &lgr; is used for step (ii) and (v). A suitable value of &lgr; is in the range 0.7 to 0.8.
To a first approximation, the step of comparing the update boundary to the boundary can be replaced by comparing the centroid with an updated centroid.
REFERENCES:
patent: 4175860 (1979-11-01), Bacus
patent: 4523278 (1985-06-01), Reinhardt et al.
patent: 5745601 (1998-04-01), Lee et al.
patent: 5986771 (1999-11-01), Henderson et al.
patent: 5987158 (1999-11-01), Meyer et al.
patent: 6581011 (2003-06-01), Johnson et al.
patent: WO93/18483 (1993-09-01), None
Bamford et al., “Improving the Robustness Cell Nucleus Segmentation,” Sep. 14-17, 1998, Proc. 9th British Machine Vision Conf. pp. 518-524.*
Bamford et al., “A Two-Stage Scene Segmentation Scheme for the Automatic Collection of Cervical Cell Images,” 1997, IEEE TENCON, pp. 683-686.*
Bamford et al., “Bayesian Analysis of Cell Nucleus Segmentation by a Viterbi Search Based Active Contour,” Aug. 16-20, 1998, IEEE, vol. 1, pp. 133-135.*
Lee et al., “A Fast and Robust Approach for Automated Segmentation of Breast Cancel Nuclei,” Oct. 25-27, 1999, International Conference on Computer Graphics and Imaging, pp. 1-9.*
Bamford et al., “A Methodology for Quality Control in Cell Nucleus Segmentation,” Dec. 7-8, 1999, Digital Image Computing: Technique Application, pp. 1-5.*
Schupp et al., “Mathematical Morphology and Active Contours for Object Extraction and Localization in Medical Images,” pp. 317-321.*
Nasrabadi et al., “A Stereo Vis
Bamford Pascal
Lovell Brian
CSSIP (Cooperative Research Centre for Sensor Signal and Informa
Mariam Daniel G.
Oppenheimer Wolff & Donnelly LLP
Patel Shefali
LandOfFree
Method of unsupervised cell nuclei segmentation 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 unsupervised cell nuclei segmentation, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of unsupervised cell nuclei segmentation will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3197247