Image analysis – Applications – Biomedical applications
Reexamination Certificate
1999-03-26
2002-12-10
Bella, Matthew C. (Department: 2674)
Image analysis
Applications
Biomedical applications
C382S128000
Reexamination Certificate
active
06493460
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to image cytometry systems in general, and in particular to automated systems for detecting malignancy-associated changes in cell nuclei.
BACKGROUND OF THE INVENTION
The most common method of diagnosing cancer in patients is by obtaining a sample of the suspect tissue and examining it under a microscope for the presence of obviously malignant cells. While this process is relatively easy when the location of the suspect tissue is known, it is not so easy when there is no readily identifiable tumor or precancerous lesion. For example, to detect the presence of lung cancer from a sputum sample requires one or more relatively rare cancer cells to be present in the sample. Therefore patients having lung cancer may not be diagnosed properly if the sample does not accurately reflect the conditions of the lung.
Malignancy-associated changes (MACs) are subtle changes that are known to take place in the nuclei of apparently normal cells found near cancer tissue. In addition, MACs have been detected in tissue found near precancerous lesions. Because the cells exhibiting MACs are more numerous than the malignant cells, MACs offer an additional way of diagnosing the presence of cancer, especially in cases where no cancerous cells can be located.
Despite the ability of researchers to detect MACs in patients known to have cancer or a precancerous condition, MACs have not yet achieved wide acceptance as a screening tool to determine whether a patient has or will develop cancer. Traditionally, MACs have been detected by carefully selecting a cell sample from a location near a tumor or precancerous lesion and viewing the cells under relatively high magnification. However, it is believed that the malignancy-associated changes that take place in the cells are too subtle to be reliably detected by a human pathologist working with conventional microscopic equipment, especially when the pathologist does not know beforehand if the patient has cancer or not. For example, a malignancy-associated change may be indicated by the distribution of DNA within the nucleus coupled with slight variations in the shape of the nucleus edge. However, nuclei from normal cells may exhibit similar types of changes but not to the degree that would signify a MAC. Because human operators cannot easily quantify such subtle cell changes, it is difficult to determine which cells exhibit MACs. Furthermore, the changes which indicate a MAC may vary between different types of cancer, thereby increasing the difficulty of detecting them.
SUMMARY OF THE INVENTION
The present invention is a system for automatically detecting malignancy-associated changes in cell samples. The system includes a digital microscope having a CCD camera that is controlled by and interfaced with a computer system. Images captured by the digital microscope are stored in an image processing board and manipulated by the computer system to detect the presence of malignancy-associated changes (MACs). At the present state of the art, it is believed that any detection of MACs requires images to be captured at a high spatial resolution, a high photometric resolution, that all information coming from the nucleus is in focus, that all information belongs to the nucleus (rather than some background), and that there is an accurate and reproducible segmentation of the nucleus and nuclear material. Each of these steps is described in detail below.
To detect the malignancy-associated changes, a cell sample is obtained and stained to identify the nuclear material of the cells and is imaged by the microscope. The stain is stoichiometric and specific to DNA only. The computer system then analyzes the image to compute a histogram of all pixels comprising the image. First, an intensity threshold is set that divides the background pixels from those comprising the objects in the image. All pixels having an intensity value less than the threshold are identified as possible objects of interest while those having an intensity value greater than the threshold are identified as background and are ignored.
For each object located, the computer system calculates the area, shape and optical density of the object. Those objects that could not possibly be cell nuclei are ignored. Next, the image is decalibrated, i.e., corrected by subtracting an empty frame captured before the scanning of the slide from the current frame and adding back an offset value equal to the average background light level. This process corrects for any shading of the system, uneven illumination, and other imperfections of the image acquisition system. Following decalibration, the images of all remaining objects must be captured in a more precise focus. This is achieved by moving the microscope in the stage z-direction in multiple focal planes around the approximate frame focus. For each surviving object a contrast function (a texture feature) is calculated. The contrast function has a peak value at the exact focus of the object. Only the image at the highest contrast value is retained in the computer memory and any object which did not reach such a peak value is also discarded from further considerations.
Each remaining in-focus object on the image is further compensated for local absorbency of the materials surrounding the object. This is a local decalibration which is similar to that described for the frame decalibration described above, except that only a small subset of pixels having an area equal to the area of a square into which the object will fit is corrected using an equivalent square of the empty frame.
After all images are corrected with the local decalibration procedure, the edge of the object is calculated, i.e., the boundary which determines which pixels in the square belong to the object and which belong to the background. The edge determination is achieved by the edge-relocation algorithm. In this process, the edge of the original mask of the first contoured frame of each surviving object is dilated for several pixels inward and outward. For every pixel in this frame a gradient value is calculated, i.e., the sum and difference between all neighbor pixels touching the pixel in question. Then the lowest gradient value pixel is removed from the rim, subject to the condition that the rim is not ruptured. The process continues until such time as a single pixel rim remains. To ensure that the proper edge of an object is located, this edge may be again dilated as before, and the process repeated until such time as the new edge is identical to the previous edge. In this way the edge is calculated along the highest local gradient.
The computer system then calculates a set of feature values for each object. For some feature calculations the edge along the highest gradient value is corrected by either dilating the edge by one or more pixels or eroding the edge by one or more pixels. This is done such that each feature achieves a greater discriminating power between classes of objects and is thus object specific. These feature values are then analyzed by a classifier that uses the feature values to determine whether the object is an artifact or is a cell nucleus. If the object appears to be a cell nucleus, then the feature values are further analyzed by the classifier to determine whether the nucleus exhibits malignancy-associated changes. Based on the number of objects found in the sample that appear to have malignancy-associated changes and/or an overall malignancy-associated score, a determination can be made whether the patient from whom the cell sample was obtained is healthy or harbors a malignant growth.
REFERENCES:
patent: 4453266 (1984-06-01), Bacus
patent: 5016283 (1991-05-01), Bacus et al.
patent: 5099521 (1992-03-01), Kosaka
patent: 5889881 (1999-03-01), MacAulay et al.
patent: 6026174 (2000-02-01), Palcic et al.
Garner David M.
Harrison S. Alan
Jaggi Bruno W.
MacAulay Calum E.
Palcic Branko
Bella Matthew C.
MonoGen, Inc.
LandOfFree
Method and apparatus for automatically detecting... 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 and apparatus for automatically detecting..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for automatically detecting... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2924721