Image analysis – Applications – Biomedical applications
Reexamination Certificate
2000-02-04
2003-10-07
Patel, Jayanti K. (Department: 2721)
Image analysis
Applications
Biomedical applications
C128S922000, C600S454000
Reexamination Certificate
active
06631204
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to medical diagnostic methods and classification of medical images for this purpose, and more particularly, to a similarity measurement method employing an algorithm for the classification of medical images into predetermined categories.
BACKGROUND OF THE INVENTION
Numerous systems of grading of different pathologies have existed to guide medical personnel in diagnoses. Physicians and especially dermatologists have used the ABCD checklist for detecting cutaneous melanoma. The Gleason method of histologic grading is used for evaluating prostatic cancer. Many other pathological changes in tissues such as in breast, colon, uterus and others could benefit from a screening method for diagnosing borderline malignancies. Current advances in imaging processes have brought the computer more and more commonly into medical offices. Now imaging can be useful for diagnosis of skin irregularities, for examining tissue samples and for screening for many other medical conditions.
There are many advanced methods used today that attempt to classify sets of images into predefined groups. These methods are usually based on the following:
1) Selecting a set of features (area, perimeter, diameter, etc.).
2) Extracting those features from the given sets of images.
3) Assigning weights to the individual features which would result in the ability to classify the images with a degree of exactitude similar to that achieved by an expert's mind
These methods, however, have many shortcomings. For example, there is uncertainty regarding the selection of the best features actually needed to do the classification. Also, despite the performance of an impressive succession of iterative algorithms (such as back-propagation, neural-networks or others), there remains a lack of true understanding as to why the different weights were assigned a particular value in relation to the classification task. These methods seek to replace the complicated classification process of an expert so as to deliver results that are equivalent to the outcome of the thinking process. However, medical phenomena do not always appear to fall neatly into groupings. A physician who sees many cases of a certain type of pathology will learn to adjust the definition of a grouping based on the variations that he has seen.
Thus, it would be desirable to provide a similarity measurement method for the classification of medical images into predetermined categories that would aid the physician in formulating a diagnosis.
SUMMARY OF THE INVENTION
Accordingly, it is a principal object of the present invention to provide a similarity measurement method for the classification of medical images into predetermined categories.
The similarity method of the present invention takes a different approach than the prior art approaches. It accepts that in all methods the experts have the final say, and lets them create sets of classified groups. When this is done, an efficient method is found to measure similarity between the image that is in need of classification and each of the given sets.
The method first determines which features most effectively describe each image set. The standard linear method for extracting such information about a set of images is known as the Karhunen-Loe've Transform (KLT). This transform uses the eigenvectors of the covariance matrix of the set of images, i.e. it uses the principal components of the distribution of images. The eigenvectors describe the variation from the prototype (average) image of the set. These eigenvectors together characterize all of the variations between images of the set and the prototype. Each new image can now be projected to every one of the sets using their eigenvectors, and the distance from each set's prototype can indicate the amount of similarity to the set. The method uses the smallest distance to classify the image and its value to indicate the quality of that classification.
The philosophy behind the method of the present invention is as follows. One skilled in the art can perform the initial task of classifying a broad set of images into libraries of groups. Once this is accomplished, special expertise is not needed, as similarity RMS_ERROR methods can be used to continue the classifications. With time and cooperation, these libraries can expand (assuming that an agreement can be reached on acceptability criteria such as RMS_ERROR value) thus enhancing the classification potential. In the case of Gleason grading of prostatic cancer, for example, a slide could be divided into a number of areas, each classified using the method, and color-coded accordingly. This would result in a clear graphic presentation of the overall Gleason grade.
Using the method of the present invention, digital image libraries can be built per each histopathological classification. Then each screened image will be converted to a digital media and a computer will measure the similarity distance to each pre-classified set in the library. The shortest distance, assuming it's acceptably small, will provide the diagnostics.
The time required to grade a microscopic slide by a pathologist varies, depending on his experience, between 5 and 20 min. The computer implementation time of the method of the present invention is negligible in comparison. The rate determining step of the overall time performance will be the moving of the microscope stage during scanning. Thus the method of the present invention will reduce the time needed for reviewing large numbers of slides.
It will be apparent to those skilled in the art that the method of the present invention may be applied to a variety of medical images including tissue samples, CT scans, PET scans, osteoporosis screening, thallium imaging, various cardiological tests, and surgical applications, by way of example.
REFERENCES:
patent: 5303381 (1994-04-01), Yagasaki
patent: 5710833 (1998-01-01), Moghaddam et al.
patent: 5790702 (1998-08-01), Yoshimura
patent: 5987456 (1999-11-01), Ravela et al.
patent: 6064749 (2000-05-01), Hirota et al.
patent: 6144755 (2000-11-01), Niyogi et al.
patent: 6404925 (2002-06-01), Foote et al.
patent: 6418234 (2002-07-01), Whited
patent: 6430306 (2002-08-01), Slocum et al.
patent: 2001/0026631 (2001-10-01), Slocum et al.
patent: 2002/0028021 (2002-03-01), Foote et al.
patent: 2002/0181756 (2002-12-01), Shibuya et al.
Choobin Barry
Langer Edward
Patel Jayanti K.
Shiboleth Yisraeli Roberts Zisman & Co.
Yissum Research Development Company of the Hebrew University of
LandOfFree
Similarity measurement method for the classification 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 Similarity measurement method for the classification of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Similarity measurement method for the classification of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3141667