Image analysis – Applications – Biomedical applications
Reexamination Certificate
1998-02-23
2002-08-13
Au, Amelia M. (Department: 2623)
Image analysis
Applications
Biomedical applications
C382S171000, C382S172000
Reexamination Certificate
active
06434261
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to image processing of digital images, and more specifically to image processing using a wavelet transform technique.
2. Description of Related Art
Currently, breast cancer is a leading cause of death among women. Nearly 10 percent of all women in North America will develop breast cancer during their lifetime (Bassell and Gold, 1987). Earlier diagnoses of breast cancer are of great importance in modem medicine. At present, mammography is the method of choice for early breast cancer detection. Automation of mammogram analysis by computer vision technology has been of interest to more and more researchers and radiologists (Bowyer and Astley, 1994; Dengler et al., 1993; Brzakovic et al., 1990).
Although automatic analysis of mammograms cannot fully replace radiologists, an accurate and efficient computer-aided analysis method can help radiologists to make more efficient and accurate decisions. Two of the very important objects of interest in mammograms that need to be detected and segmented in more detail are tumors and microcalcifications.
Recent developments in wavelet transform techniques can be very useful for such computer aided diagnosis. Multiresolution processing techniques (Mallat, 1989; Daubechies, 1992) may be formulated and implemented by wavelet transforms. Currently, such techniques are widely used in image processing and analysis (Akansu and Smith, 1996). By using wavelet transforms and multiscale analysis, most of the gray scale information can be obtained in a large scale image, and singularity information, such as edge and texture, etc., can be detected in detailed images at different scales. A proper form of the wavelet transform and choice of a suitable mother wavelet is application-dependent. In addition, as will be discussed in detail below, the choice of pre-processing and post-processing methods may be critical to successful detection and classification of tumors and microcalcifications.
Several papers have been published in the area of application of wavelet transforms for analysis of mammograms (Yoshina et al., 1995; Strickland and Hahn, 1994; Strickland and Hahn, 1996; McLeod et al., 1996; Yoshida et al., 1996; Chitre and Dhawan, 1996). Most of these papers mainly deal with clustered microcalcifications, which appear as a relatively evident and larger bright area compared to the size of single microcalcifications.
However, the detection of single microcalcifications is also very important for the diagnoses of breast cancer, because often they are indicators of malignancy. It is easy to understand that these single microcalcifications may easily be overlooked by radiologists because of their small size. Therefore, computer-aided detection and segmentation are of great importance in these cases.
A goal of many types of image processing is to develop automated detection of objects of interest (generally referred to as “targets”) from a noisy, cluttered background.
Many methods for segmenting targets from a noisy, cluttered background have been developed. Among these, many involve greyscale-based thresholding strategies, which usually assume the image to have a uniform and stationary, or at least quasi-stationary, distribution of intensities over the target and background. Such methods are generally not very effective when images have complex, nonstationary distributions of intensities. It is usually difficult to adapt global techniques by simply using local processing, since the resulting operators are usually more sensitive to noise and generally yield desultory improvement.
However, if the image is observed at a particular scale, or range of scales, viz., by passing it through a bandpass filter, the resulting image intensity distribution will often exhibit a greater degree of local stationarity. In this way, the performance of local, adaptive segmentation/ thresholding algorithms may be improved.
SUMMARY OF THE INVENTION
The present invention presents a new method for the detection and segmentation of small, single microcalcifications.
Wavelet transform-based methods offer a natural framework for providing multi-scale image representations that may be separately analyzed. For example, through a multiscale wavelet decomposition, most of the gross intensity distribution can be isolated in a large scale image, while information about details and singularities, such as edges and textures, can be isolated in mid- to small scales.
The present invention presents a new and systematic method for the segmentation of potential target areas based on a selected wavelet decomposition and a Bayes classifier. Further, an adaptive, multi-scale threshold selection criterion is disclosed, which analyzes the image probability distribution function (PDF).
The term “bright target” as used herein means a connected, cohesive object which has an average intensity distribution above that of the rest of an image. Such targets usually appear as having a blob-like appearance. Of course, other image components may occur that are bright, but which may be poorly connected or which occur only over a small scale. Through the use of a natural multiscale decomposition, such as an appropriate wavelet transform, coherent targets may be separated and subsequently distinguished from transient bright objects.
Presented is an approach to target segmentation that involves isolating the putative target scale through selection of an appropriate scale in the wavelet decomposition, then thresholding the single-scale image using a Bayesian classifier. The method is used as a benchmark comparison for the methods of the present invention. The Bayesian approach, while optimal in a statistical sense, requires certain a priori information and suffers from some limitations.
The present invention includes a method of determining an adaptive threshold for a wavelet analysis of a digital image, which includes the steps of: decomposing the digital image by wavelet packet transform to obtain transformed images of different scale channels; obtaining a histogram of the transformed images; decomposing the histogram by wavelet transform to obtain histogram images of different scale channels; and selecting the adaptive threshold from the histogram images.
The present invention also includes a method of determining an adaptive threshold for a wavelet analysis of a digital image comprising the steps of: decomposing the digital image by wavelet packet transform to obtain transformed images of different scale channels; and selecting a minimum point of a probability distribution function of the transformed images by wavelet analysis.
The present invention also includes a method for automatically detecting an abnormal portion of a digital image comprising the steps of: obtaining the digital image; selecting a mother wavelet optimized for the digital image; decomposing the digital image by wavelet packet transform into various channels; determining an adaptive threshold from all channels; comparing the adaptive threshold to each of the channels; eliminating each of the channels less than the minimum threshold to obtain an adjusted singularity map; restoring the adjusted singularity map to a restored image; and locating regions in the digital image corresponding to the restored image.
The present invention also resides in a method of obtaining a microcalcifications map from a mammogram image, including the steps of: inputting a digitalized mammogram image; preprocessing the image (normalization is one example); obtaining wavelet packet transforms of the image; performing wavelet-based analysis of a probability distribution function of the transformed images at different channels; selecting thresholds and obtaining segmented images of each channel; mapping the segmented images of each channel to the original digital image; and combining all of the mapped images to obtain the microcalcifications map.
REFERENCES:
patent: 5576548 (1996-11-01), Clarke et al.
patent: 5598481 (1997-01-01), Nishikawa et al.
patent: 5666434 (1997-09-01), Nishikawa
Desai Mita D.
Zhang Xiao-Ping
Au Amelia M.
Board of Regents , The University of Texas System
Dastouri Mehrdad
Fulbright & Jaworski L.L.P.
LandOfFree
Method for automatic detection of targets within a digital... 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 for automatic detection of targets within a digital..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for automatic detection of targets within a digital... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2900474