You have no items in your shopping cart.

Subtotal: 0.00

Matlab code for Detection of Microcalcification

Introduction to Micro calcification in Digital Mammograms

Breast cancer is one of the major causes for the increase in mortality among women, especially in developed and under developed countries. The World Health Organization's International agency for Research on Cancer in Lyon, France, estimates that more than 150 000 women worldwide die of breast cancer each year. The breast cancer is one among the top three cancers in American women. In United States, the American Cancer Society estimates that, 215 990 new cases of breast carcinoma has been diagnosed, in 2004. It is the leading cause of death due to cancer in women under the age of 65 . In India, breast cancer accounts for 23% of all the female cancers followed by cervical cancers (17.5%) in metropolitan cities such as Mumbai, Calcutta, and Bangalore. However, cervical cancer is still number one in rural India. Although the incidence is lower in India than in the developed countries, the burden of breast cancer in India is alarming. Organ chlorines are considered a possible cause for hormone-dependent cancers . Detection of early and subtle signs of breast cancer requires high-quality images and skilled mammographic interpretation. In order to detect early onset of cancers in breast screening, it is essential to have high-quality images. Radiologists reading mammograms should be trained in the recognition of the signs of early onset of, which may be subtle and may not show typical malignant features. Mammography screening programs have shown to be effective in decreasing breast cancer mortality through the detection and treatment of early onset of breast cancers.

Emotional disturbances are known to occur in patient's suffering from malignant diseases even after treatment. This is mainly because of a fear of death, which modifies Quality Of Life (QOL). Desai et al.,reported an immuno histo chemical analysis of steroid receptor status in 798 cases of breast tumors encountered in Indian patients, suggests that breast cancer seen in the Indian population may be biologically different from that encountered in western practice. Most imaging studies and biopsies of the breast are conducted using mammography or ultrasound, in some cases, magnetic resonance (MR) imaging . Although by now some progress has been achieved, there are still remaining challenges and directions for future research such as developing better enhancement and segmentation algorithms.

Demonstration Video


Image Features For Detection Of Microcalcification

Hojjatoleslami et al segment the image into suspected regions and then classify them into normal and abnormal regions. The segmentation technique uses a seed point detection algorithm followed by region growing using pixel aggregation. Suspected regions are classified based on the following seven features

  • Area of the object (number of pixels)
  • Average gray level of the border of the object
  • Average gray level of the object
  • Gradient strength of object's perimeter (Morphological gradient is used)
  • Contrast: difference between the maximum gray level in the region and the mean of the border.
  • Maximum gray level inside the region
  • Maximum edge gradient of the border

Features for detection also could include schemes based on:

  • templates of calcifications
  • morphological features
  • Statistical properties (as in the above example)

In fact, in some studies, as many as 42 features have been used including fractal based textures

Mammogram Databases

The following is a list of some of the databases that are commonly used

Nijmegen Database

Contains 40 images, showing one or more cluster of microcalcifications. The mammograms are from 21 different patients. This database is available freely for research purposes. The scanning resolution is 100 microns per pixel, at 12 bits per pixel.

Mias Database

Images scanned at a resolution of 50um x 50um, at 8 bits/pixel. A small subset with lower resolution can be downloaded for research purposes.

Digital Database for Screening Mammography

A new project for use by the mammography's image analysis research community. This will be a collaborative effort involving Massachusetts General Hospital, the University of South Florida, and Sandia National Laboratories. The database will contain approximately 3,000 studies.

LLNL/UCSF DATABASE

Lawrence Livermore National Laboratories (LLNL) and University of California at San Francisco (UCSF) Radiology Dept. have developed a 12 volume CD Library of digitized mammograms featuring microcalcifications. For each digitized film image, 2 associated “truth” images (full sized binary images) that show the extent of calcification clusters and the contour and area of a few individual calcifications in each cluster, and contains 198 films from 50 patients

How it works





Fig 1 shows a schematic diagram of our methods for detection of microcalcification clusters in mammograms. The mammogram images were first decomposed into several subimages at different scales from 1 to 4 by a novel filter bank .These subimages were the horizontal subimage for the second difference in the vertical direction, the vertical subimage for the second difference in the horizontal direction, and the diagonal subimage for the first difference in the vertical direction followed by the first difference in the horizontal direction. The subimages for NC and the subimages for NLC were obtained from analysis of the Hessian matrix. Many regions of interest (ROIs) of 5 mm 5 mm were then selected from the mammogram image automatically. In each ROI, eight features were determined from the subimages for NC at scales from 1 to 4 and the subimages for NLC at scales from 1 to 4 (Section II-D). The Bayes discriminant function with these eight features was employed for distinguishing among abnormal ROIs with a microcalcification cluster and two different types of normal ROIs without a microcalcification cluster (Section II-E). The region connecting the ROIs that were classified as abnormal was considered to be region of potential microcalcification clusters. A 115 115 matrix (approximately 5 mm 5 mm) was chosen as the ROI size, since the microcalcification cluster was defined as a region containing three or more microcalcifications per 5 mm 5 mm area . When the ROIs were selected at intervals of 5 mm so as to border on the adjacent ROIs, some microcalcification clusters may have been divided across two or more ROIs. These microcalcification clusters were not detected correctly because each ROI included only the information on the divided cluster. Therefore, we need to select ROIs at a shorter interval so that the center of a microcalcification cluster will be at the center of one of the ROIs. Although we must select ROIs at intervals of 1 pixel (0.0435 mm) to analyze a mammogram in detail, there were no large differences between adjacent ROIs selected at intervals of 1 mm. Therefore, we selected the ROIs at intervals of 23 pixels (approximately 1 mm) so that one ROI would overlap with the adjacent ROIs.

Filter Bank for Detection of Nodular Components and

Linear Components

1) Hessian Matrix Classifying Nodular Structures and

Linear Structures: For the distinction between microcalcification clusters and normal tissues in mammograms, it may be important to detect both nodular components, such as microcalcifications, and linear components, such as blood vessels and mammary ducts. To detect these components, we employed the second derivative. The values of the second derivatives for the nodular structure in all directions tend to become negative. However, the value of the second derivative for the linear structure tends to become zero in the direction of the axis of the linear structure, whereas it tends to become negative in the direction perpendicular to the axis of the linear structure. Therefore, the filters based on the second derivatives can be used for the detection or enhancement of the nodular structure and the linear structure. Shimizu et al. [24], [25] developed a minimum directional difference filter (Min-DD Filter) based on the smallest value of the second derivatives in all directions, and a maximum directional difference filter (Max-DD Filter) based on the largest value of the second derivatives in all directions. They applied the Min-DD Filter for detection of large lung nodules in chest X-ray images.

Filter Bank for Detection of Nodular Components and

Linear Components: As we described in the previous section, the nodular component and the linear component can be detected by using the value of the second derivative or the eigenvalue of the Hessian matrix. Although lung nodules have various sizes, the length of the filter for the second derivative was constant in the method [25] of Shimizu. Therefore, it might be possible to detect nodular structures and linear structures of various sizes more accurately by using filters for the second derivative of various sizes. In addition, it might be necessary to properly shape the nodular structure and the linear structure by a smoothing operator, because the second derivative is usually in- fluenced by noise. These issues are easily solvable by a filter bank which consists of high-pass filters and low-pass filters of various lengths. Once a microcalcification cluster is detected, the next problem is to determine whether the lesion is benign or malignant. Many investigators have developed computerized schemes for distinguishing between benign and malignant micro calcifications [29]-[31]. In these schemes, it is important to enhance microcalcifications while maintaining their shapes, since image features such as size and irregularity are used to determine the likelihood that a microcalcification cluster is malignant. However, this issue was not taken into account in the methods [26]-[28] of Li and Sato. This issue is solvable by use of a filter bank that satisfies the requirement for perfect reconstruction. Therefore, we introduced the concept of the Hessian Matrix into a filter bank that satisfies the requirement for perfect reconstruction.





Fig. 2 shows a two-channel filter bank. The analysis bank on the left has a low-pass filter , HL(Z) a high-pass filter , HH(Z) and a downsampling operator ( ) which removes the odd numbered components after filtering. The synthesis bank on the right has a low-pass filter FL(Z) , a high-pass filter FH(Z) , and an upsampling operator ( ) which inserts a zero in the odd components. The filter bank for the one-dimensional (1-D) DWT is usually given by iterating the low-pass channel of the two-channel filter bank, as is illustrated in Fig. 3. For the perfect reconstruction with an -step delay, the filters of the two-channel filter bank must satisfy the following conditions [32].

HL(Z) FL(Z) + HH(- Z) FH (- Z) = 0

 

Enhancement Technique

In order to improve the contrast enhancement, first, the original mammogram is wavelet transformed by using filter bank, yields a smooth (low pass) component and three detail (high pass) components. These processes were applied without 2-factor down sampling from wavelet transform coefficients. Second, global nonlinear operator was applied on decomposed detail subband images (highpass components) using multiscale adaptive gain method. In this technique, highpass components will be suppressed if it is value less than the threshold and will be increased if it is greater than threshold. The formula used to accomplish this nonlinear operation is given by





where b and c are defined as threshold and rate of enhancement respectively. It can be shown that f(y) is continuous and monotonically increasing within interval [-I, 71. In addition, f(y) satisfies the condition f(0) = 0 and f(7) = 7. For an input image y with maximum absolute Ymaw, we map the image range [-ymax,ymaJ onto interval [--?,?I. This is accomplished by using ymax as a normalizing factor in (5)

 

Detection of Microcalcification Algorithm

The microcalcification detection algorithm were proposed by using statistical methods such as skewness, kurtosis and boxplot outlier. DetectioN methods were applied on modification decomposed image as a sum result of matrixes on detail component orientation (horizontal, vertical and diagonal). In this paper, detection is carried out in two steps. First, the detail-image is divided into same square region size, i.e. n x n pixels. In these regions, skewness and kurtosis, measures of the asymmetry and impulsiveness of the distribution are estimated.

Skewness (y,) and kurtosis (y2) parameters are defined as :


where is k-moment of variable x.

If a region contains microcalcification then due to theimpulsive nature of microcalcification the symmetry of the distribution of detail-image coefficients is destroyed. It is also evident that the tails of the distribution are heavier and hence the kurtosis assumes a high value. Therefore a statistical test based on skewness and kurtosis is effective in finding regions with asymmetrical and heavier tailed distributions [3]. The detection problem is posed as a hypothesis testing problem in which the null hypothesis, where can be describe as :



where TI and T2 are skewness and kurtosis thresholds value, respectively. Value "0" signs there is no microcalcification in the region, and "1" signs there is microcalcification in the region. In the second stage, an outlier labeling method [7] is used to find the locations of microcalcification in these regions. In this method, the data x (ROI data matrixes) is first rank ordered, x = {x7, xz, ..., XN). Next, the median, the lower quartile (a,) and the upper quartile (a3) values are determined. The interquartile range Rf is defined to be Q3-Q7. The boxplot method determines the outliers to be the part of data, which is outside the region (Ql -kR,,Q, + kR,). Figure I illustrates the boxplot outlier labeling method.

Implementation and Results

The simulation is carried out by using the following conditions

1) Test mammogram images were obtained by scanned as raw format with 8-bit grayscale and 256x256 pixels size. These mammograms have been chosen by the radiologist and suspected as mammograms with micro calcification. In this simulation, 30 variation of image as part of 18 digitized mammograms is used.

2) The chosen wavelet basis function is the Daubechies with four coefficients as a filter banks. These processes were applied without 2-factor down sampling from wavelet transform coefficients. Its used to reduce lost information and maintain size of images.

3)Global image enhancement procedure was applied only on 4-level decomposed detail sub band image (high pass components) using multiscale adaptive gain method. In this technique, high pass components will be suppressed if it's value less than the threshold and will be increased if it's greater than threshold.

4)The detection of micro calcification algorithm will be done as described above

Matlab code for micro calcification detection









Conclusion and Extension of Work

The regions of clustered microcalcification can be detected and the presence another location of clustered microcalcification could be considered to clarify the diagnoses. In order to test the detection method, we used the visual analysis to detect presence microcalcification in mammograms based on comparison between the result images and the original ones. The result of test images shown effectiveness simulation on microcalcification detection, even there are some result could not detect the clustered microcalcification. Fail of detection process will reduce the calculation of Simulation effectiveness. From the 30 test images, there were the 29 test images result a good detection process and just one image was fail. Additionally the processing is simple and does not require a full decomposition and reconstruction.

References

[ I ] R.N. Strickland and H.I. Hahn, "Wavelet Transform for Detecting Microcalcification in Mammograms", IEEE Transaction on Medical Imaging, Vol. 15, No.2, pp.218-229, April 1996.

[2] H. Yoshida, K. Doi, R.M. Nishikawa, "Automated detection . of clustered microcalcification in digital mammograms using wavelet transform techniques", Proc. SPIE, 21 67:868-886, 1994.

[3] M.N. Gurcan, Y. Yardimci, A.E. Cetin and R. Ansari, "Automated Detection and Enhancement of Microcalcification on Digital Mammograms using Wavelet Transform Techniques", Dept. of Radiology, Univ, of Chicago, 1997.

[4] S. Mallat, "A Theory for Multiresolution Signal Decomposition : The Wavelet Representation", IEEE Trans. Pattern. Annal. Machine Intell.,

[5] I. Daubechies, "Orthonormal Bases of Compactly Supported Wavelet", Comm. On Pure and Applied Mathematics , Vol. 41, pp.

[6] A.F. Laine, S. Schuler, J. Fan and W. Huda, "Mammographic Feature Enhancement by Multiscale Analysis", /€E€ Transaction on Medical Imaging , Vol. 13, No.4, pp 725-740, Dec., 1994.

[7] Y. Barnett, T. Lewis, "Outliers in Stafistical Data", 3w Ed. New York: John Wiley & Sons, 1994. Vol. 11, NO. 7, pp. 674-693, Jul. 1989. 906-966, 1988. 697