RU136612U1 - BREAST CANCER TYPE "BREAST CANCER" TYPE DIAGNOSTICS DEVICE ON MAMMOGRAPHIC PICTURES - Google Patents

Abstract

A device for diagnosing a neoplasm area of the type "Breast Cancer" in mammography images, comprising: a digitizing unit that receives a digital representation of an analog mammography image and transmits it to the input of the preprocessing unit; a preprocessing unit performing contrast and noise reduction operations; a neoplasm region allocation unit performing texture segmentation and subsequent binarization operations; block for selecting contours and determining the coordinates of its points; a buffer of the LIFO type, which stores the coordinates of the points of the neoplasm contour obtained from the block for selecting the contours, and transfers them to the input of the block for calculating the area and coordinates of the points of the center of the found neoplasm, and then to the input of the block for determining the perimeter and shape factor of the neoplasm; a buffer of the FIFO type, which receives the brightness values of the points of the selected neoplasm; a unit for calculating the texture characteristics of the neoplasm that calculates the values of the texture features of the found area; the unit for the formation of the neoplasm of the area "Breast Cancer", copying an array of brightness points of the area of the found neoplasm; a control device that determines the parameters of the pre-processing operations, texture segmentation and binarization, as well as the required characteristics of the selected neoplasm.

Description

The device relates to computer technology and information technology in medicine. It can be used to visually improve the quality of digitized images, and also as a screening method for identifying areas of breast cancer at various stages.

Currently, automated technologies for obtaining and processing data are being introduced in all areas of medicine. The best method for determining breast tumors is mammography. With the advent of computer technology, it has become possible to process and analyze digital and analog mammograms.

The well-known "Method of obtaining an informational image of breast cancer by mammograms" (patent of the Russian Federation No. 2235361, class G06K 9/50, G06T 7/60, 2002) used as a non-invasive biopsy at the marker level at any stage of development of breast cancer in the form of detection by mammograms of foci of malignant growth through abnormal asymmetry of the cell masses of the malignant tumor relative to the masses of healthy and non-cancerous cells [1]. The technical result achieved by the implementation of the claimed invention is the identification by mammograms of foci of malignant growth through abnormal asymmetry of the cell masses of the malignant tumor relative to the masses of healthy and non-cancerous cells.

The problem is solved by a method designed to detect a cancer tumor and its most intense growth points, including obtaining mammograms, presenting it in numerical form as an initial matrix of numbers at the sampling steps, obtaining a final image using a known technological technique for detecting the density distribution of the desired type of information with using a dedicated information window in the form of a square of at least nine sampling steps and not more than one quarter of the original matrix, and the difference In order to obtain a specific informational image of breast cancer by a mammogram, they form in the original numerical matrix through an n-cell square template with a single cell size equal to the sampling step and the size of the template itself of at least four cells and not more than a quarter of the original matrix, information cells from n new numbers in the cells of each cell obtained by shifting the information template by one sampling step over the entire original matrix and determine the difference between the numbers that fall into notch pattern, and the number with the minimum value in one of the pattern of cells, according to the formula:

where Hi is the initial value of the number that fell into the i-th cell of the information template,

H'i - the new value of the number in the i-th cell of the information cell,

i - cell numbers of the cell corresponding to the cell numbers of the template (1, 2, 3, ..., n);

among the new H'i numbers in the cell, find the maximum H'i, according to which all the numbers H'i are normalized, of the information cell, getting a new set of normalized numbers H "i, for all cells of the cell; then the deviation for all H" i is calculated, according to the formula:

where ± Vi, is the deviation from the level of 0.5,

0,5 - the level obtained by dividing in half the maximum value among the normalized numbers H "i, which is equal to unity; then positive deviations are summarized separately:

and separately negative deviations:

where k + l = n;

after these calculations, the indicator of the abnormal asymmetry of the information cell is found by the formula:

all positive A + values are considered a highlighted informative sign of breast cancer;

then, using the well-known technology for working with the information window, a new information image is obtained in the form of a matrix of numbers reflecting the density distribution of the selected sign, the resulting new information image is analyzed from the point of view of revealing the relief of the studied anomalies corresponding to the foci of the cancerous tumor, on the mammogram, if necessary, the identification is repeated density distribution through the old or new size of the information window to obtain the relief of the studied anomalies of the required scale.

At the same time, a new information image in the form of a matrix of numbers reflecting the density distribution of the selected attribute is represented in color, according to the sign and density value.

This method allows, according to the mammogram, to organize the normalized allocated spaces on the span of the two-dimensional signal through the n-cell template, to look at them in the observation interval in numerical form by comparing the deviations in the positive and negative regions of the normalized numerical values of the signal in the cells from half the span in the cell and present the signal on the observation interval in the form of information cells with a certain asymmetry value of each cell, reflecting the presence of anomalies and constituting image information of the breast. Information on abnormal asymmetry allows detecting cancer foci in the form of a relief by the relative specific gravity of a specific informative sign in the information image of anomalies in the mammogram.

The problem is also solved by a method designed to identify a clearer border of a cancer tumor by a mammogram, characterized in that in the proposed method described above, the calculations are adjusted to the sum of the positive S + deviations, which are considered as an isolated sign of breast cancer, then all the described operations to obtain a new information image in the form of a matrix of numbers reflecting the density distribution of the selected feature S +, having the shape of the anomaly relief, corresponds to those foci of a cancer tumor on a mammogram; The obtained relief of anomalies serves to analyze, together with the relief obtained above, the features of the manifestation of breast cancer.

At the same time, a new information image in the form of a matrix of numbers reflecting the density distribution of the selected attribute is represented in gray scale colors according to the density value.

The authors tried to use this method to identify foci of malignant growth by mammogram. However, they received an unsatisfactory result. A significant drawback in the utility model is the use of only one information feature - an indicator of anomalous asymmetry.

Researchers R. Lukac, B. Smolka, K.N. Plataniotis in their work [2] proposed a new filtering scheme for the detection and removal of impulse noise in digital color images. Since color images are fundamentally different from x-ray images, such a filter cannot be directly applied to a digitized mammography image, since it has individual characteristics due to the selection of voltage and current on the mammograph, as a result of which, in addition to pulsed noise, interference from the mammography apparatus is added .

Researchers S. Saheb Basha, Dr. K. Satya Prasad in his work [3] used the morphological operators of erosion and dilatation, opening and closing to detect the area of breast cancer:

Applying only texture segmentation to an untreated mammographic image does not give the desired result of isolating the breast cancer region, especially at an early stage when the tumor is up to 1 cm in size.

The proposed utility model is based on the diagnosis of the neoplasm of the type of "Breast Cancer" (BC) in accordance with the International Classifier of Disease based on the texture map of the mammogram. There are no complete analogues of the proposed model. Therefore, the most interesting is the creation of such diagnostic devices - assistants to the doctor in the diagnosis of breast cancer.

Closest to the proposed device in its technical essence is the "Method of obtaining an informational image of breast cancer by mammograms" [1]. The proposed device allows you to most accurately identify areas of breast cancer on mammograms at an early stage (tumor diameter from 7 mm) due to the preliminary processing of the digitized mammographic image.

The technical task of this device is to identify areas on the mammogram containing breast cancer.

The device implements the following calculation steps:

1. An analog mammogram is digitized using a scanner having the following characteristics: resolution 6400 × 9600 dpi, optical density 4D. The digitized mammogram l (x, y) has a size of m * n pixels. The brightness of the points is distributed in the range [0, 255].

2. Since, as mentioned above, for each mammary gland, as well as for a different mammographic film and protective screens, the laboratory assistant selects the voltage and current strength on the mammograph, the result of the mammogram display depends on these characteristics (it can be very dark, low-contrast), remaking images creates additional exposure for patients, which is extremely undesirable.

The authors developed an algorithm for histogram transformations of the mammogram, which normalizes the brightness and contrast of the image, and also makes it possible to isolate breast cancer against the background of mastopathy:

where g (x, y) is the output image, g _max is the maximum brightness value of the output image, g _min is the minimum brightness value of the output image,

K = 1.5 - coefficient (1 <= K => 2)

l (x, y) - the current brightness value of the original image,

l _max , l _min , l _cf - maximum, minimum and average values of the initial mammogram, respectively,

where i is the value of the brightness gradation of the point of the output image (i = 0..255),

j is the value of the brightness gradation of the point of the original image (j = 0..255, j <= i,),

v [j] is the value of the histogram element of the original image with brightness j (v [j] ∈V),

T is the region of variation of the elements of the histogram of the output image g (x, y),

t [i] is the value of the histogram element of the output image at brightness i (t [i] ∈T),

t [0] = t [l (x, y) = 0] is the value of the histogram element of the processed image at a point brightness of 0,

m * n - image size.

3. Image filtering unit.

Usually, histogram transformation causes some additional image noise, so the image g (x, y) is filtered using a 3 * 3 sigma filter. The algorithm of the authors R. Lukac, V. Smolka, K.N. was taken as a basis Plataniotis from [2].

where d (x, y) is the output image, S is the sigma filtering operator.

4. The block selection area of breast cancer carries out texture segmentation.

The texture segmentation was based on the algorithm of the authors S. Saheb Basha, Dr. K. Satya from [3].

where e (x, y) is the output image, C is the texture segmentation operator.

5. The block for selecting contours in the image localizes the desired area with clear boundaries.

Before highlighting the contour in the image, the texture map of the image is binarized. The binarization algorithm is as follows:

where b (x, y) is the output image,

e (x, y) - texture map of the image,

where i is the value of the brightness gradation of the point of the output image (i = 0..255),

j is the value of the brightness gradation of the point of the original image (j = 0..255, j <= i,),

z [j] is the value of the histogram element of the original image with brightness j (z [j] ∈Z),

u [i] is the value of the histogram element of the output image at brightness i (u [i] ∈U),

u [0] = u [e (x, y) = 0] - the value of the histogram element of the processed image with a point brightness of 0,

w is the coefficient on which the mapping of various types of tissues depends (if w = 0.01, then only the areas of the pectoral muscle and breast cancer remain in the picture) (0 <w <= 10)

m * n - image size.

6. The selection of circuits is carried out using the Sobel operator.

where r (x, y) is the output image, L is the contouring operator.

The Sobel operator is an overlay of two 3 * 3 masks at each point in the image. These masks reveal the boundaries of objects located vertically and horizontally in the image.

7. Block characterization performs the calculation of:

- area of breast cancer region S _k (the number of points in the region);

- the perimeter of the region P - approximately equal to the number of points in the contour of the region;

- form factor:

The drawing shows a structural diagram of a device for diagnosing areas of neoplasms of the type "Breast Cancer" on mammograms. The device consists of: a unit for obtaining a digital representation of a mammogram (block for digitizing a mammogram) - block 1; input preprocessing unit - block 2; block allocation of the neoplasm - block 3; neoplasm contour highlighting unit - block 4; LIFO type buffers (the principle of technical data processing on the principle of who comes last is served first) - block 5; block for calculating the area and coordinates of the neoplasm center - block 6; unit for calculating the perimeter and shape factor of the neoplasm - block 7; a buffer of the FIFO type, receiving the coordinates of the contour of the object - block 8; a unit for calculating the texture characteristics of the neoplasm — block 9, a block for forming the neoplasm region “Breast Cancer” —block 10 and a control device — block 11.

The device operates as follows: with the help of the digitizing unit, a digital representation of the analog mammogram picture is obtained (block 1). It is transmitted to the pre-processing unit (block 2). In it, the operation of contrasting and noise reduction is performed on the obtained image to improve the quality of the mammogram, as well as to isolate small tumors among the breast tissue by increasing the contrast and brightness of these areas. If a high quality mammogram is processed, then block 2 is optional, block 3 can immediately be performed - texture segmentation - this block allows you to group similar areas on the mammogram: neoplasms, pectoral muscles, background, adipose tissue, where the neoplasms will be assigned to the first class, which hereinafter, it is considered as the main one, then binarization of the texture map of the mammographic image with the coefficient w is performed, with the help of which the neoplasms are highlighted in black white, the rest of the picture. After binarization, the outlines of the found neoplasms are selected, which are highlighted in the original image — block 4. The number of points N belonging to the object’s contour determines the size of the LIFO buffer (block 5), in which their coordinates are written. From this buffer, they enter the input of the block for calculating the area and coordinates of the neoplasm center (block 6). These parameters are transmitted to the input of the perimeter determination unit and the neoplasm shape coefficient (block 7). The FIFO buffer (block 8) records the brightness values of the points of the selected neoplasm (from the original digitized mammogram). From the output of the buffer, the brightness values are sent to the texture characteristics calculation unit, which form the values of the feature vector (block 9). After calculating the characteristics of the found neoplasm region, the brightness values of the points of the found neoplasm (included in the closed region of the neoplasm contour) are copied to the array (block 10). Calculation of geometric characteristics such as the perimeter, area and shape factor allows you to track the dynamics of tumor growth. Calculation of texture characteristics (18 texture features) allows us to further classify selected areas using recognition algorithms. The control device determines the operations of pre-processing, texture segmentation and binarization, as well as the values of the required characteristics of the selected neoplasm.

Information sources

1. Patent of the Russian Federation for invention No. 2235361 A method for obtaining an information image of breast cancer by mammograms. 11/20/2002

2. R. Lukac, B. Smolka, K.N. Plataniotis et al. Generalized adaptive vector sigma filers, International Conference on Multimedia Computing and Systems / International Conference on Multimedia and Expo - ICME (ICMCS), vol. 1, 2003, pp.537-540

3. S. Saheb Basha, Dr. K. Satya Prasad Automatic Detection of Breast Cancer Mass in Mammograms Using Morphological Operators and Fuzzy C-Means Clustering, Journal of Theoretical and Applied Information Technology, Vol.5 No 6, June 2009, pp.704-709

Claims (1)

A device for diagnosing a neoplasm area of the type "Breast Cancer" in mammography images, comprising: a digitizing unit that receives a digital representation of an analog mammography image and transmits it to the input of the preprocessing unit; a preprocessing unit performing contrast and noise reduction operations; a neoplasm region allocation unit performing texture segmentation and subsequent binarization operations; block for selecting contours and determining the coordinates of its points; a buffer of the LIFO type, which stores the coordinates of the points of the neoplasm contour obtained from the block for selecting the contours, and transfers them to the input of the block for calculating the area and coordinates of the points of the center of the found neoplasm, and then to the input of the block for determining the perimeter and shape factor of the neoplasm; a buffer of the FIFO type, which receives the brightness values of the points of the selected neoplasm; a unit for calculating the texture characteristics of the neoplasm that calculates the values of the texture features of the found area; the unit for the formation of the neoplasm of the area "Breast Cancer", copying an array of brightness points of the area of the found neoplasm; a control device that determines the parameters of the pre-processing operations, texture segmentation and binarization, as well as the required characteristics of the selected neoplasm.

Images