KR101622256B1

KR101622256B1 - Method for speckle pattern feature extraction for the computer-aided diagnosis of ultrasound image

Info

Publication number: KR101622256B1
Application number: KR1020140079468A
Authority: KR
Inventors: 최동혁; 김영모
Original assignee: 건양대학교 산학협력단; 한국디지털병원수출사업협동조합
Priority date: 2014-06-27
Filing date: 2014-06-27
Publication date: 2016-05-18
Also published as: KR20160001261A

Abstract

The present invention analyzes a speckle pattern of an ultrasound image and uses it as a feature of a computer assisted diagnosis. However, the present invention extracts speckle by localizing a patch image and extracts the speckle size, number, The present invention relates to a method for extracting speckle pattern characteristics for computer assisted diagnosis of ultrasound images, which can directly extract shape characteristics of speckle.

Description

TECHNICAL FIELD [0001] The present invention relates to a method for extracting speckle pattern characteristics for computer assisted diagnosis of ultrasound images,

The present invention relates to a method of extracting speckle pattern characteristics. Specifically, a speckle pattern of an ultrasound image is analyzed and used as a feature of a computer-assisted diagnosis. However, speckle extraction is performed by localizing a patch image The present invention relates to a method for extracting speckle pattern characteristics for computer assisted diagnosis of ultrasound images, which can directly extract shape characteristics of speckles by analyzing the size, number and cumulative shape of extracted speckles.

By analyzing the ROI patch image, the computer-assisted diagnosis of various diseases such as cirrhosis, chronic pancreatitis, pancreatic cancer and prostate cancer is possible by manually or automatically extracting ROI (region of interest) image from ultrasound image.

1 is a reference diagram showing an example in which a 128x128 ROI image is manually selected for an liver ultrasound image.

Computer assisted diagnosis can be roughly divided into feature extraction and classification. In the present invention, the speckle pattern of the ultrasound image is analyzed and used as a feature of computer assisted diagnosis . For computer assisted diagnosis of ultrasound images, we identify that the pattern of speckle in the ROI patch image is an important factor for the diagnosis of the disease, and extract features that are important elements of computer assisted diagnosis from the speckle pattern.

2 (a) and 2 (b) illustrate normal and cirrhotic ultrasound images. FIG. 2 (a) shows a normal liver ultrasound image, (c) shows rough speckle patterns due to liver and ultrasound images.

3 (a) and 3 (b) show a normal pancreatic ultrasound image and a chronic pancreatitis ultrasound image, respectively, and FIG. 3 (b) It can be confirmed that the speckle pattern is rough.

In the conventional feature extraction method, mainly, a co-occurence matrix of an image is obtained from a patch image, n-order motions and entropy are obtained from the matrix, an average level and dispersion of a patch image are obtained, The texture characteristic of the speckle and the histogram of the patch.

Korean Patent No. 10-0760251 (September 19, 2007)

The object of the present invention is to extract local speckle by local binarization of a patch image and to extract the shape of the speckle by analyzing the size, A speckle pattern for computer assisted diagnosis of ultrasound images that can directly extract the characteristics, obtain a covariance matrix from the accumulated speckle shape, and obtain two eigenvalues and two eigenvectors of the covariance matrix and use it as a computer assisted diagnostic feature And a characteristic extracting method.

The method for extracting speckle pattern characteristics for computer-assisted diagnosis of an ultrasound image of the present invention for the above-mentioned purpose includes: a region selection step of selecting a region of interest among acquired ultrasound images; An image adjusting step of detecting a connected blob by partially binarizing an image of the selected region of interest and adjusting the size of the block by changing the threshold level so that the face fit of the detected block is set to a predetermined ratio with respect to the entire image size; Sorting the areas of the chunks in the selected area in order of magnitude and excluding upper and lower 20%; A first feature extracting step of calculating the number of remaining chunks and the area average through the aligning step; An accumulation image generation step of accumulating the remaining masses through the sorting step to generate an accumulated image; A second feature extraction step of obtaining a covariance matrix from the cumulative image and calculating two eigenvalues and two eigenvectors of the covariance matrix; .

In the image adjustment step, binarization of the image is performed through the Niblack method, and the adjustment threshold level is adjusted so that the surface fitness of the detected mass is 20% of the total image size. X 64 area.

Conventional methods for extracting the speckle pattern characteristics mainly use the cooccurence matrix of the patch image, the average level of the patch image, the texture characteristic of the ultrasound image such as dispersion and edge detection, and the histogram of the image In contrast, the present invention extracts speckle by local binarization of a patch image, extracts characteristics directly reflecting the shape characteristics of the speckle by analyzing the size, number, and cumulative shape analysis of the extracted speckle. It is possible to make the assist diagnosis of the ultrasound image more accurate.

FIG. 1 is a reference view showing an example in which an ROI image of 128 × 128 pixels is manually selected for an liver ultrasound image,
2 is a reference diagram showing normal and cirrhotic ultrasound images,
3 is a reference view showing a normal and chronic pancreatitis ultrasound image,
4 is a flowchart showing a procedure according to a preferred embodiment of the present invention,
5 is a reference diagram showing a partial binarization result when N = M = 7 and? = 0.8,
FIG. 6 is a reference diagram showing the size order in the interior (64x64) of the binarized image, the remaining 20% leftover blobs,
7 is a reference view showing a cumulative image of blobs,
8 is a reference diagram showing the eigenvalues and eigenvectors of the cumulative blobs.

FIG. 4 is a flowchart illustrating a procedure according to a preferred embodiment of the present invention. Hereinafter, a speckle pattern characteristic extraction method for computer-assisted diagnosis of the ultrasound image of the present invention will be described in detail with reference to the accompanying drawings.

In the first region selection step (S 110), the ROI is selected from among the acquired ultrasound images and can be manually or automatically selected. In FIG. 1, a user manually selects an area of interest.

In the second image adjustment step (S 120), the image of the selected region of interest is partially binarized to detect a connected blob, the size of the mass is adjusted by changing the threshold level, So that the set ratio is adjusted.

At this time, partial binarization of the image can be performed by Niblack's method, which binarizes using the mean and standard deviation of the level of the local region. The effect of this binarization scheme is that it is binarization independent of image level changes. In other words, the local averages are used to reflect the local changes, the standard deviations of the local regions are used, and the standard deviations of the local regions are used to distinguish the characteristic micropatterns and speckle patterns from the ultrasound images. Express.

M, and N are respectively the width and height of the local image, the local average at the position of the image (i, j) is expressed by Equation (1).

In addition, the local standard deviation (i, j)

Is calculated through the equations (2) and (3).

Further, the binarization threshold value at the position of the image (i, j)

Is expressed by Equation (4).

FIG. 5 is a reference diagram showing a partial binarization result when N = M = 7 and α = 0.8. FIG. 5 (a) shows an image before binarization, FIG. 5 (b) shows an image after binarization, Are detected. The alpha value is used to determine the binarization level.

To detect the shape characteristics of the speckle pattern, we detect the blob by local binarization and find the connected blob for the binary image according to the 8 connectivity diagram.

At this time, by changing the alpha value in the program, the size of the detected mass is increased or decreased by raising or lowering the level of the threshold, and the sum of the detected masses of blobs adjusted by adjusting the threshold level is 20% The binary image is used.

In the third alignment step (S 130), the areas of the chunks in the selected area are sorted in order of magnitude, and upper and lower 20% are excluded.

FIG. 6 is a reference diagram showing the blobs left in the lower 20% in size order in the interior (64 × 64 pixels) of the binarized image. FIG. 6 (a) shows the image before binarization, (b) shows the image after the binarization, FIG. 6 (c) shows the image showing the blobs left in the lower 20% of the size (64 × 64 pixels).

In the fourth first feature extraction step (S 140), the number of remaining chunks and the area average through the alignment step are calculated. For the plurality of blobs detected in the previous step, the areas for only the blobs of the selected 64x64 pixel area in the ROI are calculated and sorted in the order of size, and the upper and lower 20% are discarded and the remaining lumps ( the number of blobs, and the area average of the remaining blobs are used as the speckle number and size characteristics of the ROI ultrasound image.

In the fifth cumulative image generation step (S 150), the remaining clusters through the alignment step are accumulated to generate an accumulated image.

That is, the remaining blobs are accumulated as shown in FIG. 6 (c) to obtain a shape characteristic of the speckle accumulation pattern. The cumulative method obtains an average size for the blob and accumulates the shape of a plurality of blobs around the average size.

FIG. 7 is a reference view showing an accumulated image of blobs, which is an enlarged image after accumulating in this manner.

Then, in the sixth feature extraction step (S 160), a covariance matrix is obtained from the cumulative image, and two eigenvalues and two eigenvectors of the covariance matrix are calculated.

The speckle accumulation pattern analysis is to obtain the covariance for the cumulative mass (blob) image, obtain the eigenvalue of the covariance, and find the minimum eigenvalue ratio to the maximum eigenvalue. In order to obtain the covariance of the cumulative image, a point of the image is represented by a second random vector of the following formula (5).

The mean vector is the expected value of the vector as shown in the following equation (6).

The covariance matrix of vectors is shown in Equation (7).

If we obtain the eigenvector and eigenvalue of the covariance,

. Where χ is an eigenvector, λ is an eigenvalue,

The eigenvalues of

The eigenvalue ratio is

.

FIG. 8 is a reference diagram expressing the eigenvalues and eigenvectors of the cumulative blobs. Two eigenvalues and eigenvectors are indicated, and the eigenvalues represent the elongation of the blobs.

It is to be understood that the invention is not limited to the disclosed embodiment, but is capable of many modifications and variations within the scope of the appended claims. It is self-evident.

Claims

An area selection step (S110) of selecting an area of interest of the acquired ultrasound image;
An image adjusting step of detecting a connected blob by partially binarizing an image of the selected region of interest and adjusting the size of the block by changing the threshold level so that the face fit of the detected block is set to a predetermined ratio with respect to the entire image size S 120);
An alignment step (S 130) of arranging the areas of the chunks in the selected area in order of magnitude and excluding upper and lower 20%;
A first feature extracting step (S 140) of calculating the number and area average of the remaining 60% of the lumps excluding the upper and lower 20% through the aligning step;
An accumulated image generation step (S 150) of accumulating the remaining 60% of the lumps excluding the upper and lower 20% through the alignment step to generate an accumulated image;
A second feature extraction step (S 160) of obtaining a covariance matrix from the cumulative image, and calculating two eigenvalues and two eigenvectors of the covariance matrix; And extracting speckle pattern characteristics for computer assisted diagnosis of ultrasound images.

The method according to claim 1,
In the image adjustment step (S 120), binarization of the image is performed through the Niblack method, and the threshold level is adjusted so that the face fit of the detected mass is 20% of the total image size,
Wherein the region selected in the aligning step (S 130) is an image of a region selected from a user. A method for extracting a speckle pattern characteristic for computer assisted diagnosis of an ultrasound image.