RU2530300C1 - Method of improving quality of structural image of biological object in optical coherence tomography - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: disclosed is a method of obtaining a structural image of a biological object in optical coherence tomography. The method includes breaking down a source colour vide frame into non-overlapping spatial blocks consisting of more than one pixel. A structural image is obtained via small-angle raster scanning in the arm of an optical coherence tomography sample. The obtained image with a size of P_iskh bytes is broken down into non-overlapping spatial blocks only on columns; adjacent column blocks are averaged pixel by pixel to form a new image with a size of P_stl bytes; the new image is broken down into non-overlapping spatial blocks only on rows; adjacent row blocks are averaged pixel by pixel to form a resultant image with a size of P_res bytes, and the averaging process is controlled based on an exponential relationship P_stl from the number of averaging column blocks U_stl and P_res from the number of averaging row blocks - U_str.

EFFECT: high quality of the structural image of a biological object in optical coherence tomography.

7 dwg

Description

The invention relates to the field of image coding, in particular, can be used for compressing coding of a structural image of a biological object in optical coherent tomography with an increase in its contrast and information content.

A known method of compression and decompression of the image [see Patent No. 2461977 (RF), H04N 7/26 (2006.01), H04N 7/64 (2006.01), No. 2009127749/07 / Zuo F .; De B. C; Brules B.X.A .; Hinnen K.Y.G .; Verberne M.Y. - 2007], owned by the CONINCLEKE PHILIPS ELECTRONICS N.V. corporation, in which pixel clusters are specified for use in image compression and decompression, while the image information used to specify the clusters may include pixel values at a predetermined position relative to a pixel or corresponding motion vectors, gradients, texture, etc. The technical result is to provide the weakening of compression artifacts.

The disadvantages of this method are a partial loss of image quality and focus on compression with the least loss in quality of existing images.

The prototype adopted a method of compressing images and video sequences [see Patent No. 2420021 (RF), H04N 7/26 (2006.01), Н03М 7/34 (2006.01), H04N 11/04 (2006.01), No. 2009110512 / Mishurovsky M.N .; Josan O.V .; Levers M.N .; Rybakov O.S .; Lee S.-S. - 2009], owned by SAMSUNG ELECTRONICS Co., Ltd. (KR), in which it is proposed that each color pixel be represented by three color components, each of which is initially encoded with ten bits. Encoding is carried out by splitting the original color video frame into non-overlapping spatial blocks and then dividing the bit representation of each of the color components of the pixel into the highest part, consisting of more than one high-order bit, and the low-order part, consisting of at least one low-order bit, then separate encoding of the older and younger parts, and the encoding of the older part is carried out by applying more than one encoding method, each of which takes into account inter-pixel communication then only within the spatial block being processed, evaluating the encoding error, choosing the encoding method that gives the smallest error, sending data on the encoding method by transmitting the prefix code, encoding the lower part, which is carried out by averaging more than one value included in the lower part, and the size of the averaging regions within the younger part depend on the selected encoding method of the senior part, set the fixed number of bits in advance required for compact a new representation of the original spatial color block. The technical result is the effective compression of color high-quality images without visible visual distortion.

The disadvantages of this method include a partial loss of image quality and the fact that it is focused on compression with the least loss in quality of video sequences and existing images.

The technical objective of the method is to improve the quality of the structural image of a biological object in optical coherence tomography, namely, the signal-to-noise ratio due to raster averaging.

The stated technical problem is achieved by the fact that in the method of obtaining a structural image of a biological object in optical coherence tomography, the source color video frame is divided into non-overlapping spatial blocks consisting of more than one pixel, and in contrast to the prototype, in order to improve the quality of the structural image, namely, the ratio signal / noise, to obtain a structural image, the method of small-angle raster scanning in the shoulder of an optical sample is coherently th tomograph obtained image size P _ref byte is divided into non-overlapping spatial blocks only on columns adjacent blocks columnar pixel-averaged, thereby forming a new image, a new image is divided into non-overlapping spatial blocks only by rows, adjacent blocks row pixel by pixel are averaged, creating wherein the resulting image is of size P _Res bytes, and the averaging process is controlled by the exponential dependence of the output image size P _Res Ait averages of the number of adjacent non-overlapping spatial blocks of columnar U:

,

,

- размер одного пикселя после усреднения, байт; i и j - индексы строки и столбца, являющиеся координатами пикселя в изображении размером Р_рез, байт.where n and m are, respectively, the number of rows and columns in the image with the size P _ref ; $$P_{i,j}^n$$

- the size of one pixel after averaging, bytes; i and j are the row and column indices, which are the coordinates of the pixel in the image size P _rez , bytes.

Various approaches are possible for constructing a structural image of a biological object in optical coherence tomography, differing in the filtering of the signal in the time and frequency domain, the types of window functions for the Fourier transform, the features of reading and splitting the signal, the order of operations, and other similar subtleties that are not relevant to the invention . In this regard, we consider the essence of the method, relying on a specific block diagram shown in figure 1.

The blocks responsible for controlling the process of constructing a structural image of a biological object in optical coherence tomography are shown in red and marked with a - * symbol.

In the “Data Entry” block, the user is given the opportunity to choose the image construction path:

1) with a given compression ratio relative to the uncompressed image,

2) with a given averaging index with respect to the non-averaged image.

If the 1st path is selected, then the control logic parameter F is assigned the value 1, and if the 2nd path is selected, the value 0 is assigned.

After the cycle of processing data segments is completed, the value of parameter F is checked.

If the condition F> 0 is not fulfilled, then in the “Averaging Calculation” block, based on the compression index specified by the user, the averaging index is calculated, then in the “Raster Averaging” block, compression is performed using averaging.

If the condition F> 0 is fulfilled, then in the “Compression Calculation” block, based on the averaging index specified by the user, the compression index is calculated, then in the “Raster Averaging” block, compression is performed using averaging.

Obviously, pixel-by-pixel averaging of image pixels, in this case it is more correct - the spectrogram matrix data, can be done both by rows and columns, dividing the image into non-overlapping spatial block-rows and column-blocks, respectively. Moreover, in both cases, the size of the structural image obtained after averaging will decrease almost exponentially with an increase in the number of averages (the exact pattern depends on the selected graphic format and on the image itself).

The exponential dependence is explained by the fact that when saving a structural image from an optical coherent tomograph in non-vector graphic formats, each image pixel will correspond to the cell of the spectrogram matrix, i.e. the dimension of the image will coincide with the dimension of the matrix. With averaging, this dimension will be inversely proportional to averaging, but with rounding down. Figure 2 shows the dependence of the number of pixels in the image on the averaging index for an image of 500 rows and 180 columns. If the number of pixels in an image decreases almost exponentially, then its size will change in much the same way, adjusted only for color coding.

The presence of a clearly traceable regularity between the element-wise averaging of non-overlapping spatial data blocks and image compression is most pronounced with the remote small-angle raster scanning method in the shoulder of a sample of an optical coherent tomograph (using a galvanic scanner) and is of great practical importance, since allows you to improve image quality, while reducing the size of its file.

Let us consider this regularity in detail for averaging U-neighboring non-overlapping spatial column blocks, spectrogram matrices of size n · m, followed by saving the received image data in t × t format.

We accept the following conventions:

data compression rate - S;

size (bytes) of the original image, i.e. images before averaging - P _ref ;

size (bytes) of the result image, i.e. images after averaging - P _res ;

size (bytes) of one pixel before averaging - P _{i, j} .

.size (bytes) of one pixel after averaging - $$P_{i,j}^n$$

.

Obviously, the data compression ratio will be equal to the ratio of the image size before averaging to the image size after averaging:

For the t × t format, the file size before and after averaging is easy to calculate. This is due to the fact that when writing all the characters (including the delimeter) that we will use in this format occupy 1 byte. At the same time, there are no columns in the file structure, the record goes line by line, each line is separated from the previous one by a space. From the above it follows that the file size of the source image will be equal to:

The image size after averaging is calculated by a similar formula, with the only difference being that the image size will be smaller due to averaging and we will talk about completely different values for pixels.

New values of intensities arise precisely in the process of cyclic averaging (finding the arithmetic mean) of each U-adjacent non-overlapping spatial column blocks into one. Conditionally, this process can be expressed by the following formula:

It should not be forgotten that the averaging of neighboring column blocks is performed cyclically, with a shift by U, and the calculation of the image file sizes before and after averaging is performed once, therefore these formulas cannot be combined. Formula (4) only demonstrates averaging by the example of the first U-columns and is valid only when m = U.

Thus, combining formulas (1) - (3), we obtain the final formula for the dependence of file compression on the averaging index for the case under consideration:

There is no point in deriving a similar formula for averaging over adjacent non-overlapping spatial block rows it is much easier to transpose the matrix of the spectrogram, perform averaging and calculations using these formulas (given the fact that the lines in the spectrogram are not n, but U times smaller), and then transpose again for the usual kind of image.

If the image is averaged over both non-overlapping spatial blocks of rows and non-overlapping spatial blocks of columns, averaging is performed sequentially: first, averaging over blocks of columns, then transposing the result, once again averaging (already over blocks of rows), re-transposing, and finally , output the result.

As an example, clearly illustrating the above, Fig. 3 shows the dependence of the image file size on the value of the averaging index over non-overlapping spatial column blocks. Characteristics of the structural image from an optical coherence tomograph: in vivo nail and nail bed of a human thumb, 180 A-scans (vertical lines), 35,000 points in A-scan. This dependence was obtained both experimentally and according to the above formulas - for the t × t format, almost one hundred percent coincidence was obtained, that is, both the absolute and relative errors are almost zero.

The number of averaged in one block column is determined depending on the number of A-scans, and, for example, for 900 A-scans is 3-10. The number of averaged over one block line is determined depending on the depth of coherent sounding and most often is 2-3. As a result of this processing - averaging over two coordinates, the image contrast increases and the level of speckle noise decreases.

The advantages of the proposed method include the fact that with the selected rational compression, the quality of the structural image is not only not lost, but even increased. This is primarily due to the fact that raster averaging significantly (by 10–20 dB) reduces the phase noise of a low coherent infrared radiation source from an optical coherent tomograph and greatly reduces speckle noise. This effect is partly due to the fact that compression is part of the image construction process, and therefore operations are performed on the intensities for the source data of each point of the structural image from an optical coherent tomograph, and not with pixel color codes, and partly due to the features of the small-angle raster scanning and averaging process itself . As an example, illustrating the foregoing, FIGS. 4 and 5 show structural images of a human nail and nail bed in vivo with different values of the averaging index over block rows and column blocks: (a) - image constructed without averaging - 500 × 900 pixels; (b), (c) and (d) - with averagings of 2, 3, 4 neighboring non-overlapping spatial blocks-columns, respectively, (e) and (g) - with averaging of 2 and 4 neighboring non-overlapping spatial blocks-columns, respectively; (f) - optimal image (the averaging parameter for block rows is 4 and for block columns is 2); (h) - an excessively averaged image (the averaging parameter for both block rows and column blocks is 4) - as a result, the quality is partially lost. Image size 2 × 2 mm ² . The blue circle marks the capillaries of the nail bed, which become visible as a result of applying averaging methods.

To confirm such a high efficiency of averaging in small-angle raster scanning, we analyze how the signal-to-noise ratio changes by the example of several structural images from an optical coherent tomograph. Figures 6 and 7 show graphs of the dependence of the signal-to-noise ratio on the averaging index over column blocks for the first and last 5000 points taken from 180 A scans of two identical scans of two different in vivo bioobjects: a human nail and a blood vessel human palms, respectively. In each of the A-scans, there are only 35,000 points, the first and last 5,000 points have a maximum of noise. On the graphs, you can see a sharp increase in the signal-to-noise ratio for small values of the averaging index (up to 30) and a smooth growth further up to averaging all vertical lines into one. Also, judging by these graphs, there is no strong saturation in the signal-to-noise ratio. Moreover, with an increase in the number of A-scans, this trend continues.

The present invention can be used in optical coherent tomographs, scanners for ultrasound studies and in other similar devices for rational visualization of the internal structure of a biological object.

Thus, the use of the small-angle raster scanning method in the arm of a sample of an optical coherent tomograph to obtain a structural image of size P _ex bytes, element-wise averaging of non-overlapping spatial column blocks and row blocks and control over this process based on the exponential dependence of the size of the resulting image P _res ( byte) from the number of averages of neighboring non-overlapping spatial column blocks U, increase the signal-to-noise ratio and, as a result s, improve the quality of the structural image bioobject in optical coherence tomography.

Claims (1)

A method of obtaining a structural image of a biological object in optical coherence tomography, which consists in dividing the original color video frame into non-overlapping spatial blocks consisting of more than one pixel, characterized in that a small-angle raster scanning method in the shoulder of an optical coherent sample is used to obtain a structural image tomograph of size P received _outgoing byte image is divided into non-overlapping spatial blocks only under Art lbtsam, neighboring columnar blocks pixel by pixel are averaged, thereby forming a new image, a new image is divided into non-overlapping spatial blocks only by rows, adjacent blocks row pixel by pixel are averaged to form an output image size P _Res bytes, and the averaging process is controlled by an exponential function the size of the resulting image P _res bytes from the number of averages of neighboring non-overlapping spatial column blocks U:

,
where n and m are, respectively, the number of rows and columns in the image with the size P _ref ; $$P_{i,j}^n$$

- the size of one pixel after averaging, bytes; i and j are the row and column indices, which are the coordinates of the pixel in the image size P _rez , bytes.

Images