Actuality

     Main property of objects on medical images is complication. Requirements are therefore pulled out to exactness, reliability and authenticity of results.

     It is for this purpose necessary to appeal to the mathematical methods with the use of the computing engineering, that will allow considerably to accelerate the process of treatment and promote research reliability.

     Automation of image processing allows to accelerate diagnostics of diseases on the early stages, does research more high-quality, enables to specify treatment and management therapeutic processes.

Scientific novelty

     In this work attention is mainly concentrated on the pathological changes of cellular kernel. Under a light microscope the kernel of living cage appears a homogeneous little body in a state of gel with a small spot inwardly - by a karyonucleus. An electronic microscopy brought in a valuable contribution to the study of ultrastructure of nuclear membrane (which appeared double) and chromoplasm - difficult combination of the thinnest filaments with dense granules. However to the real moment there are little developments on automation of treatment of image of cells and construction on this basis of some conclusions in relation to pathological changes in a cell.

The review of the developments

     Tasks of image processing enormous amount of works. Most known labours there are the followings books in this area: U. Prett the «Digital processing of images», Soyfer V.A. the «Computer processing of images». There is possibility of processing of images in the package of Matlab.

     Also there is a row of software developments in this area. For example, SOFTWARE “Videotest-morphology 5.0”, which enables to decide the great number of tasks. Programmatic system of analysis of medical images of "DiaMorf Lens And" allows automatically or semi-automatic device to expect morfo- and densitometric parameters of preparation objects. One of universal instruments is the program “Imageexpertpro”, which allows to do a numerical analysis.

Methods of object detection

     On the circuit all stages of image processing used now are presented.

     1. Colour image analysis.

     We will consider structure of page (frame) of the digital colour map. The colour model of page of the map looks like the following:

     M [ f (m,n)] = { p₀ [ x₀ (m), y₀ (n), c₀ ( r(l), g(l), b(l) ) ],
     p₁[ x₁(m), y₁(n), c₁( r(l), g(l), b(l) ) ],
     . . . ,
     p_k[ x_k (m), y_k (n), c_k( r(l), g(l), b(l) ) ] },
     M - the array of points;
     f(m, n) - function of layout of points;
     p - point consisting of three colours (red, green, blue);
     x - location of point on horizontal string of the map;
     y - location of point on vertical string of the map;
     c - the colour number; l - number of colour gradation level; k - number of points in the map;
     r - red colour; g - green colour; b - blue colour.

     Objects on the map can cause in the sizes, in a configuration and in a colour. The number of objects on one map can fluctuate largely. Following algorithms are offered: 1) definition of presence of seven pure colours on scale RGB; 2) sharing of the initial (complete) map on two maps (grey and purely colour, not having shades of grey colour from black to brightly white); 3) lowering of number of gradation of colour by offset of gradation to a lower layer, an average level, a top level or certain level of colour lost-free points the given colour; 4) search of areas of an agglomerate points the selected colour on the map by usage of spiral scanning; 5) revealing of an outline of the colour object in the selected area.

2. Morphological analysis of image form.

     A task of image form construction is substantial part of morphological analysis. From that, as far as a form is high-quality built, the result of decision of morphological analysis tasks depends substantially.

     One of image form construction methods consists of task of areas an isoluminance on physical properties of object, on the location of homogeneously luminous or reflecting verges or scopes in relation to an observer. Adding these areas various a brightness, will get the form of image as great number of images. If we do not have so detailed information about the object of research, we can build a form on some to one image, knowing, to what transformations of brightness of this image can bring changing terms over of supervision.

     How can we approximate the form of image f, if the image f is no-observed, and his version g distorted noise is accessible to the supervision only? For the decision of this problem will take advantage of by a circumstance that great number piece permanent images is everywhere dense. It means that exists sequence {f_N}, converging to f. Approximation of form V_f consists now of that, to use the form of V_fN piece permanent image f_N, near enough to f, in place of form of image f. The closeness of images f_N and f is here understood in that sense, that their difference f-f_N must be small as compared to noise of g.

     1st method. Let the sequence {A₁,:,A_N}_N, N=1,2,..., more shallow breaking up of eyeshot X, and brightness of C_i i=1,...,N, gets out from the decision of task of the best approaching

     images g piece by a permanent image. The decision of this task are numbers

     2nd method. There is a sequence {C₁,:,C_N}_N, N=1,2,..., of brightnesses of piece permanent image, and the proper {A₁,:,A_N}N breaking up of eyeshot X, and the brightness of C_i gets out from the decision of next task of the best approaching

     3. Method of homogeneous areas search.

     Segmentation is breaking up of image on component parts which have semantic essence. There is plenty of algorithms of image segmentation, but majority from them it is possible to divide into two groups, each of which uses fundamental property of image - likeness and difference. In accordance with it there are two basic going near segmentation: 1) method of homogeneous areas search; 2) method of contour lines selection.

     Segmentation on the method of homogeneous areas search is possible to conduct on some property of S, which characterizes likeness of elements of every area between itself. It can be color, texture, grey level.

     The increase of areas consists of that nearby elements group with identical or near grey levels, uniting them in homogeneous areas. It is thus necessary to avoid the errors of incorrect determination of nearby elements depending on the chosen vinicity Neyman's Background (fourcoherent) or Mur vinicity (eightcoherent). Importance of exact determination is rotined on a fig. 1.

     For the account of errors during segmentation it is necessary to take into account the followings situations:

     1. Three areas is a black contour and two white areas (Fig. 1, a).

     2. There is an ambiguousness: if to choose neighbours the vinicity of Mura, black elements make a joint contour, and on a Neyman's Background vinicity - 4 separate rectangle. A paradox consists of that a white element (on the Mur vinicity) inwardly will be related to the external white area (Fig. 1, b).

     Image segmentation on the parameters of brightness takes into account that every display group moves away from other nearby. For segmentation of the threshold distributing a method it is necessary to get a binary image from a half-tone. Some threshold value is for this purpose set. After a quantum a function of image G (i, j) = k (integer values) is at T_k > G(i,j) >= T_k-1, k ª (0, k_max), where T_k is a value of k threshold level. In the case of k_max = 1 digitised image is named binary. Display with the level of brightness elements anymore threshold take on a value 1, less threshold - 0.

Review of results

     As a result of implementation of magistr's work is the method of determination of object, which was got on a cytologycal pictures, and also calculation of row of parameters which allow to conclude about a possible disease.

     The example of painted kernels determination is presented on a histological picture:

Literature

Biography    Library    References    Search    Individual task    Russian    Ukrainian

Up