RU2585700C2 - Method of diagnosing pathology of paranasal sinuses by recognising images - Google Patents

Abstract

FIELD: medicine.

SUBSTANCE: invention relates to medicine, X-ray imaging, digital image processing techniques and statistical methods of image recognition, can be used for diagnosing pathologies of maxillary and frontal sinuses. On X-ray image separating an “area of interest” by contours of sinus (P) and same orbit (O), latter is used as a reference image (EO) for identification and estimation of degree of pathology P. Digitising image contours of sampling and quantisation amplitude intensity pixel brightness on basis of resolution of image source and capacity of computer. Constructing histograms - statistical portraits images of contours P and O with evaluation of their brightness densities M, dispersions D, mean-square deviations (MSD) σ, density factor K_pl (ratio of density M_p of analysed contour P to density M_o of contour of same O, expressed in units of gray scale). Identifying state P from value K_pl= M_p/M_o. For K_pl < 1±σ fixing norm, for K_pl> 1 +σ - pathology P. To increase resolution of diagnosis of condition P image of selected contours of similar P and O are presented as a result of digitisation of same-size matrices of brightness. Forming variation series (VR) with evaluation of span of every VR Δx = x_max-x_min. Determining ratio of extreme elements of series η= x_max/_min, in primary approach by estimating condition P: norm - for Δx≅0, η≅1; pathology- Δx≠0, η> 1. Further, nature of pathology P is assessed quantitatively with coefficient of variation of V (ratio of SKO σ_x to density M_x of corresponding histogram). Pairs of similar contours P and O having a multimodal histogram, segmented by decomposition of each corresponding contour of matrix of intensity values of brightness into L submatrices size n_l×n, where n_l= n/L, n×n - dimension of initial matrix of contour. Constructing L corresponding histograms and estimate their quantitative characteristics M_l, D_l, σ_l, V_l, where l is number of segment. By comparative analysis of histograms segments of contours O are selected as OE of such l-th segment, histogram of which is homogeneous with minimum dispersion D_ol→min and coefficient of variation of V_ol→min. Fixing parameters of its histogram, performing differential topological decomposition of overall K_pl of analysed P into L coefficients of density of segments, each of which is defined by ratio of density M_pl of histogram of l-th segment of loop P to density of histogram of segment of contour O M_ol, selected as OE: K_pll=M_pl/M_ol. Based on topologic series of values K_pl and V segments of contour P {K_pl1, K_pl2, …, K_plL}; {V₁, V₂, …, V_L} estimating state P by local interpretation of its content.

EFFECT: method provides quantitative assessment of content paranasal P, degree of pathology thereof, localisation accuracy and differential diagnostics, possibility of classifying analysed pathology.

1 cl, 17 dwg, 3 ex, 2 tbl

Description

The present invention relates to medicine, namely to otorhinolaryngology, and can be used to identify inflammatory pathology of the paranasal sinuses (SNPs).

Inflammatory diseases of ONP - paranasal sinusitis (PNS) are the most common inflammatory pathology in otorhinolaryngology (Piskunov GZ, Piskunov SZ Clinical rhinology / M .: Miklosh, 2002. - 390 p.). The number of orbital complications of PNS in the last decade does not have a steady downward trend, especially in the region of the North Caucasus (Volkov A.G., Gyusan A.O. The course and treatment of exudative frontal sinusitis in the North Caucasus // Basic Research. - 2005. - No. 2. - S. 115-116), where, for a number of reasons, PNS of the anterior sinus group is especially common. In recent years, there has been a significant increase in inflammatory diseases of the nose and SNPs, both in absolute numbers and in the specific gravity in the structure of the total ENT morbidity and is estimated at 52.7% (Ternovoi S.K., Arablinsky A.V., Sinitsin V .E. Modern radiation diagnosis of diseases of the paranasal sinuses. - M .: 2004, - 120 p.).

The main methods in the diagnosis of SNPs are radiation methods. For many decades, in the diagnosis of these diseases, a panoramic radiography of the paranasal sinuses has been used (Piskunov I.S., Piskunov B.C. Clinical anatomy of the ethmoid and sphenoid bones and the sinuses that form in them. - Kursk, 2001. - 296 p.).

Conducted by research on medical, scientific and patent literature, we have identified various methods for the diagnosis of maxillary and frontal sinuses. Imagine their analysis of the essences, advantages and disadvantages.

A known X-ray method for the diagnosis of PNS (Fayzullin M.Kh. Radiodiagnosis of diseases and injuries of the adnexal nasal cavities. - M .: Medicine. - 1969. - S. 5-8), which consists in obtaining a shadow image of the paranasal sinuses - radiographs on a photosensitive film during transmission sinuses by x-ray radiation and subsequent analysis of the results of the study The method is implemented as follows: perform the laying of the patient’s head so that the central radiation flux of the x-ray tube passes through the sagittal the plane of the skull at the height of the palpebral fissure, exposure is performed, and then the image is processed. The state of the SNP is assessed by visual comparison of the x-ray transparency in the sinus and the same orbit, the transparency of which is taken as a conditional (reference) norm.

An essential sign, common with the essential features of the proposed method, is the selection of contours on the x-ray image from the area of the sinus and orbit images, determining the state of the sinus through a comparative analysis of the transparency of the sinus region and the same orbit.

The reasons that impede the achievement of the technical result and are the disadvantages of this method are as follows:

- intuitive-quality, based on professional experience, the principle of making decisions on assessing the condition of the sinus;

- the dependence of the reliability of the diagnosis on the quality of the image taken and the technology for developing the film;

- subjectivity when a doctor identifies the contours of diagnosed fragments.

The advent of digital x-ray machines compared with those of traditional technology with the fixation of the image of an object on a film during panoramic x-ray analysis made it possible to obtain a number of advantages in diagnostics (Kirillov S.N., Oreshkov V.I. Application of combined processing methods to improve the quality of digital diagnostic images // Biomedical technologies and radio electronics. - 2005. - No. 7. - S. 37-40):

- weakening the radiation load on the patient by 50-70%, which is associated with high sensitivity of digital x-ray sensors, as well as a decrease in environmental impact;

- simplification of image processing, in particular, a long and complicated film development technology and the need for the disposal of chemical waste, and accordingly the presence of a darkroom, are excluded;

- high speed of obtaining a digital image (less than one minute);

- the image in digital form can be stored for an arbitrarily long time on the computer, CD-ROM, flash memory with fast access;

- the ability to calibrate, filter, standardize a digital image;

- Easily and quickly transfer images over computer networks, the Internet.

The latter contributed to the discovery of a new direction in medicine - telemedicine, when a diagnostic image can be transferred for consultation from the district hospital to the medical centers of large cities.

Over the past two decades, computed tomography (CT) has found wide application in the diagnosis of diseases of SNPs (Ternovoi S.K., Arablinsky A.V., Sinitsin V.E. Modern radiation diagnosis of diseases of the paranasal sinuses. - M. - 120 p .; Trufanov G.E., Alekseev K.N. Radiation diagnosis of diseases of the paranasal sinuses. - 2nd edition, ELBI: St. Petersburg, 2011 .-- 234 p.). This highly informative research tool allows you to visualize pathological changes in SNP in 2D, 3D technologies. In this regard, the role in the preoperative diagnosis of pathological changes in SNP, the determination of the topographic anatomy and options for the anatomical structure of the nose and SNP in mandatory x-ray styling for their study in coronary and axial projections has increased.

However, both in digital radiography and CT, X-ray images are recorded and processed to facilitate visualization of images in the non-optical range, and they do not provide the selection of information necessary for solving diagnostic problems.

The semantics of biomedical images on X-ray and CT images are examined and interpreted visually by the radiologist and provides information to the clinician. Such a sequence of image analysis can be a source of false interpretations not only in the analysis of the material, but also in its transfer.

It should be noted the high cost and inaccessibility of CT equipment to a wide mass of patients of the Russian Federation. The subjectivity of the interpretation of survey X-ray and CT data significantly reduces their diagnostic value and is the source of 30% of erroneous decisions (Korolyuk I.P. X-ray atlas of the skeleton (norm, options, interpretation errors) / M .: Vidar, 1996. - 192 p .; Volkov A.G., Krasnopolsky OV Additional opportunities to reduce errors in the radiological diagnosis of frontal sinusitis // Collection of medical and scientific-practical conf., Dedicated to the 80th anniversary of the city hospital No. 1. - Rostov n / D, 2002 . - C 31).

, где m, n - строк и столбцов массива, i, j - индексы массива изображения по строкам и столбцам соответственно. Такое представление снимков открывает большие возможности в плане их математической обработки с целью разработки новых методов поиска смысловой информации для достижения высокого уровня достоверности диагноза воспалительных патологий ОНП.In relation to the proposed method, film, digital x-ray machines and a computer tomograph, as well as electronic, magnetic and optical storage media are image sources for subsequent diagnostic analysis. The image is described by the function of two coordinates f (y, z), stored and processed in a computer in the form of an array of points (pixels) quantized by brightness and discretized by the topology of the image

, where m, n are the rows and columns of the array, i, j are the indices of the image array by rows and columns, respectively. Such a presentation of images opens up great opportunities in terms of their mathematical processing in order to develop new methods of searching for semantic information in order to achieve a high level of reliability of the diagnosis of inflammatory pathologies of SNPs.

A known method for the study and diagnosis of a biological object and its parts (application No.201515575, publication date 10.07.2012, authors: Natarovsky S.N. (RU), Belyakov V.A. (RU), Manturova HE (RU)), including yourself the following steps:

1. obtaining an image of a biological object or part thereof by means of an optical-mechanical device, a microscope, an X-ray apparatus, an ultrasonograph, or by taking an image from a magnetic, electrical or optical information carrier;

2. The image of the object or its part is obtained by sampling in space and quantization in amplitude;

3. calibration, standardization and filtering of the received image;

4. computer morphometric and densitometric image research;

4.1. identification of "zones of interest" in the resulting image;

4.2. the establishment of the necessary ranges of characteristics of the signal intensity according to the fixed distribution of the characteristics of the interaction of radiation with matter in the image;

4.3. dividing the "zone of interest" into areas with the definition of morphometric and densitometric image parameters within the areas;

4.4. comparing the data of the investigated object with similar indicators of the comparison group;

5. Based on the results of the comparison, a decision is made about the state of the investigated object or its part.

The reasons that impede the achievement of a technical result are:

- low diagnostic level of the investigated object. In accordance with the restrictive part of the claims of the known method, the identification of the state class of the object under study (or part thereof) is carried out by comparing the parameters (apparently densitometric and morphometric) characterizing the objects forming the comparison group and belonging to different classes of states. The state of the investigated object (or its part) is identified by the class of the object (its part) of the group with the closest (equal) parameters to the studied. The number of objects in the comparison group is integer and presumably can be estimated as N = 2 ^m , where m is the number of parameters.

- in the presented claims of the known method does not emphasize the nature of the comparison (multiplicative or additive) indicators of the studied object (or part thereof) and objects of the comparison group:

- in case of additive comparison, the error value is “secretive” and depends on the image quality of the studied object and the compared ones;

- with a multiplicative nature, partial compensation of these errors is possible;

- unfortunately, the functional relationship between the values and the number of parameters with state classes has not been disclosed (there is no diagnostic model);

- the authors of the known method do not indicate in which area (spatial, temporal or frequency) of the image the proposed method works;

- as follows from the distinctive part of the claims of the known method, obtaining series of images of the investigated object (its part) in various radiation spectra receive reflection from its appearance, which limits the scope, since our proposed method is designed to diagnose the state of the internal structure of the bony tissues of the paranasal sinuses ;

- the method in the implementation, in our opinion, is quite inaccurate, structurally difficult to implement, because:

- calibration of the image of the object or its part is necessary for different radiation spectra, which will certainly lead to the need to recalculate the error in determining the ranges of densitometric parameters, in particular, when determining the quantization step, this action will lead to a change in the discharge grid for each spectral region;

- the claims of the known method are generalized, in our opinion, incorrect, for example, how to link paragraph 1 and paragraph 4: according to paragraph 1 receive an image of the object in various spectra, and according to paragraph 4, the image is obtained through a digital x-ray apparatus?

- the incorrectness of clauses 7 and 8: the source of interfering interference and the nature of their manifestation are not clear, therefore, the implementation of filtering is difficult.

The essential features common with the essential features of the proposed method are:

- obtaining an image of an object through a digital x-ray machine or tomograph;

- the image is obtained by discretization by space (morphometric parameter) and quantization by the amplitude of the brightness intensity (densitometric parameter);

- determine the range of characteristics according to the distribution of the densitometric parameter (the intensity of the brightness of the area corresponding to the normal state of the SNP and the brightness intensity for various classes of sinus conditions with inflammatory pathologies);

- a comparative comparison of the indicators of the study area with the indicators of the reference, followed by identification of the state of the study area.

Closest to the proposed method is a method for the diagnosis of inflammatory pathology of the maxillary sinuses (Patent RU No. 2234859, IPC A61B 6/00 of 10/21/2002, publ. 10/27/2004 // Palchun V.T., Magomedov M.M., Petukhova P. B.), which consists in the fact that it represents the following sequence of actions.

Survey radiographs of the paranasal sinuses are performed in the nasal-chin projection. On the roentgenogram, “zones of interest” are distinguished, the so-called contours of the maxillary sinus and the same orbit. The method is implemented by means of a scanner (if the image of the diagnosed object is, for example, on the developed photosensitive film and the memory of a personal computer), a printer is also possible for printing the results of the diagnostic process. Using a special computer program, densitometry of the sinus and orbit of the same name is performed, i.e. display the density of their tissues in gray scale units. The contour areas are displayed respectively by the number of pixels.

Thus, discretization is performed in the space of images of the contours of the sinus and the orbit of the same name. The contours are displayed in pixels (elementary points of the corresponding images), each of which has spatial coordinates in the Cartesian planar system (Y, Z). The contour image can be displayed by a matrix of two-dimensional functions f (y _i , z _j ); i, j ∈n, the value of which at each point will display the intensity (x) of the brightness of the monochrome image in the range of 2 ^k , where k is the bit depth, 2 ^k is the number of gray scale levels.

Then, histograms are constructed - statistical images of the images of the selected contours that display the structure of the sinus tissue and the same orbit. The histogram of the sinus contour determines the arithmetic mean of the intensity of brightness, which characterizes the density of the tissue, for example, the maxillary sinus, in units of the gray scale, similarly determine the density of the intensity of the brightness of the same orbit in gray scale units.

It should be noted that the value of the brightness density of the maxillary sinus, expressed in units of the gray scale, cannot reliably reflect the state of the sinus, since the difference between the density indices for various sinus pathologies is not reliable. The density parameters of the sinus brightness in a X-ray image are also affected by the technical parameters of the X-ray examination (X-ray stiffness, film quality, exposure method, etc.), which equally change the histogram values in the region of the sinus and orbit contours.

And therefore, in order to exclude the influence of the above parameters on the reliability of diagnosis of sinus pathologies, the authors of the known invention proposed to evaluate the condition of the sinuses with a density coefficient. The density coefficient (Kp) is defined as the ratio of the image density of the sinus under study to the image density of the orbit of the same name, taken as a reference image.

In this method, when computer processing the X-ray of SNPs, a quantitative value is obtained, called the density coefficient, which characterizes the presence and degree of pathology in the maxillary sinus. Moreover: at values of the coefficient of density Kp = 1.03 ± 0.12, the norm is diagnosed; Kp = 1.78 ± 0.2 - polyps of the mucosa of the maxillary sinus without obstruction; with Kp = 1.83 ± 0.21 - hyperplastic mucous membrane; the value of Kp = 2.28 ± 0.19 - corresponds to the purulent contents in the maxillary sinus, that is, the severity of the pathological process and the nature of the exudate can be quantified in the form of a class scale.

In addition to the quantitative value characterizing the presence of a pathological process in the maxillary sinus in the form of a histogram (a curve of the intensity density of pixel brightness), a qualitative intuitive analysis of the contents of the maxillary sinus is carried out.

A histogram with one extremum (single-humped) displays the homogeneity (homogeneity) of the contents (or exudate, or air, or polyps), and the density coefficient for the entire maxillary sinus is taken into account in the diagnosis of sinusitis and is a reliable sign only if the histogram has the form " one-humped curve.

Inhomogeneity (heterogeneity) of the contents of the maxillary sinus in the known method is detected in the form of various curves in the histogram, that is, the histogram has several extrema and the corresponding values of the densitometric parameter (mode). The histogram in this case is called multimodal (multi-humped - in the author’s edition).

The presence of a "multi-humped" histogram that displays the image of the sinus contour indicates the presence in the sinus of several structures (tissues) of different densities (air + exudate, air + polyps, etc.), the heterogeneity of the sinus density makes it necessary to perform a computer analysis of each of humps ”histograms with determination of density coefficients for a given section of the maxillary sinus. A comparison of the density coefficients obtained by computer processing the “multi-humped” histogram with the values of the density coefficient corresponding to the state of the sinuses in the “normal” allows us to judge the nature and form of the inhomogeneous contents of the maxillary sinus and the urgent need to continue the study.

The reasons that impede the achievement of a technical result are:

- a qualitative analysis of "one-humped" histograms leads to ambiguity in the diagnosis;

- poor-quality processing of the photosensitive film, heterogeneity of the tissues of the orbit of the same name during digital radiography leads to the appearance of a multi-hump histogram (statistical image) of the orbit, which is used as a reference image in assessing the condition of the sinus and quantifying the degree of its pathology, which leads to a decrease in the reliability of diagnosis, as well as to unreliable diagnostics assessment of the degree of inflammatory processes;

- the heterogeneity of the tissue structure displayed by the multi-hump histogram is evaluated intuitively, that is, in a qualitative manner.

Common signs with the claimed method are:

- X-ray method for obtaining images of parts of the investigated object, in the form of survey radiographs;

- the allocation of "zones of interest" on them through the contours of the sinus and the same orbit;

- the use of the orbit contour as a reference image in identifying the class and assessing the degree of pathology;

- obtaining a digital model of the sinus and orbit by sampling the contour plane in space and quantizing it according to the amplitude of the brightness intensity;

- construction of histograms of the contours of the studied sinus and the same orbit with subsequent assessment of their quantitative characteristics (density, that is, arithmetic mean, and dispersion);

- assessment of the condition of the sinus: with Kp <1.03 ± 0.12, the condition of the sinus as normal, with Kp≥1.15 - the presence of pathology;

- the formation of a quantitative scale of density coefficients for the subsequent classification of pathologies and identification of the class of pathology.

A common integral feature of the known and proposed methods is the process of recognizing the inflammatory pathology of the paranasal sinuses based on a comparison of a posteriori information (histograms, density coefficients) of the diagnosed object and objects with fixed, a priori and a posteriori characteristics based on the pattern recognition principle.

The problem to which the claimed invention is directed, is to increase the effectiveness of decision-making by an ENT doctor in the differential diagnosis of pathology of the paranasal (maxillary and frontal) sinuses.

Recall that in accordance with the work (Gorelik A.Ya., Skripkin V.A. Recognition methods. - M.: Higher school, - 2004. - 261 p .; Mazurov Vl.D. Mathematical methods of pattern recognition. - Yekaterinburg. - 2010. - 101 p.) In medical diagnostics, the purpose of the recognition system is to obtain the information necessary to make certain decisions about whether an unknown object (sinus pathology) belongs to a particular class (pathology). The rule that each object (its state) is associated with a particular class name is called a decision rule. In the method of medical diagnostics, recognition is not an end in itself, but a means of obtaining initial information necessary for making managerial decisions by a doctor, in other words, it is a system of measures of information support for an otorhinolaryngologist.

The technical result achieved by the implementation of the claimed method is:

- the possibility of a quantitative assessment of the contents of the SNP by introducing a coefficient of variation that allows you to assess the degree of uniformity of the contents of the investigated circuit;

- to increase the efficiency of differential diagnostics with uniformity of the contents of the SNP by increasing the accuracy of localization of the area of the alleged pathology by scalable segmentation of the sinus and orbit contours of the same name in the X-ray image under study.

The technical result is achieved due to the fact that in the proposed method for diagnosing inflammatory pathology of the maxillary sinuses, the images of the selected contours of the same pair of sinus and orbit contours are presented as the result of digitization of the intensities of the brightness in the decimal number system for visual analysis of the topological distribution of the densitometric parameter, respectively forming the variation series followed by evaluation of a number of variations of each amplitude Δh = x _max -x _min, GER fissile element number ratio extreme η = x _max / x _min, the approximation is performed in the primary evaluation of the sinuses: OK - at Δh≅0, η≅1; the presence of pathology Δx ≠ 0, η> 1, when confirming the diagnosis, the nature of the sinus pathology is additionally quantified by the coefficient of variation V, determined by the ratio of the standard deviation σ to the density M of the corresponding histogram, thereby increasing the resolution of the diagnosis of sinus condition.

The technical effect of the claimed method is represented as an increase in the pathology image recognition effect due to segmentation of a pair of sinus and orbit contours of the same name having a multimodal histogram by decomposition of each corresponding brightness matrix intensity matrix into L submatrices of size n _l × n, where n _l = n / L, n × n is the dimension of the initial contour matrix; according to densitometric data of the submatrices, L histograms corresponding to them are constructed and evaluated by their quantitative characteristics M _l , D _l , σ _l , V _l , where l is the segment number nt, and by means of a quantitative and qualitative comparative analysis of the histograms of the segments of the orbit contours, the lth segment is selected as a reference image, the histogram of which has a homogeneous character with a minimum dispersion value D _ol → min and the coefficient of variation V _ol → min and fix the parameters of its histogram, then a differential topological decomposition of the total density coefficient K _{pl of} the sinus under study into L segment density coefficients is made, each of which respectively determines sheniem density M _pl, histograms l-th segment sinus contour to the contour segment density histograms orbit M _ol, selected as the reference image K _{pl l} = M _pl / M _ol, for topological row density values of the coefficients and variations sinus path segments {K _PL1 , K _pl2 , ..., K _plL }; {V ₁ , V ₂ , ..., V _L } edit the sinus condition assessment and thereby localize the contents of the diagnosed sinus.

The method is supplemented by drawings explaining all stages of the differential diagnosis of SNPs based on pattern recognition of pathology.

In FIG. 1. An example of the allocation of "zones of interest" in a survey roentgenogram by contouring the studied SNPs and homolateral orbits of the same name.

In FIG. 2. An example of segmentation of the studied orbit (A) and SNP (B).

In FIG. 3. Histograms and their numerical characteristics corresponding to segmented areas.

In FIG. 4. Survey radiograph of patient K., on which the contour of the studied areas (sinuses) and standards (orbits) was performed.

In FIG. 5. Histogram models corresponding to left-side SNP and patient K.'s orbit of the same name.

In FIG. 6. Histogram models corresponding to the right-side SNP and the orbit of patient K.

In FIG. 7. Survey radiograph of patient A., on which the contour of the studied areas (sinuses) and standards (orbits) was performed.

In FIG. 8. Histogram models corresponding to the right-side ONP and the same orbit of patient A.

In FIG. 9. Segmentation of the right-side ONP of patient A., with the subsequent obtaining of histogram models for each segment

In FIG. 10. Histogram models corresponding to the left-side SNP and the same orbit of patient A.

In FIG. 11. Segmentation of the left-sided orbit of patient A., followed by histogram models for each segment.

In FIG. 12. Segmentation of the left-side SNP of patient A., with the subsequent obtaining of histogram models for each segment

In FIG. 13. Survey radiograph of patient M., on which the contour of the studied areas (sinuses) and standards (orbits) was performed.

In FIG. 14. Histogram models corresponding to the right-side ONP and the same orbit of patient M.

In FIG. 15. Segmentation of the right-sided orbit of patient M., followed by obtaining histogram models for each segment.

In FIG. 16. Segmentation of the right-side ONP of patient M., with the subsequent obtaining of histogram models for each segment.

In FIG. 17. Histogram models corresponding to left-side SNP and patient’s orbit of the same name M.

The method is implemented as follows.

An overview radiograph (computed tomogram) is made in the projection necessary for the clinical study of SNPs. If the X-ray image is obtained on a physical medium (film), it is digitized based on well-known scanning technologies.

On the digital image obtained by a clinician using specialized software, “zones of interest” are selected by contouring the studied SNP and the orbit of the same name (Fig. 1), while the orbit acts as a reference sample for identifying and assessing the degree of pathology of the sinus of the same name.

The highlighted “zones of interest” of the digital X-ray image are processed in the form of discretization according to the space of the selected contours and quantized according to the degree of brightness of pixels on a gray scale and are ultimately presented as one-dimensional matrices of pixel brightness values

where the size of the matrix n is determined by the discretization step of the studied area of the image, and the values of the elements of the matrix x ₁₁ ... x _{nn are} determined by the quantization step or the selected number of levels of the gray scale range.

(Вентцель Е.С., Овчаров Л.А. Теория вероятностей, 1969; Кобзарь А.И. Прикладная математическая статистика. - М.: ФИЗМАТЛИТ, 2006).According to the data of the corresponding matrices, a variational (increasing) series of values of matrix elements is found

(Ventzel E.S., Ovcharov L.A. Probability Theory, 1969; Kobzar A.I. Applied Mathematical Statistics. - M.: FIZMATLIT, 2006).

Then, for each variational series, the amplitude is determined as the difference between the extreme values of the elements of the series Δx = x _max- x _min and their ratio η = x _max / x _min , which in the primary approximation allows us to estimate the state of the studied area: for Δх≅0, η ≅1 suggest the norm; when Δx ≠ 0, η> 1 - the presence of pathology.

Each variation series in the subsequent action is divided into a finite number of intervals

where h = 1 + log ₂ n is the value of the interval according to the Sturgess formula (Kremer N.Sh. Probability theory and mathematical statistics. - M.: UNITY-DANA, 2003, - 543 p., p. 265)

After that, we count the number of values n _{i of the} densitometric feature that fall into each of k intervals:

Further, on the basis of the obtained data, histograms of the sinus and the same orbit of the densitometric parameter distribution are constructed depending on the level of pixel brightness intensity on a gray scale (from 0 to 255), while the histograms become the most complete statistical portrait of the selected contours and allow:

- visualize the studied fragments;

- evaluate their brightness probability distribution;

- compare with other diagnostic and informative areas of the x-ray and assess the condition of the patient;

- by the type of histogram to judge the presence or absence of extraneous inclusions in the field of research: with a single-mode histogram - homogeneity (homogeneity), with a multimodal - heterogeneity (heterogeneity) of the content.

To classify the contents of the studied "zones of interest", each histogram is evaluated by quantitative characteristics:

- the density of the intensity of brightness, respectively, of the sinus and orbit:

- variances

и

and

- standard deviations

Then determine, as in the prototype, the density coefficient (K _p ) the ratio of the density of the intensity of the brightness (M [X _p ]) of the values of the studied and reference (M [X _about ]) of the same name contours:

By the value of the density coefficient K _PL , the state of the sinus is identified (with a value of K _PL <1.02 ± σ, the condition is assessed as normal, with K _PL ≥1.02 ± σ, the presence of pathology in the corresponding sinus is diagnosed).

For a visual quantitative analysis of the spatial distribution of the contents of the pathological sinus and an assessment of the degree of uniformity of brightness of the pixels of the image of the contour of the orbit, the degree of heterogeneity of the images of the sinus and orbit is quantified respectively according to the variation coefficients:

where σ [X _p ], M [X _p ] is the standard deviation and density density of the brightness of the pixels of the contour of the studied sinus.

The larger the value of the coefficient of variation, the relatively greater the spread and less uniformity of the studied values of the densitometric parameter. Based on statistical studies (Samoilenko A.P., Volkov A.G., Puzhayev S.I., Pribylsky A.V. Statistical method for pattern recognition in the diagnosis of patients' frontal sinuses. - In the book: Mater. Inter-republic. Practical conference of the otorhinolar of Siberia and the Far East with the international participation “Actual issues of the otorhinolar” (July 1, 2013). - P. 12-14) the following scale for assessing the degree of heterogeneity of the contents of the studied contours by the value of the variation coefficients can be established: the content of the studied area is considered uniform if the values of k coefficients of variation are in the range from 0 to 0.2; when the coefficient of variation exceeds a threshold value of 0.2, we can talk about the heterogeneity of the content and, accordingly, the presence of extraneous artifacts in this area.

The multimodal histogram by the number of modes is characterized by a set of values of the coefficient of variation V [x] _i (i = 1 ... n, where n is the number of modes) and density coefficients K _pi .

This raises the problem in the classification of the obtained groups of density coefficients and variation coefficients in order to localize the alleged pathology against the general background of the contour of the corresponding sinus with heterogeneity of its contents. To resolve the issue of differential distinguishing between the nature and degree of pathologies in the proposed method, a pair of the same sinus and orbit contours of the X-ray image is segmented, the histogram of which is heterogeneous.

This operation is implemented as follows.

. Числом сегментов можно варьировать.The original matrices of densitometric attribute values of the matrix of the same orbit (S ^о ) are decomposed into L submatrices (Fig. 2a) of dimension n _L ∗ n, where

. The number of segments can vary.

и производят последовательность действий, аналогичных при анализе всего контура в целом.According to the data of the obtained segments (submatrices), variational series of values of the elements of the submatrices are found

and produce a sequence of actions similar to the analysis of the entire circuit as a whole.

при этом посредством количественного и качественного сравнительного анализа гистограмм сегментов контуров орбиты выбирают в качестве эталонного образа такой l-й сегмент, гистограмма которого имеет гомогенный характер с минимальным значением дисперсии D[X_ol]→min и коэффициента вариации, V[X_ol]→min, где l≡Adress l-го сегмента эталонного образа), и фиксируют соответствующее значение плотности интенсивности яркости пикселей M[X_ol]|_{min Vol}, тем самым принимая его за эталонное значение в последующих исследованиях.At the next stage, L histograms are constructed from the found variational series (Fig. 3) for each segment of the orbit, respectively, the density of the brightness intensity M [X _ol ], the variance D [X _ol ], the standard deviation σ [X _ol ], and the coefficient of variation V _{ol are determined} histograms of each l-th segment, where

in this case, by means of a quantitative and qualitative comparative analysis of the histograms of the segments of the orbit contours, the lth segment is selected as a reference image, the histogram of which has a homogeneous character with a minimum dispersion value D [X _ol ] → min and coefficient of variation, V [X _ol ] → min , where l≡Adress of the lth segment of the reference image), and fix the corresponding value of the density of pixel brightness intensity M [X _ol ] | _{min Vol} , thereby taking it as a reference value in subsequent studies.

At the next stage, by analogy with the informative segmentation of the orbit, the segmentation of the zone of the studied sinus is made.

The original matrix of values of the densitometric characteristic of the matrix of the same sinus (S ^P ) by decomposition is divided into L submatrices (Fig. 2b).

According to the data of the obtained segments (submatrices), variational series of values of the elements of the submatrices are found

At the next stage, L histograms are constructed for each sinus segment from the variational series found, respectively, the density of the brightness intensity M [ _Xpl ], the variance D [ _Xpl ], the standard deviation σ [ _Xpl ], the coefficient of variation _{Vpl of the} histogram of each l-th are _determined segment.

Each _lth segment of the sinus contour is evaluated by the _lth density coefficient K _pll , as the ratio of the intensities of the brightness intensities of the _lth segment of the sinus contour M [ _Xpl ] and the selected segment of the orbit contour M [X _ol ] | _{min Vol} as an updated reference image. Therefore, the contour of the studied pathological sinus is evaluated by a sequence of density coefficients, coefficients of variation of the sinus segments:

{K _{PL PL2} , K _{PL PL2} , ..., K _{PL L} },

{V ₁ , V ₂ , ..., V _L }.

By analyzing the sequence of density coefficients and variations, an assessment is made of the distribution of pathology according to the topology of the sinus contour.

According to the proposed method, 251 survey radiographs of the paranasal sinuses of patients with various forms of sinusitis (sinusitis, frontal sinusitis) verified by various clinical methods (puncture of the paranasal sinuses, sinusoscopy, operations) were studied. The control group consisted of panoramic radiographs of the paranasal sinuses of 46 healthy volunteers. The results are presented in table 1.

The foregoing is illustrated by the following clinical examples.

Example 1

Patient K., 32 years old, medical history No. 21271, underwent a voluntary study, had no complaints.

A standard x-ray scan was performed in the nasal-chin projection. As a result of the study by the clinician, ENT pathology was not detected.

According to the proposed method, X-ray data were processed in the DiaLPro v.2.0 program (certificate of state reg. Of October 8, 2012 No. 2012619118 Software processor of a mobile diagnostic terminal for information support of an otorhinolaryngologist doctor // Samoilenko A.P., Pribylsky A.V. and etc.). In the presented X-ray image, “zones of interest” of the corresponding SNPs and orbits of the same name are highlighted (Fig. 4).

Analyzing the histogram parameters for the left-side SNP (Fig. 5), we see that the obtained values of the density coefficient K _PL = 1.07 and the coefficient of variation V = 0.19 clearly indicate the absence of foreign inclusions in the study area, and thereby allow us to diagnose the absence of pathologies (normal ) (Table 1).

The processing results of the right SNP group (Fig. 6), namely, the obtained intensity coefficient K _PL = 1.02, V = 0.18, allowed, in accordance with the data in Table 1, to conclude that the contents of the right SNP are uniform and thereby diagnose the absence of pathologies (normal) (Table 1).

Example 2

Patient A., 31 years old, case history No. 12421, was admitted to the hospital with complaints of difficulty in nasal breathing, hyposmia, severe pain when the head was tilted forward, purulent runny nose, fever up to 37.5 ° C.

A panoramic x-ray of the patient's paranasal sinuses was performed in the nasal-chin projection. Visually on the roentgenogram, filling of the right-sided and left-sided maxillary sinus with extraneous inclusion (exudate) was observed, which, combined with the provided symptoms, made it possible to make a diagnosis: acute bilateral purulent maxillary sinusitis.

According to the proposed method, the X-ray data were processed in the DiaLPro program (v.2.0.) The clinician identified “zones of interest”: the contours corresponding to the X-ray contours of the ONP and the contours of the same orbits (Fig. 7).

For each region, histogram models were constructed and numerical parameters were found that characterize the distribution of the intensity of the brightness of pixels in the selected zone of interest: expectation, variance, standard deviation, coefficient of variation, density coefficient.

The visual superposition of the histograms of the right-side ONP and the corresponding orbit of the same name, as well as the obtained numerical characteristics of these contours (Fig. 8), in particular, the coefficient of the density of pixel brightness intensities К _П = 1.77, made it possible to conclude the presence of foreign inclusions in the studied area and the need for further diagnostic actions to localize pathology against the general background of the sinus contour.

In accordance with the proposed method, an assessment was made of the uniformity of the contents of the studied orbit contour. The obtained value of the coefficient of variation of the orbit V _o = 0.13 indicates that there is no need to edit the reference image, so the previously recorded value of the mathematical expectation M _o = 62.7 remains unchanged.

The coefficient of variation of the right-side SNP V _P = 0.28 indicates the heterogeneity of its contents. Decomposed the right-side ONP of patient A (Fig. 9) into 7 segments, for each of which a histogram was constructed, its mathematical expectation and coefficient of variation were determined. To assess the topological distribution of pathology along the sinus contour, we calculated, taking into account the fixed value of the mathematical expectation of the standard M _o = 62.7, the sequence of density coefficients {K _PL1 = 2.50, K _PL2 = 2.20, K _PL3 = 2.20, K _PL4 = 2.36, K _PL5 = 2.45 , K _pl6 = 2.48, K _pl7 = 2.47}; {V ₁ = 0.19, V ₂ = 0.2, V ₃ = 0.16, V ₄ = 0.13, V ₅ = 0.14, V ₆ = 0.15, V ₇ = 0.09}. The data obtained, in accordance with table 1, allowed to conclude that: from 3 to 7 segment of the sinus, the contents are uniform, have no foreign inclusions and are diagnosed as an acute form of purulent maxillary sinusitis; heterogeneity of the contents is observed in the 1st and 2nd segments of the sinus: air / exudate, however, in general, the contents are diagnosed as an acute form of purulent maxillary sinusitis.

The visual superposition of the histograms of the left-side ONP and the corresponding orbit of the same name, as well as the obtained numerical characteristics of these contours (Fig. 10), in particular, the coefficient of density of pixel brightness intensities К _П = 2.17, allowed us to conclude that further diagnostic actions are necessary to localize the pathology on the general background of the contour of the sinus.

In accordance with the proposed method, an assessment was made of the uniformity of the contents of the studied pair of sinus and orbit contours. The obtained value of the coefficient of variation of the orbit V _o = 0.24 suggests the need to edit the reference image.

Produced decomposition orbit circuit 4 segments (. Figure 11), receiving each segment orbit histogram determined mathematical expectation {M _o1 = 59.19, M = 62.40 _{o2, o3} M = 65.32, F = 60.14 _a4}; coefficient of variation {V _o1 = 0.089, V _o2 = 0.087, V _o3 = 0.089, V _o4 = 0.1} histograms of each of the 4 segments, segment No. 2 with the lowest coefficient of variation V _o2 = 0.087 was chosen as the updated reference image this fixed the corresponding mathematical expectation M _a2 = 62.40.

The coefficient of variation of the right-hand SNP V _P = 0.26 indicates the heterogeneity of its contents. After decomposition of the left-side SNP of patient A (Fig. 12) into 7 segments, for each of which a histogram was constructed, its mathematical expectation and coefficient of variation were determined. To assess the topological distribution of pathology along the sinus contour, we calculated, taking into account the fixed value of the mathematical expectation of the standard M _o2 = 62.4, the sequence of density coefficients {K _PL1 = 2.21, K _PL2 = 1.89, K _PL3 = 1.82, K _PL4 = 2.16, K _PL5 = 2.30 , K _pl6 = 2.38, K _pl7 = 2.43} and coefficient of variation {V ₁ = 0.13, V ₂ = 0.2, V ₃ = 0.19, V ₄ = 0.14, V ₅ = 0.12, V ₆ = 0.15, V ₇ = 0.17} for each segment. The data obtained, in accordance with table 1, allowed to conclude that: in 1 and 4 to 7 segments of the sinus, the contents are homogeneous, have no foreign inclusions and are diagnosed as an acute form of purulent maxillary sinusitis; In 2 and 3 segments of the sinus, heterogeneity of the contents is observed: air / exudate, however, in general, the contents are diagnosed as an acute form of purulent maxillary sinusitis.

As a result of the study of the SNP of patient A. by the proposed method, they diagnosed an acute form of bilateral purulent maxillary sinusitis, and also performed a local structuring of the contents of the SNP, revealing the presence of inclusions in the form of air located in the upper sinus.

Example 3

Patient M., 25 years old, medical history No. 12315, was admitted to the hospital with complaints of difficulty in nasal breathing, nasal mucous membranes, temperature 36.9.

An overview radiograph of the patient's SNP in the nasal-chin projection was performed. Visually, on the roentgenogram, the filling of the right maxillary sinus was observed with extraneous inclusion (pus + mucus), which, combined with the symptoms provided, made it possible to make a diagnosis: right-sided maxillary sinusitis.

According to the developed method, the review x-ray of the patient was processed in order to localize and classify the alleged pathology. “Zones of interest” were distinguished: the contours corresponding to the X-ray contours of the SNP and the contours of the same orbits (Fig. 13).

The visual superposition of the histograms of the right-side ONP and the corresponding orbit of the same name, as well as the obtained numerical characteristics of these contours (Fig. 14), in particular, the density coefficient of pixel brightness intensities K _P = 2.16, allowed us to conclude that further diagnostic actions are necessary to localize the pathology on the general background of the contour of the sinus.

In accordance with the proposed method, an assessment was made of the uniformity of the contents of the contour of the right-handed orbit. The obtained value of the coefficient of variation of the orbit V _o = 0.2, indicates the need to edit the reference image. The orbit contour was decomposed into 6 segments (Fig. 15), having obtained histograms of each segment of the orbit, the mathematical expectation {M _o1 = 50.5, M _o2 = 56.5, M _o3 = 59.2, Mo ₄ = 57.3, M _o5 = 55.8, M _o6 = 54.4} and the coefficient of variation {V _o1 = 0.068, V _o2 = 0.084, V _o3 = 0.073, V _o4 = 0.066, V _o5 = 0.074, V _o6 = 0.062}, segment No. was chosen as the updated reference image 4 with the lowest coefficient of variation V _o4 = 0.066, while fixing the corresponding mathematical expectation M _o4 = 57.3.

The coefficient of variation of the right-hand SNP V _P = 0.26 indicates the heterogeneity of its contents. After decomposition of the right-side ONP of patient M (Fig. 16) by 7, for each of which a histogram was built, its mathematical expectation and coefficient of variation were determined. To assess the topological distribution of the pathology along the sinus contour, we calculated, taking into account the fixed value of the mathematical expectation of the standard M _o2 = 57.3, the sequence of density coefficients {K _PL1 = 2.12, K _PL2 = 2.03, K _PL3 = 2.02, K _PL4 = 2.08, K _PL5 = 2.17 , K _pl6 = 2.25, K _pl7 = 2.35} and coefficient of variation {V ₁ = 0.18, V ₂ = 0.24, V ₃ = 0.23, V ₄ = 0.17, V ₅ = 0.19, V ₆ = 0.2, V ₇ = 0.17} . The data obtained, in accordance with table 1, allowed us to conclude that in the sinus segments No. 1, 4, 5, 7, the contents are homogeneous, have no foreign inclusions and are diagnosed as purulent maxillary sinusitis, and in the sinus segments No. 2, 3, 6 heterogeneity of the contents: mucus + pus, however, in general, the contents are diagnosed as purulent maxillary sinusitis.

Analyzing the parameters of the histograms for the left-side SNP (Fig. 17), we see that the obtained values of the density coefficient K _PL = 1.14 and the coefficient of variation V = 0.19 clearly indicate the absence of foreign inclusions in the studied zone and, thus, make it possible to diagnose the absence of pathologies ( norm) in the left-side SNP (table 1).

As a result of the study of the SNP of patient M. by the proposed method, right-sided maxillary sinusitis was diagnosed, and local contents of the right-sided SNP were localized, revealing the presence of inclusions in the form of pus + mucus.

The given examples clearly demonstrate the implementation of the sequence of actions of the proposed method for diagnosing pathologies of the paranasal sinuses by pattern recognition, as well as a high assessment of the effectiveness of making a diagnostic decision for various degrees of ENT pathology.

List of sources used

1. Piskunov G.Z., Piskunov S.Z. Clinical Rhinology / M .: Miklosh, 2002 .-- 390 p.

2. Volkov A.G., Gyusan A.O. The course and treatment of exudative frontal sinusitis in the North Caucasus // Basic research. - 2005. - No. 2. - S. 115-116

3. Ternovoi S.K., Arablinsky A.V., Sinitsin V.E. Modern radiation diagnosis of diseases of the paranasal sinuses - M .: 2004, - 120 p.

4. Piskunov I.S., Piskunov B.C. The clinical anatomy of the ethmoid and sphenoid bones and the sinuses that form in them. - Kursk, 2001 .-- 296 p.

5. Fayzullin M.Kh. X-ray diagnosis of diseases and injuries of the adnexa nasal cavities. - M.: Medicine. - 1969. - S. 5-8

6. Kirillov S.N., Oreshkov V.I. The use of combined processing methods to improve the quality of digital diagnostic images // Biomedical Technologies and Radio Electronics. - 2005. - No. 7. - S. 37-40

7. Ternovoi S.K., Arablinsky A.V., Sinitsin V.E. Modern radiation diagnosis of diseases of the paranasal sinuses. - M. - 120 s .;

8. Trufanov G.E., Alekseev K.N. Radiation diagnosis of diseases of the paranasal sinuses. - 2nd edition, ELBI: St. Petersburg, 2011 .-- 234 p.

9. Korolyuk I.P. X-ray anatomical atlas of the skeleton (norm, variants, interpretation errors) / M .: Vidar, 1996. - 192 p.

10. Volkov A.G., Krasnopolsky O.V. Additional opportunities to reduce errors in the radiological diagnosis of frontal sinusitis // Sat. Mater. honey. scientific-practical conf. 80th anniversary of the city hospital №1. - Rostov n / a, 2002. - C 31.

11. Application for invention No.20154675, publication date 10.07.2012, authors: Natarovsky S.N. (RU), Belyakov V.A. (RU), Manturova H.E. (RU)

12. Patent RU No. 2234859, IPC АВВ 6/00 dated 10.21.2002, publ. 10.27.2004 // Palchun V.T., Magomedov M.M., Petukhova P.V.

13. Gorelik A.Ya., Skripkin V.A. Recognition methods. - M .: Higher school, - 2004. - 261 p.

14. Mazurov Vl.D. Mathematical methods for pattern recognition. - Yekaterinburg. - 2010 .-- 101 s.

15.Ventzel E.S., Ovcharov L.A. Probability Theory, 1969;

16. Kobzar A.I. Applied Mathematical Statistics. - M .: FIZMATLIT, 2006.

17. Kremer N.Sh. Theory of Probability and Mathematical Statistics. - M.: UNITY-DANA, 2003, - 543 p., P. 265.

18. Samoilenko A.P., Volkov A.G., Puzhayev S.I., Pribylsky A.V. A statistical method for pattern recognition in the diagnosis of patients' frontal sinuses. - In the book: Mater. intersub. scientific and practical. conf. otorhinolar. Siberia and the Far East with int. participation “Act. questions otorinolar. "(July 1, 2013). - S. 12-14.

19. Glushakov S.V., Lomotko D.V. Database. - M .: AST, 2002.

20. Mokrozub V.G. Graph structures and relational databases in automated intelligent information systems. - M .: Publishing house "Spectrum", 2011.

21. Certificate of state. reg. dated October 8, 2012 No. 2012619118 The software processor of the mobile diagnostic terminal for information support of an otorhinolaryngologist // Samoilenko A.P., Pribylsky A.V. and etc.

Claims (1)

A method for diagnosing pathologies of the paranasal sinuses by pattern recognition, which includes highlighting “areas of interest” on the X-ray image by means of the sinus and the orbit of the same name, the latter is used as a reference image for identifying and assessing the degree of sinus pathology, digitizing the image of the contours by spatial sampling and quantization by space the amplitude of the intensity of the brightness of the pixels based on the resolution of the image source and the resolution of the computer, the construction of histograms - statistical portrait images sinus contours and orbit estimate of their densities luminance M, dispersions D, standard deviations σ, tightness factor K _pl as the relationship density _Mn investigated sinus circuit to the density M _of circuit homonymous orbit, expressed in terms of gray scale identification the sinus state according to the value of the density coefficient K _PL = M _p / M _o , at values of K _PL < 1 ± σ, the norm is fixed, at K _PL > 1 + σ - sinus pathology, characterized in that in order to increase the resolution of the diagnosis To determine the sinus state, the images of the selected contours of the sinus and orbit contours of the same name are presented as the result of digitization of the intensities of the brightness in the decimal number system for analyzing the topological distribution of the densitometric parameter, and variational series are formed respectively, followed by an estimate of the magnitude of each variation series Δx = x _max -x _min determine the ratio of extreme elements of the series η = x _max / x _min , in the primary approximation, assess the condition of the sinus : normal - at Δх≅0, η≅1; the presence of pathology Δx ≠ 0, η> 1, when confirming the diagnosis, the nature of the sinus pathology is additionally quantified by the coefficient of variation V, determined by the ratio of the standard deviation σ _x to the density M _{x of the} corresponding histogram, pairs of the same sinus and orbit contours having a multimodal histogram are segmented by decomposition each respective contour matrix brightness intensity values for l submatrices of size n _l × n, where _{n l = n / l, n} × n - dimension of the contour of the original matrix, by densitometric d nnym submatrices building L corresponding histograms and evaluate their quantitative characteristics M _l, D _l, σ _l, V _l, where l a segment number, wherein by means of a comparative analysis of the histograms of the orbit path segment is selected as a reference image a l-th segment, which histogram has a homogeneous nature with a minimum value of dispersion D _ol → min and the coefficient of variation V _ol → min and fixed parameters of its histogram, then produce a differential topology decomposition overall density factor K we investigate _mp sinus by L segment density coefficients, each of which is respectively determined by the ratio of the density M _PL , histograms of the l-th sinus segment segment to the histogram density of the orbit contour segment M _ol selected as a reference image K _PL = M _PL / M _ol , according to the topological a number of values of the density coefficients and variations of the sinus contour segments {K _pl1 , K _pl2 , ..., K _plL }; {V ₁ , V ₂ , ..., V _L } edit the sinus condition assessment and thereby localize the contents of the diagnosed sinus.

Images