RU2843800C1 - Method for three-dimensional heart reconstruction for non-invasive electrophysiological examination in atrial and ventricular tachyarrhythmias - Google Patents

Abstract

FIELD: medical science.

SUBSTANCE: invention refers to medicine, namely to image processing and data analysis systems and methods, and can be used for medical imaging image data processing. Disclosed is a method in which a non-contrast plain computed tomography is performed, data from the DICOM medical image transmission international standard are decoded and a tomogram spatial matrix is obtained, then the reference point of the heart region is selected in the spatial matrix in the XOZ plane, then the mask is created by the algorithm at the zero iteration, in which the zero pixel is selected, and a neighbouring pixel having a modulus value close to zero is placed into a mask of candidates for use in the next iteration, and for the next iteration, and continue to poll all pixels around the zero pixel, and upon completion of the poll, the algorithm switches to the first iteration, wherein the zero pixel is transferred to the mask of the final image, and from the mask of zero iteration is removed, after which the front of satisfying pixels around the initial zero pixel is obtained, next, around each of these pixels, checking for satisfying neighbours search, at that, at the N-th iteration, the algorithm polls the neighbours of the front formed at the N-1 iteration and the cardiac muscle is selected in the YOZ plane, then this algorithm is repeated for each tomogram slice in the XOY plane and a three-dimensional computer model of the heart is obtained.

EFFECT: invention provides more effective non-invasive electrophysiological examination and further catheter therapy of atrial and ventricular tachyarrhythmias.

1 cl, 7 dwg

Description

The invention relates to medicine, namely to systems and methods for image processing and data analysis, and can be used to process medical imaging image data. The method of three-dimensional reconstruction of the heart is based on non-contrast, overview computed tomography (CT) and is essentially "growing" tissues by assessing shades of gray in the gray-scale range, on the basis of which the cardiac muscle is isolated from other tissues in each section of the tomogram. The resulting 3D models of the heart can subsequently be used in cardiology and cardiac surgery. This method is implemented in the operation of the developed system for visualizing electrophysiological processes both in non-invasive electrophysiological examination of the heart and in combined invasive and non-invasive three-dimensional electrophysiological and electroanatomical mapping in an operating room, including in real time. In general, the developed method of three-dimensional visualization will allow for a wider use of non-contrast, overview CT in the context of developing technologies for personalized diagnostics and treatment of various diseases, including in patients with contraindications for contrast-enhanced CT.

A method of three-dimensional mapping of the heart chambers using the Astrocard navigation system for treating patients with cardiac arrhythmia is known [1]. Multispiral computed tomography (MSCT) or magnetic resonance imaging (MRI) of the heart is performed to obtain a 3D image of the heart. An electroanatomical model of the heart chamber is obtained intraoperatively using the Astrocard navigation mapping system. The obtained electroanatomical model of the heart chambers is combined with the processed MSCT or MRI image of the heart. However, this method can only be used intraoperatively during catheter treatment of arrhythmias, and protocols with contrast enhancement are used when performing cardiac CT.

A method of non-invasive electrophysiological examination of the heart is known [2] implemented in the creation of the hardware and software complex "Amycard™" (Russia, Switzerland). The presented method is capable of restoring the epicardial surface of the heart, but the basis for its operation is data obtained from MSCT with contrast enhancement.

The disadvantage of this method is that it is based on data obtained during contrast-enhanced MSCT, and this study is not always possible, especially in patients with allergic reactions to iodine-containing drugs and chronic kidney disease with a significant decrease in glomerular filtration rate. It is also worth noting that in order to obtain a "satisfactory" three-dimensional model of the heart, necessary for a non-invasive electrophysiological study, it is desirable that during MSCT there is equal filling of all chambers of the heart with contrast, which is quite difficult to achieve during the study and which requires a larger amount of contrast agent and is accompanied by an increase in the radiation load on the patient.

No adequate prototype was found in the analyzed patent and scientific-medical literature.

The objective of the invention is to develop a method for three-dimensional reconstruction of the heart based on non-contrast survey computed tomography to improve the effectiveness of non-invasive electrophysiological research and subsequent catheter treatment of atrial and ventricular tachyarrhythmias, including in patients with allergic reactions to iodine-containing drugs and in chronic kidney disease with a significant decrease in the glomerular filtration rate.

The problem is solved by performing a non-contrast overview computed tomography, decoding data from the international standard for transmitting medical images DICOM and obtaining a spatial matrix of the tomogram, then selecting a reference point of the heart region in the spatial matrix in the XOZ plane, then the algorithm creates a mask at the zero iteration, in which the zero pixel is selected, and the neighboring pixel, which has a value close to zero in modulus, is placed in the candidate mask for use in the next iteration, and for the next iteration, they continue to poll all pixels around the zero pixel, and upon completion of the poll, the algorithm goes to the first iteration, while the zero pixel is transferred to the mask of the final image, and it is removed from the mask of the zero iteration, after which a front of satisfying pixels is obtained around the original zero pixel, then a check is carried out around each of these pixels to find satisfying neighbors, while at the N-th iteration the algorithm polls the neighbors of the front formed at the N-1 iteration and selects the heart muscle in the YOZ plane, then this algorithm is repeated for each slice of the tomogram in the XOY plane and a three-dimensional computer model of the heart is obtained.

Thus, by segmenting the plane of the heart in XOZ based on a point selected by the operator on the plane passing through the X axis and the Z axis, dividing it into sections for analysis and identifying significant points belonging to the heart. Further iterations - segmentation in the XOY planes: transition from plane to plane perpendicular to XOZ, and identifying heart points, where the initial points are the segmentation points along XOZ. Then this algorithm is repeated for each slice of the tomogram in the XOY plane, and a point cloud is obtained, on the basis of which a three-dimensional model of the plane of the heart will be built.

The invention will be clear from the following description and the accompanying figures.

Fig. 1 shows a schematic representation of a study using the computed tomography method.

Fig. 2 shows a - 21st slice of the tomogram; b - 61st slice of the tomogram; c - 150th slice of the tomogram.

Fig. 3 shows the image reconstructed into the YOZ plane.

Fig. 4 shows the result of the segmentation algorithm in the YOZ plane.

Fig. 5 shows an illustration of the algorithm operation in the XOY plane at cut 96: A - 1st iteration; B - 20th iteration; C - 40th iteration; G - 60th iteration, D - 71st and last iteration.

Fig. 6 shows an example of a three-dimensional reconstruction of the heart based on the results of the algorithm.

Fig. 7 shows an example of the reconstructed heart surface after simplification and smoothing.

The method is implemented as follows: the basis for modeling the epicardial surface of the heart should be the data of a non-invasive study of the heart, for example, an ultrasound study of the heart, X-ray tomography, magnetic resonance imaging. Most of these studies are recorded by the tomograph for subsequent transmission to the attending physician or patient. The DICOM (Digital Imaging and Communications in Medicine) standard, developed by the American College of Radiology and the National Association of Electronic Manufacturers [3], is most widely used for transmitting medical images. Images obtained from the tomograph are in DICOM format and represent a set of image slices following one another with a given step. The minimum possible step is equal to the resolution of the tomograph, however, in order to reduce the radiation dose, recording with a decrease in resolution is possible. Fig. 1 shows a schematic representation of the types of data within the DICOM format, and also shows the main types of compatible tomography types.

The data obtained as a result of any tomographic examination are presented in a single style, or more precisely as a set of graphic images in grayscale, where each pixel represents an unsigned sixteen-bit number, the value of which depends on the corresponding examination scale and contrast settings in the tomograph. Also, a specification header is written to each raster image. This header contains patient data that is filled in by the tomograph operator. The header includes: the name of the tomograph manufacturer, tomograph data, specific scanning mode, radiation dose, spatial data of the scanning area, tomograph resolution, maximum available tomograph resolution, resolution of the selected scanning mode, slice serial number within the set, and other columns important for specialists. The header is created automatically during the noninvasive examination. The specification describes the relationship between the physical size of the scanning area and the digital size of the images in pixels. When creating a model, it is necessary to take into account that the tomographic image is three-dimensional, which means that a pixel has a "size" not only in width and length, but also in thickness. When examining the specification, it is necessary to take into account the pixel size in three axes in order to avoid distortion of the subsequent model.

The scanning results in a set of layered chest slices in the X0Y projection with a given step (thickness) of slices (Z axis). These images have the format of the medical industry standard DICOM. To build a 3D model of the heart surface, it is necessary to isolate the contours of the outer walls of the heart from the general chest image for each slice. This procedure is called segmentation or organ contour selection. To implement segmentation algorithms, it is most convenient to work with the form of raster images, therefore it is necessary to apply the conversion of DICOM standard images into commonly used raster graphic formats without compression and distortion, taking into account spatial dimensions, bringing the size in pixels to a physical value, for example, 1 pixel is equal to 1 mm.

The result of the algorithm's work on each slice should be a contour that will contain the area of space where the patient's heart muscle is located. Having contours for each slice, you can proceed to building a heart model. Since the contours of each slice are used to build the model, the first step will be building a "wire" model of the heart. As a result, you get a wireframe model consisting of "wires" of contours located along the Z coordinate in accordance with their slice number. To do this, normalization of coordinates is carried out, i.e. all contours are brought to a common zero (in the x and y coordinates, but with different z coordinates).

Let us consider the obtained tomographic data presented in the form of digital images, the dynamic range of which was normalized within the limits of [0-255], for this it is necessary to switch from a sixteen-bit to an eight-bit unsigned image format. The images after normalization are presented in ( Fig. 2 ) and were obtained on the basis of X-ray MSCT. In such data, each pixel represents not the relative intensity of light that hit the semiconductor matrix of the camera or scanner, as in a classic digital image, but the relative X-ray permeability. It is worth noting that the reduction to a normal dynamic range is necessary for the convenience of the operator, in the algorithm itself, full-fledged data without interpolation will be used further, so as not to worsen the situation with contrast.

Looking at (Fig. 2), we can conclude that the internal organs are poorly distinguishable and the boundaries between them are poorly contrasting. For navigation, it is more convenient for the operator to use the YOZ plane rather than the XOY plane. Such an image can be reconstructed without losses into another plane. An example of a reconstructed image is shown in (Fig. 3). The proposed algorithm, like most segmentation methods, requires defining the starting point of the work, the zero iteration, a certain area that, in the operator's opinion, exactly corresponds to the selected organ, in our case, the heart. It is also necessary to select a layer that reflects the longest projection of the heart for segmentation into the YOZ plane. It is desirable, but not essential, to select a slice close to the axis in the center of the heart. After selecting the zero iteration, the starting point, the algorithm queries neighboring values based on the rule - the coincidence of points corresponding to a given range.

Let us consider the algorithm in more detail, the results of its work in the YOZ plane are presented in (Fig. 4), this is its first stage. At the zero iteration, a binary mask is created for candidate pixels and a binary mask for precisely selected pixels. Masks are three-dimensional images filled with zeros. In the candidate mask, there is one pixel, which is equal to one in the area with the coordinates of the starting point, selected by the operator. The final mask at the zero iteration remains filled with zeros. Around the operator point, neighboring 8 pixels will be checked and if a neighboring pixel has a value close in modulus to the original, then such a pixel is placed in the candidate mask. Pixels are selected according to the rule

, zeroing the binary mask matrix.

Step 1 - initialization of the algorithm.

initialization point specified by the operator.

Step 2: creation of the YOZ plane of the heart, inside which lies the initialization point selected by the operator.

Step 3 Create heart planes in each XOY plane that contains the points defined as the heart muscle in step 2.

Where - overview tomographic image, - binary matrix of pixels of a tomographic image belonging to the heart muscle, - matrix of candidate pixels.

After polling all neighboring pixels around the zero one, the algorithm moves from the zero iteration to the first one. The zero pixel is transferred to the final mask of the final image. It will be removed from the candidate mask of the first iteration, so as not to poll the already tested area several times. After this, a front of candidate pixels will be obtained for testing at the first iteration, except for the pixels contained inside the final mask. After polling, the candidate pixels that passed the test will be included in the final mask, and a new front of candidates for the second iteration will be formed around them.

At the N-th iteration, the algorithm will poll the neighbors of the front formed at the N-1 iteration, excluding the polling of already approved values to save computational time. Thus, it is possible to select the cardiac muscle in the YOZ plane. The result of such selection is shown in (Fig. 4).

Then this algorithm will be repeated for each slice of the tomogram in the XOY plane. Here it is possible to expand or narrow the boundary of the interval of the permissible value of the pixels of the three-dimensional image, if it is necessary to make the algorithm more "aggressive" and vice versa. It is worth noting that in the zero iteration of the second stage of the algorithm, the points recognized as the heart muscle at the previous stage will be used as a starting point or front, and these points in each slice in the YOX plane will automatically become the first points in the candidate matrix. An example of the algorithm's operation at different iterations in one slice is shown in Fig. 6. It should be noted that the layers in which no points were found at the first stage of "growing" will not be used in the second stage to save computing time. That is, the working area of the algorithm will be reduced automatically.

In (Fig. 5) it is seen that the last iteration for the slice is the 71st iteration. The algorithm can stop on its own if it has reached the growth limit and is surrounded by "strong" boundaries. In other words, the algorithm has finished its work due to the lack of new points for "growing" the area. It can also be stopped forcibly if the operator considers the grown area sufficient, since in case of insufficient quality of MSCT images, the algorithm can "grow" beyond the desired zone. Images obtained with MSCT can hardly be called ideal due to noise associated with the data acquisition process, in general, their quality leaves much to be desired, it is difficult, if not impossible, to influence it.

After layer-by-layer selection of the heart area in each section, you can proceed to constructing an STL surface model. From the area selected in each section, you need to construct perimeter points, these polynomials will become the basis for creating a three-dimensional STL model of the heart by "adding" them like a coin column and forming a wire model. Then, using the sphere morphing algorithm, you can obtain an STL model from the cloud of boundary points for the doctor's work, and can also be used in other software packages or directly shown to the patient for demonstration purposes.

An example of the reconstructed three-dimensional model is shown in (Fig. 6). As can be clearly seen, the algorithm failed to avoid the main error of model construction - "growth" beyond the desired zone, when a fragment of the liver was perceived as part of the heart, and in this case the operator will have to "cut" it off using the tools of a three-dimensional graphic editor.

As shown in (Fig. 6), as a result of the algorithm's operation, it "grew" beyond the heart, which is a consequence of the close density of structures in the range of adjacent tissues selected by the algorithm. It is obvious that the organs themselves have a physical boundary, but this boundary is not reflected in the MSCT slices and the algorithm has not yet "grown" an obvious difference in density. This error can be solved with further development of the algorithm by introducing additional "complex" growth conditions or by improving the quality of MSCT. At this stage, the problem is solved by the final formation of a three-dimensional model manually by the operator, removing unnecessary elements. After forming an array of reference points on the reconstructed surface, the surface of the heart model is approximated. Computer graphics has a rich set of tools for constructing surfaces based on reference points. The most common is Delaunay triangulation [4]. As a result, the heart model consists of many triangles and points (Fig. 7). Then, depending on the available computing power, the STL model is simplified to the required level of complexity.

Sources of information:

1. Patent. No. 2724191 Russian Federation Method of three-dimensional mapping of heart chambers using the Astrocard navigation system for treating patients with heart rhythm disorders [Text] / Rivishvili A.Sh., Artyukhina E.A., Yashkov M.V., Popov A.Yu., Vasin V.A.; applicant and patent holder: Federal State Budgetary Institution "A.V. Vishnevsky National Medical Research Center of Surgery" (RU) - No. 2019141386; declared.12.13.2019; published.06.22.2020 Bull. No. 18.

2. Patent. No. 2409313 Russian Federation Method of non-invasive electrophysiological study of the heart [Text] / Rivishvili A.Sh., Kalinin V.V., Kalinin A.V.; applicant and patent holder: Revishvili Amiran Shotaevich (RU) - No. 2008146994; declared 27.11.2008; published 20.01.2011 Bulletin No. 2.

3. GOST R ISO 12052 - 2009 Digital imaging and communications in medicine (DICOM), including document and data management: national standard of the Russian Federation: date of introduction 2009-09-14 / Federal Agency for Technical Regulation and Metrology. - Official ed. - Moscow: Standartinform, 2010. - 16 p.

4. Skvortsov A.V., Mirza N.S. Algorithms for constructing and analyzing triangulation. - Tomsk: Publishing House Tom. Univ., 2006. - 168 p.

Claims (1)

A method for three-dimensional reconstruction of the heart for conducting a non-invasive electrophysiological study in atrial and ventricular tachyarrhythmias, characterized by the fact that non-contrast survey computed tomography is performed, data from the international standard for the transmission of medical images DICOM are decoded and a spatial tomogram matrix is obtained, then a reference point of the heart region is selected in the spatial matrix in the XOZ plane, then the algorithm creates a mask at the zero iteration, in which the zero pixel is selected, and the neighboring pixel, having a value in modulus close to zero, is placed in the candidate mask for use in the next iteration, and for the next iteration, and continue to poll all pixels around the zero pixel, and upon completion of the poll, the algorithm goes to the first iteration, wherein the zero pixel is transferred to the mask of the final image, and it is removed from the zero iteration mask, after which a front of satisfying pixels is obtained around the original zero pixel, then a check is carried out around each of these pixels to find satisfying neighbors, wherein At the N-th iteration, the algorithm queries the neighbors of the front formed at the N-1 iteration and selects the cardiac muscle in the YOZ plane, then this algorithm is repeated for each slice of the tomogram in the XOY plane and a three-dimensional computer model of the heart is obtained.