Image enhancement methods for chest X-rays used to assess heart failure

Abstract

This invention relates to the field of image enhancement processing technology, specifically to an image enhancement method for chest X-rays used to assess heart failure. The invention first obtains the occluded region of the lung area from an orthogonal X-ray and projects it onto a target image to obtain the occluded portion. Further, based on the similarity between the skeleton lines of the foreground connected components and the edge lines of the occluded portion, combined with the edge blurring features of the connected components and local gray-level differences, the pulmonary vascular information of each pixel is obtained, and the pulmonary vascular region is acquired. Further, based on the similarity of gray-level changes in the vertical direction, pixels in the pulmonary vascular region of left and right anterior oblique X-rays are matched. Finally, based on the overlapping pixel groups and the pulmonary vascular information, the left and right anterior oblique X-rays are translated and superimposed, and the fused pulmonary vascular portion is fitted to the occluded portion to obtain a pulmonary vascular X-ray image. Matching, translation, and superimposition repair the occluded area, overcomes oblique distortion, and improves the display effect of pulmonary vessels.

Description

Technical Field

[0001] This invention relates to the field of image enhancement processing technology, and more specifically to an image enhancement method for chest X-rays used to assess heart failure. Background Technology

[0002] X-rays, using low-dose X-rays, provide doctors with images of the chest cavity, allowing them to assess the condition of the heart, lungs, and other structures within the chest.

[0003] When assessing heart failure using changes in pulmonary blood flow on chest X-rays, the central region of the lungs inevitably casts a shadow over the heart. This area is crucial for observing pulmonary vessels and hilar edema. Therefore, it is necessary to project the frontal (PA) image onto the left anterior oblique (LAO) and right anterior oblique (RAO) images to correct the obscured area. However, distortions during LAO and RAO acquisition can alter the actual state of the pulmonary vessels during fusion projection, affecting the accuracy of subsequent heart failure assessments. Summary of the Invention

[0004] To address the technical problem of distortion in lateral chest X-rays affecting the repair of obstructed pulmonary vessels in the anterior view, this invention aims to provide an image enhancement method for chest X-rays used to assess heart failure. The specific technical solution adopted is as follows:

[0005] Acquire chest X-ray images of the current patient from three perspectives: frontal, left anterior oblique, and right anterior oblique. In the lung region of the frontal X-ray image, obtain the occluded area based on the gradient distribution of pixels.

[0006] Select left and right anterior oblique X-ray images one by one as target images; project the occluded area onto the target image to obtain the occluded part; extract the foreground part of the target image, and based on the similarity features between the skeleton line of the connected domain in the foreground part and the edge line of the occluded part, combined with the edge blurring features of the connected domain and the local gray level difference, obtain the lung vascular information of the pixels in the connected domain;

[0007] The lung vascular region is obtained based on the distribution difference of the lung vascular information. According to the similarity of gray-level changes on both sides of the pixel in the vertical direction, the pixels of the lung vascular region in the left anterior oblique and right anterior oblique X-rays are matched to obtain overlapping pixel groups. Based on the overlapping pixel groups and the lung vascular information, the left anterior oblique and right anterior oblique X-rays are translated and superimposed, and the fused lung vascular part is attached to the occluded part to obtain the lung vascular X-ray image.

[0008] Furthermore, the method for obtaining the occluded area includes:

[0009] Extract the edge line of the lung region; take the two pixels with the largest gradient value in the preset neighborhood of each pixel as the marked pixels; connect the marked pixels with the marked pixels in the preset neighborhood, and obtain the splitting factor of each connecting line segment according to the overall gradient value of each connecting line segment and the number of edge pixels;

[0010] Select the closed connecting segment with the largest severance factor to obtain the occlusion area.

[0011] Furthermore, the method for obtaining the pulmonary vascular information includes:

[0012] Obtain the DTW distance between the edge lines of the occluded portion. When the DTW distance between two edge lines is the minimum of all their corresponding DTW distances, assign the two edge lines to an edge line group and match the edge line group with the connected components in the foreground portion between the edge lines.

[0013] Based on the skeleton line of the connected region in the foreground portion, the similarity features of the two edge lines in the nearest matching edge line group, and the edge blurring features of the connected region, the lung vascular coefficient of the corresponding pixel in the connected region is obtained.

[0014] The texture information is obtained based on the local grayscale differences of pixels within the connected domain; the lung vascular information is obtained by fusing the lung vascular coefficient and the texture information of the pixels.

[0015] Furthermore, the method for obtaining the pulmonary vascular coefficient includes:

[0016] The lung vascular coefficient of the corresponding pixel in the connected region is obtained by fusing the average DTW distance between the skeleton line of the connected region and the two edge lines in the nearest matching edge line group, and the average gradient value of the edge line of the connected region.

[0017] Furthermore, the method for obtaining the texture information includes:

[0018] The variance of the grayscale values ​​of pixels in the preset neighborhood of a pixel within a connected domain is used as the texture information of the corresponding pixel.

[0019] Furthermore, the method for obtaining the pulmonary vascular region includes:

[0020] Threshold segmentation is performed based on the amount of lung vascular information in the pixels of the target image, and the region formed by the pixels with values ​​higher than the threshold is taken as the lung vascular region.

[0021] Furthermore, the method for obtaining the overlapping pixel group includes:

[0022] Select each pixel in the lung vascular region of the target image as a target pixel; on a vertical line passing through the target pixel and perpendicular to the bottom of the image, obtain the pixels in the lung vascular region that the vertical line passes through, and record them as collinear pixels.

[0023] Based on the grayscale values, curve fitting is performed on the collinear pixels on both sides of the target pixel. According to the similarity features of the two fitted curves of the target pixel and the pixels of the lung vascular region in another X-ray, the matching factor of the corresponding two pixels is obtained.

[0024] The target pixel and the pixel corresponding to the largest matching factor are combined to form an overlapping pixel group.

[0025] Furthermore, the method for obtaining the matching factor includes:

[0026] Based on the positional relationship between the fitted curve and the target pixel, the fitted curve is divided into an upper curve and a lower curve; the sum of the absolute residuals between the upper curves and the lower curves of the two pixels is fused to obtain the matching factor for the corresponding two pixels.

[0027] Furthermore, the method for translating and superimposing X-ray images of the left anterior oblique and right anterior oblique based on the overlapping pixel group and the amount of pulmonary vascular information includes:

[0028] The edges of the left and right anterior oblique X-ray films are aligned, and the overlapping pixel group with the farthest distance in each row is marked as the marked pixel group. The pixel distance between the marked pixel groups is obtained. Within the range of the pixel distance, the target translation distance is selected one by one.

[0029] Based on the difference between the pixel distance between each group of marked pixels and the target translation distance, and combined with the average value of the pulmonary vascular information of two pixels within the group of marked pixels, the superposition loss coefficient of the target translation distance is obtained.

[0030] Select the target translation distance corresponding to the smallest superposition loss coefficient, and perform translation superposition on the X-ray images of the left and right anterior oblique angles.

[0031] Furthermore, the method for obtaining the occluded portion includes:

[0032] An x-axis is established by connecting the tops of the bilateral diaphragms in an orthogonal X-ray, and a y-axis is established by the midline of the spine. The target image is superimposed on the orthogonal X-ray, and the tops of the bilateral diaphragms and the midline of the spine are matched. The occluded area is projected onto the target image based on the position of the occluded area on the coordinate axis to obtain the occluded portion.

[0033] The present invention has the following beneficial effects:

[0034] Acquire chest X-ray images of the current patient from three perspectives: frontal, left anterior oblique, and right anterior oblique. In the lung region of the frontal X-ray image, obtain the occluded area based on the gradient distribution of pixels.

[0035] Select left and right anterior oblique X-ray images one by one as target images; project the occluded area onto the target image to obtain the occluded part; extract the foreground part of the target image, and based on the similarity features between the skeleton line of the connected component in the foreground part and the edge line of the occluded part, combined with the edge blurring features of the connected component and the local gray level difference, obtain the lung vascular information of the pixels in the connected component.

[0036] The lung vascular region is obtained based on the distribution difference of lung vascular information. According to the similarity of gray-level changes on both sides of the pixel in the vertical direction, the pixels of the lung vascular region in the left anterior oblique and right anterior oblique X-rays are matched to obtain the overlapping pixel group. Based on the overlapping pixel group and lung vascular information, the left anterior oblique and right anterior oblique X-rays are translated and superimposed, and the fused lung vascular part is attached to the occluded part to obtain the lung vascular X-ray image.

[0037] This invention first obtains the occluded area of ​​the lung region in an orthogonal X-ray film based on the gradient distribution of pixels, and locates the area that needs to be repaired. Next, the occluded area is projected onto the target image to obtain the occluded portion, and the corresponding occluded area in the oblique view is precisely located. Further, based on the characteristic that pulmonary vessels are part of the image foreground and have blurred edges, and combined with local gray-level differences to represent local texture content, the amount of pulmonary vessel information in each pixel is obtained, quantifying the useful information carried by each pixel. Further, utilizing the characteristic that distortion only occurs in the horizontal direction, pixels in the pulmonary vessel region of the left anterior oblique and right anterior oblique X-ray films are matched based on the similarity of gray-level changes in the vertical direction. Finally, based on the overlapping pixel groups and the amount of pulmonary vessel information, the left anterior oblique and right anterior oblique X-ray films are translated and superimposed, and the fused pulmonary vessel portion is attached to the occluded portion to obtain a pulmonary vessel X-ray image, providing clearer and more accurate data support for relevant personnel. This invention obtains three-view chest radiograph registration, detects cardiac occlusion areas based on gradient detection, projects them onto oblique radiographs, and extracts pulmonary vascular information by combining skeleton lines, edge blurring, and grayscale differences. It uses matching translation and superposition to repair occlusion areas, overcomes oblique distortion, and improves the display effect of pulmonary vascularity. Attached Figure Description

[0038] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 A flowchart illustrating an image enhancement method for evaluating heart failure on a chest X-ray, provided as an embodiment of the present invention;

[0040] Figure 2 An orthopedic chest X-ray provided as an embodiment of the present invention;

[0041] Figure 3 A coordinate system within an orthotopic chest X-ray provided in one embodiment of the present invention;

[0042] Figure 4 This is a flowchart of a method for acquiring pulmonary vascular information according to an embodiment of the present invention. Detailed Implementation

[0043] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an image enhancement method for evaluating chest X-ray films of heart failure according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0045] The following describes in detail, with reference to the accompanying drawings, a specific scheme of the image enhancement method for chest X-rays used to assess heart failure provided by the present invention.

[0046] Please see Figure 1 The diagram illustrates a flowchart of an image enhancement method for evaluating heart failure on a chest X-ray, according to an embodiment of the present invention, specifically including:

[0047] Step S1: Obtain chest X-rays of the current patient from three perspectives: frontal, left anterior oblique, and right anterior oblique; in the lung region of the frontal X-ray, obtain the occluded area based on the gradient distribution of pixels.

[0048] In one embodiment of the present invention, an intelligent DR system integrating ECG gating and respiratory sensing is used to complete a standard anteroposterior (PA)-left anterior oblique (LAO)-right anterior oblique (RAO) three-view sequence exposure within a single breath-holding cycle (≤6 seconds) for the patient. The oblique view film adopts a 15° tube rotation + 4mAs low-dose protocol (total effective dose ≤315μSv). Simultaneously, a low-dose CT scan (radiation dose ≤1.5mSv) is acquired within 48 hours, and a virtual multi-angle X-ray film with geometry consistent with the actual imaging is generated using a three-dimensional projection simulation algorithm.

[0049] A neural network was used to segment chest X-ray images to obtain the lung region.

[0050] It should be noted that the standard angle for left anterior oblique is usually 45°~60° (body rotated to the left), and the standard angle for right anterior oblique is usually 30°~45° (body rotated to the right). The specific angle can be adjusted according to the patient's body shape and clinical needs. The acquisition method of X-ray and the method of training neural networks to extract the lung area in X-ray are well-known technologies and will not be elaborated further.

[0051] Please see Figure 2 This image shows an orthopedic chest X-ray provided in one embodiment of the present invention. In the image, L represents the left lateral view, PA (Posteroanterior) indicates that the X-rays penetrate the patient's body from the posterior to the anterior, and UPRIGHT indicates that the patient is in an upright (standing) position when the image is taken.

[0052] Since the heart in the chest cavity casts a shadow on an X-ray, and this shadow is located in the hilar region of the lungs, and the shadow cast by the heart in the lung region will appear as a highlight in the image, causing gradient changes in the lung region, the occluded area is obtained based on the gradient distribution of pixels in the lung region of an orthogonal X-ray, and the area that needs to be repaired is located first.

[0053] Preferably, in one embodiment of the present invention, Canny edge detection is performed on the lung region to extract the edge lines and gradient values ​​of pixels in the lung region;

[0054] Since the shadows created by the heart covering will appear as highlights in the image, the two pixels with the largest gradient values ​​in the preset neighborhood of each pixel are used as marked pixels.

[0055] Furthermore, since the shadows caused by the occlusion of the lung and heart areas can lead to the rupture of pulmonary blood vessels in the image, abnormal edge breaks will appear in the image. By connecting the marked pixels with the marked pixels in the preset neighborhood, the larger the number of edges the connecting line segment passes through and the higher the overall gradient, the more pulmonary blood vessel ruptures the connecting line segment may have occurred. Therefore, the rupture factor of each connecting line segment is obtained based on the overall gradient value of each connecting line segment and the number of edge pixels.

[0056] As an example, the product of the overall average gradient value of each connecting line segment and the number of edge pixels is used as the splitting factor for each connecting line segment. The overall gradient value is represented by the average gradient value, and the gradient distribution of the pixels is represented by the splitting factor.

[0057] The larger the severance factor, the greater the number of edges traversed and the stronger the connectivity of the edges, resulting in more pulmonary vascular ruptures. Therefore, the closed connecting line segment with the largest severance factor is selected to obtain the occluded area.

[0058] It should be noted that "closed" means connecting the beginning and end; the preset neighborhood uses the commonly used eight-neighborhood. Canny edge detection is a well-known technique in the field and will not be elaborated further.

[0059] Step S2: Select the left anterior oblique and right anterior oblique X-ray films one by one as the target images; project the occluded area onto the target image to obtain the occluded part; extract the foreground part of the target image, and based on the similarity features between the skeleton line of the connected component in the foreground part and the edge line of the occluded part, combined with the edge blurring features of the connected component and the local gray level difference, obtain the lung vascular information of the pixels in the connected component.

[0060] After obtaining the obstructed area in the orthogonal X-ray, it is necessary to analyze the corresponding parts in the X-rays acquired on the left and right sides to restore and repair the area before obstruction. Since the analysis process for the obstruction information of the X-rays on the left and right sides is the same, the X-rays of the left anterior oblique and right anterior oblique are selected as target images, and only one example is described without repeating the explanation.

[0061] First, the occluded area is projected onto the target image to obtain the occluded part, and then the corresponding occluded area in the oblique image is accurately located.

[0062] Preferably, in one embodiment of the present invention, the x-axis is established by connecting the tops of the bilateral diaphragms in the positive X-ray film, and the y-axis is established by the midline of the spine. The target image is superimposed on the positive X-ray film, and the tops of the bilateral diaphragms and the midline of the spine are matched. The occluded area is projected into the target image based on the position of the occluded area on the coordinate axis to obtain the occluded part.

[0063] Please see Figure 3It illustrates a coordinate system within a chest X-ray taken from an orthogonal view, as provided in one embodiment of the present invention, defining image edges parallel or nearly parallel to the human spine as vertical edges (e.g., Figure 2 and Figure 3 The left and right sides), corresponding to the vertical direction, and the image edges that are perpendicular or nearly perpendicular to the human spine are horizontal edges (such as...). Figure 2 and Figure 3 The upper and lower sides of the spine correspond to the horizontal direction; the line connecting the centroids of the vertebrae at the upper and lower ends of the spine is the midline of the spine, and the y-axis is established from bottom to top, and the line connecting the tops of the diaphragms on both sides is established from left to right.

[0064] The centroid of the vertebrae can be extracted by threshold segmentation or neural network to obtain the connected domains of the vertebrae at the upper and lower ends of the spine. Then, a two-dimensional rectangular coordinate system is established based on the lower left corner of the image. The average value of the pixel coordinates in the connected domain is obtained in the two-dimensional rectangular coordinate system as the centroid of the vertebrae.

[0065] It should be noted that the spine and the top of the bilateral diaphragms are well-known features. When acquiring the occluded part, feature points such as the top of the bilateral diaphragms and the midline of the spine can be extracted. The homography matrix can be calculated by feature matching (such as SIFT and ORB), and finally the geometric transformation projection of the occluded area can be achieved to obtain the occluded part. This method is a commonly used technique in the field of science and can spatially register X-ray or CT images from different perspectives, and will not be elaborated further.

[0066] Considering that distortion may occur during the acquisition of LAO and RAO positions, the target image needs to be corrected. In order to preserve the information in the image to the greatest extent and reduce information loss, it is necessary to first quantify the amount of lung vascular information in the pixels.

[0067] Since pulmonary vessels appear milky white in the image, belonging to the foreground and exhibiting a blurred, band-like edge, the foreground of the image contains pulmonary vessels. Extracting the foreground of the target image reveals that it contains several connected components. Because pulmonary vessels have a clear direction and blurred edges, and local grayscale differences reflect local texture features and represent local information, the lung vessel information of pixels within the connected components is obtained by combining the similarity between the skeleton lines of the connected components in the foreground and the edge lines of the occluded portion, along with the edge blurring features and local grayscale differences. This quantifies the useful information carried by each pixel, providing a basis for the final occlusion repair.

[0068] Preferably, in one embodiment of the present invention, please refer to Figure 4 The flowchart illustrates a method for acquiring pulmonary vascular information according to an embodiment of the present invention, specifically including:

[0069] Step S201: Obtain the DTW distance between the edge lines of the occluded part. When the DTW distance between two edge lines is the minimum of all their corresponding DTW distances, assign the two edge lines to an edge line group and match the edge line group with the connected components in the foreground part between the edge lines.

[0070] Considering that the DTW algorithm can be used to measure the similarity of changes between two curves, the smaller the obtained DWT distance, the higher the similarity. Therefore, the DTW distance between the edge lines of the occluded part is obtained. At this time, each edge line has a DTW distance with other edge lines. When the DTW distance between two edge lines is the minimum of all their corresponding DTW distances, it means that the trend between the two edge lines is the most consistent and the similarity feature is the strongest. The two edge lines are then assigned to an edge line group.

[0071] Simultaneously, analyze whether there is a connected component in the foreground portion between the two edge lines of each edge line group. If so, match the edge line group with the connected component in the foreground portion between the edge lines.

[0072] It should be noted that the DTW algorithm is a well-known technology. For complex folded edge lines, such as zigzag edge lines or forked edge lines, the edge line can be segmented at the folding point, the DTW distance of the edge line segment to other edge lines can be calculated, and finally, the DTW distance of each edge line segment is weighted and summed according to the length ratio of each edge line segment, which is used as the DTW distance corresponding to the complex folded edge line.

[0073] Meanwhile, other edge lines for comparison may also have multiple segments, such as A1, A2 and B1, B2. A and B represent edge lines, and 1 and 2 represent the segment numbers of the edge lines. The minimum DTW distance between the segments is then summed with weights. For example, if the DTW distance between A1 and B1 is less than the DTW distance between A1 and B2, and the DTW distance between A2 and B2 is less than the DTW distance between A2 and B1, then the DTW distances between A1 and B1 and between A2 and B2 are finally summed with weights. The proportions of the lengths of the two edge lines occupied by A1 and B1, and A2 and B2 are used as the weights.

[0074] The turning point indicates that the curve has turned back or forked. A rectangular coordinate system can be established in the lower left corner of the target image to analyze the coordinate changes of the edge line. When the horizontal coordinate increases, there are points on the edge line where the horizontal coordinate decreases; or when the horizontal coordinate decreases, there are points where the horizontal coordinate increases, indicating that the curve has turned back. When an edge point connects to three or more edge points, it is a fork point on the edge line.

[0075] Step S202: Based on the skeleton line of the connected region in the foreground part and the similarity features of the two edge lines in the nearest matching edge line group, and combined with the edge blurring features of the connected region, obtain the lung vascular coefficient of the corresponding pixel in the connected region.

[0076] To facilitate the analysis of the trend of connected regions, morphological skeletons (a known technique) are used to extract the skeleton lines of connected regions in the foreground. Considering that the stronger the similarity between the skeleton lines of connected regions and the two nearest matched edge lines, the more obvious the trend is, and the more likely it is to be a pulmonary vascular region. At the same time, edge blurring is also an important feature of pulmonary vessels. Therefore, the pulmonary vascular coefficients of pixels in connected regions are obtained by combining the two.

[0077] As an example, the DTW distance is also used to show the similarity between the skeleton line of the connected region and the edge line of the occluded part. Considering that the smaller the gradient value of the edge line, the more blurred the edge is, the average gradient value is used to represent the overall edge blurring feature. The average DTW distance between the skeleton line of the connected region and the two edge lines in the nearest matching edge line group is fused with the average gradient value of the edge line of the connected region to obtain the lung vascular coefficient of the corresponding pixel in the connected region.

[0078] Specifically, the average of the two DTW distances between the skeleton line of the connected component and the two nearest matching edge lines in the same edge line group, and the average gradient value of the edge lines of the connected component are multiplied, and then a preset division-by-zero positive parameter of 0.01 is added. The reciprocal of this sum is used as the pulmonary vascular coefficient of the corresponding pixel in the connected component. Negative correlation mapping is performed by taking the reciprocal, adjusting the logical relationship. The preset division-by-zero positive parameter is used before taking the reciprocal to prevent the denominator from being zero.

[0079] Step S203: Obtain texture information based on the local grayscale differences of pixels within the connected domain; fuse the lung vascular coefficient and texture information of the pixels to obtain lung vascular information.

[0080] Considering that the greater the local grayscale difference of a pixel, the greater the amount of texture information, and the greater the lung vascular coefficient, it indicates that the pixel is more likely to be a lung vascular pixel, and is more likely to carry useful information, thus having a greater amount of lung vascular information.

[0081] As an example, considering that the larger the variance of the gray values ​​of pixels in the preset neighborhood of a pixel, the greater the local gray value difference and the greater the amount of texture information, the variance of the gray values ​​of pixels in the preset neighborhood of a pixel in the connected domain is used as the amount of texture information of the corresponding pixel, and the local gray value difference is represented by the local variance.

[0082] The product of the pulmonary vascular coefficient and the texture information of a pixel is used as the pulmonary vascular information.

[0083] It should be noted that the preset neighborhood uses the commonly used eight-neighborhood. In other embodiments of the present invention, a four-neighborhood or a preset neighborhood of other sizes such as 5x5 can be used instead. The foreground part of the image can be obtained by binarizing the image through the maximum inter-class variance. This is a well-known technique in the art and will not be described in detail here.

[0084] Step S3: Obtain the lung vascular region based on the distribution difference of lung vascular information; match the pixels of the lung vascular region in the left anterior oblique and right anterior oblique X-rays according to the similarity of gray-level changes on both sides of the pixel in the vertical direction to obtain overlapping pixel groups; perform translation and superposition on the left anterior oblique and right anterior oblique X-rays based on the overlapping pixel groups and lung vascular information, and attach the fused lung vascular part to the occluded part to obtain the lung vascular X-ray image.

[0085] Since the amount of lung vascular information varies among different pixels in the foreground connected region of the target image, and the amount of lung vascular information in pixels in the lung vascular region differs significantly from that in pixels in other regions, the lung vascular region is obtained based on the distribution difference of lung vascular information.

[0086] Preferably, in one embodiment of the present invention, threshold segmentation is performed based on the amount of lung vascular information in the pixels of the target image, and the region composed of some pixels with values ​​higher than the threshold is regarded as the lung vascular region.

[0087] One approach is to use the maximum inter-class variance method to determine the segmentation threshold, and then use the region formed by the pixels above the threshold as the lung vascular region. This is a well-known technique in the art and will not be elaborated further.

[0088] The lung vascular regions in the acquired LAO and RAO images of the patient are represented by the left and right side views of the partially obscured blood vessels in the PA image of the patient. Due to the side view, some blood vessels may appear repeatedly in the LAO and RAO images. Therefore, when overlaying the PA images, it is necessary to first match the lung vascular images in the images to determine the overlapping parts in the two images.

[0089] Since the distortions in LAO and RAO images are only in the horizontal direction, the vertical changes of the matched lung vessel pixels are consistent. Pixel matching is then performed to avoid interference from horizontal distortion. Considering that the gray-level changes on both sides of the overlapping matched pixels are similar, the pixels of the lung vessel region in the left anterior oblique and right anterior oblique X-rays are matched based on the similarity of gray-level changes on both sides of the pixels in the vertical direction to obtain overlapping pixel groups.

[0090] Preferably, in one embodiment of the present invention, pixels within the lung vascular region of the target image are selected one by one as target pixels; in order to analyze the grayscale changes in the vertical direction, pixels within the lung vascular region that pass through the target pixels and are perpendicular to the bottom of the image are obtained on a vertical line and recorded as collinear pixels.

[0091] Among them, the image edge that is parallel or nearly parallel to the human spine is the vertical edge, corresponding to the vertical direction; the bottom of the image refers to the image edge that is perpendicular or nearly perpendicular to the human spine and is located below it.

[0092] Considering that grayscale values ​​are regarded as the amplitude of data, curve fitting can be performed on the collinear pixels on both sides of the target pixel based on the grayscale values. The similarity of grayscale changes of the two pixels in the vertical direction can be analyzed by using the fitted curves. Therefore, based on the similarity features of the two fitted curves of the target pixel and the pixels in the lung vascular region of another X-ray, the matching factor of the corresponding two pixels can be obtained.

[0093] As an example, the fitted curve is first divided into an upper curve and a lower curve based on the positional relationship between the fitted curve and the target pixel. The upper and lower curves of the two curves are then compared separately. Considering that the larger the sum of the absolute residuals between the curves, the greater the difference between the fitted curves, indicating that the similarity of grayscale changes on both sides of the pixel in the vertical direction is smaller, the sum of the absolute residuals between the upper curves and the lower curves of the two pixels is fused to obtain the matching factor for the corresponding two pixels.

[0094] Specifically, the product of the sum of the absolute residuals between the target pixel and the pixels in the pulmonary vascular region of another X-ray image, and the sum of the absolute residuals between the upper and lower curves, is used as the independent variable. After negative correlation mapping using the exp(-x) function, the mapped value is used as the matching factor.

[0095] The fitting curve is obtained by directly connecting adjacent data values. The residual is obtained by taking one curve as the original curve and the other curve as the predicted curve, taking the difference of data values ​​in the same order as the residual, filling missing zeros, and finally summing the absolute values ​​of all residuals on one side of the curve as the absolute residual sum. The residuals are then multiplied and fused, and then mapped by a negative exponential function with the natural constant e as the base to adjust the logical relationship. x is the independent variable. The larger the matching factor, the greater the similarity of gray-level changes on both sides of the pixel in the vertical direction.

[0096] Then, the target pixel and the pixel corresponding to the largest matching factor are combined to form an overlapping pixel group.

[0097] Change the target pixel to obtain all overlapping pixel groups.

[0098] After obtaining the overlapping features of the pulmonary vascular region pixels in the occluded parts of the LAO and RAO images by overlapping pixel groups, it is necessary to perform translation and superposition on the left anterior oblique and right anterior oblique X-rays based on the overlapping pixel groups and pulmonary vascular information to obtain the unoccluded pulmonary vascular part obtained from the side. The fused pulmonary vascular part is then attached to the occluded part to repair the frontal X-ray and obtain pulmonary vascular X-ray images. This provides clearer and more accurate data support for relevant personnel, making it easier for them to analyze changes in pulmonary blood flow and assess the patient's cardiac function using X-rays.

[0099] Preferably, in one embodiment of the present invention, the edges of the left anterior oblique and right anterior oblique X-ray films are first bonded together. When bonding the edges, it is necessary to ensure that the overlapping part is in the center of the bonding. For example, if the overlapping part is on the right side of Figure A and the left side of Figure B, then the right side of Figure A and the left side of Figure B are bonded together, and the most matching pixels are on the same horizontal line.

[0100] Considering that when two images have overlapping areas, the overlapping pixels between the farthest overlapping pixel groups in each row are duplicates and need to be discarded during translation and overlay. The farthest overlapping pixel groups in each row are marked as marked pixel groups, and the pixel distances between the marked pixel groups are obtained. Within the range of pixel distances, the target translation distance is selected one by one.

[0101] In order to ensure the maximum preservation of information in the image during translation and superposition, priority should be given to the pixel group with a large amount of lung vascular information. The pixel distance between the marked pixel group is taken as its optimal translation distance. The difference between the pixel distance and the target translation distance represents the translation deviation. At this time, the average amount of lung vascular information of two pixels in the marked pixel group reflects the translation priority.

[0102] Based on this, the superposition loss coefficient of the target translation distance is obtained by combining the difference between the pixel distance between each group of marked pixels and the target translation distance, and the mean value of the lung vascular information of two pixels within the group of marked pixels.

[0103] As an example, the absolute value of the difference between the pixel distance between each group of marked pixels and the target translation distance is used as the first factor, the mean value of the lung vascular information of two pixels within each group of marked pixels is used as the second factor, and the product of the first and second factors is used as the loss factor for each group of marked pixels. The average value of the loss factors of all groups of marked pixels under the target translation distance is used as the superposition loss coefficient of the target translation distance.

[0104] The difference between the pixel distance and the target translation distance is represented by the absolute value of the difference. The larger the difference, the greater the difference between the actual translation and the theoretical optimal translation. At the same time, the average value of the lung vascular information of two pixels in the pixel group is marked. The more lung vascular information a pixel carries, the more loss factor is obtained by multiplication and fusion. The smaller the loss factor, the smaller the translation loss.

[0105] Finally, the target translation distance corresponding to the smallest stacking loss coefficient was selected, and the X-ray images of the left and right anterior oblique angles were stacked by translation.

[0106] During the translation and overlay process, an X-ray image is used as a reference, and another image is horizontally translated and overlaid onto the reference image to obtain the fused pulmonary vascular portion. Finally, the fused pulmonary vascular portion is attached (covered) to the obscured part, thereby using chest X-ray images from both left and right perspectives to complete the correction of the frontal X-ray image and obtain the pulmonary vascular X-ray image.

[0107] In summary, to address the technical problem of distortion in lateral chest X-rays affecting the restoration of pulmonary vessels in the anterior view, this invention proposes an image enhancement method for evaluating chest X-rays of patients with heart failure. First, the occluded region of the lung area in the anterior view X-ray is obtained based on the gradient distribution of pixels. Then, the occluded region is projected onto the target image to obtain the occluded portion. Next, based on the similarity between the skeleton lines of the foreground connected region and the edge lines of the occluded portion, combined with the edge blurring features of the connected region and local gray-level differences, the pulmonary vessel information of each pixel is obtained, and the pulmonary vessel region is acquired. Furthermore, based on the similarity of gray-level changes in the vertical direction, pixels in the pulmonary vessel region of the left and right anterior oblique X-rays are matched. Finally, based on the overlapping pixel groups and the pulmonary vessel information, the left and right anterior oblique X-rays are translated and superimposed, and the fused pulmonary vessel portion is attached to the occluded portion to obtain a pulmonary vessel X-ray image. This invention obtains three-view chest radiograph registration, detects cardiac occlusion areas based on gradient detection, projects them onto oblique radiographs, and extracts pulmonary vascular information by combining skeleton lines, edge blurring, and grayscale differences. It uses matching translation and superposition to repair occlusion areas, overcomes oblique distortion, and improves the display effect of pulmonary vascularity.

[0108] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0109] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An image enhancement method for chest X-rays used to assess heart failure, characterized in that, The method includes: Acquire chest X-ray images of the current patient from three perspectives: frontal, left anterior oblique, and right anterior oblique. In the lung region of the frontal X-ray image, obtain the occluded area based on the gradient distribution of pixels. Select left and right anterior oblique X-ray images one by one as target images; project the occluded area onto the target image to obtain the occluded part; extract the foreground part of the target image, and based on the similarity features between the skeleton line of the connected domain in the foreground part and the edge line of the occluded part, combined with the edge blurring features of the connected domain and the local gray level difference, obtain the lung vascular information of the pixels in the connected domain; The lung vascular region is obtained based on the distribution difference of the lung vascular information. According to the similarity of gray-level changes on both sides of the pixel in the vertical direction, the pixels of the lung vascular region in the left anterior oblique and right anterior oblique X-rays are matched to obtain overlapping pixel groups. Based on the overlapping pixel groups and the lung vascular information, the left anterior oblique and right anterior oblique X-rays are translated and superimposed, and the fused lung vascular part is attached to the occluded part to obtain the lung vascular X-ray image. The method for obtaining pulmonary vascular information includes: Obtain the DTW distance between the edge lines of the occluded portion. When the DTW distance between two edge lines is the minimum of all their corresponding DTW distances, assign the two edge lines to an edge line group and match the edge line group with the connected components in the foreground portion between the edge lines. Based on the skeleton line of the connected region in the foreground portion, the similarity features of the two edge lines in the nearest matching edge line group, and the edge blurring features of the connected region, the lung vascular coefficient of the corresponding pixel in the connected region is obtained. The texture information is obtained based on the local grayscale differences of pixels within the connected domain; the lung vascular information is obtained by fusing the lung vascular coefficient and the texture information of the pixels. The lung vascular coefficient of the corresponding pixel in the connected domain is obtained by fusing the average DTW distance between the skeleton line of the connected domain and the two edge lines in the nearest matching edge line group, and the average gradient value of the edge lines of the connected domain. The variance of the grayscale values ​​of pixels in the preset neighborhood of a pixel within a connected domain is used as the texture information of the corresponding pixel.

2. The image enhancement method for chest X-rays used to assess heart failure according to claim 1, characterized in that, The method for obtaining the occluded area includes: Extract the edge line of the lung region; take the two pixels with the largest gradient value in the preset neighborhood of each pixel as the marked pixels; connect the marked pixels with the marked pixels in the preset neighborhood, and obtain the splitting factor of each connecting line segment according to the overall gradient value of each connecting line segment and the number of edge pixels; Select the closed connecting segment with the largest severance factor to obtain the occlusion area.

3. The image enhancement method for chest X-rays used to assess heart failure according to claim 1, characterized in that, The method for obtaining the pulmonary vascular region includes: Threshold segmentation is performed based on the amount of lung vascular information in the pixels of the target image, and the region formed by the pixels with values ​​higher than the threshold is taken as the lung vascular region.

4. The image enhancement method for chest X-rays used to assess heart failure according to claim 1, characterized in that, The method for obtaining the overlapping pixel group includes: Select each pixel in the lung vascular region of the target image as a target pixel; on a vertical line passing through the target pixel and perpendicular to the bottom of the image, obtain the pixels in the lung vascular region that the vertical line passes through, and record them as collinear pixels. Based on the grayscale values, curve fitting is performed on the collinear pixels on both sides of the target pixel. According to the similarity features of the two fitted curves of the target pixel and the pixels of the lung vascular region in another X-ray, the matching factor of the corresponding two pixels is obtained. The target pixel and the pixel corresponding to the largest matching factor are combined to form an overlapping pixel group.

5. The image enhancement method for chest X-rays used to assess heart failure according to claim 4, characterized in that, The method for obtaining the matching factor includes: Based on the positional relationship between the fitted curve and the target pixel, the fitted curve is divided into an upper curve and a lower curve; the sum of the absolute residuals between the upper curves and the lower curves of the two pixels is fused to obtain the matching factor for the corresponding two pixels.

6. The image enhancement method for chest X-rays used to assess heart failure according to claim 1, characterized in that, The method for superimposing and shifting X-ray images of left and right anterior oblique based on the overlapping pixel group and the amount of pulmonary vascular information includes: The edges of the left and right anterior oblique X-ray films are aligned, and the overlapping pixel group with the farthest distance in each row is marked as the marked pixel group. The pixel distance between the marked pixel groups is obtained. Within the range of the pixel distance, the target translation distance is selected one by one. Based on the difference between the pixel distance between each group of marked pixels and the target translation distance, and combined with the average value of the pulmonary vascular information of two pixels within the group of marked pixels, the superposition loss coefficient of the target translation distance is obtained. Select the target translation distance corresponding to the smallest superposition loss coefficient, and perform translation superposition on the X-ray images of the left and right anterior oblique angles.

7. The image enhancement method for chest X-rays used to assess heart failure according to claim 1, characterized in that, The method for obtaining the obscured portion includes: An x-axis is established by connecting the tops of the bilateral diaphragms in an orthogonal X-ray, and a y-axis is established by the midline of the spine. The target image is superimposed on the orthogonal X-ray, and the tops of the bilateral diaphragms and the midline of the spine are matched. The occluded area is projected onto the target image based on the position of the occluded area on the coordinate axis to obtain the occluded portion.

Images