A cartilage recognition system in medical images based on image enhancement

Abstract

The application relates to the technical field of image enhancement, and discloses a cartilage recognition system in medical images based on image enhancement; the system comprises an image segmentation module, an image denoising module, a region recognition module and an image enhancement module; after connecting lines of region critical points, the image segmentation module segments region images; the image denoising module selects an independent denoising mode or a mixed denoising mode to denoise the region images; the region recognition module recognizes a target region where cartilage is located; the image enhancement module enhances and stretches the target region to generate a cartilage enhanced image; and the cartilage recognition module recognizes cartilage information; the application effectively avoids mutual interference phenomena existing in a whole unified mode during noise source analysis, and can provide a continuous and accurate stretching position for the enhancement and stretching of the target region; the application can not only realize enlargement processing effect of the cartilage part, but also can highlight the cartilage state recognition locally, thereby effectively improving the recognition accuracy of cartilage characteristics, details and state information in cartilage recognition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image enhancement, more particularly, the present application relates to a kind of cartilage recognition system in medical image based on image enhancement. BACKGROUND

[0002] In the MRI image of knee joint, due to the interference of various factors, the obtained MRI image has various noises, which will cause the edge of the region image of the cartilage part in the MRI image of knee joint to be blurred, the resolution to be low and the details to be unclear, and thus hinder the further recognition and analysis of the cartilage. Therefore, it is necessary to perform image enhancement processing on the MRI image of knee joint to improve the image clarity and recognition accuracy of the cartilage in the MRI image.

[0003] The patent application with publication number CN110443790A discloses a cartilage recognition method and system in medical image, which uses the mutation edge of single quantitative information to form automatic processing and classification, improves the recognition accuracy of tissue edge, uses different contour generation methods for verification to avoid overfitting, so that the cartilage tissue can realize automatic segmentation, accurate contour positioning and automatic three-dimensional modeling in the gray image, and effectively improves the recognition efficiency of professional manpower recognition resources.

[0004] In the image enhancement processing of existing medical image, the image is usually recognized and analyzed by a whole unified analysis method for the noise type in the image, and the subsequent whole denoising and enhancement processing effect of the image is finally realized. However, when there are noises generated by multiple different noise source types in the image, the multiple different types of noises will interfere with each other in the whole unified analysis, resulting in inaccurate analysis results of the noise source, so that the position of the cartilage region with characteristics in the subsequent denoised image cannot be accurately enhanced and optimized, the characteristics of the cartilage region are not obvious and prominent enough from other characteristics, and thus the recognition accuracy of the cartilage in the MRI image of knee joint is reduced.

[0005] In view of this, the present application provides a cartilage recognition system in medical image based on image enhancement to solve the above problems. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purpose, the present application provides the following technical solution: a cartilage recognition system in medical image based on image enhancement, applied to an image processor, comprising:

[0007] An image segmentation module is used to import the MRI image of knee joint sagittal position into the image processor, identify the region critical point in the MRI image, and segment the region image after connecting the region critical point.

[0008] The image denoising module is configured to analyze the noise dimension of the region image, analyze the noise pollution source of the region image, and select an independent denoising mode or a hybrid denoising mode to denoise the region image, so as to generate a denoised image;

[0009] The region identification module is configured to determine a knee joint region in the denoised image, mark a target node in the knee joint region, and identify a target region where the cartilage is located.

[0010] The image enhancement module is configured to draw stretch lines of the target region, perform enhancement stretching on the target region based on an enhancement stretching criterion, and generate a cartilage enhanced image, wherein the enhancement stretching criterion is that all the stretch lines are synchronously stretched along a vertical direction.

[0011] The cartilage identification module is configured to input the cartilage enhanced image into a cartilage identification model to identify cartilage information corresponding to the cartilage enhanced image.

[0012] Further, the region critical points include first critical points and second critical points.

[0013] The identification method of the region critical points is as follows:

[0014] Draw lines at positions of boundaries along a vertical direction of the MRI image, draw two perimeter lines, and mark pixel values of each pixel point in the MRI image.

[0015] Take the pixel points at the top ends of the two perimeter lines as starting points, and mark A non-adjacent points on the two perimeter lines at equal distances according to a preset marking length, to obtain A perimeter points.

[0016] Draw extension lines from the A perimeter points as starting points to the inside of the MRI image along a horizontal direction, until the pixel values of the end points of the extension lines are not 0, mark the end points of the extension lines as region critical points, to obtain A first critical points and A second critical points.

[0017] Further, the segmentation method of the region image is as follows:

[0018] Start from the first critical point and the second critical point corresponding to the starting point, sequentially connect the remaining first critical points and second critical points to form a first auxiliary line and a second auxiliary line.

[0019] Along the vertical direction, extend the first auxiliary line and the second auxiliary line downward to the bottom edge of the MRI image to draw a first critical line and a second critical line.

[0020] Mark a region between the first critical line and the second critical line in the MRI image as a target region, and segment the target region from the MRI image to generate a region image.

[0021] Further, the analysis method of the thermal noise source, the bias noise source and the volume noise source is:

[0022] The four corner points of the connection area image form two diagonal lines, the lengths of the two diagonal lines are measured, and the sixth of the minimum value of the diagonal line length is recorded as the standard length;

[0023] A pixel point is randomly selected as the center of the circle in the area image, and a B pixel circle with overlapping area is drawn with the standard length as the radius, and the pixel point with a pixel value greater than the calibrated pixel value is recorded as a noise point;

[0024] C noise points are respectively taken as the center of the circle, and a line is drawn with the fifth of the standard length as the radius, and C noise circles are drawn in the pixel circle with adjacent positions tangent to each other;

[0025] The distance value of any two noise points in the C noise circles is measured along the horizontal direction and the vertical direction respectively, and the maximum value of the distance value is recorded as the horizontal value and the vertical value, and the noise shape index is calculated after comparing the horizontal value and the vertical value;

[0026] The noise circle with a noise shape index less than the noise shape threshold is recorded as a demand circle, and the number of demand circles in the pixel circle is counted, and the pixel circle with a demand circle number greater than one-third of the number of noise circles is recorded as a thermal noise circle;

[0027] When there are two thermal noise circles with overlapping areas, there is a thermal noise source in the area image;

[0028] A horizontal line is drawn in the area image along the horizontal direction, and one end of the horizontal line is taken as the starting point and the other end is taken as the ending point, when the pixel value of the pixel point on the horizontal line increases or decreases in turn, there is a bias noise source in the area image;

[0029] The femur, patella and tibia in the area image are identified through computer vision technology, and the femur line, patella line and tibia line are drawn by drawing a line along the position of the outer edge of the femur, patella and tibia;

[0030] The first line segment, the second line segment and the third line segment are respectively cut out equidistantly on the femur line, the patella line and the tibia line, the bending degree of the first line segment, the second line segment and the third line segment is measured, and the first line segment, the second line segment and the third line segment with a bending degree greater than the calibrated bending degree are recorded as the first bending line, the second bending line and the third bending line;

[0031] When there are three continuous first line segments, second line segments or third line segments, there is a volume noise source in the area image.

[0032] Further, the selection method of the independent denoising mode or the mixed denoising mode is:

[0033] counting the number of noise pollution sources, when the number of noise pollution sources is 1, selecting an independent denoising mode;

[0034] when the number of noise pollution sources is 2 or 3, a mixed denoising mode is selected.

[0035] Further, the determination method of the knee joint region is:

[0036] Along the vertical direction, the distance value of the pixel point on the femur line to the pixel point on the tibia line is measured, and the two pixel points corresponding to the minimum distance value are recorded as the femur reference point and the tibia reference point, respectively;

[0037] Along the horizontal direction, the distance values of the two pixel points on the femur line and the tibia line are measured, respectively, and the maximum values of the two distance values are recorded as the femoral span value and the tibial span value, respectively;

[0038] Along the vertical direction, the distance value of the two pixel points on the patella line is measured, and the maximum value of the distance value is recorded as the patellar span value;

[0039] The sum of the femoral span value and the patellar span value is taken as the upper span length, and the upper span point position which is one upper span length away from the femur reference point is marked above the femur reference point in the denoised image along the vertical direction. After drawing a horizontal line through the position where the upper span point position is located, an upper region line is drawn.

[0040] The tibial span value is taken as the lower span length, and the lower span point position which is one lower span length away from the tibia reference point is marked below the tibia reference point in the region image along the vertical direction. After drawing a horizontal line through the position where the lower span point position is located, a lower region line is drawn, and the region between the upper region line and the lower region line is recorded as the knee joint region.

[0041] Further, the identification method of the target region is:

[0042] Mark the two pixel points corresponding to the femoral span value and the tibial span value on the femur line and the tibia line, respectively, as the femoral boundary point and the tibial boundary point;

[0043] Randomly select a pixel point between the femur line and the tibia line as the center of a circle to draw a target circle, and continuously increase the radius of the target circle until both the two femoral boundary points and the two tibial boundary points are located inside the target circle. When the target circle is increased, the region inside the target circle is recorded as the to-be-verified region;

[0044] Convert the highlighted to-be-verified region into a highlighted signal graph, mark the signal value of each signal point in the highlighted signal graph, and record the signal point whose signal value is between the first signal threshold value and the second signal threshold value as the target node;

[0045] The target nodes located at the edge are identified by an edge detection technique, the target nodes located at the edge are connected to form a closed area, and the area inside the closed area is recorded as a target area.

[0046] Further, when drawing the stretching line, first, any three adjacent target nodes in the target area are combined to form a triangular mesh, the lengths of the three mesh edges of the triangular mesh are measured and recorded as mesh edge lengths, then the difference values of any two mesh edge lengths in the triangular mesh are calculated one by one, and the triangular mesh with two inconsistent difference values is recorded as a stretching mesh, then a target node is randomly marked inside the stretching mesh, the position of the target node is continuously adjusted so that the perpendicular distances of the target node to the three mesh edges are equal, and the adjusted target node is recorded as a stretching point, finally, the stretching points in adjacent triangular meshes are sequentially connected to draw the stretching line.

[0047] Further, the generation method of the cartilage enhanced image is:

[0048] The number of stretching points contained in all stretching lines is counted one by one, and the number of stretching points is compared with the number of stretching lines to calculate a stretching ratio;

[0049] The actual stretching value is calculated by multiplying the stretching ratio by the calibrated stretching value;

[0050] Taking the actual stretching value as a standard, all the stretching lines are synchronously enhanced and stretched to both sides along the direction perpendicular to the stretching line, and the target area after stretching is recorded as a cartilage enhanced image.

[0051] Further, the cartilage information includes a local thickness value, an elastic modulus value and a smooth area;

[0052] The identification method of the cartilage information is:

[0053] A plurality of groups of cartilage enhanced images and cartilage information are collected in advance, the plurality of groups of cartilage enhanced images are converted into a plurality of groups of feature vectors, the cartilage information is converted into labels, and a plurality of groups of labels are obtained;

[0054] One feature vector corresponds to one label to form a group of training data, and a plurality of groups of training data form a training set, and the labeled training data are divided into a training set and a test set;

[0055] The feature vector is used as the input of a deep learning model, the label corresponding to the feature vector is used as the output of the deep learning model, the training set is used to train the deep learning model, the test set is used to test the deep learning model, a preset error threshold is set, when the mean value of the prediction errors of all training data in the test set is less than the preset error threshold, a cartilage recognition model is trained;

[0056] The collected cartilage enhanced image is converted into a feature vector and input into a cartilage recognition model to recognize cartilage information.

[0057] The technical effect and advantages of the cartilage recognition system in medical images based on image enhancement are as follows:

[0058] (1) By drawing a pixel circle with an overlapping area and a tangent state noise circle in the regional image, and combining the measurement of the bending degree of the femoral line, the patellar line and the tibial line in a certain length range, the noise form generated by the thermal noise source and the volume noise source in the regional image can be accurately identified, realizing independent and orderly identification of the noise caused by the thermal noise source, the bias noise source and the volume noise source from the positional relationship and the structural form, effectively avoiding the mutual interference phenomenon existing in the overall unified manner in the noise source analysis, and thus providing accurate noise source data support for the subsequent denoising processing in image enhancement.

[0059] (2) By constructing a triangular mesh and dynamically adjusting the production of stretching points in the triangular mesh, accurate and independent point indications can be provided for the enhancement stretching of the target area, and by drawing the stretching line by connecting the stretching points, the continuous and accurate stretching positions can be provided for the enhancement stretching of the target area under the restriction of the enhancement stretching criterion, which can not only realize the enlargement processing effect of the part corresponding to the cartilage in the target area, but also highlight the local effect of the cartilage state, thereby effectively improving the identification accuracy of the cartilage features, details and state information in cartilage identification. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 A schematic diagram of the architecture of the cartilage recognition system in medical images based on image enhancement is provided for the first embodiment of the present application.

[0061] Figure 2 A flowchart of the cartilage recognition method in medical images based on image enhancement is provided for the second embodiment of the present application. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0063] Embodiment one: please refer to Figure 1 The cartilage recognition system in medical images based on image enhancement described in the present embodiment is applied to an image processor and includes:

[0064] The image segmentation module imports the MRI image of the knee joint sagittal position into the image processor, identifies the region critical point in the MRI image, and after connecting the region critical point, the region image is segmented out;

[0065] The MRI image of the knee joint sagittal position is an image obtained by a nuclear magnetic resonance imaging instrument from the sagittal position of the knee joint, that is, the imaging operation can be performed on the bones, cartilages, muscles, ligaments and other tissues at the knee joint of the photographer, and used as the target medical image for subsequent identification of cartilage.

[0066] Limited by the medical image imaging management rules, the specifications of the MRI image are usually consistent, and the region in the middle position in the MRI image is the trunk where the knee joint is located, and the regions on both sides are the background without substantial meaning. In order to improve the subsequent specific identification of the cartilage in the knee joint in the MRI image, it is necessary to divide the knee joint region and the background region in the MRI image;

[0067] When dividing the knee joint region and the background region, the region critical point is identified at the position between the two regions to ensure that the region critical point is the point of the boundary position of the knee joint region and the background region in the MRI image.

[0068] The region critical point includes a first critical point and a second critical point; wherein the first critical point and the second critical point are respectively the segmentation points of the connection between the knee joint corresponding region and the two side background regions, and the region between the first critical point and the second critical point is taken as the knee joint region.

[0069] The identification method of the region critical point is:

[0070] The MRI image is imported into the image processor, a line is drawn along the position of the boundary in the vertical direction of the MRI image, two perimeter lines are drawn, and the pixel value of each pixel point in the MRI image is marked;

[0071] Taking the pixel points at the top of the two perimeter lines as the starting points and taking the preset marking length as the standard, A non-adjacent points are marked on the two perimeter lines respectively, A perimeter points are obtained; the preset marking length is used to represent the distance between the two adjacent perimeter points, so as to ensure that the two adjacent perimeter points are kept equidistant and independent, and to provide reasonable position limitation for the first critical point and the second critical point;

[0072] On the two perimeter lines, respectively taking the A perimeter points as the starting points, the extension lines extending to the inside of the MRI image are drawn along the horizontal direction;

[0073] When the pixel value of the pixel point at the end of the extension line is not 0, the pixel point at the end of the extension line is recorded as a region critical point, and A first critical points and A second critical points are obtained. When the pixel value is 0, it indicates that the region color corresponding to the pixel point at this time is black, and it is a background region.

[0074] After marking the region critical points, the region between the region critical points at this time is the overall region of the knee joint where the cartilage is located, and the image corresponding to the region is recorded as a region image, and finally the region image is segmented and divided from the MRI image;

[0075] The segmentation method of the region image is:

[0076] Starting from the first critical point and the second critical point corresponding to the starting point, the remaining first critical points and the second critical points are sequentially connected to form a first auxiliary line and a second auxiliary line;

[0077] Along the vertical direction, the first auxiliary line and the second auxiliary line are extended to the bottom edge of the MRI image to draw a first critical line and a second critical line;

[0078] The region between the first critical line and the second critical line in the MRI image is recorded as a target region, and the target region is segmented from the MRI image to generate a region image.

[0079] It should be noted that the segmentation of the region image is not physically cutting, but virtually hiding the part of the MRI image that does not belong to the region image in the image processor, so that only the required part of the image is retained in the MRI image.

[0080] The image denoising module analyzes the noise dimension of the region image, analyzes the noise pollution source of the region image, and selects an independent denoising mode or a mixed denoising mode based on the noise pollution source to denoise the region image, and generates a denoised image;

[0081] During the MRI image shooting process of the knee joint, the quality of the MRI image is affected by the external environment, the device magnetic field, and the patient himself, etc. The noise phenomenon exists in the shot MRI image, so that the specific content information in the segmented region image cannot be clearly and accurately obtained, and therefore the noise of the region image needs to be denoised.

[0082] Before denoising the region image, it is necessary to determine which types of noise exist in the region image, so it is necessary to analyze the noise dimension of the region image, identify the form of the noise in the region image, and determine the noise pollution source that brings noise pollution to the region image according to the result of the noise dimension analysis;

[0083] Specifically, noise pollution sources include thermal noise sources, bias noise sources, and volumetric noise sources. Among them, thermal noise sources refer to the noise pollution sources generated by the dynamic changes in the thermal motion of electrons in the radio frequency coil of the MRI scanner; bias noise sources refer to the noise pollution sources generated by the non-uniform magnetic field of the MRI scanner; and volumetric noise sources refer to the noise pollution sources generated by the accumulation of mixed tissue fluids within the patient's knee joint.

[0084] The analysis methods for thermal noise sources, bias noise sources, and volumetric noise sources are as follows:

[0085] The four corner points of the connected area image form two diagonals. The lengths of the two diagonals are measured, and one-sixth of the minimum diagonal length is recorded as the standard length.

[0086] In the regional image, a pixel is randomly selected as the center, and B pixel circles with overlapping regions are drawn with a standard length as the radius. A pixel circle with an overlapping region means that there is an overlapping area between two adjacent pixel circles. It is not that there is an overlapping area between two pixel circles at any position. By drawing pixel circles, the area of ​​noise recognition in the regional image can be increased, avoiding the problems of randomness and limitations. At the same time, it can also ensure that two adjacent pixel circles maintain a dynamic relationship, avoiding the subsequent data being too discrete.

[0087] The pixel values ​​of all pixels within the pixel circle are marked one by one, and pixels with pixel values ​​greater than the calibrated pixel values ​​are marked as noise points; the calibrated pixel values ​​are used to represent the minimum pixel values ​​corresponding to noise points, thus providing a basis for noise point identification;

[0088] Using C noise points as centers and lines with a radius of one-fifth of the standard length as radii, draw C adjacent tangent noise circles within the pixel circle. The tangent noise circles ensure that two adjacent noise circles remain independent, preventing data duplication and overlap within adjacent noise circles, thus ensuring the uniqueness of the data.

[0089] The distance between any two noise points in the C noise circles is measured along the horizontal and vertical directions respectively. The maximum value of the distance is recorded as the horizontal value and the vertical value. The noise shape index is calculated by comparing the horizontal value and the vertical value.

[0090] The formula for calculating the noise shape index is:

[0091] ;

[0092] In the formula, This is the noise shape index of the noise circle. The horizontal value of the noise circle. The vertical value of the noise circle;

[0093] Noise circles with noise shape indexes less than a noise shape threshold are recorded as demand circles, and the number of demand circles in a pixel circle is counted, and when the number of demand circles in the pixel circle is greater than one third of the number of noise circles, the pixel circle is recorded as a hot noise circle; the noise shape threshold is used to represent the maximum value of the noise shape index of the noise circle identified as a demand circle;

[0094] The positional relationship of all hot noise circles in the region image is identified, and when there are two hot noise circles with overlapping areas, there is a hot noise source in the region image at this time;

[0095] A horizontal line is drawn in the region image along the horizontal direction, and one end of the horizontal line is the starting point and the other end is the ending point, and when the pixel values of the pixel points on the horizontal line increase or decrease in turn, there is a bias noise source in the region image at this time;

[0096] The femur, patella and tibia in the region image are identified by computer vision technology, and lines are drawn along the positions of the outer edges of the femur, patella and tibia to draw femur lines, patella lines and tibia lines;

[0097] Continuous first line segments, second line segments and third line segments are respectively cut out on the femur lines, patella lines and tibia lines, the curvatures of the first line segments, second line segments and third line segments are measured, and the first line segments, second line segments and third line segments with curvatures greater than a calibration curvature are recorded as first curved lines, second curved lines and third curved lines; the calibration curvature is used to represent the minimum value of the curvature of the first line segments, second line segments and third line segments identified as curved lines;

[0098] When there are three continuous first line segments, second line segments or third line segments, there is a volume noise source in the region image at this time.

[0099] It should be noted that the analysis and judgment process of the hot noise source, the bias noise source and the volume noise source is not limited in the order, and the analysis and judgment order can be customized according to actual needs, and the analysis results of the hot noise source, the bias noise source and the volume noise source will not change with the change of the analysis and judgment order.

[0100] When the noise pollution source is analyzed, the noise in the region image needs to be denoised according to the type and number of noise pollution sources, and when the type and number of noise pollution sources are different, the denoising mode of the region image is also different;

[0101] Specifically, the denoising mode of the regional image includes an independent denoising mode and a mixed denoising mode; the independent denoising mode is a mode for denoising the regional image with only one noise pollution source, and the mixed denoising mode is a mode for denoising the regional image with two or more noise pollution sources; and the independent denoising mode and the mixed denoising mode cannot be selected at the same time, but only one of them can be selected.

[0102] The selection method of the independent denoising mode or the mixed denoising mode is as follows:

[0103] The number of noise pollution sources is counted, and when the number of noise pollution sources is 1, there is only one type of noise pollution source in the regional image, and the independent denoising mode is selected.

[0104] When the number of noise pollution sources is 2 or 3, there are two or three types of noise pollution sources in the regional image, and the mixed denoising mode is selected.

[0105] After the independent denoising mode or the mixed denoising mode is selected, the regional image needs to be denoised under the limitation of the mode.

[0106] For example, when the independent denoising mode is selected, the type of noise pollution source needs to be determined first. When the noise pollution source is thermal noise, there are "snowflake-shaped" white noise points in the regional image, and wavelet denoising technology or NLM denoising technology can be selected for denoising. When the noise pollution source is bias noise, there are "strip-shaped" white noise points in the regional image, and N4 correction technology or deep learning correction technology can be selected for denoising. When the noise pollution source is volume noise, the boundary of the regional image is not clear, and thin layer reconstruction technology can be selected for denoising.

[0107] In this embodiment, the wavelet denoising technology is based on multi-scale analysis, which separates noise and effective signal by decomposing the image into subbands of different frequencies to achieve the denoising effect. The NLM (non-local mean) denoising technology uses the redundancy of similar structures in the image to suppress noise. The N4 correction technology is a classic bias field correction method in MRI, which is used to eliminate the low-frequency brightness gradient caused by magnetic field inhomogeneity. The deep learning correction technology is a method of jointly correcting bias field and noise through data-driven. The thin layer reconstruction technology is a denoising technology that optimizes MRI imaging protocol and image post-processing to reduce partial volume effect and improve signal-to-noise ratio. The above-mentioned wavelet denoising technology, NLM denoising technology, N4 correction technology, deep learning correction technology and thin layer reconstruction technology are all prior art in the field, and are not optimized in this application. Therefore, they will not be described in detail here.

[0108] A region identification module is configured to determine a knee joint region in the denoised image, mark a target node in the knee joint region, and identify a target region where the cartilage is located.

[0109] After the region image is denoised to generate the denoised image, noise pollution in the denoised image can be effectively removed, so that the denoised image has better clarity, and the boundaries and shapes of different positions of the knee joint in the denoised image can be more clearly and accurately observed, and the region position of the knee joint can be accurately determined, so as to determine the knee joint region from the denoised image.

[0110] Since the denoised image contains more information, including part of the region on the femur above the knee joint and the tibia below the knee joint, and these regions have limitations for subsequent identification of the cartilage in the knee joint, it is necessary to determine the knee joint region containing only the position of the cartilage in the knee joint from the denoised image.

[0111] The method for determining the knee joint region comprises the following steps.

[0112] In the vertical direction, the distance value between the pixel point on the femur line and the pixel point on the tibia line is measured, and the two pixel points corresponding to the minimum distance value are recorded as the femur reference point and the tibia reference point, respectively. In general, the two femur reference points correspond to the left and right endpoints of the femur head, which is a spherical structure at the bottom of the femur, and the tibia reference point corresponds to the left and right endpoints of the tibia head, which is a spherical structure at the top of the tibia.

[0113] In the horizontal direction, the distance values of the two pixel points on the femur line and the tibia line are measured, respectively, and the maximum values of the two distance values are recorded as the femur span value and the tibia span value, respectively.

[0114] In the vertical direction, the distance value of the two pixel points on the patella line is measured, and the maximum value of the distance value is recorded as the patella span value.

[0115] The sum of the femur span value and the patella span value is taken as the upper span length, and an upper span point is marked at a distance of the upper span length from the femur reference point in the vertical direction above the femur reference point. After drawing a horizontal line through the position of the upper span point, an upper region line is drawn.

[0116] The tibia span value is taken as the lower span length, and a lower span point is marked at a distance of the lower span length from the tibia reference point in the vertical direction below the tibia reference point. After drawing a horizontal line through the position of the lower span point, a lower region line is drawn, and the region between the upper region line and the lower region line is recorded as the knee joint region.

[0117] After the knee joint region is determined, the knee joint region at this time includes all cartilage regions and a small amount of other torso tissues such as tibia and femur, thereby reducing the influence and burden of useless information and regions in the knee joint region.

[0118] In order to better and more accurately identify the cartilage from the knee joint region, the region belonging to the cartilage in the knee joint region needs to be further displayed, and the region where the cartilage region is located is recorded as a target region. When identifying the target region, the target node needs to be marked in the knee joint region, so that the target node can be used as a point basis for subsequent determination of the cartilage region and other regions.

[0119] Specifically, the identification method of the target region is:

[0120] Mark two pixel points corresponding to the femoral span value and the tibial span value on the femoral line and the tibial line, respectively, as femoral boundary points and tibial boundary points;

[0121] Randomly select a pixel point between the femoral line and the tibial line as the center of a circle to draw a target circle, and continuously increase the radius of the target circle until the two femoral boundary points and the two tibial boundary points are located inside the target circle. The region inside the increased target circle is recorded as a to-be-verified region. At this time, the area corresponding to the to-be-verified region can wrap the two femoral boundary points and the two tibial boundary points inside, and at the same time, it can also ensure that the area of the to-be-verified region is the minimum area that can wrap the two femoral boundary points and the two tibial boundary points, so as to ensure the accuracy of subsequent target region identification and achieve the minimum representation effect of the target region.

[0122] Convert the highlighted to-be-verified region into a highlighted signal graph, mark the signal value of each signal point in the highlighted signal graph, and mark the signal point whose signal value is between the first signal threshold value and the second signal threshold value as a target node. The first signal threshold value and the second signal threshold value are used to represent the minimum value and the maximum value of the signal value of the signal point identified as the target node, so as to provide an interval range for the signal value of the target node.

[0123] Identify the target nodes located at the edge through the edge detection technology, connect the target nodes located at the edge to form a closed region, and record the region inside the closed region as a target region.

[0124] It should be noted that through the identification of the target region, the position of the cartilage between the femur and the tibia in the knee joint can be fully represented and used as a direct object for further enhancement and other processing of the cartilage.

[0125] The image enhancement module draws a stretch line of the target region, enhances and stretches the target region based on an enhancement and stretching criterion, and generates a cartilage enhanced image.

[0126] After the target region is acquired, the target region at this time has and only has the cartilage in the knee joint, and the area occupied by the target region at this time can only occupy a very small part of the original MRI image, so that the specific texture features and other details in the target region cannot be intuitively displayed, and therefore, it is necessary to perform an enhanced stretching operation on the target region, so that the local position of the target region can be dynamically and reasonably enhanced and stretched, so as to facilitate subsequent re-identification processing of the cartilage;

[0127] The stretching line refers to a line formed by the point positions in the target region that need to be enhanced and stretched, so that the stretching line can provide a basis for subsequent enhanced stretching;

[0128] Specifically, when drawing the stretching line, first, any three adjacent target nodes in the target region are combined to form a triangular mesh, and the lengths of the three mesh edges of the triangular mesh are measured one by one and recorded as mesh edge lengths, then the difference values of any two mesh edge lengths in the triangular mesh are calculated one by one, and the triangular mesh with two inconsistent difference values is recorded as a stretching mesh, then a target node is randomly marked inside the stretching mesh, the position of the target node is continuously adjusted so that the perpendicular distances from the target node to the three mesh edges are consistent, and the adjusted target node is recorded as a stretching point, and finally, the stretching points in the triangular meshes in adjacent positions are connected in sequence to draw the stretching line.

[0129] In the above drawing of the stretching line, the triangular meshes in adjacent positions need to be limited in a basic direction. In general, in principle, the stretching points in the triangular meshes in adjacent positions in the same horizontal direction need to be connected in sequence, but in actual operation, it is not necessary to maintain the same horizontal direction. The upward inclination angle and the downward inclination angle of the horizontal direction can be preset, and the stretching points in the triangular meshes in adjacent positions within the upward inclination angle and the downward inclination angle can also be connected in sequence.

[0130] It should be noted that the number of stretching lines can be 1 or more than 1. When there are multiple characteristic protrusions or irregular phenomena in the cartilage, the range of the target region that needs to be enhanced and stretched is also larger, and therefore, the number of stretching lines can also be larger.

[0131] After the stretching line is determined, the stretching line can be used as a stretching reference to perform enhanced stretching processing on the stretching line in the target region, and the image corresponding to the stretched target region is recorded as a cartilage enhanced image;

[0132] When stretching the stretching line, the stretching needs to be performed under the limitation of the enhanced stretching criterion to ensure the stability and accuracy of the enhanced stretching of the stretching line;

[0133] The enhanced stretching criterion is to stretch all the stretching lines in the vertical direction synchronously, so as to ensure that the target region can be enhanced and stretched as a whole and synchronously, and improve the stability and accuracy of the enhanced stretching.

[0134] The method for generating the cartilage enhanced image is:

[0135] The number of stretching points contained in all the stretching lines is counted one by one, and the stretching ratio is calculated after comparing the number of stretching points with the number of stretching lines.

[0136] The calculation formula of the stretching ratio is:

[0137] ;

[0138] In the formula, is the stretching ratio, is the number of stretching points, is the number of stretching lines;

[0139] The actual stretching value is calculated by multiplying the stretching ratio by the calibrated stretching value. The calibrated stretching value is the amplitude of the enhanced stretching of the stretching line, and is a constant value that can be set according to actual needs.

[0140] The calculation formula of the actual stretching value is:

[0141] ;

[0142] In the formula, is the actual stretching value, is the calibrated stretching value;

[0143] The actual stretching value is calculated by multiplying the stretching ratio by the calibrated stretching value. The calibrated stretching value is the amplitude of the enhanced stretching of the stretching line, and is a constant value that can be set according to actual needs.

[0144] Through the operation of enhancing and stretching the stretching line, the part corresponding to the cartilage in the target region can be enlarged, the recognition accuracy of the detailed features in the cartilage can be improved, and the difference between the cartilage and the non-cartilage can be enlarged, so as to highlight the information affecting the quality of the cartilage.

[0145] The cartilage recognition module inputs the cartilage enhanced image into the trained cartilage recognition model to recognize the cartilage information corresponding to the cartilage enhanced image.

[0146] After obtaining the cartilage enhanced image, the cartilage enhanced image can be input into the pre-trained cartilage recognition model, and the information affecting the quality of the cartilage in the cartilage enhanced image can be learned deeply by the cartilage recognition model, and the cartilage information directly reflecting the actual cartilage quality can be generated.

[0147] Cartilage information is used to represent the quality of cartilage state in cartilage enhanced images. Specifically, the cartilage information includes local thickness value, elastic modulus value and smooth area.

[0148] The local thickness value is used to represent the minimum thickness value of the cartilage contour, and is used as a basis for judging the thickness of the cartilage. The elastic modulus value is used to represent the compressive elastic properties of the cartilage contour. The smooth area is used to represent the total area value of the contour region in the smooth state of the cartilage contour.

[0149] In this embodiment, the cartilage recognition model is based on a deep learning model, combined with current big data and big computing power, and trained by deep learning and iterative optimization of a large number of different types of cartilage enhanced images and corresponding cartilage information. The cartilage recognition model can recognize the corresponding cartilage information according to the input cartilage enhanced image.

[0150] The method for identifying cartilage information is as follows:

[0151] A plurality of sets of cartilage enhanced images and cartilage information are collected in advance, the plurality of sets of cartilage enhanced images are converted into a plurality of sets of feature vectors, and the cartilage information is converted into labels to obtain a plurality of sets of labels.

[0152] One feature vector corresponds to one label to form a set of training data, and a plurality of sets of training data form a training set. The labeled training data is divided into a training set and a test set. 70% of the training data is used as the training set, and 30% of the training data is used as the test set.

[0153] The feature vector is used as the input of the deep learning model, and the label corresponding to the feature vector is used as the output of the deep learning model. The training set is used to train the deep learning model, and the test set is used to test the deep learning model. A preset error threshold is set. When the average prediction error of all training data in the test set is less than the preset error threshold, the cartilage recognition model is trained.

[0154] The collected cartilage enhanced image is converted into a feature vector and input into the cartilage recognition model to identify the cartilage information.

[0155] After obtaining the cartilage information, the corresponding local thickness value, elastic modulus value and smooth area in the cartilage information are directly exported from the image processor and sent to the client receiving the cartilage information, so that the client can intuitively and accurately display the specific state of the cartilage in the MRI image.

[0156] In this embodiment, by drawing the pixel circle with the overlapping area and the noise circle in the tangential state in the region image, and combining the measurement of the femoral line, the patellar line and the tibial line in the bending degree of a certain length range, the noise pattern generated by the thermal noise source and the volume noise source in the region image can be accurately identified, the noise caused by the thermal noise source, the bias noise source and the volume noise source is independently and orderly identified from the positional relationship and the morphological structure, the mutual interference phenomenon existing in the overall unified manner in the noise source analysis is effectively avoided, and accurate noise source data support is provided for the subsequent image enhancement denoising processing.

[0157] By constructing a triangular mesh and dynamically adjusting the production of stretching points in the triangular mesh, accurate and independent point indications can be provided for the enhancement stretching of the target region, and by drawing the stretching line by connecting the stretching points, the continuous and accurate stretching positions can be provided for the enhancement stretching of the target region under the restriction of the enhancement stretching criterion, which can not only realize the enlargement processing effect of the part corresponding to the cartilage in the target region, but also highlight the local effect of the cartilage state, thereby effectively improving the identification accuracy of the cartilage characteristics, details and state information in the cartilage identification.

[0158] Embodiment two: please refer to Figure 2 As shown in the figure, the part not described in detail in this embodiment is described in embodiment one, and a cartilage identification method in medical images based on image enhancement is provided, which is applied to an image processor and realized based on a cartilage identification system in medical images based on image enhancement, and includes:

[0159] S01: import the MRI image of the knee joint sagittal position into the image processor, identify the region critical point in the MRI image, and after connecting the region critical point, segment the region image;

[0160] S02: analyze the noise dimension of the region image, analyze the noise pollution source of the region image, and select an independent denoising mode or a mixed denoising mode to denoise the region image, to generate a denoised image;

[0161] S03: determine the knee joint region in the denoised image, mark the target node in the knee joint region, and identify the target region where the cartilage is located;

[0162] S04: draw the stretching line of the target region, perform enhancement stretching on the target region based on the enhancement stretching criterion, and generate a cartilage enhanced image;

[0163] S05: input the cartilage enhanced image into the cartilage identification model, and identify the cartilage information corresponding to the cartilage enhanced image.

[0164] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application.

Claims

1. An image enhancement based cartilage recognition system in medical images, applied to an image processor, characterized in that, The method comprises the following steps: An image segmentation module is used to import a knee joint sagittal MRI image into an image processor, identify regional critical points in the MRI image, and segment the regional image after connecting the regional critical points; An image denoising module is used to analyze the noise dimension of the regional image, analyze the noise pollution source of the regional image, and select an independent denoising mode or a mixed denoising mode to denoise the regional image, thereby generating a denoised image, wherein the noise pollution source includes a thermal noise source, a bias noise source, and a volume noise source; The analysis method of the thermal noise source, the bias noise source, and the volume noise source is as follows: A pixel point is randomly selected as the center of a circle in the regional image, a standard length is used as the radius to draw B pixel circles with overlapping regions, and a pixel point with a pixel value greater than a standard pixel value is recorded as a noise point; C noise points are respectively used as the center of a circle, and a standard length of one-fifth is used as the radius to draw C noise circles that are tangent to each other within the pixel circle; The distance between any two noise points in the C noise circles is measured along the horizontal direction and the vertical direction, respectively, and the maximum value of the distance is recorded as the horizontal value and the vertical value. After comparing the horizontal value and the vertical value, the noise shape index is calculated; Noise circles with a noise shape index less than a noise shape threshold are recorded as demand circles, the number of demand circles in the pixel circle is counted, and the pixel circle with a demand circle number greater than one-third of the number of noise circles is recorded as a thermal noise circle; When there are two thermal noise circles with overlapping regions, there is a thermal noise source in the regional image; A horizontal line is drawn in the regional image along the horizontal direction, and one end of the horizontal line is used as the starting point and the other end is used as the ending point. When the pixel values of the pixel points on the horizontal line increase or decrease sequentially, there is a bias noise source in the regional image; The femur, patella, and tibia in the regional image are identified through computer vision technology, and the femur line, patella line, and tibia line are drawn along the positions of the outer edges of the femur, patella, and tibia; The first, second, and third line segments are continuously cut on the femur line, patella line, and tibia line at equal distances, the curvatures of the first, second, and third line segments are measured, and the first, second, and third line segments with a curvature greater than a standard curvature are recorded as the first, second, and third curved lines; When there are three continuous first, second, or third line segments, there is a volume noise source in the regional image; A regional identification module is used to determine the knee joint region in the denoised image, mark the target node in the knee joint region, and identify the target region where the cartilage is located; An image enhancement module is used to draw the stretching line of the target region, perform enhancement stretching on the target region based on an enhanced stretching criterion, and generate a cartilage enhanced image, wherein the enhanced stretching criterion is that all stretching lines are synchronously stretched along the vertical direction; A cartilage identification module is used to input the cartilage enhanced image into a cartilage identification model to identify the cartilage information corresponding to the cartilage enhanced image.

2. The cartilage recognition system in medical images based on image enhancement according to claim 1, characterized in that, The regional critical points include first critical points and second critical points; The identification method of the regional critical points is as follows: Draw a line along the position of the boundary in the vertical direction of the MRI image, draw two perimeter lines, and mark the pixel value of each pixel point in the MRI image; Take the pixel points at the top of the two perimeter lines as the starting points, and mark A non-adjacent points on the two perimeter lines at equal distances with a preset marking length to obtain A perimeter points; Draw an extension line from each of the A perimeter points in the horizontal direction until the pixel point at the end of the extension line has a pixel value of 0, and mark the pixel point at the end of the extension line as a region critical point to obtain A first critical point and A second critical point.

3. The cartilage recognition system in medical images based on image enhancement according to claim 2, characterized in that, The segmentation method of the region image is: From the first critical point and the second critical point corresponding to the starting point, sequentially connect the remaining first critical points and the second critical points to form a first auxiliary line and a second auxiliary line; Along the vertical direction, extend the first auxiliary line and the second auxiliary line to the bottom edge of the MRI image to draw a first critical line and a second critical line; Mark the region between the first critical line and the second critical line in the MRI image as a target region, and segment the target region from the MRI image to generate a region image.

4. The cartilage recognition system in medical images based on image enhancement according to claim 3, characterized in that, The selection method of the independent denoising mode or the mixed denoising mode is: Count the number of noise pollution sources, and when the number of noise pollution sources is 1, select the independent denoising mode; When the number of noise pollution sources is 2 or 3, select the mixed denoising mode.

5. The cartilage recognition system in medical images based on image enhancement according to claim 4, characterized in that, The determination method of the knee joint region is: Along the vertical direction, measure the distance value from the pixel point on the femur line to the pixel point on the tibia line, and mark the two pixel points corresponding to the minimum distance value as the femur reference point and the tibia reference point, respectively; Along the horizontal direction, measure the distance values of the two pixel points on the femur line and the tibia line, and mark the maximum of the two distance values as the femoral span value and the tibial span value, respectively; Along the vertical direction, measure the distance value of the two pixel points on the patella line, and mark the maximum distance value as the patellar span value; Take the sum of the femoral span value and the patellar span value as the upper span length, mark an upper span point in the denoised image above the femur reference point at a distance of one upper span length from the femur reference point, and draw an upper region line by drawing a horizontal line through the position of the upper span point; Take the tibial span value as the lower span length, mark a lower span point in the region image below the tibia reference point at a distance of one lower span length from the tibia reference point, draw a lower region line by drawing a horizontal line through the position of the lower span point, and mark the region between the upper region line and the lower region line as the knee joint region.

6. The cartilage recognition system in medical images based on image enhancement according to claim 5, characterized in that, The identification method of the target region is: Mark the two pixel points corresponding to the femoral span value and the tibial span value on the femur line and the tibia line as the femoral boundary point and the tibial boundary point, respectively; Draw a target circle with a randomly selected pixel point between the femur line and the tibia line as the center, draw a target circle, and continuously increase the radius of the target circle until both femoral boundary points and both tibial boundary points are located inside the target circle, and mark the region inside the increased target circle as a to-be-verified region. The highlight conversion of the to-be-verified region into a highlight signal graph, marking the signal value of each signal point in the highlight signal graph, and marking the signal point with a signal value between the first signal threshold and the second signal threshold as a target node; Identifying the target nodes located at the edge through the edge detection technology, connecting the target nodes located at the edge to form a closed region, and marking the region inside the closed region as a target region.

7. The cartilage recognition system in medical images based on image enhancement according to claim 6, characterized in that, In drawing the stretching line, first, any three adjacent target nodes in the target region form a triangular mesh, and the lengths of the three mesh edges are measured and marked as mesh edge lengths, then the difference between any two mesh edge lengths is calculated one by one, and the triangular mesh with two inconsistent differences is marked as a stretching mesh, then a target node is randomly marked inside the stretching mesh, the position of the target node is adjusted so that the perpendicular distances from the target node to the three mesh edges are equal, and the adjusted target node is marked as a stretching point, and finally, the stretching points in adjacent triangular meshes are connected in turn to draw the stretching line.

8. The cartilage recognition system in medical images based on image enhancement according to claim 7, characterized in that, The generation method of the cartilage enhanced image is: The number of stretching points contained in all the stretching lines is counted one by one, and the number of stretching points is compared with the number of stretching lines to calculate the stretching ratio; The actual stretching value is calculated by multiplying the stretching ratio and the calibrated stretching value; Taking the actual stretching value as the standard, all the stretching lines are synchronously stretched to both sides in the direction perpendicular to the stretching line, and the target region after stretching is marked as the cartilage enhanced image.

9. The cartilage recognition system in medical images based on image enhancement according to claim 8, characterized in that, The cartilage information includes local thickness value, elastic modulus value and smooth area; The identification method of the cartilage information is: A plurality of groups of cartilage enhanced images and cartilage information are collected in advance, the plurality of groups of cartilage enhanced images are converted into a plurality of groups of feature vectors, the cartilage information is converted into labels, and a plurality of groups of labels are obtained; One feature vector corresponds to one label to form a group of training data, and a plurality of groups of training data form a training set, and the labeled training data are divided into a training set and a test set; The feature vector is used as the input of the deep learning model, the label corresponding to the feature vector is used as the output of the deep learning model, the training set is used to train the deep learning model, the test set is used to test the deep learning model, a preset error threshold is set, when the mean of the prediction errors of all training data in the test set is less than the preset error threshold, the cartilage recognition model is trained; The collected cartilage enhanced image is converted into a feature vector and input into the cartilage recognition model to identify the cartilage information.

Images