Medical image intelligent analysis method

Abstract

The invention relates to the field of medical image processing, in particular to a medical image intelligent analysis method. The method comprises the steps that original medical image data are collected and preprocessed, and an image matrix is obtained; respectively calculating a gradient value and a density value based on the image matrix, calculating a partition regulation and control value, and adaptively adjusting partitions; and on the basis of the adjusted partitions, calculating a feature matrix of the partitions, performing similarity clustering, calculating a clustering center of each cluster, performing abnormal partition detection, and generating an analysis report. The problems that traditional image analysis methods mostly depend on a fixed partition strategy, the size of a partition cannot be adjusted in a self-adaptive mode according to changes of local features in an image, and diagnosis accuracy is affected are solved; generally, a non-linear processing mechanism is lacked, the feature extraction process is easily interfered, and subsequent analysis and judgment are influenced. The problems that missing detection or false detection is easily caused and reliable diagnosis support cannot be provided due to the fact that an abnormal area is judged by simple threshold setting are solved.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a medical image intelligent analysis method. Background Art

[0002] With the continuous development of medical imaging technology, imaging devices such as MRI (magnetic resonance imaging), CT (computed tomography) and ultrasound imaging have been widely used in clinical diagnosis. The imaging devices can generate high-resolution tissue images to help doctors identify lesion areas and formulate treatment plans. In recent years, with the advancement of computer vision and artificial intelligence technology, automated medical image analysis methods based on deep learning and machine learning have gradually emerged. The method can extract features from a large amount of medical image data, realize automatic detection of lesion areas, and significantly improve diagnostic efficiency and accuracy.

[0003] However, the above-mentioned image analysis methods still have some urgent problems to be solved in medical image analysis: first, most of them rely on fixed partitioning strategies, which are limited when processing complex medical images. They cannot adaptively adjust the size of the partitions according to the changes in local features in the image. As a result, they may not be sensitive enough to the detection of subtle lesions, or cannot perform fine processing on lesions with complex structures, thus affecting the accuracy of diagnosis; secondly, in terms of feature extraction, they usually lack nonlinear processing mechanisms. When faced with noise and outliers in medical images, the feature extraction process is easily disturbed, resulting in the inability to accurately capture truly meaningful feature information, thereby affecting subsequent analysis and judgment; finally, they often rely on simple threshold settings to judge abnormal areas, which can easily lead to missed detections or false detections, with low accuracy and inability to provide reliable diagnostic support. Summary of the invention

[0004] The present invention provides a medical image intelligent analysis method to solve the problems that most traditional image analysis methods rely on fixed partitioning strategies, are limited when processing complex medical images, and cannot adaptively adjust the size of the partition according to changes in local features in the image. As a result, the detection of subtle lesions may be insufficiently sensitive, or lesions with complex structures cannot be finely processed, thereby affecting the accuracy of diagnosis; in terms of feature extraction, there is usually a lack of nonlinear processing mechanism. When faced with noise and outliers in medical images, the feature extraction process is easily disturbed, resulting in the inability to accurately capture truly meaningful feature information, thereby affecting subsequent analysis and judgment; and rely on simple threshold settings to judge abnormal areas. When processing subtle differences between normal tissues and lesion areas, it is easy to cause missed detection or false detection, with low accuracy and inability to provide reliable diagnostic support.

[0005] A medical image intelligent analysis method of the present invention specifically includes the following technical solutions:

[0006] A medical image intelligent analysis method comprises the following steps:

[0007] S1: Collect and preprocess the original medical image data to obtain an image matrix; calculate the gradient value and density value based on the image matrix; calculate the partition control value based on the gradient value and the density value; adaptively adjust the partition based on the partition control value to obtain the adjusted partition;

[0008] S2: Based on the adjusted partitions, calculate the feature matrix of the partitions; based on the feature matrix of the partitions, perform similarity clustering and calculate the cluster center of each cluster; based on the feature matrix of the partitions and the cluster center of the cluster to which the partitions belong, perform abnormal partition detection and generate an analysis report.

[0009] Preferably, the S1 specifically includes:

[0010] Based on the image matrix, the second-order derivative and logarithmic function are introduced to calculate the gradient value.

[0011] Preferably, the S1 specifically includes:

[0012] Set the sliding window size, move the sliding window pixel by pixel on the image matrix, introduce nonlinear activation function and gradient product, extract density features of the neighborhood of each pixel, and calculate the density value.

[0013] Preferably, the S1 specifically includes:

[0014] Based on the gradient value and density value, a dynamic partition control function is constructed to calculate the partition control value; the dynamic partition control function is as follows:

[0015]

[0016] Wherein, P(i, j) is the partition control value of the pixel at position (i, j); i, j are the row index and column index in the image matrix respectively; k is the height of the sliding window; l is the width of the sliding window; m is the row index in the sliding window; n is the column index in the sliding window; G(m, n) is the gradient value of the pixel at position (m, n) in the sliding window; D(m, n) is the density value of the pixel at position (m, n) in the sliding window; α and β are the gradient adjustment parameters and density adjustment parameters respectively; D(i, j) is the density value of the pixel at position (i, j); D avg is the average density value of the image.

[0017] Preferably, the S1 specifically includes:

[0018] The image matrix is ​​divided into rectangular partitions according to a preset size, and the average of the partition control values ​​of all pixels in the partition is calculated to obtain the average control value of the partition; when the average control value of the partition is lower than the preset control threshold, the partition is refined and the partition is adaptively adjusted to obtain the adjusted partition.

[0019] Preferably, the S2 specifically includes:

[0020] Based on the adjusted partitions, the gradient value, density value and distance between pixels of the partitioned pixels are fused to calculate the eigenvalues ​​by combining logarithmic operation, exponential decay and fractional structure. Based on the eigenvalues, the feature matrix of the partitions is constructed. The formula for generating the feature matrix is ​​as follows:

[0021]

[0022] Among them, F r (p, q) is the feature value at position (p, q) in the rth partition; r is the index variable of the partition; (p, q) is the coordinate of the pixel in the rth partition; N r is the number of pixels in the rth partition; Represents all pixel positions in the rth partition exp(·) is the exponential operation; is the position in the rth partition The gradient value of the pixel at ; is the position in the rth partition The density value of the pixel at .

[0023] Preferably, the S2 specifically includes:

[0024] Based on the feature matrix of the partition, the k-means clustering algorithm is used to perform similarity clustering; based on the feature matrix of the partition in the cluster, the nonlinear combination of matrix product and determinant is introduced to obtain the cluster center of the cluster to which the partition belongs.

[0025] Preferably, the S2 specifically includes:

[0026] Based on the feature matrix of the partition and the cluster center of the cluster to which the partition belongs, the abnormal measurement value of the partition is calculated; based on the abnormal measurement value of the partition, the abnormal partition is detected; when the abnormal measurement value is greater than the preset abnormal threshold, the partition to which the abnormal measurement value belongs is marked as an abnormal area, all abnormal areas are visualized on the original medical image data, and an analysis report is generated.

[0027] The beneficial effects of the technical solution of the present invention are:

[0028] 1. The present invention constructs a dynamic partition control function, combines the gradient features and density features of the image, and adaptively adjusts the partition size according to the local changes in the image, thereby avoiding the limitations of the traditional fixed-size partition strategy; especially when processing medical imaging data with complex structures or high noise, the adaptive partition method can accurately capture local features, improve the detection accuracy of abnormal areas, avoid over-reliance on global fixed parameter settings, and make image analysis more robust and adaptable to more practical clinical scenarios.

[0029] 2. The present invention introduces second-order derivatives and nonlinear functions (such as logarithmic and hyperbolic tangent functions) in the calculation of gradient features and density features, which effectively enhances the ability to capture subtle changes in the image and suppresses the interference of outliers on the calculation results; by combining the gradient values ​​and density values ​​of local pixels, a feature matrix generation formula is proposed that can more accurately reflect the complex local features in the image, so that edge detection and feature extraction have higher accuracy and robustness, which helps to improve the diagnostic value of the image.

[0030] 3. The present invention can effectively detect abnormal areas in images by measuring the difference between feature matrices and cluster centers; the introduction of Euclidean distance and logarithmic processing, combined with the determinant of the feature matrix, further improves the accuracy and stability of abnormality detection, and can quickly and accurately identify the feature differences of different partitions. Especially when used to mark lesion areas in clinical practice, it can intuitively present abnormal areas in images, reduce the workload of doctors during the diagnosis process, and provide more accurate diagnostic references. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of a medical image intelligent analysis method described in the present invention. DETAILED DESCRIPTION

[0032] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0033] Unless defined otherwise, 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 belongs.

[0034] The specific scheme of the medical image intelligent analysis method provided by the present invention is described in detail below with reference to the accompanying drawings.

[0035] Refer to the attached Figure 1 , which shows a flow chart of a medical image intelligent analysis method provided by an embodiment of the present invention, the method comprising the following steps:

[0036] S1: Collect and preprocess the original medical image data to obtain an image matrix; calculate the gradient value and density value based on the image matrix; calculate the partition control value based on the gradient value and the density value; adaptively adjust the partition based on the partition control value to obtain the adjusted partition;

[0037] First, collect original medical image data, such as MRI (magnetic resonance imaging) or CT (computed tomography) image data, and preprocess the original medical image data; the preprocessing specifically includes: first graying the original medical image data, then using Gaussian filtering to perform denoising, and finally using Sobel operator to perform edge enhancement operation to obtain an image matrix for subsequent analysis;

[0038] Then, based on the image matrix, the gradient value is calculated to reflect the intensity change rate and intensity of the pixel. The larger the gradient value, the higher the pixel intensity change rate and the more intense the change.

[0039] The gradient calculation formula is as follows:

[0040]

[0041] Among them, G(i, j) is the gradient value of the pixel at position (i, j), that is, the gradient feature of the pixel; i, j are the row index and column index in the image matrix, respectively, indicating the position of the image in the horizontal and vertical directions; I(i, j) is the intensity value of the pixel at position (i, j); are the first-order derivatives of the image in the horizontal and vertical directions, respectively, indicating the intensity change rate of the pixel intensity value I(i, j) in the row and column directions; are the second-order derivatives of the image in the horizontal and vertical directions, respectively, indicating the rate of change of the intensity value I(i, j) of the pixel in the row and column directions;

[0042] The gradient calculation formula introduces second-order derivatives and logarithmic functions, which can accurately capture edge changes in the image and reduce the impact of outliers on gradient calculation through nonlinear operations;

[0043] Next, the expert experience method is used to flexibly set the sliding window size according to the characteristic scale, resolution and noise level of the image in the specific implementation scenario, and the sliding window is moved pixel by pixel on the image matrix, and the density feature is extracted from the neighborhood of each pixel to calculate the density value;

[0044] Specifically, the neighborhood of the pixel refers to the local area around the pixel, which is determined by the size of the sliding window and is used to analyze the intensity distribution and changes of the pixels in the local area; the density value reflects the intensity distribution and changes of the pixel in the local area by performing nonlinear processing on the intensity values ​​of all pixels in the neighborhood. The larger the density value, the more uneven the pixel intensity distribution and the more drastic the change in the local area where the pixel is located;

[0045] The density calculation formula is as follows:

[0046]

[0047] Where D(i, j) is the density value of the pixel at position (i, j), that is, the density feature of the pixel. The boundary pixels of the image are processed by mirror filling operation, that is, the part outside the edge of the image is first filled by mirror copying, and then the density value is calculated; k is the height of the sliding window; l is the width of the sliding window; k×l represents the size of the sliding window; m is the row index in the sliding window; n is the column index in the sliding window; tanh is the hyperbolic tangent function, which is used for nonlinear transformation and compresses the input value range to avoid the influence of outliers; I(m, n) is the intensity value of the pixel at position (m, n) in the sliding window, It is a double summation operation, which is used to sum all pixels in the neighborhood window centered at (i, j); are the first-order derivatives of the image in the sliding window in the horizontal and vertical directions respectively;

[0048] The density calculation formula uses nonlinear activation function and gradient product to improve the ability to distinguish density features in local areas, enhance the sensitivity to local density changes, and avoid the influence of outliers on density feature extraction;

[0049] Next, based on the gradient value and density value, a dynamic partition control function is constructed to calculate the partition control value for adjusting the partition size. The dynamic partition control function is as follows:

[0050]

[0051] Wherein, P(i, j) is the partition control value of the pixel at position (i, j), which is used to control the size and shape of the partition. A small partition control value indicates that the area needs a finer division; G(m, n) is the gradient value of the pixel at position (m, n) in the sliding window; D(m, n) is the density value of the pixel at position (m, n) in the sliding window; α and β are the gradient adjustment parameters and density adjustment parameters, respectively, which are used to control the weights of the pixel gradient value and the pixel density value on the partition control, and are obtained through experimental tuning; D avgis the average density value of the image, obtained by averaging the density values ​​of all pixels in the image matrix;

[0052] The dynamic partition control function introduces the density value and gradient value of pixels in the neighborhood, calculates the partition control value, dynamically adjusts the partition size, and avoids the fixed-size partition strategy;

[0053] Finally, based on the partition control value, the partition is adaptively adjusted to obtain the adjusted partition. Specifically, the image matrix is ​​divided into rectangular partitions according to a preset size, for example, the size of each rectangular area is 32 pixels by 32 pixels, and then the average of the partition control values ​​of all pixels in the partition is calculated to obtain the average control value of the partition; the control threshold is set according to the specific implementation scenario. If the average control value of the partition is lower than the control threshold, it means that the gradient and density of the partition change more drastically, and the partition needs to be refined to capture details, such as refining a 32 pixel by 32 pixel partition into a 16 pixel by 16 pixel partition. The preset size, control threshold and specific refinement level are set according to the specific implementation scenario.

[0054] S2: Based on the adjusted partitions, calculate the feature matrix of the partitions; based on the feature matrix of the partitions, perform similarity clustering and calculate the cluster center of each cluster; based on the feature matrix of the partitions and the cluster center of the cluster to which the partitions belong, perform abnormal partition detection and generate an analysis report.

[0055] First, based on the adjusted partitions, the feature matrix of the partitions is calculated to capture the characteristics of the partitions;

[0056] The feature matrix of the partition is composed of eigenvalues, which are obtained by calculating the gradient value, density value and distance from the central pixel (i.e., the pixel at the position (p, q)) of the pixel, and reflect the characteristic change of the pixel in the partition; the feature matrix of the partition reflects the overall characteristic pattern of the entire partition, and is used to capture the image feature distribution in the partition;

[0057] The formula for generating the feature matrix is ​​as follows:

[0058]

[0059] Among them, F r (p, q) is the feature value at position (p, q) in the rth partition; r is the index variable of the partition; (p, q) is the coordinate of the pixel in the rth partition, indicating the pixel position in the partition; N r is the number of pixels in the rth partition; Represents all pixel positions in the rth partition exp(·) is the exponential operation; is the position in the rth partition The gradient value of the pixel at ; is the position in the rth partition The density value of the pixel at .

[0060] The feature matrix generation formula combines logarithmic operations, exponential decay and fractional structure to fuse the gradient value, density value and distance between pixels of the partitioned pixels, which not only enhances the nonlinear expression of local features, but also suppresses the influence of pixels far away from the center point. The introduction of nonlinear operations (such as logarithms and exponentials) can avoid the excessive influence of outliers on the capture of partitioned features, increase the robustness and precision of the feature matrix generation formula in edge detection and feature extraction, and improve the recognition and classification capabilities of local features of images.

[0061] Then, based on the feature matrix of the partitions, similarity clustering is performed, and the partitions are classified into different partition sets, i.e., different clusters, according to the similarity of the features in each partition using the k-means clustering algorithm. The k-means clustering algorithm is a prior art and will not be described in detail here. Based on the feature matrix of the partitions in the cluster, the cluster center of each cluster is calculated, and the cluster center calculation formula is as follows:

[0062]

[0063] Among them, C k is the cluster center of the kth cluster, and represents the comprehensive feature center of the feature matrix of all partitions in the kth cluster; |S k | is the number of partitions contained in the kth cluster; S k is the k-th cluster, representing the k-th partition set; r∈S k Indicates that the rth partition belongs to the kth partition set S k ; F r is the feature matrix of the rth partition; is the feature matrix F of the rth partition r The transposed matrix of the partition is used to calculate the correlation of the feature matrix of the partition; det(F r ) is the feature matrix F of the rth partition r The determinant of is used to reflect the overall feature strength of the feature matrix of the partition;

[0064] The cluster center calculation formula, by introducing the nonlinear combination of matrix product and determinant, can capture the correlation and global structural characteristics between the feature matrices of the partitions and obtain the cluster center of the cluster to which the partition belongs;

[0065] Next, based on the feature matrix of the partition and the cluster center of the cluster to which the partition belongs, the partition's anomaly metric is calculated to detect abnormal partitions. The anomaly metric indicates the degree of difference between the partition and the cluster center. The larger the anomaly metric, the more obvious the feature difference between the partition and the cluster center, the greater the degree of deviation, and the greater the probability of belonging to an abnormal area.

[0066] The anomaly detection formula is as follows:

[0067]

[0068] Among them, A r is the anomaly measure of the rth partition; ||·|| 2 is the square of the Euclidean distance; |F r | is the feature matrix F of the rth partition r The dimension of is the number of elements of the feature matrix of the partition;

[0069] An abnormal threshold is set according to the specific implementation scenario. When the abnormal metric value is greater than the abnormal threshold, the partition to which the abnormal metric value belongs is marked as an abnormal area. All abnormal areas are visualized on the original medical image data. By intuitively marking the abnormal areas, medical staff can intuitively see the specific location and range of the abnormal areas. The annotations on the image are set according to the specific implementation scenario using borders, color filling or other visual prompts;

[0070] Finally, an analysis report is generated, which includes the number of abnormal areas and their specific locations in the image, and records the abnormal measurement value and partition size of each abnormal area, and provides a comparison between the original medical image data and the labeled medical image data. The analysis report is used as a diagnostic reference to assist doctors in determining whether further examinations, treatment plans, or follow-up clinical interventions are needed.

[0071] In summary, a medical image intelligent analysis method is completed.

[0072] The order of the embodiments of the invention is for description only and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0073] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A medical image intelligent analysis method, characterized in that: The following steps are involved: S1: Collect and preprocess the original medical imaging data to obtain an image matrix; calculate the gradient value and density value based on the image matrix; calculate the partition control value based on the gradient value and density value; Based on the partition control value, the partition is adaptively adjusted to obtain the adjusted partition; S2: Based on the adjusted partitions, calculate the feature matrix of the partitions; Based on the partitioned feature matrix, similarity clustering is performed and the cluster center of each cluster is calculated; Based on the feature matrix of the partition and the cluster center of the cluster to which the partition belongs, abnormal partition detection is performed and an analysis report is generated.

2. A medical image intelligent analysis method according to claim 1, characterized in that: The S1 specifically includes: Based on the image matrix, the second-order derivative and logarithmic function are introduced to calculate the gradient value.

3. A medical image intelligent analysis method according to claim 2, characterized in that: The S1 specifically includes: Set the sliding window size, move the sliding window pixel by pixel on the image matrix, introduce nonlinear activation function and gradient product, extract density features of the neighborhood of each pixel, and calculate the density value.

4. A medical image intelligent analysis method according to claim 3, characterized in that: The S1 specifically includes: Based on the gradient value and density value, a dynamic partition control function is constructed to calculate the partition control value; the dynamic partition control function is as follows: Wherein, P(i, j) is the partition control value of the pixel at position (i, j); i, j are the row index and column index in the image matrix respectively; k is the height of the sliding window; l is the width of the sliding window; m is the row index in the sliding window; n is the column index in the sliding window; G(m, n) is the gradient value of the pixel at position (m, n) in the sliding window; D(m, n) is the density value of the pixel at position (m, n) in the sliding window; α and β are the gradient adjustment parameters and density adjustment parameters respectively; D(i, j) is the density value of the pixel at position (i, j); D avg is the average density value of the image.

5. A medical image intelligent analysis method according to claim 4, characterized in that: The S1 specifically includes: The image matrix is ​​divided into rectangular partitions according to a preset size, and the average of the partition control values ​​of all pixels in the partition is calculated to obtain the average control value of the partition; when the average control value of the partition is lower than the preset control threshold, the partition is refined and the partition is adaptively adjusted to obtain the adjusted partition.

6. A medical image intelligent analysis method according to claim 1, characterized in that: The S2 specifically includes: Based on the adjusted partitions, the gradient value, density value and distance between pixels of the partitioned pixels are fused to calculate the eigenvalues ​​by combining logarithmic operation, exponential decay and fractional structure. Based on the eigenvalues, the feature matrix of the partitions is constructed. The formula for generating the feature matrix is ​​as follows: Among them, F r (p, q) is the feature value at position (p, q) in the rth partition; r is the index variable of the partition; (p, q) is the coordinate of the pixel in the rth partition; N r is the number of pixels in the rth partition; Represents all pixel positions in the rth partition exp(·) is the exponential operation; is the position in the rth partition The gradient value of the pixel at ; is the position in the rth partition The density value of the pixel at .

7. A medical image intelligent analysis method according to claim 6, characterized in that: The S2 specifically includes: Based on the feature matrix of the partition, the k-means clustering algorithm is used to perform similarity clustering; based on the feature matrix of the partition in the cluster, the nonlinear combination of matrix product and determinant is introduced to obtain the cluster center of the cluster to which the partition belongs.

8. A medical image intelligent analysis method according to claim 7, characterized in that: The S2 specifically includes: Based on the feature matrix of the partition and the cluster center of the cluster to which the partition belongs, the abnormal measurement value of the partition is calculated; based on the abnormal measurement value of the partition, the abnormal partition is detected; when the abnormal measurement value is greater than the preset abnormal threshold, the partition to which the abnormal measurement value belongs is marked as an abnormal area, all abnormal areas are visualized on the original medical image data, and an analysis report is generated.

Images