Medical image intelligent analysis method, system and device based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based medical image intelligent analysis method, system and device, which relates to the field of medical image analysis technology and is used to improve the problem that the complexity and diversity of medical image data leads to inaccurate detection area division, resulting in increased abnormality recognition errors and reduced image processing intelligence. By dividing the system into main and secondary areas, image data is collected and feature weights are set respectively, texture features are extracted from the detection area image, abnormal areas are judged, and the feature weights are combined to determine whether to enter the abnormality control mechanism. After entering the control mechanism, the grayscale value and noise value of the abnormal area are collected, and the error is analyzed in combination with the baseline value. An image optimization plan is formulated based on the error, feedback data is collected, and it is determined whether to adjust the detection area division. This method realizes differentiated modeling and dynamic control, improves the sensitivity of early lesion and minor abnormality recognition and the level of personalized intelligent processing, and thus improves the accuracy of medical image analysis.

Description

Technical Field

[0001] The present invention relates to the field of medical image analysis technology, and more specifically, to a medical image intelligent analysis method, system, and device based on artificial intelligence. Background Art

[0002] With the continuous development of medical imaging equipment and image processing technology, structural images obtained through multimodal image acquisition methods such as CT, MRI, and ultrasound are widely used in scientific research, structural identification, image management, and other scenarios. To improve the efficiency and accuracy of image processing, artificial intelligence-related technologies (such as image recognition, feature extraction, structural segmentation, and image enhancement) have been gradually embedded in various image analysis systems to achieve automatic recognition of image structures and judgment of regional changes.

[0003] The existing technology has the following deficiencies:

[0004] At present, the complexity and diversity of medical imaging data lead to indetailed detection area division, insufficient utilization of texture and local features, lack of hierarchical regulation of abnormal processing due to heterogeneity of abnormal areas, frequent ignoring of local noise distribution in error assessment, and lack of effective feedback on optimization effects, which lead to increased errors in abnormal recognition and decreased intelligence in image processing. Therefore, an intelligent analysis method, system and device for medical images based on artificial intelligence are proposed. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a medical image intelligent analysis method, system and device based on artificial intelligence. By dividing the detection area into master and slave areas, different feature weights are collected and assigned respectively, texture feature extraction and abnormal area identification are combined, the grayscale value and noise value of the abnormal area are dynamically adjusted, the error degree is analyzed based on the benchmark value, a targeted image optimization plan is formulated, and the detection area division is intelligently adjusted according to the feedback data to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The medical image intelligent analysis method based on artificial intelligence includes the following steps:

[0008] Step S1: Obtain an image of a detection area of ​​the detection object through a medical imaging device, divide the detection area of ​​the detection object into a main area and a slave area, collect image data of the main area and the slave area respectively, and set feature weights of different areas according to the image data;

[0009] Step S2: Extract texture features from the detection area image of the detection object, determine whether there is an abnormal area in the detection area based on the texture features, lock the area classification to which the abnormal area belongs, and determine whether to enter the abnormal control mechanism based on the feature weight of the corresponding area;

[0010] Step S3: When entering the abnormality control mechanism, the grayscale value and noise value of the abnormal area are collected, the grayscale reference value and noise reference value of the detection area in the image database are called, and the error degree of the abnormal area is analyzed in combination with the grayscale value and noise value of the abnormal area;

[0011] Step S4: Formulate an image optimization plan based on the error degree of the abnormal area, collect feedback data of the optimized abnormal area, and determine whether to divide and adjust the detection area based on the feedback data.

[0012] In a preferred embodiment, in step S1, the medical imaging device is an X-ray machine, and an image of a detection area of ​​the detection object is acquired by the X-ray machine, and the detection area of ​​the detection object is divided into a main area and a slave area. The specific steps are as follows:

[0013] Identify the center point of the detection area image, rotate the partition edge according to the preset initial value of the area division length at the center point, and obtain a circular area with the center point as the center and the area division length as the radius. The obtained circular area is used as the main area of ​​the detection area, and the area within the detection area but outside the circular area is used as the slave area of ​​the detection area;

[0014] The image data are the maximum brightness difference and radiation dose of the master region and the slave region respectively.

[0015] In a preferred embodiment, in step S1, feature weights of different regions are set according to the maximum brightness difference between the master region and the slave region and the radiation dose. The specific steps are as follows:

[0016] The ratio of the maximum brightness difference between the master area and the slave area to the sum of the maximum brightness differences of the two areas is used as the brightness difference weight of the master area and the brightness difference weight of the slave area respectively;

[0017] The ratios of the radiation doses of the main area and the slave area to the sum of the radiation doses of the two areas are used as the dose weight of the main area and the dose weight of the slave area respectively;

[0018] The multiplication result of the brightness difference weight and the dose difference weight of the main area is set as the feature weight of the main area, and the multiplication result of the brightness difference weight and the dose difference weight of the slave area is set as the feature weight of the slave area.

[0019] In a preferred embodiment, in step S2, when extracting texture features, a plurality of extraction sub-regions of different areas are randomly set in the detection area, and the regional entropy value of each extraction sub-region is calculated. If the regional entropy value exceeds the entropy value threshold, the corresponding extraction sub-region is judged to be an abnormal sub-region, and the number of abnormal sub-regions is counted. When the number of abnormal sub-regions is not 0, it is judged that there is an abnormal region in the detection area, and all abnormal sub-regions are regarded as abnormal regions;

[0020] Call the position of each abnormal sub-area. When all the abnormal sub-area positions belong to the main area, the detection area is judged to be the main area abnormal; when all the abnormal sub-area positions belong to the slave area, the detection area is judged to be the slave area abnormal; otherwise, the detection area is judged to be the entire area abnormal.

[0021] In a preferred embodiment, in step S2, when the detection area is abnormal in the main area or the sub-area, the average regional entropy value of all abnormal sub-areas is calculated and multiplied by the feature weight of the corresponding area, and the result of the calculation is used as the regional abnormality coefficient;

[0022] When the detection area is abnormal in the entire region, the average regional entropy value of the abnormal sub-region of the main region and the average regional entropy value of the abnormal sub-region of the slave region are calculated respectively, and weighted with the feature weight of the corresponding region to obtain the regional anomaly coefficient;

[0023] When the regional anomaly coefficient of the detection area exceeds the preset control threshold, the abnormal control mechanism is entered.

[0024] In a preferred embodiment, in step S3, when the abnormality control mechanism is entered, the corresponding grayscale value and noise value of the identified abnormal area are collected;

[0025] Through an algorithm based on texture analysis and edge gradient detection, the grayscale values ​​of all pixels in the locked abnormal area are extracted and the average value is calculated to obtain the grayscale value of the abnormal area;

[0026] By traversing the pixel grayscale in the abnormal area, the fluctuation degree of the pixel grayscale in the abnormal area is calculated to obtain the noise value of the abnormal area.

[0027] In a preferred embodiment, in step S3, the grayscale reference value and the noise reference value of the detection area are obtained by calling the image database;

[0028] Calculate the absolute difference between the grayscale value of the abnormal area and the grayscale reference value of the detection area to obtain the grayscale value difference of the abnormal area;

[0029] The absolute difference between the noise value of the abnormal area and the noise baseline value of the detection area is calculated to obtain the noise value difference of the abnormal area.

[0030] In a preferred embodiment, in step S3, the grayscale value difference of the abnormal area and the noise value difference of the abnormal area are normalized and substituted into the bivariate tangent function coupling mapping function to obtain the error degree of the abnormal area. The specific formula is expressed as follows:

[0031] ;

[0032] Where, is the error degree of the abnormal area, is the gray value difference of the abnormal area after normalization, is the noise value difference of the abnormal area after normalization, is the scale factor, is the inverse tangent function, is the normalization coefficient.

[0033] In a preferred embodiment, in step S4, a corresponding image optimization scheme is formulated according to the error degree of the abnormal area to eliminate the error degree of the abnormal area;

[0034] The image optimization solution is the bilateral filtering solution in the adaptive filtering algorithm;

[0035] After the image optimization scheme is optimized, feedback data of the abnormal area is collected, namely the local Fourier phase consistency index;

[0036] Based on the images before and after optimization of the abnormal area, two-dimensional Fourier transform is performed in the local window to extract the frequency domain phase information of the image. The local Fourier phase consistency index is obtained by calculating the consistency degree of the local phase.

[0037] Comparing the local Fourier phase consistency index with a preset division threshold;

[0038] If the local Fourier phase consistency index exceeds the division threshold, it means that there is no need to adjust the division of the detection area. If the local Fourier phase consistency index is lower than the division threshold, it means that the division of the detection area needs to be adjusted.

[0039] When the detection area is divided and adjusted, an expansion coefficient is randomly generated and multiplied by the initial value of the area division length to update the initial value of the area division length.

[0040] The medical image intelligent analysis system based on artificial intelligence is used to implement the above-mentioned medical image intelligent analysis method based on artificial intelligence, and includes the following devices:

[0041] X-ray machine: used to obtain images of the inspection area of ​​the inspection object;

[0042] Computer workstation: used to divide the detection area into regions, extract texture features, process data and determine whether to enter the abnormal control mechanism;

[0043] Image data sensor: used to collect grayscale value, noise value and frequency domain phase information of abnormal areas;

[0044] Image database: used to store and call grayscale reference values ​​and noise reference values;

[0045] Optimization software: built-in image optimization solution, generates feedback data to determine whether to divide and adjust the detection area, and generates expansion coefficients to adjust the divided area.

[0046] The technical effects and advantages of the medical image intelligent analysis method, system and device based on artificial intelligence of the present invention are as follows:

[0047] The present invention obtains an image of a detection area of ​​a detection object through a medical imaging device, divides the detection area of ​​the detection object into a main area and a slave area, collects image data of the main area and the slave area respectively, and sets feature weights of different areas according to the image data, extracts texture features from the image of the detection area of ​​the detection object, judges whether there is an abnormal area in the detection area according to the texture features, and chooses whether to enter an abnormal regulation mechanism in combination with the feature weight of the area where the abnormal area is located. When entering the abnormal regulation mechanism, the grayscale value and noise value of the abnormal area are collected, and the error degree of the abnormal area is analyzed in combination with the grayscale reference value and the noise reference value of the detection area. An image optimization plan is formulated according to the error degree of the abnormal area, and feedback data of the abnormal area after optimization is collected to judge whether to divide and adjust the detection area. Differential modeling and dynamic regulation of different structural areas can be achieved, and the recognition sensitivity of early lesions or minor abnormalities and the personalized intelligent level of image processing are improved, thereby improving the analysis accuracy of medical images. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Schematic diagram of the medical image intelligent analysis method based on artificial intelligence of the present invention.

[0049] Figure 2 This is a flow chart of the medical image intelligent analysis system based on artificial intelligence of the present invention. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0051] The present invention obtains an image of a detection area of ​​a detection object through a medical imaging device, divides the detection area of ​​the detection object into a main area and a slave area, collects image data of the main area and the slave area respectively, and sets feature weights of different areas according to the image data, extracts texture features from the image of the detection area of ​​the detection object, judges whether there is an abnormal area in the detection area according to the texture features, and chooses whether to enter an abnormal regulation mechanism in combination with the feature weight of the area where the abnormal area is located. When entering the abnormal regulation mechanism, the grayscale value and noise value of the abnormal area are collected, and the error degree of the abnormal area is analyzed in combination with the grayscale reference value and the noise reference value of the detection area. An image optimization plan is formulated according to the error degree of the abnormal area, and feedback data of the abnormal area after optimization is collected to judge whether to divide and adjust the detection area, thereby improving the analysis accuracy of the medical image.

[0052] Example 1, medical image intelligent analysis method based on artificial intelligence, such as Figure 1 As shown, the following steps are included:

[0053] Step S1: Obtain an image of a detection area of ​​the detection object through a medical imaging device, divide the detection area of ​​the detection object into a main area and a slave area, collect image data of the main area and the slave area respectively, and set feature weights of different areas according to the image data;

[0054] Step S2: Extract texture features from the detection area image of the detection object, determine whether there is an abnormal area in the detection area based on the texture features, lock the area classification to which the abnormal area belongs, and determine whether to enter the abnormal control mechanism based on the feature weight of the corresponding area;

[0055] Step S3: When entering the abnormality control mechanism, the grayscale value and noise value of the abnormal area are collected, the grayscale reference value and noise reference value of the detection area in the image database are called, and the error degree of the abnormal area is analyzed in combination with the grayscale value and noise value of the abnormal area;

[0056] Step S4: Formulate an image optimization plan based on the error degree of the abnormal area, collect feedback data of the optimized abnormal area, and determine whether to divide and adjust the detection area based on the feedback data.

[0057] The specific implementation is as follows:

[0058] In step S1 , the medical imaging device is an imaging device that acquires images of a detection area, such as an X-ray machine, a computer tomography scanner, etc.

[0059] The X-ray machine is used to obtain an image of the inspection area of ​​the inspection object, and the inspection area of ​​the inspection object is divided into a main area and a slave area. The specific steps are as follows:

[0060] Identify the center point of the detection area image, rotate the partition edge according to the preset initial value of the area division length at the center point, and obtain a circular area with the center point as the center and the area division length as the radius. The obtained circular area is used as the main area of ​​the detection area, and the area within the detection area and outside the circular area is used as the slave area of ​​the detection area.

[0061] It should be explained that the X-ray machine is a widely used medical imaging device used to obtain images of the results inside the body of the test object. The preset initial value of the area division length is not unique and can be set according to actual conditions. For example, if the initial value of the area division length is set to 10 cm, the center point is used as the center of the circle, and a circle with a radius of 10 cm is drawn, and the circular area is used as the main area of ​​the detection area.

[0062] The image data are the maximum brightness difference and radiation dose of the master area and the slave area respectively. When collecting the maximum brightness difference of the master area or the slave area, all pixels in the corresponding area are traversed to obtain the brightness values ​​of all pixels, and the difference between the maximum brightness value and the minimum brightness value is used as the maximum brightness difference of the corresponding area.

[0063] The greater the brightness difference in the detection area, the more likely it is to have motion artifacts. Motion artifacts are interference caused by the movement of the detection object during the imaging process. Artifacts will increase the brightness difference of the image in the detection area, resulting in image blur.

[0064] Radiation dose refers to the amount of radiation used by the X-ray machine when collecting images. The greater the radiation dose, the higher the image quality and the higher the image clarity.

[0065] Set the feature weights of different areas based on the maximum brightness difference between the master area and the slave area and the radiation dose. The specific steps are as follows:

[0066] The ratio of the maximum brightness difference of the main area to the sum of the maximum brightness differences of the two areas is used as the brightness difference weight of the main area, and the ratio of the maximum brightness difference of the slave area to the sum of the maximum brightness differences of the two areas is used as the brightness difference weight of the slave area;

[0067] The ratio of the radiation dose of the main area to the sum of the radiation doses of the two areas is used as the dose difference weight of the main area, and the ratio of the radiation dose of the slave area to the sum of the radiation doses of the two areas is used as the dose difference weight of the slave area;

[0068] The multiplication result of the brightness difference weight and the dose difference weight of the main area is set as the feature weight of the main area, and the multiplication result of the brightness difference weight and the dose difference weight of the slave area is set as the feature weight of the slave area.

[0069] In step S2, when extracting texture features, a plurality of extraction sub-regions of different areas are randomly set in the detection area, and the regional entropy value of each extraction sub-region is calculated. The regional entropy value is compared with a preset entropy value threshold. When the regional entropy value exceeds the entropy value threshold, the corresponding extraction sub-region is judged to be an abnormal sub-region, and the number of abnormal sub-regions is counted. When the number of abnormal sub-regions is not 0, it is judged that there is an abnormal area in the detection area, and all abnormal sub-regions are regarded as abnormal areas.

[0070] Call the position of each abnormal sub-area and lock the area where it is located. When all the abnormal sub-area positions belong to the main area, the detection area is judged to be abnormal in the main area; when all the abnormal sub-area positions belong to the slave area, the detection area is judged to be abnormal in the slave area; otherwise, the detection area is judged to be abnormal in the entire area;

[0071] When the detected area is abnormal in the main area, the average regional entropy value of all abnormal sub-areas is calculated and multiplied by the feature weight of the main area, and the result is used as the regional anomaly coefficient; when the detected area is abnormal in the slave area, the average regional entropy value of all abnormal sub-areas is calculated and multiplied by the feature weight of the slave area, and the result is used as the regional anomaly coefficient; when the detected area is abnormal in the entire area, the average regional entropy value of the abnormal sub-area of ​​the main area and the average regional entropy value of the abnormal sub-area of ​​the slave area are calculated respectively, and the maximum value is taken after weighted summation with the feature weight of the corresponding area as the regional anomaly coefficient.

[0072] When the regional anomaly coefficient of the detection area exceeds the preset control threshold, the abnormal control mechanism is entered.

[0073] It should be noted that the higher the regional anomaly coefficient of the detection area, the lower the image accuracy, and the more it needs to enter the anomaly control mechanism.

[0074] In step S3, when entering the abnormality control mechanism, the corresponding grayscale value and noise value of the identified abnormal area are collected;

[0075] The grayscale value of the abnormal area is the average grayscale value of the pixels in the abnormal area, assuming the abnormal area is a set of coordinates in the image. The acquisition logic is to extract the grayscale values ​​of all pixels in the locked abnormal area and calculate the average value through an algorithm based on texture analysis and edge gradient detection to obtain the grayscale value of the abnormal area.

[0076] The grayscale value calculation formula is as follows:

[0077] ;

[0078] Where, is the grayscale value of the abnormal area, For the image at coordinate point The gray value at is the total number of pixels in the abnormal area;

[0079] It should be noted that the algorithms for texture analysis and edge gradient detection can be Gabor filters, gray-level co-occurrence matrices, Sobel edge detection, etc. In this example, a texture direction response extraction method based on Gabor filters combined with an edge gradient enhancement method using the Sobel operator is used to accurately identify suspected structural abnormalities and assist in delineating the grayscale calculation area. This will not be described in detail here.

[0080] The coordinate set in the image is obtained by a region growing method, a K-means algorithm based on pixel clustering, or a region discrimination method based on a convolutional neural network. The specific acquisition method and method are not limited and will not be described in detail here.

[0081] Furthermore, when the grayscale value of the abnormal area is larger, it means that the overall brightness of the abnormal area is higher;

[0082] The noise value of the abnormal area is defined as the standard deviation or local variance of the pixel grayscale in the abnormal area. The acquisition logic is to traverse the pixel grayscale in the abnormal area, calculate the fluctuation degree of the pixel grayscale in the abnormal area, and obtain the noise value of the abnormal area.

[0083] The noise value of the abnormal area is obtained based on the fluctuation degree of the grayscale of the pixels in the abnormal area. Furthermore, the fluctuation degree of the grayscale of the pixels in the abnormal area is calculated by the standard deviation calculation formula of the grayscale value in the abnormal area, which is expressed as follows:

[0084] ;

[0085] Where, The noise value of the abnormal area, specifically the standard deviation of the grayscale value, reflects the discrete degree of pixel grayscale or texture complexity in the area. The larger the value, the greater the grayscale fluctuation within the image area, the more likely there is high noise or mixed tissue boundaries.

[0086] To detect the abnormal area coordinate set locked in the image, it contains the two-dimensional coordinates of all pixels belonging to the abnormal area;

[0087] For the image at pixel points Gray value at ;

[0088] is the average gray value of the abnormal area, and its calculation method is shown in the above average gray calculation formula;

[0089] It should be noted that the degree of fluctuation of the grayscale value can be extracted by an image statistics method, a spatial local feature enhancement method, or a noise estimation model based on an adaptive filtering window. In this embodiment, a sliding window combined with a median filter preprocessing method is preferably used to suppress background interference, and then a local standard deviation method is used for calculation in the noise enhancement area, thereby improving the stability and accuracy of edge structure recognition in abnormal areas. The specific implementation method is not limited and will not be described in detail here.

[0090] By calling the image database, the grayscale reference value and noise reference value of the detection area are obtained;

[0091] It should be noted that the image database is preset in an external image processing server. The data in the database is called through index matching or key feature retrieval to obtain the grayscale reference value and noise reference value of the detection area that matches the characteristics of the current detection object. The specific calling method is not described in detail here.

[0092] The grayscale reference value and noise reference value of the detection area are obtained by setting the image training samples or historical image statistical model;

[0093] More specifically, the grayscale baseline value and noise baseline value of the detection area are not kept as fixed constant values, but are updated according to factors such as image category, imaging device model, tissue type and area location;

[0094] The execution time interval and update intensity parameters of the update rule are determined by our experimenters based on factors such as the number of image samples, historical anomaly distribution, and actual detection accuracy requirements, and will not be elaborated here;

[0095] Calculate the absolute difference between the grayscale value of the abnormal area and the grayscale reference value of the detection area to obtain the grayscale value difference of the abnormal area;

[0096] Calculate the absolute difference between the noise value of the abnormal area and the noise baseline value of the detection area to obtain the noise value difference of the abnormal area;

[0097] It should be noted that, given that the abnormal region is located within the detection region, those skilled in the art should understand that the grayscale reference value and noise reference value corresponding to the detection region can be used as reference values ​​for the abnormal region to assess the degree of deviation of the image characteristics of the abnormal region. The relevant corresponding relationship is something that can be directly inferred by those skilled in the art and is not elaborated here.

[0098] Normalize the grayscale value difference and the noise value difference of the abnormal area so that the grayscale value difference and the noise value difference of the abnormal area are kept in the same dimension.

[0099] It should be noted that the standardization methods include but are not limited to standard linear transformation based on interval scaling, Z-Score standardization method based on statistics, or normalization method based on nonlinear mapping function. The application methods of standardization are not described in detail here.

[0100] Substitute the grayscale value difference of the abnormal area after normalization and the noise value difference of the abnormal area into the bivariate tangent function coupling mapping function to obtain the error degree of the abnormal area. The specific formula is expressed as follows:

[0101] ;

[0102] Where, is the error degree of the abnormal area;

[0103] is the gray value difference of the abnormal area after normalization;

[0104] is the noise value difference of the abnormal area after normalization;

[0105] is the scale factor;

[0106] is the inverse tangent function, with a domain of real numbers and a range of , which has monotonically increasing and saturation characteristics, can map arbitrarily large differences to a finite interval and suppress the influence of extreme outliers;

[0107] is the normalization coefficient, which ensures that the maximum value of the error degree is 1 after the product of the two inverse tangent function values ​​is normalized, thus realizing the standardized expression of the error degree;

[0108] Specifically, the above formula comprehensively considers the differences in grayscale and noise in abnormal areas through the coupled mapping of the bivariate inverse tangent function, avoids the error judgment dominated by a single indicator, and enhances the robustness and sensitivity of the comprehensive evaluation of the abnormal area error;

[0109] It should be noted that the greater the difference between the grayscale value of the abnormal area after normalization and the noise value of the abnormal area, the greater the error degree of the abnormal area, and the more obvious the characteristics of the abnormal area;

[0110] In step S4, a corresponding image optimization scheme is formulated according to the error degree of the abnormal area to eliminate the error degree of the abnormal area;

[0111] Among them, the image optimization scheme is a bilateral filtering scheme in the adaptive filtering algorithm set by people in this field;

[0112] Specifically, those skilled in the art implement image optimization using a bilateral filtering scheme in an adaptive filtering algorithm based on the error degree of the abnormal area;

[0113] Specific implementation steps include:

[0114] First, determine the spatial neighborhood range and filter radius of the abnormal area;

[0115] Secondly, the spatial weight and pixel value similarity weight are calculated based on the pixel grayscale difference within the abnormal area;

[0116] Then the two weights are combined to form a joint weight coefficient, and weighted smoothing is performed on the pixels in the abnormal area;

[0117] Finally, the filtering parameters are dynamically adjusted to adapt to different error levels, achieving a balance between noise elimination and edge protection, thereby improving image quality and diagnostic accuracy.

[0118] It should be noted that the specific implementation and method of the image optimization scheme are selected and adjusted by the experimenters according to the error degree of the abnormal area and the requirements of the specific application scenario. The optimization schemes adopted include but are not limited to the bilateral filtering scheme in the adaptive filtering algorithm;

[0119] After the image optimization scheme is optimized, feedback data of abnormal areas is collected;

[0120] The feedback data from the abnormal area is a dimensionless numerical indicator used to characterize the stability of image structure and texture, namely the local Fourier phase consistency index. Its acquisition logic is based on the images before and after optimization of the abnormal area. A two-dimensional Fourier transform is performed in the local window to extract the frequency domain phase information of the image. The local Fourier phase consistency index is obtained by calculating the consistency of the local phase.

[0121] In order to capture the local structural information of the image, the image is divided into several overlapping or non-overlapping local windows (such as pixel blocks, etc.). The window size is determined according to the texture details and resolution of the abnormal area, which will not be described in detail here;

[0122] Furthermore, the two-dimensional Fourier transform is a mathematical transformation that converts a two-dimensional spatial domain image signal into a frequency domain signal, which is defined as:

[0123] ;

[0124] Where, The grayscale value of the spatial domain image pixel, is a complex value in the frequency domain, containing amplitude and phase information, and are the number of pixels of the image in the horizontal and vertical directions, is the frequency domain coordinate, is an imaginary unit;

[0125] Specifically, the amplitude and phase information are the results of Fourier transform and are in complex form, and the formula is expressed as:

[0126] ;

[0127] Where, is the amplitude, is the phase, frequency domain phase information The geometric features that reflect the local structure and texture of the image will not be described in detail here;

[0128] Furthermore, the local Fourier phase consistency index is based on the idea of ​​phase synchronization and measures whether the phases of different frequency components in a local window are consistent.

[0129] The specific calculation steps include: applying a set of bandpass filters (such as Gabor filters) of different frequencies and directions to the image in the local window to extract the phase response; calculating the conjugate consistency measurement formula of the phase of each frequency component is:

[0130] ;

[0131] Where, For location The phase consistency at the position ranges from 0 to 1. For the The amplitude weight of each frequency component;

[0132] Specifically, the numerator is the vector and amplitude of different frequency phases, and the denominator is the amplitude sum to achieve normalization; the values ​​of all pixels in the local window are The local Fourier phase consistency index is obtained by averaging. The higher the local Fourier phase consistency index, the more complete the image structure information, the higher the texture recovery, and the better the image optimization effect. Otherwise, it indicates that the abnormal area still has problems such as structural distortion or detail loss.

[0133] Comparing the local Fourier phase consistency index with a preset division threshold;

[0134] If the local Fourier phase consistency index exceeds the division threshold, it means that there is no need to adjust the division of the detection area. If the local Fourier phase consistency index is lower than the division threshold, it means that the division of the detection area needs to be adjusted.

[0135] When the detection area is divided and adjusted, an expansion coefficient is randomly generated and multiplied by the initial value of the area division length to update the initial value of the area division length.

[0136] It should be noted that the range of values ​​for the expansion coefficient and the setting of the division threshold are set by the experimenters based on factors such as the resolution level of different types of medical images, texture complexity, or recognition accuracy requirements of tissues and organs, and will not be elaborated here.

[0137] Example 2, medical image intelligent analysis system based on artificial intelligence, such as Figure 2 As shown, a method for realizing intelligent analysis of medical images based on artificial intelligence includes the following devices:

[0138] X-ray machine: used to obtain images of the inspection area of ​​the inspection object;

[0139] Computer workstation: used to divide the detection area into regions, extract texture features, process data and determine whether to enter the abnormal control mechanism;

[0140] Image data sensor: used to collect grayscale value, noise value and frequency domain phase information of abnormal areas;

[0141] Image database: used to store and call grayscale reference values ​​and noise reference values;

[0142] Optimization software: built-in image optimization solution, generates feedback data to determine whether to divide and adjust the detection area, and generates expansion coefficients to adjust the divided area.

[0143] The above embodiments may be implemented in whole or in part through software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0144] Those skilled in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application of the technical solution and the invention constraints. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0145] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0146] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

[0147] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An artificial intelligence-based medical image intelligent analysis method, characterized in that: The following steps are involved: Step S1: Acquire an image of a detection area of ​​a detection object through a medical imaging device, divide the detection area of ​​the detection object into a master area and a slave area, collect image data of the master area and the slave area respectively, and set feature weights of different areas according to the image data; the image data is the maximum brightness difference and the radiation dose; Step S2: Extract texture features from the detection area image of the detection object, determine whether there is an abnormal area in the detection area based on the texture features, lock the area classification to which the abnormal area belongs, and determine whether to enter the abnormal control mechanism based on the feature weight of the corresponding area; Step S3: When entering the abnormality control mechanism, the grayscale value and noise value of the abnormal area are collected, the grayscale reference value and noise reference value of the detection area in the image database are called, and the error degree of the abnormal area is analyzed in combination with the grayscale value and noise value of the abnormal area; Step S4: Formulate an image optimization plan based on the error degree of the abnormal area, collect feedback data of the optimized abnormal area, and determine whether to divide and adjust the detection area based on the feedback data; The specific process of formulating the corresponding image optimization plan based on the error degree of the abnormal area is as follows: The image optimization solution is the bilateral filtering solution in the adaptive filtering algorithm; After the image optimization scheme is optimized, feedback data of the abnormal area is collected, namely the local Fourier phase consistency index; Based on the images before and after optimization of the abnormal area, two-dimensional Fourier transform is performed in the local window to extract the frequency domain phase information of the image. The local Fourier phase consistency index is obtained by calculating the consistency degree of the local phase. Comparing the local Fourier phase consistency index with a preset division threshold; If the local Fourier phase consistency index exceeds the division threshold, it means that there is no need to adjust the division of the detection area. If the local Fourier phase consistency index is lower than the division threshold, it means that the division of the detection area needs to be adjusted. When the detection area is divided and adjusted, an expansion coefficient is randomly generated and multiplied by the initial value of the area division length to update the initial value of the area division length.

2. The medical image intelligent analysis method based on artificial intelligence according to claim 1, characterized in that: In step S1, the medical imaging device is an X-ray machine, which acquires an image of a detection area of ​​the detection object through the X-ray machine and divides the detection area of ​​the detection object into a main area and a slave area. The specific steps are as follows: Identify the center point of the detection area image, rotate the partition edge according to the preset initial value of the area division length at the center point, and obtain a circular area with the center point as the center and the area division length as the radius. The obtained circular area is used as the main area of ​​the detection area, and the area within the detection area and outside the circular area is used as the slave area of ​​the detection area.

3. The medical image intelligent analysis method based on artificial intelligence according to claim 2, characterized in that: In step S1, the feature weights of different regions are set according to the maximum brightness difference between the master region and the slave region and the radiation dose. The specific steps are as follows: The ratio of the maximum brightness difference between the master area and the slave area to the sum of the maximum brightness differences of the two areas is used as the brightness difference weight of the master area and the brightness difference weight of the slave area respectively; The ratios of the radiation doses of the main area and the slave area to the sum of the radiation doses of the two areas are used as the dose weight of the main area and the dose weight of the slave area respectively; The multiplication result of the brightness difference weight and the dose weight of the main area is set as the feature weight of the main area, and the multiplication result of the brightness difference weight and the dose weight of the slave area is set as the feature weight of the slave area.

4. The medical image intelligent analysis method based on artificial intelligence according to claim 3, characterized in that: In step S2, when extracting texture features, a plurality of extraction sub-regions of different areas are randomly set in the detection area, and the regional entropy value of each extraction sub-region is calculated. If the regional entropy value exceeds the entropy value threshold, the corresponding extraction sub-region is judged to be an abnormal sub-region, and the number of abnormal sub-regions is counted. When the number of abnormal sub-regions is not 0, it is judged that there is an abnormal area in the detection area, and all abnormal sub-regions are regarded as abnormal areas; Call the positions of each abnormal sub-region. When all abnormal sub-region positions belong to the main region, the detection region is judged to be abnormal in the main region. When all abnormal sub-region locations belong to the slave region, the detection region is judged to be a slave region abnormality; otherwise, the detection region is judged to be a full region abnormality.

5. The medical image intelligent analysis method based on artificial intelligence according to claim 4, characterized in that: In step S2, when the detection area is abnormal in the main area or the secondary area, the average regional entropy value of all abnormal sub-areas is calculated and multiplied by the feature weight of the corresponding area, and the result of the calculation is used as the regional abnormality coefficient; When the detection area is abnormal in the entire region, the regional entropy value average of the abnormal sub-region of the main region and the regional entropy value average of the abnormal sub-region of the slave region are calculated respectively, and the weighted summation is performed with the feature weight of the corresponding region to obtain the regional anomaly coefficient; When the regional anomaly coefficient of the detection area exceeds the preset control threshold, the abnormal control mechanism is entered.

6. The medical image intelligent analysis method based on artificial intelligence according to claim 1, characterized in that: In step S3, when entering the abnormality control mechanism, the corresponding grayscale value and noise value of the identified abnormal area are collected; Through an algorithm based on texture analysis and edge gradient detection, the grayscale values ​​of all pixels in the locked abnormal area are extracted and the average value is calculated to obtain the grayscale value of the abnormal area; By traversing the pixel grayscale in the abnormal area, the fluctuation degree of the pixel grayscale in the abnormal area is calculated to obtain the noise value of the abnormal area.

7. The medical image intelligent analysis method based on artificial intelligence according to claim 6, characterized in that: In step S3, the grayscale reference value and noise reference value of the detection area are obtained by calling the image database; Calculate the absolute difference between the grayscale value of the abnormal area and the grayscale reference value of the detection area to obtain the grayscale value difference of the abnormal area; The absolute difference between the noise value of the abnormal area and the noise baseline value of the detection area is calculated to obtain the noise value difference of the abnormal area.

8. The medical image intelligent analysis method based on artificial intelligence according to claim 7, characterized in that: In step S3, the grayscale value difference of the abnormal area and the noise value difference of the abnormal area are normalized and substituted into the bivariate tangent function coupling mapping function to obtain the error degree of the abnormal area. The specific formula is expressed as follows: ； Where, is the error degree of the abnormal area, is the gray value difference of the abnormal area after normalization, is the noise value difference of the abnormal area after normalization, is the scale factor, is the inverse tangent function, is the normalization coefficient.

9. An artificial intelligence-based medical image intelligent analysis system, based on the artificial intelligence-based medical image intelligent analysis method according to any one of claims 1 to 8, characterized in that: Includes the following devices: X-ray machine: used to obtain images of the inspection area of ​​the inspection object; Computer workstation: used to divide the detection area into regions, extract texture features, process data and determine whether to enter the abnormal control mechanism; Image data sensor: used to collect grayscale value, noise value and frequency domain phase information of abnormal areas; Image database: used to store and call grayscale reference values ​​and noise reference values; Optimization software: built-in image optimization solution, generates feedback data to determine whether to divide and adjust the detection area, and generates expansion coefficients to adjust the divided area.

Images