A method, system and storage medium for intelligent analysis of complex aortic lesions

Abstract

This application relates to the field of medical image analysis and processing technology, and discloses a method, system, and storage medium for intelligent analysis of complex aortic lesions. The method includes: collecting several aortic images of various image types containing different lesion types; collecting lesion information as label information; and annotating the aortic images with lesions based on the label information; extracting lesion regions and key regions; generating first training data and training a first recognition model; acquiring aortic images of various image types from a patient and performing enhancement processing; using the trained first recognition model to identify each key region in the aortic images to be analyzed; obtaining a lesion score for each lesion type; calculating a first lesion score for each lesion type in the aortic images to be analyzed; and determining the corresponding lesion type based on the first lesion score. This invention can effectively analyze multiple image types and improve the accuracy of aortic lesion type identification.

Description

Technical Field

[0001] This application relates to the field of medical image analysis and processing technology, and in particular to an intelligent analysis method, system and storage medium for complex aortic lesions. Background Technology

[0002] Complex aortic lesions, such as Stanford type B aortic dissection and complex abdominal aortic aneurysms, are serious vascular diseases requiring precise vascular reconstruction techniques for treatment. Currently, extracorporeal pre-fenestration combined with stent placement has become an effective treatment for these lesions. However, this technique demands extremely high precision in fenestration location and stent deployment. Furthermore, in clinical application, lesion analysis and surgical planning rely heavily on the surgeon's experience and subjective judgment, lacking systematic and intelligent analytical methods. This, to some extent, affects the success rate of the surgery and the patient's prognosis.

[0003] A similar prior art patent application, CN110675375A, provides an automatic method for identifying aortic dissection in thoracic and abdominal aortic images. This method involves projecting and statistically analyzing the contour of a medical aortic region using Radon transform, and combining this with the area and perimeter characteristics of the aortic contour to quickly and accurately determine whether the aortic region contains a dissection. However, this method relies solely on CTA images and does not involve the fusion of other modalities (such as MRI, echocardiography, etc.). A single modality image may not fully reflect the complexity of the lesion, and its discriminative ability may be limited for images where dissection features are not obvious.

[0004] Similar prior art includes Chinese patent application CN118735797A, which provides a method for assessing aortic lesions based on multi-scale feature fusion. The method includes steps S1: acquiring plain CT images of the chest and abdomen to be assessed, extracting and reconstructing the three-dimensional structure image of the aorta; step S2: obtaining multi-scale image features based on the three-dimensional aortic structure image, including multi-scale fused image features and three-dimensional aortic image features; step S3: obtaining an aortic lesion assessment model and inputting the multi-scale image features into the aortic lesion assessment model; and step S4: the aortic lesion assessment model assessing the aortic lesion assessment result based on the multi-scale image features. However, this method is mainly based on plain CT images of the chest and abdomen. For certain lesion types (such as early dissections or small lesions), the resolution and contrast of the plain CT images may be insufficient to provide adequate diagnostic information.

[0005] Therefore, the present invention provides an intelligent analysis method, system and storage medium for complex aortic lesions. Summary of the Invention

[0006] This application provides a method, system, and storage medium for intelligent analysis of complex aortic lesions, which can improve the accuracy of lesion identification.

[0007] In a first aspect, this application provides an intelligent analysis method for complex aortic lesions, the method comprising:

[0008] Step S1: Collect a number of aortic images of various image types containing different lesion types. Each image type of aortic image includes normal aortic images and diseased aortic images. Collect the corresponding lesion information as label information. The label information includes lesion type and lesion extent. Label the diseased aortic images based on the label information.

[0009] Step S2: Use the first extraction model to extract the lesion area from the aortic image, extract the key area of ​​a preset first size from the lesion area, combine the key area and the corresponding lesion information based on the annotation information to generate the first training data, train the first recognition model based on the first training data, and the first recognition model can output the corresponding lesion type and lesion score according to the key area of ​​the aortic image, and the lesion score represents the lesion probability of the corresponding lesion type.

[0010] Step S3: Obtain aortic images of various image types from the patient to be analyzed, enhance all the aortic images to be analyzed, input the enhanced aortic images of each image type to be analyzed into the first extraction model to obtain the corresponding lesion regions, extract key regions of a predetermined size from each lesion region, use the trained first recognition model to recognize each key region in the aortic images to be analyzed, obtain the lesion score for each lesion type, calculate the first lesion score for each lesion type in the aortic images to be analyzed, and determine the corresponding lesion type based on the first lesion score.

[0011] In conjunction with the first aspect, in a first implementation of the first aspect of this application, a first extraction model is used to extract the lesion region from the aortic image, including:

[0012] The brightness value of each pixel in the aortic image is acquired, the frequency distribution of each brightness value is statistically analyzed, the maximum and minimum frequencies are obtained, and a first value is preset. The minimum frequency is subtracted from the frequency of each brightness value to obtain a first result value, the maximum frequency is subtracted from the minimum frequency to obtain a second result value, the first result value is divided by the second result value to obtain a third result value, the third result value is multiplied by the first value to obtain a fourth result value, and the fourth result value is used as the normalized frequency of the corresponding brightness value. A histogram of the brightness values ​​of the aortic image is generated based on the normalized frequency. Multiple different mathematical models are used to fit the normalized frequency distribution, the parameters of each mathematical model are adjusted, and the multiple mathematical models are combined to fit the distribution of the normalized frequency. Based on the mathematical model with adjusted parameters, the pixels in the aortic image are divided into different groups, and pixels in the same group are considered to belong to the same region. The lesion region is identified by analyzing multiple different regions.

[0013] In conjunction with the first aspect, in the second implementation of the first aspect of this application, the lesion area is identified by analyzing multiple different regions, including:

[0014] Brightness values ​​of pixels are extracted from each region. The average brightness value and first standard deviation of the brightness value of each region are calculated, and a second value of each region is also calculated. The average brightness value, the first standard deviation, and the second value are used as feature values. The feature values ​​are normalized. Based on the normalized feature values, feature polygons of each region are drawn. First, an origin is set, and multiple line segments are drawn along a first number of directions from the origin. Each line segment represents a feature value. For each region, the normalized feature values ​​are marked on the corresponding line segments. All feature values ​​corresponding to each region are connected sequentially to form a polygon. The area of ​​each polygon is calculated. A first polygon with an area greater than a preset first threshold is obtained. The second standard deviation of the feature values ​​corresponding to each first polygon is calculated. A first polygon with a second standard deviation greater than a second threshold is called a second polygon. The region corresponding to the second polygon is taken as the lesion region.

[0015] In conjunction with the first aspect, in the third implementation of the first aspect of this application, calculating the second value for each region includes:

[0016] Obtain a small region containing a preset number of pixels from each region, obtain the center pixel of each small region, compare the grayscale value of other pixels in the small region with that of the center pixel, if the grayscale value of other pixels is greater than or equal to that of the center pixel, mark them as 1, otherwise mark them as 0, combine the marked values ​​of all other pixels contained in each small region into a binary number, count the binary numbers of all small regions, obtain the total number of all non-repeating binary numbers, and use the total number as the second value.

[0017] In conjunction with the first aspect, in the fourth implementation of the first aspect of this application, a first lesion score is calculated for each lesion type in the aortic image to be analyzed, including:

[0018] For each lesion type, the lesion score for each key region is obtained, the average score of the lesion score is calculated, and the standard deviation of the average score is calculated as the third standard deviation. If the third standard deviation is greater than or equal to a preset third threshold, the area ratio of the lesion region corresponding to each key region to the aortic image to be analyzed is also calculated. All the area ratios are normalized, and the normalized area ratios are used as the first weight of the corresponding lesion score. Based on the first weight and the lesion score, the weighted average score of the lesion score is calculated, and the weighted average score is used as the first lesion score of the lesion type. If the third standard deviation is less than the third threshold, the average score is used as the first lesion score of the lesion type.

[0019] In conjunction with the first aspect, in the fifth implementation of the first aspect of this application, determining the corresponding lesion type based on the first lesion score includes:

[0020] Calculate the first lesion score for the lesion type of the aortic image to be analyzed for each image type. Set a corresponding second weight for each image type based on historical data. Calculate the comprehensive lesion score for each lesion type in multiple image types based on the second weight and the first lesion score. Obtain the maximum comprehensive lesion score among all the comprehensive lesion scores. Determine whether the maximum comprehensive lesion score is greater than or equal to a preset fourth threshold. If yes, the lesion type corresponding to the maximum comprehensive lesion score is taken as the lesion type of the corresponding patient. If not, the results of each step are further reviewed by professionals.

[0021] In conjunction with the first aspect, in the sixth implementation of the first aspect of this application, all aortic images to be analyzed are enhanced, including:

[0022] Multiple first-class and second-class images are prepared in advance. The first-class and second-class images are correlated to generate image combinations. A first model is trained based on the image combinations. The first model can generate second-class images based on the first-class images. After the first model generates second-class images, the second-class images generated by the first model are used as third-class images. The third-class images are compared with the corresponding second-class images to calculate the difference between the third-class images and the second-class images. If the difference is greater than or equal to a preset fifth threshold, the first model adjusts its model parameters based on the difference and repeats this step until the difference is less than the fifth threshold.

[0023] In conjunction with the first aspect, in the seventh implementation of the first aspect of this application, generating a first model based on the image combination training includes:

[0024] During the training and generation of the first model, the feature maps of the first type of image are divided into several groups. Each group is convolved using multiple convolution kernels of different sizes to obtain intermediate feature maps. All intermediate feature maps are concatenated together to generate the third type of image. Weights for each group are generated based on the difference. The intermediate feature maps of each group are multiplied by the corresponding weights to obtain new intermediate feature maps. The new intermediate feature maps are combined to generate the third type of image.

[0025] Secondly, this application provides an intelligent analysis system for complex aortic lesions, the system comprising:

[0026] The image acquisition module is used to collect aortic images of various image types containing different lesion types. Each image type of aortic image includes normal aortic images and diseased aortic images. The corresponding lesion information is collected as label information, which includes lesion type and lesion extent. The diseased aortic images are labeled based on the label information.

[0027] The model training module is used to extract the lesion area from the aortic image using a first extraction model, extract a key area of ​​a preset first size from the lesion area, combine the key area and the corresponding lesion information to generate first training data based on the annotation information, and train a first recognition model based on the first training data. The first recognition model can output the corresponding lesion type and lesion score according to the key area of ​​the aortic image, and the lesion score represents the lesion probability of the corresponding lesion type.

[0028] The lesion analysis module is used to acquire aortic images of various image types from the patient, enhance all aortic images, input the enhanced aortic images of each image type into the first extraction model to obtain the corresponding lesion regions, extract key regions of a predetermined size from each lesion region, use a trained first recognition model to recognize each key region in the aortic images to obtain a lesion score for each lesion type, calculate the first lesion score for each lesion type in the aortic images to be analyzed, and determine the corresponding lesion type based on the first lesion score.

[0029] A third aspect of this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the aforementioned intelligent analysis method for complex aortic lesions.

[0030] Compared with the prior art, the beneficial effects of the present invention are at least as follows:

[0031] The technical solution provided in this application involves collecting several aortic images of various image types containing different lesion types, collecting lesion information as label information, and annotating the lesion aortic images based on the label information; extracting lesion areas and key areas, generating first training data and training a first recognition model; acquiring aortic images of various image types from the patient to be analyzed and performing enhancement processing; using the trained first recognition model to identify each key area in the aortic images to be analyzed, obtaining a lesion score for each lesion type, calculating the first lesion score for each lesion type in the aortic images to be analyzed, and determining the lesion type based on the first lesion score.

[0032] By combining multiple image types, the system acquires characteristic information of aortic lesions from different angles, thus reflecting the complexity of the lesions more comprehensively. Through the fusion of multimodal data, the system can utilize the advantages of different modalities to make up for the shortcomings of a single modality, significantly improving the accuracy of lesion diagnosis. By extracting the lesion area, the system also combines multi-dimensional information such as texture features and grayscale features to make lesion identification more accurate. Attached Figure Description

[0033] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0034] Figure 1 This is a schematic diagram of an embodiment of an intelligent analysis method for complex aortic lesions in this application.

[0035] Figure 2 This is a schematic diagram of an embodiment of a polygon drawn based on the normalized feature values ​​of the first region in this application.

[0036] Figure 3 This is a schematic diagram of an embodiment of a polygon drawn based on the normalized feature values ​​of the second region in this application.

[0037] Figure 4 This is a schematic diagram of one embodiment of an intelligent analysis system for complex aortic lesions in this application. Detailed Implementation

[0038] This application provides a method, system, and storage medium for intelligent analysis of complex aortic lesions. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0039] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent analysis method for complex aortic lesions in this application includes:

[0040] Step S1: Collect several aortic images of various image types containing different lesion types. Each image type of aortic image includes normal aortic images and diseased aortic images. Collect the corresponding lesion information as label information. The label information includes lesion type and lesion extent. Label the diseased aortic images based on the label information.

[0041] Specifically, traditional diagnosis of aortic lesions mainly relies on medical imaging techniques such as CTA, MRA, and echocardiography. However, traditional methods have limitations due to their single-modality nature, making it difficult to fully reflect the complex characteristics of aortic lesions. To accurately analyze complex aortic lesions, we collected aortic images of various image types, including CTA, ARI, MRA, echocardiography, and PET-CT, covering different lesion types such as Stranford type B aortic dissection, complex aortic aneurysm, and aortic stenosis. Each image type included images of normal aorta and images of diseased aorta. For diseased aortic images, we also collected corresponding lesion information as label information, including lesion type and lesion extent. The lesion extent refers to the location and size of the lesion in the diseased aortic image. Based on the label information, we annotated the diseased aortic images to facilitate subsequent model training.

[0042] Step S2: Use the first extraction model to extract the lesion area from the aortic image, extract the key area of ​​a preset first size from the lesion area, combine the key area and the corresponding lesion information based on the annotation information to generate the first training data, train the first recognition model based on the first training data, and the first recognition model can output the corresponding lesion type and lesion score according to the key area of ​​the aortic image. The lesion score represents the lesion probability of the corresponding lesion type.

[0043] Specifically, to facilitate the subsequent identification of lesion types by the first recognition model, a first extraction model is used to extract the lesion region from the collected aortic images. The specific extraction method will be explained in detail later. After extracting the lesion region, the extracted lesion region may include blurred areas with boundaries, that is, areas where lesions have occurred, or it may include the boundary area between the lesion region and the non-lesion region. To improve the recognition accuracy of the subsequent first recognition model, a key region of a preset first size is extracted from the lesion region. The preset first size may be, for example, a key region containing 256 pixels located in the middle of the lesion region. The key region is generally located in the middle of the lesion region. After identifying the key regions, the key regions are combined with the labeling information of the diseased aortic image to generate the first training data. Based on the first training data, the first recognition model is trained. The first recognition model can output the corresponding lesion type and lesion score according to the key regions of the input aortic image. For example, if the input is an aortic image with marked key regions, the first recognition model can output the corresponding lesion type and lesion score. For example, if image A includes three key regions and there are three lesion types a, b, and c, for each key region, the first recognition model can output the lesion score as follows: if the key region is lesion type a, the lesion score is 0.75; if it is lesion type b, the lesion score is 0.6; and if it is lesion type c, the lesion score is 0.5.

[0044] Step S3: Obtain aortic images of various image types from the patient to be analyzed. Enhance all aortic images to be analyzed. Input the enhanced aortic images of each image type to be analyzed into the first extraction model to obtain the corresponding lesion region. Then extract a key region of a predetermined size from each lesion region. Use the trained first recognition model to recognize each key region in the aortic images to be analyzed and obtain the lesion score for each lesion type. Calculate the first lesion score for each lesion type in the aortic images to be analyzed and determine the corresponding lesion type based on the first lesion score.

[0045] Specifically, to improve the accuracy of lesion type identification, multiple aortic images of the patient are acquired for analysis. Some aortic images may not be clear enough, so the aortic images to be analyzed are first enhanced to improve their quality, helping to more clearly obtain the lesion area and identify the lesion type, thereby improving the accuracy of identification. The enhanced aortic images of each image type are input into the first extraction model to obtain the corresponding lesion area. Then, a key region of a predetermined size is extracted from the lesion area. The first recognition model trained in the previous step is used to identify each key region in the aortic images to be analyzed, and a lesion score for each lesion type is obtained. Then, based on the recognition results of the first recognition model, the first lesion score for each lesion type in the aortic images to be analyzed is calculated. The specific process of calculating the first lesion score will be explained in detail later. The corresponding first lesion score is obtained for each image type of the aortic images to be analyzed, and the final lesion type is determined by combining the first lesion scores of all image types.

[0046] The above methods can effectively utilize various image types and artificial intelligence technologies to provide more accurate and efficient support for the analysis and diagnosis of aortic lesions.

[0047] In one specific embodiment, the lesion region is extracted from the aortic image using a first extraction model, and the following steps are also performed:

[0048] The process involves acquiring the brightness value of each pixel in an aortic image, statistically analyzing the frequency distribution of each brightness value, obtaining the maximum and minimum frequencies, and setting a first value. The minimum frequency is subtracted from the frequency of each brightness value to obtain a first result value. The maximum frequency is subtracted from the minimum frequency to obtain a second result value. The first result value is divided by the second result value to obtain a third result value. The third result value is multiplied by the first value to obtain a fourth result value. This fourth result value is used as the normalized frequency for the corresponding brightness value. A histogram of the brightness values ​​in the aortic image is generated based on the normalized frequency. Multiple different mathematical models are used to fit the normalized frequency distribution. The parameters of each mathematical model are adjusted, and the multiple mathematical models are combined to fit the normalized frequency distribution. Based on the adjusted mathematical model, the pixels in the aortic image are divided into different groups. Pixels in the same group are considered to belong to the same region. The lesion region is identified by analyzing multiple different regions.

[0049] Specifically, to accurately extract lesion areas from aortic images, the brightness value of each pixel in the aortic image is first obtained. The frequency distribution of each brightness value is then statistically analyzed, for example, the distribution of brightness values ​​from 0 to 255, obtaining the maximum and minimum frequencies. A first value is preset. The minimum frequency is subtracted from the frequency of each brightness value to obtain a first result value. The maximum frequency is subtracted from the minimum frequency to obtain a second result value. The first result value is divided by the second result value to obtain a third result value. The third result value is multiplied by the first value to obtain a fourth result value. The fourth result value is used as the normalized frequency for the corresponding brightness value. The frequency distribution is normalized using the above method to enhance the image contrast. Then, based on the normalized frequency, [the following is a separate process:] generating... A histogram of the brightness values ​​in the aortic image is generated. Then, multiple different mathematical models are used to fit the normalized frequency distribution. The parameters of each mathematical model are adjusted, and the multiple mathematical models are combined to fit the normalized frequency distribution. For example, the Gaussian mixture distribution method can be used to divide the normalized frequency distribution into multiple overall distributions. First, the initial Gaussian distribution parameters are selected, and the parameters of the Gaussian distribution are adjusted through an iterative optimization algorithm. The normalized frequency distribution is fitted using a linear combination of multiple Gaussian distributions, and finally multiple overall distributions are determined. Each normal distribution corresponds to a region in the aortic image. Pixels belonging to the same region are grouped together, and the lesion areas are further analyzed and identified from the multiple different regions.

[0050] In one specific embodiment, the lesion area is identified by analyzing multiple different regions, specifically including the following steps:

[0051] The brightness values ​​of pixels are extracted from each region. The average brightness value and the first standard deviation of the brightness value for each region are calculated. The second value of each region is also calculated. The average brightness value, the first standard deviation, and the second value are used as feature values. The feature values ​​are normalized. Based on the normalized feature values, feature polygons for each region are drawn. First, an origin is set. Multiple line segments are drawn along a first number of directions from the origin. Each line segment represents a feature value. For each region, the normalized feature values ​​are marked on the corresponding line segments. All feature values ​​corresponding to each region are connected sequentially to form a polygon. The area of ​​each polygon is calculated. The first polygon with an area greater than a preset first threshold is obtained. The second standard deviation of the feature values ​​corresponding to each first polygon is calculated. The first polygon with a second standard deviation greater than the second threshold is called the second polygon. The region corresponding to the second polygon is taken as the lesion region.

[0052] Specifically, in order to identify lesion areas, a series of feature values ​​are extracted from each area. The feature values ​​include the average brightness, the first standard deviation, the irregularity of the area shape, and the second value. The second value represents the texture features (such as smoothness or roughness) of each area. The specific method for calculating the second value will be explained in detail later. The multiple feature values ​​are normalized, and a feature variation image of each area is drawn based on the normalized feature values. The first quantity is the total number of feature values, 4. Table 1 shows the feature values ​​of two areas and the corresponding maximum and minimum values.

[0053] Eigenvalues First District Second Zone Maximum value Minimum value Average brightness 120 150 200 100 First standard deviation 10 20 30 5 Second value 0.5 0.8 0.9 0.4 Region shape irregularity 0.3 0.6 0.7 0.2

[0054] Table 1

[0055] In reality, there are multiple regions. For clarity, this embodiment only lists two regions. Normalizing the feature values ​​of the first region yields normalized feature values ​​of 0.2, 0.2, 0.2, and 0.2. Normalizing the feature values ​​of the second region yields corresponding normalized feature values ​​of 0.5, 0.6, 0.8, and 0.8. These normalized feature values ​​are then marked on the corresponding line segments, as shown below. Figure 2 As shown, this is a polygon drawn based on the normalized eigenvalues ​​of the first region. First, an origin O is set, and line segments are drawn along the origin in four directions. The normalized eigenvalues ​​0.2, 0.2, 0.2, and 0.2 are marked on the corresponding line segments. Then, all the normalized eigenvalues ​​are connected sequentially to form a polygon, as shown below. Figure 3 As shown, polygons are drawn based on the normalized eigenvalues ​​of the second region. After drawing the polygons corresponding to each region, by comparing the feature polygon images of different regions (such as lesion regions and normal tissue regions), their feature differences can be intuitively seen. For example, the feature polygon image of a lesion region may be larger than that of a normal region in some features (such as texture roughness or irregularity), while the feature polygon image of a normal region may be closer to the center in some features (such as average brightness or region area). Calculate the area of ​​each polygon; the size of the area can reflect the comprehensive strength of the region's features. The larger the area of ​​the polygon image, the higher the feature value of that region. For example, the feature value of a lesion region... The area of ​​the feature polygon may be larger than that of the normal area, which may mean that the features of the lesion area are more significant. Therefore, the first polygon with an area greater than the first threshold is obtained. The shape of the polygon can reflect the relationship between different features. For example, if the feature polygon of the lesion area is more prominent in texture roughness, it may indicate that the texture of the lesion area is more complex. For each first polygon, the second standard deviation of its corresponding feature value is calculated. The second standard deviation is used to measure the dispersion of the feature value. A second standard deviation threshold is set (second threshold). The first polygon with a second standard deviation greater than the threshold is selected (called the second polygon). The area corresponding to the second polygon is marked as the lesion area.

[0056] In one specific embodiment, calculating a second value for each region includes the following steps:

[0057] From each region, obtain a small region containing a preset number of pixels, obtain the center pixel of each small region, and compare the gray values ​​of other pixels in the small region with those of the center pixel. If the gray values ​​of other pixels are greater than or equal to the gray values ​​of the center pixel, mark them as 1; otherwise, mark them as 0. Combine the marked values ​​of all other pixels in each small region into a binary number, count the binary numbers of all small regions, obtain the total number of all unique binary numbers, and use the total number as the second value.

[0058] Specifically, to calculate the second value of each region, which is the texture feature value of each region, a small area containing a preset number of pixels (e.g., 3*3) is obtained from each region. The center pixel of each small area is obtained, and the pixels in the small area other than the center pixel are called other pixels. The gray values ​​of other pixels and the center pixel are compared. If the gray value of other pixels is greater than or equal to the gray value of the center pixel, it is marked as 1; otherwise, it is marked as 0. The corresponding marked values ​​of all other pixels contained in each small area are sorted and combined into a binary number according to the position of the pixel in the small area from left to right and top to bottom, for example, 11101100. The binary numbers of all small areas are counted, and the total number of binary numbers in all small areas is obtained. Different small areas may have the same binary value. The same binary value means that the small areas have the same texture features. Therefore, the total number of all non-repeating binary numbers is counted, and the total number is used as the second value.

[0059] In one specific embodiment, calculating a first lesion score for each lesion type in the aortic image to be analyzed specifically includes the following steps:

[0060] For each lesion type, a lesion score is obtained for each key region. The average score is calculated, and the standard deviation of the average score is used as the third standard deviation. If the third standard deviation is greater than or equal to a preset third threshold, the area ratio of the lesion region corresponding to each key region to the area of ​​the aortic image to be analyzed is also calculated. All area ratios are normalized, and the normalized area ratio is used as the first weight of the corresponding lesion score. Based on the first weight and the lesion score, the weighted average score of the lesion score is calculated, and the weighted average score is used as the first lesion score of the lesion type. If the third standard deviation is less than the third threshold, the average score is used as the first lesion score of the lesion type.

[0061] Specifically, in order to more accurately calculate the first lesion score for each lesion type in the aortic image to be analyzed, the lesion score for each key region is obtained for each lesion type. Assuming there are a total of 4 lesion types and 3 key regions, Table 2 shows the corresponding lesion scores for each of the 3 key regions of the aortic image to be analyzed for a certain image type output by the first recognition model.

[0062] Lesion type a Lesion type b Lesion type c Lesion type d Key Area 1 0.8 0.5 0.4 0.45 Key Area 2 0.3 0.85 0.35 0.6 Key Area 3 0.4 0.5 0.45 0.7

[0063] Table 2

[0064] For each lesion type (e.g., aortic dissection, aortic aneurysm), a lesion score for each key region is obtained. To assess the central tendency and dispersion of the lesion scores and determine whether area ratio should be introduced as a weight, the mean score (0.5, 0.62, 0.4, 0.58) and third standard deviation (0.0283) are first calculated for each lesion type. The dispersion of the lesion scores is determined based on the standard deviation, and it is decided whether to introduce area ratio as a weight to more accurately identify the lesion type. A preset third threshold (assumed to be 0.02) is set to determine the lesion type. If the third standard deviation is large enough, and the third standard deviation is greater than the third threshold, the dispersion of the lesion score is considered to be large, and the area ratio needs to be introduced as a weight. For each key region, the area ratio of the lesion region to the aortic image to be analyzed is calculated. All area ratios are normalized. Based on the normalized area ratios as weights, the weighted average of the lesion scores is calculated as the first lesion score. Otherwise, the dispersion of the lesion score is considered to be small, and the average score is directly used as the first lesion score.

[0065] The above method, by comprehensively considering both lesion score and area ratio, can more accurately calculate the first lesion score and improve the accuracy of subsequent lesion type judgment.

[0066] In one specific embodiment, determining the corresponding lesion type based on the first lesion score specifically includes the following steps:

[0067] Calculate the first lesion score for the lesion type of the aortic image to be analyzed for each image type. Set a corresponding second weight for each image type based on historical data. Based on the second weight and the first lesion score, calculate the comprehensive lesion score for each lesion type of multiple image types. Obtain the maximum comprehensive lesion score among all comprehensive lesion scores. Determine whether the maximum comprehensive lesion score is greater than or equal to a preset fourth threshold. If yes, the lesion type corresponding to the maximum comprehensive lesion score is taken as the lesion type of the corresponding patient. Otherwise, the results of each step are further reviewed by professionals.

[0068] Specifically, for each image type, a first lesion score is calculated for each lesion type. Based on historical diagnostic data, the accuracy and reliability of each image type in diagnosing different lesion types are analyzed. For example, CTA images may be more sensitive to certain lesion types (such as aortic aneurysms), while MRI images may be clearer for calcified lesions. A second weight is assigned to each image type. To comprehensively consider the first lesion scores of multiple image types, a comprehensive lesion score for each lesion type is calculated. Based on the second weight and the first lesion score, a comprehensive lesion score for each lesion type of multiple image types is calculated. The maximum comprehensive lesion score among all comprehensive lesion scores is obtained. It is determined whether the maximum comprehensive lesion score is greater than or equal to a preset fourth threshold, which represents the minimum confidence level of the diagnostic result. If yes, the diagnostic result is accepted, and the lesion type corresponding to the maximum comprehensive lesion score is taken as the lesion type of the corresponding patient. Otherwise, the diagnostic result is unreliable and further examination may be required.

[0069] The method described above comprehensively considers the primary lesion scores from multiple image types by calculating a composite lesion score. By assigning weights to each image type, it can more comprehensively reflect the severity and characteristics of the lesions. This method avoids the risk of misdiagnosis that may arise from relying on a single image type.

[0070] In one specific embodiment, all aortic images to be analyzed are enhanced, specifically including the following steps:

[0071] Prepare multiple first-class and second-class images in advance. Associate the first-class and second-class images to generate image combinations. Train a first model based on the image combinations. The first model can generate second-class images based on the first-class images. After the first model generates second-class images, use the second-class images generated by the first model as third-class images. Compare the third-class images with the corresponding second-class images and calculate the difference between the third-class images and the second-class images. If the difference is greater than or equal to a preset fifth threshold, the first model adjusts the model parameters based on the difference and repeats this step until the difference is less than the fifth threshold.

[0072] Specifically, to enhance the aortic images to be analyzed and improve their quality, thus aiding in more accurate lesion identification, multiple first-class and second-class images are prepared in advance. The first-class images are low-resolution images, and the second-class images are high-resolution images. Low-resolution images (first-class images) are paired with their corresponding high-resolution images (second-class images) to form image combinations. These image combinations are used to train a first model. The first-class images are input into the first model, and by learning the feature mapping relationship of the images, the model attempts to generate results that are as close as possible to the second-class images. The difference between the generated third-class images and the real second-class images is also calculated. A preset fifth threshold is set to determine whether the generated third-class images are sufficiently close to the real second-class images. If the difference is greater than or equal to the preset fifth threshold, it indicates that the gap between the generated third-class images and the real second-class images is still large, requiring further optimization of the first model. The model parameters are adjusted according to the difference to reduce the difference between the generated images and the real images. This method optimizes the performance of the first model by adjusting its parameters based on the difference until the generated third-class images are sufficiently close to the real second-class images (high-resolution images).

[0073] In one specific embodiment, the first model is generated based on image combination training, which specifically includes the following steps:

[0074] During the training process of generating the first model, the feature maps of the first class of images are divided into several groups. Each group is convolved with multiple convolution kernels of different sizes to obtain intermediate feature maps. All intermediate feature maps are concatenated together to generate the third class of images. Weights for each group are generated based on the difference. The intermediate feature maps of each group are multiplied by the corresponding weights to obtain new intermediate feature maps. The new intermediate feature maps are combined to generate the third class of images.

[0075] Specifically, to generate high-resolution images with clearer details to aid in accurate lesion diagnosis, during the training of the first model, the feature maps of the first type of image are divided into several groups. Each group undergoes convolution operations using multiple kernels of different sizes to obtain intermediate feature maps. Through grouped convolution and kernels of different sizes, more diverse features are extracted, improving the model's expressive power. All intermediate feature maps are concatenated to generate the third type of image. Weights for each group are generated based on the difference, and the intermediate feature map of each group is multiplied by its corresponding weight to obtain a new intermediate feature map. These new intermediate feature maps are then combined to generate the third type of image. Dynamically adjusting the weights based on the difference between the generated image and the real image better highlights important features and suppresses unimportant ones. Through the combination and reconstruction of weighted feature maps, the generated high-resolution image is closer to the real image, with richer details. The combination of dynamic weight adjustment and difference feedback ensures the stability and convergence of the training process.

[0076] The above method extracts more diverse features through grouped convolution and dynamic weight adjustment, and dynamically optimizes the generator's performance based on the difference between the generated image and the real image. This method not only improves the quality of the generated image, but also has an efficient training process and wide applicability.

[0077] The above describes an intelligent analysis method for complex aortic lesions in the embodiments of this application. The following describes an intelligent analysis system for complex aortic lesions in the embodiments of this application. Please refer to [link / reference]. Figure 4 One embodiment of the intelligent analysis system for complex aortic lesions in this application includes:

[0078] The image acquisition module is used to collect aortic images of various image types containing different lesion types. Each image type of aortic image includes normal aortic images and diseased aortic images. The corresponding lesion information is collected as label information, which includes lesion type and lesion extent. The diseased aortic images are labeled based on the label information.

[0079] The model training module is used to extract the lesion area from the aortic image using the first extraction model, extract the key area of ​​a preset first size from the lesion area, combine the key area and the corresponding lesion information based on the annotation information to generate the first training data, train the first recognition model based on the first training data, and the first recognition model can output the corresponding lesion type and lesion score according to the key area of ​​the aortic image. The lesion score represents the lesion probability of the corresponding lesion type.

[0080] The lesion analysis module is used to acquire aortic images of various image types from the patient, enhance all aortic images, input the enhanced aortic images of each image type into the first extraction model to obtain the corresponding lesion regions, extract key regions of a predetermined size from each lesion region, use the trained first recognition model to recognize each key region in the aortic images to obtain a lesion score for each lesion type, calculate the first lesion score for each lesion type in the aortic images to be analyzed, and determine the corresponding lesion type based on the first lesion score.

[0081] This application also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, storing instructions that, when executed on a computer, cause the computer to perform the steps of a complex aortic lesion intelligent analysis method.

[0082] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0083] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0084] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent analysis of complex aortic lesions, characterized in that, The method includes: Step S1: Collect a number of aortic images of various image types containing different lesion types. Each image type of aortic image includes normal aortic images and diseased aortic images. Collect the corresponding lesion information as label information. The label information includes lesion type and lesion extent. Label the diseased aortic images based on the label information. Step S2: Use the first extraction model to extract the lesion area from the aortic image, extract the key area of ​​a preset first size from the lesion area, combine the key area and the corresponding lesion information based on the annotation information to generate the first training data, train the first recognition model based on the first training data, and the first recognition model can output the corresponding lesion type and lesion score according to the key area of ​​the aortic image, and the lesion score represents the lesion probability of the corresponding lesion type. The first extraction model is used to extract the lesion region from the aortic image, including: The process involves acquiring the brightness value of each pixel in an aortic image, statistically analyzing the frequency distribution of each brightness value, obtaining the maximum and minimum frequencies, and setting a first value. The minimum frequency is subtracted from the frequency of each brightness value to obtain a first result value. The minimum frequency is subtracted from the maximum frequency to obtain a second result value. The first result value is divided by the second result value to obtain a third result value. The third result value is multiplied by the first value to obtain a fourth result value. The fourth result value is used as the normalized frequency for the corresponding brightness value. A histogram of the brightness values ​​in the aortic image is generated based on the normalized frequency. Multiple different mathematical models are used to fit the normalized frequency distribution. The parameters of each mathematical model are adjusted, and the multiple mathematical models are combined to fit the distribution of the normalized frequency. Based on the adjusted mathematical model, the pixels in the aortic image are divided into different groups. Pixels in the same group are considered to belong to the same region. Multiple different regions are analyzed to identify lesion areas. Step S3: Obtain aortic images of various image types from the patient to be analyzed, enhance all the aortic images to be analyzed, input the enhanced aortic images of each image type to be analyzed into the first extraction model to obtain the corresponding lesion regions, extract key regions of a predetermined size from each lesion region, use the trained first recognition model to recognize each key region in the aortic images to be analyzed, obtain the lesion score for each lesion type, calculate the first lesion score for each lesion type in the aortic images to be analyzed, and determine the corresponding lesion type based on the first lesion score.

2. The method according to claim 1, characterized in that, The analysis identified lesion areas by examining multiple different regions, including: Brightness values ​​of pixels are extracted from each region. The average brightness value and first standard deviation of the brightness value of each region are calculated, and a second value of each region is also calculated. The average brightness value, the first standard deviation, and the second value are used as feature values. The feature values ​​are normalized. Based on the normalized feature values, feature polygons of each region are drawn. First, an origin is set, and multiple line segments are drawn along a first number of directions from the origin. Each line segment represents a feature value. For each region, the normalized feature values ​​are marked on the corresponding line segments. All feature values ​​corresponding to each region are connected sequentially to form a polygon. The area of ​​each polygon is calculated. A first polygon with an area greater than a preset first threshold is obtained. The second standard deviation of the feature values ​​corresponding to each first polygon is calculated. A first polygon with a second standard deviation greater than a second threshold is called a second polygon. The region corresponding to the second polygon is taken as the lesion region.

3. The method according to claim 2, characterized in that, Calculate the second value for each region, including: Obtain a small region containing a preset number of pixels from each region, obtain the center pixel of each small region, compare the grayscale value of other pixels in the small region with that of the center pixel, if the grayscale value of other pixels is greater than or equal to that of the center pixel, mark them as 1, otherwise mark them as 0, combine the marked values ​​of all other pixels contained in each small region into a binary number, count the binary numbers of all small regions, obtain the total number of all non-repeating binary numbers, and use the total number as the second value.

4. The method according to claim 1, characterized in that, Calculate the first lesion score for each lesion type in the aortic images to be analyzed, including: For each lesion type, the lesion score for each key region is obtained, the average score of the lesion score is calculated, and the standard deviation of the average score is calculated as the third standard deviation. If the third standard deviation is greater than or equal to a preset third threshold, the area ratio of the lesion region corresponding to each key region to the aortic image to be analyzed is also calculated. All the area ratios are normalized, and the normalized area ratios are used as the first weight of the corresponding lesion score. Based on the first weight and the lesion score, the weighted average score of the lesion score is calculated, and the weighted average score is used as the first lesion score of the lesion type. If the third standard deviation is less than the third threshold, the average score is used as the first lesion score of the lesion type.

5. The method according to claim 1, characterized in that, Based on the first lesion score, the corresponding lesion type is determined, including: Calculate the first lesion score for the lesion type of the aortic image to be analyzed for each image type. Set a corresponding second weight for each image type based on historical data. Calculate the comprehensive lesion score for each lesion type in multiple image types based on the second weight and the first lesion score. Obtain the maximum comprehensive lesion score among all the comprehensive lesion scores. Determine whether the maximum comprehensive lesion score is greater than or equal to a preset fourth threshold. If yes, the lesion type corresponding to the maximum comprehensive lesion score is taken as the lesion type of the corresponding patient. If not, the results of each step are further reviewed by professionals.

6. The method according to claim 1, characterized in that, All aortic images to be analyzed were enhanced, including: Multiple first-class and second-class images are prepared in advance. The first-class and second-class images are correlated to generate image combinations. A first model is trained based on the image combinations. The first model can generate second-class images based on the first-class images. After the first model generates second-class images, the second-class images generated by the first model are used as third-class images. The third-class images are compared with the corresponding second-class images to calculate the difference between the third-class images and the second-class images. If the difference is greater than or equal to a preset fifth threshold, the first model adjusts its model parameters based on the difference and repeats this step until the difference is less than the fifth threshold.

7. The method according to claim 6, characterized in that, The first model is generated based on the image combination training, including: During the training and generation of the first model, the feature maps of the first type of image are divided into several groups. Each group is convolved using multiple convolution kernels of different sizes to obtain intermediate feature maps. All intermediate feature maps are concatenated together to generate the third type of image. Weights for each group are generated based on the difference. The intermediate feature maps of each group are multiplied by the corresponding weights to obtain new intermediate feature maps. The new intermediate feature maps are combined to generate the third type of image.

8. A complex aortic lesion intelligent analysis system, used to implement the complex aortic lesion intelligent analysis method as described in any one of claims 1-7, characterized in that, The system includes: The image acquisition module is used to collect aortic images of various image types containing different lesion types. Each image type of aortic image includes normal aortic images and diseased aortic images. The corresponding lesion information is collected as label information, which includes lesion type and lesion extent. The diseased aortic images are labeled based on the label information. The model training module is used to extract lesion regions from an aortic image using a first extraction model, extract key regions of a preset first size from the lesion regions, combine the key regions and corresponding lesion information based on the annotation information to generate first training data, and train a first recognition model based on the first training data. The first recognition model can output corresponding lesion types and lesion scores according to the key regions of the aortic image, where the lesion score represents the probability of the corresponding lesion type. The extraction of lesion regions from the aortic image using the first extraction model includes: obtaining the brightness value of each pixel in the aortic image, statistically analyzing the frequency distribution of each brightness value, obtaining the maximum and minimum frequencies, and setting a preset first value, and subtracting the minimum frequency from the frequency of each brightness value. A first result value is obtained. The maximum frequency is subtracted from the minimum frequency to obtain a second result value. The first result value is divided by the second result value to obtain a third result value. The third result value is multiplied by the first value to obtain a fourth result value. The fourth result value is used as the normalized frequency of the corresponding brightness value. A histogram of the brightness value of the aortic image is generated based on the normalized frequency. Multiple different mathematical models are used to fit the normalized frequency distribution. The parameters of each mathematical model are adjusted. The multiple mathematical models are combined to fit the distribution of the normalized frequency. Based on the mathematical model with adjusted parameters, the pixels in the aortic image are divided into different groups. Pixels that are divided into the same group are regarded as belonging to the same region. Multiple different regions are analyzed to identify lesion areas. The lesion analysis module is used to acquire aortic images of various image types from the patient, enhance all aortic images, input the enhanced aortic images of each image type into the first extraction model to obtain the corresponding lesion regions, extract key regions of a predetermined size from each lesion region, use a trained first recognition model to recognize each key region in the aortic images to obtain a lesion score for each lesion type, calculate the first lesion score for each lesion type in the aortic images to be analyzed, and determine the corresponding lesion type based on the first lesion score.

9. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement a method for intelligent analysis of complex aortic lesions as described in any one of claims 1-7.

Images