A deep learning-based war injury ultrasonic image diagnosis system and method

Abstract

The application provides a war injury ultrasonic image diagnosis system and method based on deep learning. The system and method perform self-adaptive preprocessing on a war injury ultrasonic image to be processed to obtain a preprocessed ultrasonic image. Global context features of image texture and local high-frequency features of image edges are extracted from the preprocessed ultrasonic image, and a semantic fusion feature map containing deep image semantic information is generated. A feature correlation coefficient between different feature sub-regions in the semantic fusion feature map and a preset feature template is determined based on a preset sliding window, and a segmentation probability matrix for segmenting the preprocessed ultrasonic image is determined. The preprocessed ultrasonic image is regionally segmented according to the segmentation probability matrix to obtain a regional segmentation image corresponding to the war injury ultrasonic image. The technical solution provided by the application can accurately regionally segment a war injury ultrasonic image under the structural heterogeneity and texture complexity of war injury tissue.

Description

Technical Field

[0001] This application relates to the field of ultrasound image processing technology, and more specifically, to a deep learning-based ultrasound image diagnostic system and method for combat trauma. Background Technology

[0002] Ultrasound imaging, as a non-invasive and safe medical imaging technique, is widely used in clinical diagnosis, treatment monitoring, and scientific research due to its advantages such as portable equipment, simple operation, and low cost. It uses the reflection and refraction of high-frequency sound waves in human tissues to obtain information about internal structures. However, in practical applications, the quality of ultrasound images is constrained by various factors, such as probe selection, operating techniques, and patient physical condition, resulting in problems such as noise and artifacts in the images. Furthermore, ultrasound image analysis is highly dependent on the professional knowledge and experience of doctors, is highly subjective, and is easily affected by differences in individual judgment. Therefore, the development of ultrasound image processing is of great significance for improving image quality and diagnostic accuracy.

[0003] In existing ultrasound image processing, image quality is generally improved through preprocessing such as filtering and contrast enhancement. Then, key features are extracted through edge detection and texture analysis. Next, morphological operations or segmentation algorithms are used to optimize the target region, and machine learning is sometimes combined to achieve automatic recognition and analysis. However, in the processing of war trauma ultrasound images, the pathological morphology of war trauma tissues (such as lacerations, hematomas, and internal contusions) is characterized by high structural heterogeneity and textural complexity. In war trauma ultrasound images, this is reflected in the change of acoustic impedance difference between the trauma area and healthy tissue. Its edges often exhibit low contrast, weak boundaries, and discontinuities, rather than a uniform, continuous closed region. This leads to blurred boundary segmentation when segmenting the trauma area. Therefore, how to accurately segment war trauma ultrasound images under the structural heterogeneity and textural complexity of war trauma tissues has become a challenge for the industry. Summary of the Invention

[0004] This application provides a deep learning-based ultrasound image diagnostic system and method for war trauma, which can accurately segment war trauma ultrasound images under the structural heterogeneity and texture complexity of war trauma tissue.

[0005] In a first aspect, this application provides a deep learning-based method for processing ultrasound images of combat trauma, comprising the following steps:

[0006] Acquire the war trauma ultrasound image to be processed, and perform adaptive preprocessing based on deep learning prior knowledge on the war trauma ultrasound image to obtain the preprocessed ultrasound image.

[0007] Global context features of image texture and local high-frequency features of image edges are extracted from the preprocessed ultrasound image, and then a semantic fusion feature map containing deep image semantic information is generated based on the global context features and the local high-frequency features.

[0008] Based on a preset sliding window, the feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and the preset feature template are determined, and then the segmentation probability matrix for segmenting the preprocessed ultrasound image is determined according to the feature correlation coefficients corresponding to each feature sub-region.

[0009] The preprocessed ultrasound image is regionalized based on the segmentation probability matrix to obtain the regionalized segmented image corresponding to the combat trauma ultrasound image.

[0010] In some embodiments, adaptive preprocessing based on deep learning prior knowledge is performed on the combat trauma ultrasound image to obtain a preprocessed ultrasound image, specifically including:

[0011] Based on the imaging characteristics of war trauma ultrasound images, a deep learning prior knowledge graph containing artifact features and normal tissue features of typical war trauma ultrasound images is constructed.

[0012] Based on the deep learning prior knowledge graph, artifact noise type identification is performed on the acquired war trauma ultrasound images to determine the category of artifact noise present in the war trauma ultrasound images.

[0013] Based on the identified artifact noise category, the corresponding adaptive filtering algorithm is called to suppress noise in the war trauma ultrasound image, resulting in a preprocessed ultrasound image.

[0014] In some embodiments, extracting global contextual features of image texture and local high-frequency features of image edges from the preprocessed ultrasound image specifically includes:

[0015] The preprocessed ultrasound image is input into a pre-trained deep convolutional neural network, and the global context features of the image texture are extracted through the deep network of the deep convolutional neural network.

[0016] The preprocessed ultrasound image is subjected to multi-scale Gaussian difference filtering to generate a multi-scale edge response map;

[0017] The local high-frequency features of the image edges are extracted based on the multi-scale edge response map.

[0018] In some embodiments, generating a semantic fusion feature map containing deep image semantic information based on the global context features and the local high-frequency features specifically includes:

[0019] The global context features are subjected to dimension adaptation processing to generate adapted global features that match the dimensions of local high-frequency features;

[0020] An attention mechanism is used to calculate the feature association weights between the adaptive global features and the local high-frequency features;

[0021] Based on the feature association weights, the adaptive global features and the local high-frequency features are weighted and fused to generate an initial fused feature map;

[0022] The initial fused feature map is enhanced by the residual connection module to obtain a semantic fused feature map containing deep image semantic information.

[0023] In some embodiments, determining the feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and the preset feature template based on a preset sliding window specifically includes:

[0024] Multiple feature sub-regions are generated by sliding sampling on the semantic fusion feature map according to a preset sliding window;

[0025] Determine the feature vectors corresponding to each feature sub-region;

[0026] Calculate the cosine similarity between the feature vector corresponding to each feature sub-region and the preset feature template;

[0027] The cosine similarity is normalized to obtain the feature correlation coefficient between each feature sub-region and the preset feature template.

[0028] In some embodiments, determining the segmentation probability matrix for segmenting the preprocessed ultrasound image based on the feature correlation coefficients corresponding to each feature sub-region specifically includes:

[0029] The feature correlation coefficients corresponding to each feature sub-region are mapped to the corresponding spatial locations of the semantic fusion feature map to generate an initial probability distribution matrix;

[0030] The initial probability distribution matrix is ​​spatially smoothed to obtain a smoothed probability distribution matrix;

[0031] The smooth probability distribution matrix is ​​normalized using a flexible maximum value function to obtain the segmentation probability value of each pixel belonging to the segmented region;

[0032] A segmentation probability matrix is ​​constructed based on the segmentation probability value of each pixel when segmenting the preprocessed ultrasound image.

[0033] In some embodiments, war trauma ultrasound images to be processed are obtained from a war trauma ultrasound image storage file.

[0034] Secondly, this application provides a deep learning-based combat trauma ultrasound image diagnostic system for executing a deep learning-based combat trauma ultrasound image processing method, comprising a combat trauma ultrasound image processing unit, wherein the combat trauma ultrasound image processing unit includes:

[0035] The acquisition module is used to acquire the war trauma ultrasound image to be processed, and to perform adaptive preprocessing based on deep learning prior knowledge on the war trauma ultrasound image to obtain a preprocessed ultrasound image.

[0036] The processing module is used to extract global context features of image texture and local high-frequency features of image edges from the preprocessed ultrasound image, and then generate a semantic fusion feature map containing deep image semantic information based on the global context features and the local high-frequency features.

[0037] The processing module is further configured to determine the feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and the preset feature template based on a preset sliding window, and then determine the segmentation probability matrix for segmenting the preprocessed ultrasound image based on the feature correlation coefficients corresponding to each feature sub-region.

[0038] The execution module is used to perform regional segmentation on the preprocessed ultrasound image according to the segmentation probability matrix to obtain the regional segmentation image corresponding to the combat trauma ultrasound image.

[0039] Thirdly, this application provides a computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described deep learning-based method for processing war trauma ultrasound images.

[0040] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described deep learning-based method for processing war trauma ultrasound images.

[0041] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0042] The deep learning-based war trauma ultrasound image diagnosis system and method provided in this application firstly acquires a war trauma ultrasound image to be processed and performs adaptive preprocessing based on deep learning prior knowledge to obtain a preprocessed ultrasound image. Secondly, global context features of image texture and local high-frequency features of image edges are extracted from the preprocessed ultrasound image, and then a semantic fusion feature map containing deep image semantic information is generated based on the global context features and the local high-frequency features. Then, feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and preset feature templates are determined based on a preset sliding window, and then a segmentation probability matrix is ​​determined for segmenting the preprocessed ultrasound image based on the feature correlation coefficients corresponding to each feature sub-region. Finally, the preprocessed ultrasound image is regionalized based on the segmentation probability matrix to obtain a regionalized segmented image corresponding to the war trauma ultrasound image.

[0043] Therefore, this application can accurately segment war trauma ultrasound images under the structural heterogeneity and textural complexity of war trauma tissue. First, through adaptive preprocessing based on deep learning prior knowledge, it can perform targeted optimization according to the typical noise distribution of war trauma ultrasound images, avoiding interference from invalid information on feature analysis. Second, by extracting global contextual features of image texture and local high-frequency features of image edges, and fusing them to generate a semantic fusion feature map, it integrates global image semantic information and enhances key texture details at the local level, achieving deep coupling of features at different scales and solving the problem that a single feature cannot fully represent the deep semantics of an image. Then, based on a preset sliding window, the semantic fusion feature map is traversed, and each feature is calculated by matching it with a war trauma feature template library. The feature correlation coefficients of the sub-regions are then spatially mapped, smoothed, and normalized to generate a segmentation probability matrix. This transforms the abstract feature correlation into a pixel-level segmentation probability distribution, making the segmentation boundary of the war trauma ultrasound image more accurate and avoiding segmentation ambiguity caused by the heterogeneity of war trauma tissue structure and the complexity of texture. Finally, the preprocessed ultrasound image is regionalized based on the segmentation probability matrix to obtain the regionalized segmented image corresponding to the war trauma ultrasound image. This effectively clarifies the semantic attribution of pixel regions, filters out irrelevant background information and noise interference, highlights the segmentation features of the image, and solves the ambiguity of semantic segmentation of ultrasound images in war trauma scenarios. In summary, the technical solution provided in this application can accurately regionalize war trauma ultrasound images under the structural heterogeneity and texture complexity of war trauma tissue. Attached Figure Description

[0044] Figure 1 This is an exemplary flowchart of a deep learning-based ultrasonic image processing method for combat trauma, according to some embodiments of this application.

[0045] Figure 2 This is an exemplary flowchart illustrating the determination of a semantic fusion feature map according to some embodiments of this application;

[0046] Figure 3 This is a schematic diagram of the structure of a combat trauma ultrasound image processing unit according to some embodiments of this application;

[0047] Figure 4 This is a schematic diagram of the structure of a computer device that implements a deep learning-based method for processing ultrasonic images of combat trauma, according to some embodiments of this application. Detailed Implementation

[0048] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] refer to Figure 1 The figure is an exemplary flowchart of a deep learning-based ultrasonic image processing method for combat trauma, according to some embodiments of this application. The figure mainly includes the following steps:

[0050] In step S101, the war trauma ultrasound image to be processed is acquired, and the war trauma ultrasound image is subjected to adaptive preprocessing based on deep learning prior knowledge to obtain a preprocessed ultrasound image.

[0051] In specific implementation, the war trauma ultrasound image to be processed is obtained from the war trauma ultrasound image storage file. The war trauma ultrasound image to be processed refers to the war trauma ultrasound image that needs to be processed. The war trauma ultrasound image storage file refers to the file that stores the war trauma ultrasound image to be processed. This will not be elaborated here. By obtaining the war trauma ultrasound image to be processed, data basis can be provided for the processing of the war trauma ultrasound image.

[0052] In some embodiments, adaptive preprocessing based on deep learning prior knowledge is performed on the combat trauma ultrasound images to obtain preprocessed ultrasound images, which is achieved through the following steps:

[0053] Based on the imaging characteristics of war trauma ultrasound images, a deep learning prior knowledge graph containing artifact features and normal tissue features of typical war trauma ultrasound images is constructed.

[0054] Based on the deep learning prior knowledge graph, artifact noise type identification is performed on the acquired war trauma ultrasound images to determine the category of artifact noise present in the war trauma ultrasound images.

[0055] Based on the identified artifact noise category, the corresponding adaptive filtering algorithm is called to suppress noise in the war trauma ultrasound image, resulting in a preprocessed ultrasound image.

[0056] In specific implementation, firstly, by collecting a large amount of combat trauma ultrasound image data, artifact regions and normal tissue regions in the combat trauma ultrasound image data are manually labeled. Then, existing feature extraction algorithms are used to extract artifact features and normal tissue features of the corresponding regions, and a structured deep learning prior knowledge graph is constructed according to categories. The imaging characteristics that are prone to occur in the combat trauma ultrasound images during the imaging process include artifact features and normal tissue features. The artifact features include acoustic shadowing and speckles, and the normal tissue features include, but are not limited to, grayscale distribution and texture structure. The deep learning prior knowledge graph refers to the feature information containing various artifact features and normal tissue features in the combat trauma ultrasound images, which can be used to guide subsequent image processing. Secondly, based on the deep learning prior knowledge graph, pattern recognition methods (such as support vector machines) are used to extract the scanning features of each region in the combat trauma ultrasound images. Cosine similarity is used to classify the regions in the combat trauma ultrasound images. The features are compared and matched with the artifact features in the deep learning prior knowledge graph, and the artifact category corresponding to the largest cosine similarity is taken as the artifact noise category (such as acoustic artifact or speckle artifact) of the war trauma ultrasound image. Different artifact features correspond to different artifact noise categories. The artifact noise category refers to the image interference type in the war trauma ultrasound image that does not belong to the real tissue structure. Finally, according to the identified artifact noise category, the corresponding adaptive filtering algorithm is called. When the artifact noise category is speckle artifact, the existing speckle denoising algorithm (such as wide filtering) is used to suppress the noise of the war trauma ultrasound image to obtain a preprocessed ultrasound image. When the artifact noise category is acoustic artifact, the existing region growing-based compensation algorithm is used to suppress the noise of the war trauma ultrasound image to obtain a preprocessed ultrasound image. In addition, other adaptive filtering algorithms can also be used to suppress the noise of the war trauma ultrasound image, which is not limited here.

[0057] It should be noted that the preprocessed ultrasound image in this application refers to the war trauma ultrasound image after artifact noise suppression. By determining the preprocessed ultrasound image, the purpose is to specifically process and remove the inherent artifacts (such as spots and acoustic shadows) in the war trauma ultrasound image, optimize the contrast and clarity of the war trauma ultrasound image, and make the key features such as the texture and edges of the war trauma tissue in the image more prominent and easier to identify. This provides a more reliable basis for the subsequent extraction of global contextual features of image texture and local high-frequency features of image edges from the image.

[0058] In step S102, global context features of image texture and local high-frequency features of image edges are extracted from the preprocessed ultrasound image, and then a semantic fusion feature map containing deep image semantic information is generated based on the global context features and the local high-frequency features.

[0059] In some embodiments, the extraction of global contextual features of image texture and local high-frequency features of image edges from the preprocessed ultrasound image is achieved through the following steps:

[0060] The preprocessed ultrasound image is input into a pre-trained deep convolutional neural network, and the global context features of the image texture are extracted through the deep network of the deep convolutional neural network.

[0061] The preprocessed ultrasound image is subjected to multi-scale Gaussian difference filtering to generate a multi-scale edge response map;

[0062] The local high-frequency features of the image edges are extracted based on the multi-scale edge response map.

[0063] In specific implementation, firstly, a deep convolutional neural network (such as the existing visual geometry network) pre-trained on a large-scale war trauma ultrasound image dataset is selected. The pre-processed ultrasound images are adjusted according to the network input size requirements. The deep convolutional neural network then extracts features layer by layer from the pre-processed ultrasound images through a deep network structure composed of multiple stacked convolutional and pooling layers. This extracts the distribution and morphological features reflecting image texture in the pre-processed ultrasound images. Then, based on all the extracted features, a global context feature of the image texture is constructed. This global context feature refers to a high-dimensional feature set reflecting the distribution and morphological features of image texture in the pre-processed ultrasound images. Secondly, Gaussian kernels of different scales are used... Gaussian blurring is performed on several pairs of preprocessed ultrasound images to obtain multiple sets of blurred ultrasound images at different scales. Then, difference operations are performed on blurred ultrasound images at adjacent scales to generate multi-scale edge response maps. The multi-scale edge response map refers to the image that responds to the edge structure in the preprocessed ultrasound image at different spatial scales, which can highlight the edges of tissues of different sizes. Finally, based on the multi-scale edge response map, gradient operators (such as the Sobel operator) are used to calculate the gradient magnitude and direction of the pixels in the multi-scale edge response map. Pixels with gradient magnitudes greater than a preset gradient threshold are retained by thresholding, thereby extracting the local high-frequency features of the image edges. The gradient threshold can be set according to actual needs and is not limited here.

[0064] It should be noted that, in this application, local high-frequency features refer to features that reflect local details of the image edges in the preprocessed ultrasound image. By determining the global context features of the image texture and the local high-frequency features of the image edges, the global context features can reflect the morphology of the image texture in the preprocessed ultrasound image, providing global-level information support for the regional segmentation of the preprocessed ultrasound image. Meanwhile, the local high-frequency features can accurately capture the detailed changes in the image edges. The combination of the two can achieve multi-dimensional feature representation from the whole to the part, avoiding the loss of local details due to relying solely on global features, and preventing overall judgment bias caused by focusing only on local features. This lays a comprehensive and accurate feature foundation for the subsequent generation of semantic fusion feature maps containing deep image semantic information.

[0065] In some embodiments, reference Figure 2 As shown, this figure is an exemplary flowchart of determining a semantic fusion feature map according to some embodiments of this application. In this embodiment, generating a semantic fusion feature map containing deep image semantic information based on the global context features and the local high-frequency features can be achieved by the following steps:

[0066] In step S1021, the global context features are subjected to dimension adaptation processing to generate adapted global features that match the dimensions of local high-frequency features.

[0067] In step S1022, an attention mechanism is used to calculate the feature association weights between the adaptive global features and the local high-frequency features;

[0068] In step S1023, the adaptive global features and the local high-frequency features are weighted and fused based on the feature association weights to generate an initial fused feature map;

[0069] In step S1024, the initial fusion feature map is enhanced by the residual connection module to obtain a semantic fusion feature map containing deep image semantic information.

[0070] In specific implementation, firstly, addressing the dimensional differences between global context features and local high-frequency features, a feature mapping method (such as through a fully connected layer) is used to linearly transform the global context features, adjusting their feature dimensions to match those of the local high-frequency features, thus generating adapted global features. These adapted global features are global context features capable of subsequent fusion operations with the local high-frequency features. Secondly, an attention mechanism (such as calculating the cosine similarity between the adapted global features and the local high-frequency features) is used to quantify the correlation strength between them. Then, a softmax function is used to convert the correlation strength of each element into feature correlation weights. The feature association weight refers to the weight value used to measure the degree of association between the global features and the local high-frequency features at different positions. Then, the global features and the local high-frequency features are summed element-wise according to the feature association weight to generate an initial fusion feature map. The initial fusion feature map refers to the feature map that is initially integrated with the global features and the local high-frequency features. Finally, the initial fusion feature map is added element-wise to the features (such as the global features or the local high-frequency features) input to the residual connection module through the residual connection module. This supplements the detailed information lost during the feature fusion process, enhances the representational ability of the fusion feature map, and thus obtains a semantic fusion feature map containing deep image semantic information.

[0071] It should be noted that the semantic fusion feature map in this application refers to a feature map that can reflect the deep semantic information of war trauma ultrasound images. In the prior art, when global features and local features are simply spliced ​​or weighted and fused, feature redundancy and dilution of key information are easy to occur. This solution ensures that the two types of features are accurately associated under the unified space and channel dimensions through dimensional adaptation. By using feature association weights to dynamically quantify the cross-dimensional association strength between global context and local high-frequency features, the fusion process focuses more on preserving features that are semantically associated with the war trauma ultrasound image. Then, the residual connection is used to specifically supplement the weakened global distribution trend or local edge details in the fusion process, and finally a semantic fusion feature map containing deep image semantic information is formed to accurately preserve the edge gray-scale changes and subtle textures at the local level in the war trauma ultrasound image, thereby realizing the deep recognition of semantic information in the war trauma ultrasound image.

[0072] In step S103, the feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and the preset feature template are determined based on a preset sliding window, and then the segmentation probability matrix for segmenting the preprocessed ultrasound image is determined according to the feature correlation coefficients corresponding to each feature sub-region.

[0073] In some embodiments, the feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and the preset feature template are determined based on a preset sliding window using the following steps:

[0074] Multiple feature sub-regions are generated by sliding sampling on the semantic fusion feature map according to a preset sliding window;

[0075] Determine the feature vectors corresponding to each feature sub-region;

[0076] Calculate the cosine similarity between the feature vector corresponding to each feature sub-region and the preset feature template;

[0077] The cosine similarity is normalized to obtain the feature correlation coefficient between each feature sub-region and the preset feature template.

[0078] In specific implementation, firstly, a fixed-size sliding window (e.g., 5×5 pixels) is set and slids sequentially along the horizontal and vertical directions on the semantic fusion feature map with a preset step size (e.g., 1 pixel). Each slide extracts a corresponding region as a feature sub-region, thereby generating multiple feature sub-regions. The feature sub-region refers to the local feature region extracted from the semantic fusion feature map through the sliding window. Secondly, for each feature sub-region, the mean, variance, and entropy of the pixel grayscale values ​​in each feature sub-region are extracted using a feature vector extraction method and concatenated into a one-dimensional vector, generating the feature vector corresponding to each feature sub-region. The feature vector refers to the feature used by the feature sub-region to quantitatively represent the features of that region. Then, a preset war trauma feature template library (which contains preset feature templates constructed from historical war trauma ultrasound images, specifically based on historical war trauma ultrasound images processed by existing machine learning models) is used. The process of obtaining the preset feature template (which is a standard template image containing the morphology and texture of historical war trauma ultrasound images, which will not be elaborated here) is as follows: The preset feature template is retrieved from the image and the cosine similarity between the feature vector corresponding to each feature sub-region and the feature vector corresponding to the preset feature template is calculated. The feature vector corresponding to the preset feature template can be determined using the same method as the feature vector corresponding to the feature sub-region, which is not limited here. The cosine similarity measures the degree of similarity between the feature vector corresponding to the feature sub-region and the feature vector corresponding to the preset feature template, and is used to reflect the similarity between the image features of the current war trauma ultrasound image and the standard image features. Finally, the calculated cosine similarity is processed using a normalization method (such as mapping the cosine similarity to the [0,1] interval through linear transformation) to make the result fall within a uniform numerical range, thus obtaining the feature correlation coefficient between each feature sub-region and the preset feature template.

[0079] It should be noted that the feature correlation coefficient in this application refers to a numerical value that measures the degree of correlation between the feature sub-region and the preset feature template. By determining the feature correlation coefficient, the degree of correlation between each feature sub-region in the semantic fusion feature map and the preset feature template can be quantified. On the one hand, the magnitude of the correlation coefficient can be used to accurately locate the local area in the image that is highly correlated with the typical preset feature template, providing a quantitative basis for the regional segmentation of war trauma ultrasound images.

[0080] In some embodiments, the segmentation probability matrix for segmenting the preprocessed ultrasound image based on the feature correlation coefficients corresponding to each feature sub-region is determined by the following steps:

[0081] The feature correlation coefficients corresponding to each feature sub-region are mapped to the corresponding spatial locations of the semantic fusion feature map to generate an initial probability distribution matrix;

[0082] The initial probability distribution matrix is ​​spatially smoothed to obtain a smoothed probability distribution matrix;

[0083] The smooth probability distribution matrix is ​​normalized using a flexible maximum value function to obtain the segmentation probability value of each pixel belonging to the segmented region;

[0084] A segmentation probability matrix is ​​constructed based on the segmentation probability value of each pixel when segmenting the preprocessed ultrasound image.

[0085] In specific implementation, firstly, the feature association coefficients corresponding to each feature sub-region are mapped according to the spatial coordinates of the corresponding feature sub-region in the semantic fusion feature map. That is, each pixel covered by a feature sub-region is assigned a feature association coefficient of that feature sub-region, forming an initial probability distribution matrix. The initial probability distribution matrix refers to a matrix containing the initial association probabilities of each pixel, directly generated by mapping the feature association coefficients according to their spatial positions. Secondly, a spatial smoothing algorithm (such as Gaussian filtering) is used to process the initial probability distribution matrix. By weighting the initial association probabilities of each pixel with its neighboring pixels (the weights can be set according to actual needs, for example, decreasing with increasing distance), a smoothed probability distribution matrix is ​​obtained. The smoothed probability distribution matrix refers to a matrix that has undergone spatial smoothing. The probability distribution matrix is ​​smoothed. Then, a softmax function is applied to the probability value of each pixel in the smoothed probability distribution matrix. By performing an exponential operation on the probability values ​​of all pixels and normalizing them, the probability value of each pixel falls within the interval [0,1] and the sum of the probability values ​​of all pixels is 1. This yields the segmentation probability value of each pixel belonging to the segmentation region. The segmentation probability value of a pixel belonging to the segmentation region refers to the numerical value that quantifies the likelihood of a pixel belonging to the segmentation region. The segmentation region refers to the region in the preprocessed ultrasound image that needs to be segmented. Finally, the segmentation probability values ​​of all pixels are arranged according to their spatial position in the semantic fusion feature map to construct the segmentation probability matrix for segmenting the preprocessed ultrasound image.

[0086] It should be noted that the segmentation probability matrix in this application refers to the matrix used to guide the segmentation of preprocessed ultrasound images. The segmentation probability matrix provides a precise and quantitative basis for the segmentation and recognition of preprocessed ultrasound images. The probability value of each pixel in the matrix intuitively reflects the possibility that it belongs to the segmentation region. It can quickly locate high-probability segmentation regions by setting a probability threshold, and judge the segmentation boundary and overall shape based on the spatial distribution characteristics of the probability values. This avoids the problem of boundary blurring or region omission caused by relying on a single threshold in traditional segmentation methods, and effectively improves the objectivity and accuracy of the segmentation results.

[0087] In step S104, the preprocessed ultrasound image is regionalized according to the segmentation probability matrix to obtain the regionalized segmented image corresponding to the combat trauma ultrasound image.

[0088] In some embodiments, the preprocessed ultrasound image is regionalized based on the segmentation probability matrix to obtain the regionalized segmented image corresponding to the combat trauma ultrasound image, which is achieved through the following steps:

[0089] Based on the segmentation probability matrix, a first segmentation value and a second segmentation value are determined for regional segmentation of the preprocessed ultrasound image.

[0090] Obtain the segmentation probability value of the corresponding pixel in the preprocessed ultrasound image from the segmentation probability matrix;

[0091] Pixel regions with segmentation probability values ​​greater than or equal to the second segmentation value are classified as first-level segmentation regions, pixel regions with segmentation probability values ​​between the first segmentation value and the second segmentation threshold are classified as second-level segmentation regions, and pixel regions with segmentation probability values ​​less than or equal to the first segmentation value are classified as third-level segmentation regions, thereby obtaining the regionalized segmentation image corresponding to the combat trauma ultrasound image.

[0092] In specific implementation, firstly, the maximum and minimum probability values ​​in the segmentation probability matrix are extracted. The length of the probability interval is determined by the difference between the maximum and minimum probability values. One-third of the probability interval length is used as the baseline probability segmentation value. The sum of the minimum probability value and one times the baseline probability segmentation value is used as the first segmentation value. The sum of the minimum probability value and two times the baseline probability segmentation value is used as the second segmentation value. The first and second segmentation values ​​refer to the first and second thresholds used for segmenting the preprocessed ultrasound image. Secondly, the segmentation probability values ​​of corresponding pixels in the preprocessed ultrasound image are obtained from the segmentation probability matrix. Then, pixel regions with segmentation probability values ​​greater than or equal to the second segmentation value are divided into first-level segmentation regions, pixel regions with segmentation probability values ​​between the first segmentation value and the second segmentation threshold are divided into second-level segmentation regions, and pixel regions with segmentation probability values ​​less than or equal to the first segmentation value are divided into third-level segmentation regions, thereby obtaining the regionalized segmented image corresponding to the combat trauma ultrasound image.

[0093] It should be noted that the regionalized segmentation image in this application refers to the image after dividing a group of pixels with similar deep semantic features in a preprocessed ultrasound image. Because war trauma ultrasound images are easily affected by imaging environment interference (such as battlefield movement scenes and equipment shaking), resulting in noise and artifacts, and the boundaries between damaged and normal tissues are often blurred, directly segmenting the original war trauma ultrasound image makes it difficult to identify the semantic attribution of pixel regions. Dividing the segmentation probability values ​​into three levels (Level 1, Level 2, and Level 3 segmentation regions) is a targeted design based on the pathological characteristics of war trauma tissue. Specifically, Level 1 segmentation regions (high-probability areas) correspond to the pixel regions with the highest segmentation probability values, typically representing the core semantic regions of war trauma ultrasound images. The secondary segmentation region (medium probability region) represents a transitional region with a moderate probability value, often corresponding to the image edge region of combat trauma ultrasound images. The tertiary segmentation region (low probability region) represents normal tissue or irrelevant background with the lowest probability value. This three-level segmentation strategy overcomes the limitations of traditional binary segmentation in complex combat trauma scenarios and can more finely characterize the heterogeneity gradient of damaged tissue. Therefore, by performing regional segmentation on combat trauma ultrasound images, the semantic attribution of pixel regions can be effectively clarified, irrelevant background information and noise interference can be filtered out, the segmentation features of the image segmentation can be highlighted, and the technical problems of semantic ambiguity and unclear segmentation targets in ultrasound images in combat trauma scenarios can be solved, thereby significantly improving the effectiveness of segmentation processing of combat trauma ultrasound images.

[0094] Furthermore, in another aspect of this application, in some embodiments, this application provides a deep learning-based combat trauma ultrasound image diagnostic system, which includes a combat trauma ultrasound image processing unit, referencing... Figure 3 The figure is a schematic diagram of the structure of a combat trauma ultrasound image processing unit according to some embodiments of this application. The combat trauma ultrasound image processing unit includes: an acquisition module 201, a processing module 202, and an execution module 203, which are described below:

[0095] The acquisition module 201 in this application is mainly used to acquire the war trauma ultrasound image to be processed, and to perform adaptive preprocessing based on deep learning prior knowledge on the war trauma ultrasound image to obtain a preprocessed ultrasound image.

[0096] Processing module 202, in this application, is mainly used to extract global context features of image texture and local high-frequency features of image edges from the preprocessed ultrasound image, and then generate a semantic fusion feature map containing deep image semantic information based on the global context features and the local high-frequency features.

[0097] The processing module 202 is further configured to determine the feature correlation coefficient between different feature sub-regions in the semantic fusion feature map and the preset feature template based on a preset sliding window, and then determine the segmentation probability matrix for segmenting the preprocessed ultrasound image based on the feature correlation coefficients corresponding to each feature sub-region.

[0098] The execution module 203 in this application is mainly used to perform regional segmentation on the preprocessed ultrasound image according to the segmentation probability matrix to obtain the regional segmentation image corresponding to the combat trauma ultrasound image.

[0099] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, the processor being configured to acquire the code and execute the above-described deep learning-based war trauma ultrasound image processing method.

[0100] In some embodiments, reference Figure 4 The figure is a schematic diagram of the structure of a computer device implementing a deep learning-based ultrasonic image processing method for combat trauma, according to some embodiments of this application. The deep learning-based ultrasonic image processing method for combat trauma in the above embodiments can... Figure 4 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0101] The processor 301 can be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the deep learning-based war trauma ultrasound image processing method in this application.

[0102] The communication bus 302 can be used to transmit information between the aforementioned components.

[0103] The memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory 303 may exist independently and be connected to the processor 301 via the communication bus 302. The memory 303 may also be integrated with the processor 301.

[0104] The memory 303 stores program code for executing the scheme of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiments, the determination of the deep learning-based war trauma ultrasound image processing method can be implemented by the processor 301 and one or more software modules in the program code in the memory 303.

[0105] Communication interface 304 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0106] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0107] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0108] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described deep learning-based method for processing war trauma ultrasound images.

[0109] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0110] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A deep learning-based method for processing war trauma ultrasound images, used in a war trauma ultrasound image diagnostic system for processing war trauma ultrasound images, characterized in that, The method includes the following steps: Acquire the war trauma ultrasound image to be processed, and perform adaptive preprocessing based on deep learning prior knowledge on the war trauma ultrasound image to obtain the preprocessed ultrasound image. Global context features of image texture and local high-frequency features of image edges are extracted from the preprocessed ultrasound image, and then a semantic fusion feature map containing deep image semantic information is generated based on the global context features and the local high-frequency features. Based on a preset sliding window, the feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and the preset feature template are determined, and then the segmentation probability matrix for segmenting the preprocessed ultrasound image is determined according to the feature correlation coefficients corresponding to each feature sub-region. The preprocessed ultrasound image is regionalized based on the segmentation probability matrix to obtain the regionalized segmented image corresponding to the combat trauma ultrasound image. Specifically, extracting global contextual features of image texture and local high-frequency features of image edges from the preprocessed ultrasound image includes: The preprocessed ultrasound image is input into a pre-trained deep convolutional neural network, and the global context features of the image texture are extracted through the deep network of the deep convolutional neural network. The preprocessed ultrasound image is subjected to multi-scale Gaussian difference filtering to generate a multi-scale edge response map; Based on the multi-scale edge response map, local high-frequency features of the image edge are extracted. These local high-frequency features refer to features that reflect local details of the image edge in the preprocessed ultrasound image. Specifically, determining the feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and the preset feature template based on a preset sliding window includes: Multiple feature sub-regions are generated by sliding sampling on the semantic fusion feature map according to a preset sliding window; Determine the feature vectors corresponding to each feature sub-region; Calculate the cosine similarity between the feature vector corresponding to each feature sub-region and the preset feature template; The cosine similarity is normalized to obtain the feature correlation coefficient between each feature sub-region and the preset feature template.

2. The method as described in claim 1, characterized in that, The war trauma ultrasound images are subjected to adaptive preprocessing based on deep learning prior knowledge to obtain preprocessed ultrasound images, specifically including: Based on the imaging characteristics of war trauma ultrasound images, a deep learning prior knowledge graph containing artifact features and normal tissue features of typical war trauma ultrasound images is constructed. Based on the deep learning prior knowledge graph, artifact noise type identification is performed on the acquired war trauma ultrasound images to determine the category of artifact noise present in the war trauma ultrasound images. Based on the identified artifact noise category, the corresponding adaptive filtering algorithm is called to suppress noise in the war trauma ultrasound image, resulting in a preprocessed ultrasound image.

3. The method as described in claim 1, characterized in that, Generating a semantic fusion feature map containing deep image semantic information based on the global context features and the local high-frequency features specifically includes: The global context features are subjected to dimension adaptation processing to generate adapted global features that match the dimensions of local high-frequency features; An attention mechanism is used to calculate the feature association weights between the adaptive global features and the local high-frequency features; Based on the feature association weights, the adaptive global features and the local high-frequency features are weighted and fused to generate an initial fused feature map; The initial fused feature map is enhanced by the residual connection module to obtain a semantic fused feature map containing deep image semantic information.

4. The method as described in claim 1, characterized in that, The segmentation probability matrix for segmenting the preprocessed ultrasound image is determined based on the feature correlation coefficients corresponding to each feature sub-region. Specifically, this includes: The feature correlation coefficients corresponding to each feature sub-region are mapped to the corresponding spatial locations of the semantic fusion feature map to generate an initial probability distribution matrix; The initial probability distribution matrix is ​​spatially smoothed to obtain a smoothed probability distribution matrix; The smooth probability distribution matrix is ​​normalized using a flexible maximum value function to obtain the segmentation probability value of each pixel belonging to the segmented region; A segmentation probability matrix is ​​constructed based on the segmentation probability value of each pixel when segmenting the preprocessed ultrasound image.

5. The method as described in claim 1, characterized in that, Retrieve the war trauma ultrasound images to be processed from the war trauma ultrasound image storage file.

6. A deep learning-based combat trauma ultrasound image diagnostic system, used to execute the deep learning-based combat trauma ultrasound image processing method as described in any one of claims 1 to 5, wherein the deep learning-based combat trauma ultrasound image diagnostic system includes a combat trauma ultrasound image processing unit, characterized in that, The war trauma ultrasound image processing unit includes: The acquisition module is used to acquire the war trauma ultrasound image to be processed, and to perform adaptive preprocessing based on deep learning prior knowledge on the war trauma ultrasound image to obtain a preprocessed ultrasound image. The processing module is used to extract global context features of image texture and local high-frequency features of image edges from the preprocessed ultrasound image, and then generate a semantic fusion feature map containing deep image semantic information based on the global context features and the local high-frequency features. The processing module is further configured to determine the feature correlation coefficients between different feature sub-regions in the semantic fusion feature map and the preset feature template based on a preset sliding window, and then determine the segmentation probability matrix for segmenting the preprocessed ultrasound image based on the feature correlation coefficients corresponding to each feature sub-region. The execution module is used to perform regional segmentation on the preprocessed ultrasound image according to the segmentation probability matrix to obtain the regional segmentation image corresponding to the combat trauma ultrasound image.

7. A computer device, characterized in that, The computer device includes a memory and a processor, the memory storing code, and the processor being configured to retrieve the code and execute the deep learning-based war trauma ultrasound image processing method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the deep learning-based ultrasonic image processing method for combat trauma as described in any one of claims 1 to 5.

Images