An emergency fracture automatic recognition and diagnosis ultrasonic auxiliary system

Abstract

The application discloses an emergency fracture automatic identification and diagnosis ultrasonic auxiliary system and relates to the technical field of acoustic fracture detection. The auxiliary system comprises an ultrasonic image acquisition module for acquiring ultrasonic image data of a fracture position of a patient; a processing module for processing the ultrasonic image data and outputting identification data; and a diagnosis decision module for outputting diagnosis data based on the identification data of the processing module. The application is characterized in that a regional focus analysis unit is arranged, the calculation resource distribution is optimized by gradually narrowing the analysis range, the primary detection layer rapidly excludes the intact region, and the primary fracture region mark is output. The soft tissue reaction quantitative analysis of the primary fracture region is carried out by the intermediate classification layer, the marked region is generated, the marked region is processed by the senior positioning layer, the resource consumption is reduced by combining the sub-pixel level analysis, the system resource is rapidly concentrated on the fracture region by the architecture, and the processing time of fracture identification is reduced.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of ultrasonic fracture detection, in particular to an ultrasonic auxiliary system for automatic identification and diagnosis of emergency fractures. BACKGROUND

[0002] Emergency fractures are common traumatic diseases in clinical practice, and rapid and accurate diagnosis is crucial for treatment decision-making and prognosis evaluation.

[0003] For example, the application number "CN115272176A" and the name "A fracture identification model and method based on a hip joint DR image" capture global context information through the multi-head self-attention layer and multi-head attention layer in the fracture identification model, thereby establishing a long-distance dependence on the target and extracting more powerful features. The algorithm network proposed in the above patent significantly improves the recall rate of the fracture area, and the number of false positives of the fracture is also significantly reduced, thereby improving the accuracy of fracture identification.

[0004] The above patent identifies fractures at the hip joint, and there are complete and intact anatomical regions in the hip joint ultrasound image. Since there is no fracture phenomenon in this area, the transition analysis of these areas will significantly increase the processing time of fracture identification. Therefore, an ultrasonic auxiliary system for automatic identification and diagnosis of emergency fractures is invented. SUMMARY

[0005] The purpose of the present application is to provide an ultrasonic auxiliary system for automatic identification and diagnosis of emergency fractures to solve the problems raised in the background art.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical scheme: an ultrasonic auxiliary system for automatic identification and diagnosis of emergency fractures, the auxiliary system comprising:

[0007] an ultrasonic image acquisition module for acquiring ultrasonic image data of the fracture position of a patient;

[0008] a processing module for processing the ultrasonic image data and outputting identification data;

[0009] a diagnosis decision module for outputting diagnosis data based on the identification data of the processing module;

[0010] The processing module comprises:

[0011] a bone topology unit; the ultrasonic image data is processed by a bone analysis method, a mapping relationship between the ultrasonic coordinate system and the anatomical coordinate system is established, and a topology graph containing the mapping relationship is obtained;

[0012] The regional focus analysis unit is provided with a progressive analysis framework for step-by-step narrowing and focusing processing, and a feedback mechanism for realizing information interaction between layers is arranged in the progressive analysis framework.

[0013] The progressive analysis framework comprises:

[0014] The primary detection layer: the topological graph is analyzed by a recognition method to obtain bone cortex continuity data, bone fragment strong echo data and bone fracture broken edge feature data, obtain a bone fracture sign data set, and obtain a preliminary fracture region from the ultrasound image data based on the continuity interruption, the bone fragment strong echo area, the region where the broken edge feature of the bone fracture is located and the mapping relationship;

[0015] The intermediate classification layer: the intensity of the periosteal reaction and the range of the marked soft tissue injury in the preliminary fracture region are quantified by a quantification method to obtain a soft tissue reaction data set, and the range of the soft tissue injury and the preliminary fracture region are combined to obtain a marked region;

[0016] The advanced positioning layer: the marked region is analyzed by a sub-pixel level method, the sub-pixel level method includes analyzing the spatial arrangement structure of the trabeculae and the micro-fracture line, calculating the angle deviation value of the normal vector of the trabeculae fracture surface, marking the position of the trabeculae whose angle deviation value exceeds the deviation threshold, and obtaining a micro-injury area and a micro-injury data set;

[0017] The topological graph, the bone fracture sign data set, the soft tissue reaction data set and the micro-injury data set are combined to obtain recognition data.

[0018] Further, the bone analysis method comprises:

[0019] The bone topology unit stores a full bone anatomy segmentation model library and a bone anatomy reference database for realizing image data structure analysis;

[0020] The geometric structure feature vector of the bone region in the ultrasound image data is extracted by the full bone anatomy segmentation model, and the bone anatomy landmark points in the ultrasound image data are marked;

[0021] The bone anatomy reference database comprises a reference anatomical feature point set, the bone anatomy landmark points are corresponding to the reference anatomical feature points in the reference anatomical feature point set, and corresponding bone anatomy reference data is obtained from the bone anatomy reference database according to the corresponding relationship;

[0022] The topological graph is established based on the geometric structure feature vector, the mapping relationship and the ultrasound image data.

[0023] Further, the feedback mechanism comprises:

[0024] The micro-damage area detected by the high-level positioning layer is fed back to the primary detection layer, and the primary detection layer performs secondary scanning on the micro-damage area at a sub-pixel level to obtain a sub-pixel level fracture sign data set, and the sub-pixel level fracture sign data set replaces the fracture sign data set;

[0025] The periosteal thickening value based on the intermediate classification layer quantization adjusts the range size of the preliminary fracture area of the primary detection layer.

[0026] Further, the diagnostic decision module includes processing the recognition data by a diagnostic method to obtain a diagnostic report;

[0027] The diagnostic method includes: associating an AO typing database with a trauma scoring unit, inputting the recognition data into the AO typing database and the trauma scoring unit and performing association, and generating a diagnostic report containing fracture type and displacement risk level.

[0028] Further, the recognition method includes performing cortical bone continuity analysis on the topological graph, scanning bone cortical surface geometric features, detecting linear fracture zones with a length exceeding a length threshold, identifying broken edge features of comminuted fractures based on curvature discontinuity points, and obtaining cortical bone continuity data, bone fragment strong echo data, and broken edge feature data of fractures.

[0029] The mapping relationship between the ultrasound coordinate system and the anatomical coordinate system includes:

[0030] The three-dimensional view of the bone overlap structure in the ultrasound image data is unfolded by a surface reconstruction algorithm, and the ultrasound coordinate system is established;

[0031] Based on the bone analysis method, bone anatomical reference data associated with the ultrasound image data is obtained, and a three-dimensional view of the bone anatomical reference data is unfolded by a surface reconstruction algorithm, and an anatomical coordinate system is established,

[0032] Based on the bone anatomical landmark points and the reference anatomical feature points, the mapping relationship between the ultrasound coordinate system and the anatomical coordinate system is established.

[0033] Further, the geometric structure feature vector of the bone region includes a curvature radius R and a cortical thickness gradient ΔT, and the labeled anatomical landmark points include an epiphyseal growth plate growth line L and a joint surface boundary B.

[0034] Further, the quantification method includes:

[0035] The continuous value ΔH of the periosteal elevation height is measured, the blood clot area A is quantitatively calculated, and the region where the soft tissue swelling is located is labeled;

[0036] A tissue response feature vector of the continuous value ΔH and the blood clot area A parameters is generated;

[0037] When the ΔH is greater than the continuous value threshold, a periosteal region where the ΔH is greater than the continuous value threshold is marked;

[0038] The periosteal region and the region where the soft tissue swelling is located are combined to obtain a soft tissue injury range;

[0039] The soft tissue injury range is combined with the Doppler blood flow signal to obtain a periosteal indirect sign reaction type;

[0040] The periosteal indirect sign reaction type and the soft tissue injury range are combined to obtain a soft tissue reaction dataset.

[0041] Further, the sub-pixel level analysis comprises:

[0042] The trabecular bone spatial arrangement is analyzed, and micro-fracture lines and fracture surface normal vector angle deviation are detected;

[0043] A micro-injury graph is output, a micro-injury core region coordinate set is established in spatial association with an anatomical landmark point, and a micro-injury dataset is obtained in combination with the micro-fracture lines, the fracture surface normal vector angle difference and the spatial association.

[0044] Through the setting of the region focusing analysis unit, the region focusing analysis unit adopts a three-layer architecture of "primary detection -> intermediate classification -> high-level positioning", optimizes the allocation of computing resources by gradually narrowing the analysis range, quickly excludes intact regions in the primary detection layer, outputs a preliminary fracture region mark, performs soft tissue reaction quantitative analysis on the preliminary fracture region in the intermediate classification layer, generates a marked region, and processes the marked region in the high-level positioning layer in combination with the sub-pixel level analysis to reduce resource consumption. This architecture enables the system resources to be quickly concentrated on the fracture region, reducing the processing time of fracture identification.

[0045] By establishing a mapping relationship between the ultrasound coordinate system and the anatomical coordinate system, in combination with the automatic labeling of anatomical landmark points such as epiphyseal plates and articular surfaces, the skeletal structure is accurately positioned in complex images, the sound shadow area is reduced, the gray level co-occurrence matrix is used to quantify the texture features of the periosteal region in the intermediate classification layer, and the contrast difference between the hematoma area and the surrounding tissue is quantified, effectively distinguishing fracture features from artifacts.

[0046] The three-dimensional view of the overlapping skeletal structure is unfolded through the curved surface reconstruction algorithm, the skeletal type is automatically identified and the standard anatomical model is called, and the anatomical landmark points such as epiphyseal plates and articular surfaces are dynamically labeled. This technology solves the problem of diagnostic blind area caused by the overlapping of skeletal structures in traditional two-dimensional ultrasound, reduces the coverage of the sound shadow area on the fracture sign, and enables the observation of the spatial arrangement of the fracture and the distribution of the micro-injury, thereby improving the recognition ability of complex fractures. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 It is a schematic diagram of the system of the present application;

[0048] Figure 2A schematic diagram of a processing module of the present application;

[0049] Figure 3 A schematic diagram of a progressive analysis architecture of the present application. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0051] As shown in Figure 1 - Figure 3 The present application provides a technical solution: an ultrasonic auxiliary system for automatic identification and diagnosis of emergency fractures, the auxiliary system comprising:

[0052] an ultrasonic image acquisition module for acquiring ultrasonic image data of a fracture position of a patient;

[0053] a processing module for data processing of the ultrasonic image data and output of identification data;

[0054] a diagnosis decision module for output of diagnosis data based on the identification data of the processing module;

[0055] The processing module comprises:

[0056] a bone topology unit; the ultrasonic image data is processed by a bone analysis method, a mapping relationship between an ultrasonic coordinate system and an anatomical coordinate system is established, and a topology graph containing the mapping relationship is acquired;

[0057] a region focusing analysis unit: provided with a progressive analysis architecture for stepwise narrowing and focusing processing, and provided with a feedback mechanism for realizing information interaction between layers in the progressive analysis architecture;

[0058] The progressive analysis architecture comprises:

[0059] a primary detection layer: the topology graph is analyzed by an identification method, bone cortex continuity data, bone fragment strong echo data and fracture broken edge feature data are acquired, a fracture sign data set is obtained, and a preliminary fracture region is acquired from the ultrasonic image data based on the continuity interruption, the region where the bone fragment strong echo area and the fracture broken edge feature are located and the mapping relationship;

[0060] a middle classification layer: the synovial membrane reaction intensity in the preliminary fracture region and the marked soft tissue injury range are quantified by a quantification method, a soft tissue reaction data set is obtained, and a marked region is obtained by combining the soft tissue injury range and the preliminary fracture region;

[0061] Advanced positioning layer: analyzing the labeled region through a sub-pixel level method, the sub-pixel level method including analyzing the spatial arrangement structure of the trabecular bone and the micro-fracture line, calculating the angle deviation value of the normal vector of the trabecular bone fracture surface, labeling the trabecular bone position whose angle deviation value exceeds the deviation threshold value, obtaining the micro-damage area and the micro-damage data set;

[0062] The recognition data is obtained by combining the topological graph, the fracture sign data set, the soft tissue reaction data set and the micro-damage data set.

[0063] The bone analysis method includes:

[0064] The bone topology unit stores a full bone anatomy segmentation model library and a bone anatomy reference database for realizing image data structure analysis;

[0065] The geometric structure feature vector of the bone region in the ultrasound image data is extracted by the full bone anatomy segmentation model, and the bone anatomy landmark points in the ultrasound image data are labeled;

[0066] The bone anatomy reference database includes a reference anatomical feature point set, and the bone anatomy landmark points are corresponded to the reference anatomical feature points in the reference anatomical feature point set, and the corresponding bone anatomy reference data is obtained from the bone anatomy reference database according to the corresponding relationship;

[0067] The topological graph is established based on the geometric structure feature vector, the mapping relationship and the ultrasound image data.

[0068] The feedback mechanism includes:

[0069] The micro-damage area detected by the advanced positioning layer is fed back to the primary detection layer, and the primary detection layer performs secondary scanning on the micro-damage area at the sub-pixel level to obtain a sub-pixel level fracture sign data set of the micro-damage area, and the sub-pixel level fracture sign data set replaces the fracture sign data set;

[0070] The range size of the preliminary fracture area of the primary detection layer is adjusted based on the periostream thickening value quantified by the intermediate classification layer.

[0071] The diagnosis decision module includes processing the recognition data by a diagnosis method to obtain a diagnosis report;

[0072] The diagnosis method includes: associating the AO typing database with the trauma scoring unit, inputting the recognition data into the AO typing database and the trauma scoring unit and associating them to generate a diagnosis report containing the fracture type and the displacement risk level.

[0073] The recognition method comprises: performing cortical bone continuity analysis on the topological graph, scanning bone cortical surface geometric features, detecting linear fracture zones with a length exceeding a length threshold, identifying broken edge features of comminuted fractures based on curvature abrupt change points, and obtaining cortical bone continuity data, strong echo data of bone fragments, and broken edge feature data of fractures;

[0074] The mapping relationship between the ultrasound coordinate system and the anatomical coordinate system is established by:

[0075] A three-dimensional view of the overlapping bone structure in the ultrasound image data is unfolded by a surface reconstruction algorithm, and an ultrasound coordinate system is established;

[0076] Based on the bone analysis method, bone anatomical reference data associated with the ultrasound image data is obtained, a three-dimensional view of the bone anatomical reference data is unfolded by a surface reconstruction algorithm, and an anatomical coordinate system is established,

[0077] The mapping relationship between the ultrasound coordinate system and the anatomical coordinate system is established based on the bone anatomical landmark points and the reference anatomical feature points.

[0078] The geometric structure feature vector of the bone region includes the curvature radius R and the cortical thickness gradient ΔT, and the labeled anatomical landmark points include the epiphyseal growth plate line L and the articular surface boundary B.

[0079] The quantification method comprises:

[0080] A continuous value ΔH of the periosteal elevation height is measured, the hematoma area A is quantitatively calculated, and the region where the soft tissue swelling is located is labeled;

[0081] A tissue reaction feature vector of the continuous value ΔH and the hematoma area A parameters is generated;

[0082] When ΔH is greater than a continuous value threshold, the periosteal region where ΔH is greater than the continuous value threshold is labeled;

[0083] The periosteal region, the region where the soft tissue swelling is located, and the soft tissue injury range are obtained in combination;

[0084] The soft tissue injury range is combined with the Doppler blood flow signal to obtain a periosteal indirect sign reaction type;

[0085] The periosteal indirect sign reaction type and the soft tissue injury range are combined to obtain a soft tissue reaction data set.

[0086] The sub-pixel level analysis comprises:

[0087] The bone trabecula spatial arrangement structure is analyzed, and the micro-fracture line and the fracture surface normal vector angle deviation are detected and calculated;

[0088] A micro-damage map is output, a spatial correlation between a micro-damage core region coordinate set and an anatomical landmark point is established, and a micro-damage data set is obtained in combination with the micro-fracture line, the fracture surface normal vector angle difference, and the spatial correlation.

[0089] AO classification database is a fracture classification system established by the International Society of Internal Fixation, which uses an alphanumeric coding system to standardize the description of fracture location, morphology and severity. The AO classification database is an existing database, and the specific content of the AO classification database is not described here. Based on the AO classification database, the alphanumeric coding system in the AO classification database can be used to represent the diagnosis report. At the same time, the natural language processing technology is used in the diagnosis module to automatically convert the quantitative parameters into structured text, which includes the diagnosis conclusion and the injury mechanism analysis.

[0090] The inner layer of the cortical bone and the two ends are many irregular sheet or linear bone structures, called trabeculae. The trabeculae are abundant at the metaphyseal end, although they are continuous with the diaphysis in the inner layer of the cortex, but relatively less in the diaphysis. The trabeculae are arranged in response to the maximum stress and tension, and are connected to each other in a loose sponge-like structure, called cancellous bone.

[0091] The primary detection layer quickly excludes the intact area through the bone cortex fracture characteristics; the secondary classification layer is filtered again through the soft tissue reaction, and the high-level positioning layer only processes the area where the fracture is located. Through the setting of the primary detection layer, the secondary classification layer and the high-level positioning layer, the waste of computing resources is reduced.

[0092] The angle deviation of the fracture surface normal vector is calculated by spatial geometric analysis, and the degree of abnormality of the trabeculae is quantitatively evaluated. The angle deviation of the trabeculae fracture surface normal vector is analyzed. The signs of fracture in the ultrasonic image data include direct signs and indirect signs. The direct signs of fracture include the appearance of fracture or misplacement of the originally continuous and high-echo cortical bone line in the ultrasonic image, forming an obvious interruption gap. Irregular strong echo or low echo area can be seen at the fracture end, indicating bone separation. The indirect signs of fracture include thickening and echo enhancement of the periosteum at the fracture site, which may be accompanied by increased blood flow (visible by Doppler imaging). There are anechoic or hypoechoic areas in the soft tissue around the fracture, with clear or blurred boundaries, indicating bleeding or exudation. The echo of the surrounding tissue is enhanced and the structure is blurred, accompanied by probe compression pain. When there is a fracture near the joint, anechoic or hypoechoic areas can be seen in the joint cavity, indicating joint fluid increase or bleeding, etc.

[0093] When detecting the position of the fracture by ultrasonic wave, there will be intact areas in the obtained ultrasonic image data, which have no signs of fracture. When automatic recognition is enabled, there will be a situation of identifying and analyzing this area. Since there is no fracture phenomenon, the processing and analysis will increase the processing time of the computer.

[0094] The complete region is reduced by the regional focus analysis, and then the focus of the multiple fracture regions is realized, the range of subsequent identification and diagnosis is reduced from the complete ultrasound image data, the processing time of the computer is reduced, the primary detection layer, the intermediate classification layer and the high-level positioning layer, the region where the fracture is located is obtained through the primary detection layer, the primary detection layer, the intermediate classification layer and the high-level positioning layer are the core components of the regional focus analysis module, and the three realize the closed-loop verification from rough screening to accurate diagnosis through the hierarchical detection and feedback mechanism, and the specific functions are as follows: the primary detection layer is based on the topological graph generated by the bone topological mapping subsystem, and uses pre-labeled anatomical landmarks (such as epiphyseal plate and articular surface) and direct signs (bone cortex interruption and bone fragments) for preliminary screening, uses an edge detection algorithm to scan the bone cortex surface, identifies linear fracture zones with a length of more than 1mm and curvature mutation points (characteristics of comminuted fracture), outputs a fracture probability map and high-risk region coordinates, and provides an initial positioning framework for subsequent analysis, the primary detection layer scans the geometric features of the bone cortex surface through the edge detection algorithm, identifies linear fracture zones with a length of more than a set threshold such as 1mm, analyzes the broken edge features of comminuted fractures based on the curvature mutation points, and the region with abnormal sudden increase of curvature is marked as a suspected fracture area.

[0095] The intermediate classification layer performs cutting and focusing on the high-risk regions output by the primary detection layer, quantifies the periosteal lifting height ΔH and the hematoma area A through threshold segmentation, judges the periosteal reaction type (acute inflammation or chronic repair) in combination with the Doppler blood flow signal (resistance index RI), generates a tissue reaction heat map, and transmits the ΔH value to the high-level positioning layer to dynamically adjust the analysis intensity, the intermediate classification layer is based on the preliminary fracture region marked by the primary detection layer, analyzes the texture features of the periosteal region in the preliminary fracture region through a gray level co-occurrence matrix (GLCM), calculates parameters such as contrast and energy, and uses Otsu automatic threshold segmentation technology to quantify the contrast difference between the hematoma area in the preliminary fracture region and the surrounding normal tissue.

[0096] The high-level positioning layer applies a sub-pixel level analysis algorithm to detect the normal vector angle deviation (θ>15°) of the trabecular fracture surface in the region marked by the intermediate classification layer, generates a micro-damage cloud map and marks the micro-damage core area coordinates.

[0097] At the same time, the detected micro-damage coordinates trigger the primary detection layer to perform secondary scanning on the micro-damage coordinate area at the sub-pixel level through the feedback mechanism, obtain the fracture sign data set to replace the original fracture sign data set, form a closed loop of "preliminary screening → accurate positioning → secondary verification", and ensure that the fracture positioning error is less than 1mm. Through the setting of the regional focusing analysis unit, the regional focusing analysis unit adopts a three-layer architecture of "primary detection → intermediate classification → advanced positioning", gradually narrows the analysis range to optimize the allocation of computing resources, the primary detection layer quickly excludes the intact area, and outputs the preliminary fracture area mark, the intermediate classification layer performs soft tissue reaction quantitative analysis on the preliminary fracture area to generate a labeled area, and the advanced positioning layer processes the labeled area, combined with sub-pixel level analysis to reduce resource consumption. This architecture enables the system resources to quickly concentrate on the fracture area, reducing the processing time of fracture identification.

[0098] The gray level co-occurrence matrix (GLCM) is a core tool for image texture analysis, which quantifies the roughness and directionality of image texture by counting the co-occurrence frequency of pixel gray pairs at a specific distance and direction. In fracture detection, the gray level co-occurrence matrix identifies abnormalities by analyzing the texture changes in the periosteal area. In the intermediate classification layer, Otsu method is used to quantify the contrast difference between hematoma area and normal tissue, and through automatic threshold segmentation, the hematoma area boundary is accurately identified. When the hematoma area is greater than 5k square centimeters, the system automatically raises the classification priority of this area, triggering more detailed analysis. k is affected by the size of human bones, as the size of bones varies at different ages, the threshold of hematoma area is also affected by the size of human bones. The size of bones varies significantly, and the same hematoma area may represent different severity in people of different ages, so k is set.

[0099] The areas marked by the intermediate classification layer include: periosteal reaction area: periosteal elevation height ΔH: when ΔH > 2.5mm, the periosteum in this area is included in the soft tissue injury range, soft tissue injury area: detect the hematoma area A, the area where the soft tissue is swollen, mark the swelling area through edge detection algorithm, and assist in evaluating the injury degree. The swelling area is included in the soft tissue injury range, and the soft tissue injury range corrects the suspected fracture area to obtain the labeled area.

[0100] The phase-based sub-pixel positioning algorithm is used in the advanced positioning layer, and the gradient direction histogram (HOG) is combined with the sub-pixel edge detection algorithm to improve the positioning accuracy of the fracture line to 0.1 millimeter level. The micro-damage cloud map makes the trabecular fracture surface distribution visualized, which assists the doctor in evaluating the fracture stability. The secondary scanning of the micro-damage area at the sub-pixel level is to realize the re-operation of the primary detection layer at high pixels. The gradient direction histogram (HOG) is a statistical method for image feature description, and its core function is to capture the edge and shape features of objects by quantifying the gradient direction distribution of local image regions. The principal component analysis (PCA) is used to calculate the normal vector of the trabecular fracture surface. When the angle deviation of the fracture surface is greater than the deviation threshold, the trabecular fracture surface is labeled. In the ultrasonic fracture detection, the PCA quantifies the angle deviation of the normal vector of the trabecular fracture surface to assist in judging the fracture type (such as oblique, spiral) and stability. It is a commonly used feature analysis tool in the prior art. By establishing the mapping relationship between the ultrasonic coordinate system and the anatomical coordinate system, and combining the automatic labeling of anatomical landmarks such as epiphyseal plate and articular surface, the skeletal structure can be accurately positioned in complex images. The gray level co-occurrence matrix is used to quantify the texture features of the periosteum region in the intermediate classification layer, and the automatic threshold segmentation technology based on Otsu method is used to quantify the contrast difference between the hematoma area and the surrounding tissue, effectively distinguishing the fracture features from the artifacts.

[0101] The height of the periosteum is adjusted based on the quantification of the intermediate classification layer to adjust the size of the primary fracture region in the primary detection layer. If the periosteum cannot be included, the parameters for generating the primary fracture region size in the primary detection layer will be adjusted to include the normal periosteum and the periosteum lifted due to fracture.

[0102] The ultrasonic image acquisition module is the input end of the system, responsible for obtaining the original ultrasonic image data of the patient's fracture site. The diagnostic decision module is the output end of the system, which integrates the analysis results of the processing module and generates the final diagnosis report according to the established medical standards (such as AO classification database).

[0103] The coordinate system of the ultrasound image is a local coordinate system relative to the position of the probe, while the anatomical structure of the human body exists in a standardized three-dimensional anatomical coordinate system. First, the accurate mapping between the two needs to be realized. The system has a built-in full-skeletal anatomical segmentation model library and a skeletal anatomical reference database, which stores a large amount of standardized skeletal anatomical data, including key anatomical feature points. The image feature points are automatically extracted from the real-time acquired ultrasound image. Subsequently, by matching with the anatomical feature point set in the database, the system can identify the current scanned bone type (such as femur, tibia) and its approximate orientation. Once the bone position is identified, the system will call the corresponding skeletal anatomical segmentation model. The model further extracts more detailed geometric structure feature vectors from the ultrasound image, and the mapping relationship between the ultrasound local coordinate system and the standard anatomical coordinate system.

[0104] The skeletal topology unit fuses all the information into a topology map containing sign distribution. The topology map not only contains the type, orientation and accurate coordinate mapping relationship of the bone, but is then transmitted to the regional focus analysis unit, providing a key anatomical reference for subsequent progressive analysis.

[0105] The primary detection layer outputs one or more preliminary fracture regions, thereby narrowing the focus of analysis from the entire scanning range to a few fracture regions (preliminary fracture regions). After locating the preliminary fracture regions, the intermediate classification layer intervenes, with the focus being the surrounding tissue reaction caused by the fracture, i.e. the quantitative analysis of indirect signs. The soft tissue reaction is an indirect sign. The intermediate layer generates a tissue reaction feature vector containing parameters ΔH and A, and further accurately labels the region based on these data to provide a more focused target for high-level analysis. The labeled region of the intermediate classification layer is obtained.

[0106] The high-level positioning layer performs sub-pixel level analysis on the labeled region of the intermediate classification layer. When the normal vector angle deviation of the trabecular fracture surface exceeds the deviation threshold, the region is labeled as a micro-damage area. The deviation threshold is 13°-15°. The high-level positioning layer finally outputs a micro-damage map and establishes a spatial correlation between the micro-damage core area coordinate set and the anatomical landmark points labeled by the skeletal topology unit, which enables the micro-damage to be accurately located on the anatomical structure.

[0107] After the high-level positioning layer detects the micro-damage area, its coordinate set (micro-damage coordinate feedback) is fed back to the primary detection layer for secondary scanning and processing of the specific micro area, achieving a review of the micro-damage area to ensure accuracy and obtaining the fracture sign data set of the micro-damage area. The periosteal thickening value ΔH quantified by the intermediate classification layer will be used to expand the preliminary fracture region range delineated by the primary detection layer. Periosteal reaction is a physiological response of the skeleton to stimuli such as injury, inflammation or tumor, and its type can be classified according to pathological mechanisms, morphological characteristics and clinical significance. The periosteal reaction is the prior art.

[0108] The processing module includes a skeletal topology unit for anatomical positioning and a region-focused analysis unit for lesion analysis, the latter adopts a three-layer progressive analysis architecture (primary detection, intermediate classification, and high-level positioning) and introduces a closed-loop feedback mechanism to achieve progressive and accurate identification from macroscopic cortical disruption to microscopic trabecular damage, combined with sign-anatomy correlation analysis (sign distribution topology map is generated through the geometric structure feature vector of the bone region such as the radius of curvature R and the cortical thickness gradient ΔT), the system realizes the correlation diagnosis of direct signs (such as bone cortex interruption) and indirect signs (such as periosteal thickening), significantly improves the recognition ability of complex fractures (such as comminuted fractures and intra-articular fractures), and through the curved surface reconstruction algorithm, the three-dimensional view of the overlapping structure of the bone is developed, the bone type is automatically identified and the standard anatomical model is called, and the epiphyseal plate, articular surface and other anatomical landmark points are dynamically labeled. This technology solves the problem of diagnostic blind area caused by the overlapping structure of the bone in traditional two-dimensional ultrasound, reduces the coverage of the acoustic shadow area to the fracture sign, so as to observe the spatial arrangement and micro-damage distribution of the fracture, combined with sign-anatomy correlation analysis, the system realizes the correlation diagnosis of direct signs and indirect signs, and significantly improves the recognition ability of complex fractures.

[0109] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and variations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended embodiments and their equivalents.

Claims

1. An ultrasound-assisted system for automatic identification and diagnosis of emergency fractures, characterized in that, The auxiliary system includes: An ultrasound image acquisition module used to obtain ultrasound image data of the patient's fracture location; A processing module for processing ultrasound image data and outputting recognition data; A diagnostic decision module for outputting diagnostic data based on the identification data of the processing module; The processing module includes: Skeletal topology unit; Ultrasound image data is processed using skeletal analysis methods to establish a mapping relationship between the ultrasound coordinate system and the anatomical coordinate system, and a topology map containing the mapping relationship is obtained; Region Focused Analysis Unit: It is equipped with a progressive analysis architecture for progressively narrowing and focusing processing, and the progressive analysis architecture is equipped with a feedback mechanism to realize information interaction between each layer; The progressive analysis architecture includes: Primary detection layer: The topology map is analyzed by recognition methods to obtain data on the continuity of the bone cortex, the strong echo data of bone fragments, and the fragmentation edge feature data of the fracture, thereby obtaining a fracture sign dataset. Based on the location and mapping relationship of the continuity interruption, the strong echo area of ​​bone fragments, and the fragmentation edge feature of the fracture, the preliminary fracture area is obtained from the ultrasound image data. Intermediate classification layer: The intensity of periosteal reaction in the initial fracture area and the extent of soft tissue injury are quantified by a quantification method to obtain a soft tissue reaction dataset. The soft tissue injury extent and the initial fracture area are combined to obtain the labeled area. Advanced localization layer: The labeled area is analyzed using a sub-pixel level method. The sub-pixel level method includes analyzing the spatial arrangement structure of bone trabeculae and microfracture lines, calculating the angular deviation value of the normal vector of the bone trabeculae fracture surface, labeling the positions of bone trabeculae whose angular deviation value exceeds the deviation threshold, and obtaining the micro-damage area and micro-damage dataset. The identification data is obtained by combining topology maps, fracture sign datasets, soft tissue reaction datasets, and micro-injury datasets.

2. The ultrasound-assisted system for automatic identification and diagnosis of emergency fractures according to claim 1, characterized in that: The skeletal analysis method includes: The skeletal topology unit stores a full skeletal anatomical segmentation model library and a skeletal anatomical reference database that enable image data structure parsing; Geometric feature vectors of the skeletal region in ultrasound image data are extracted using a whole-skeletal anatomical segmentation model, and skeletal anatomical landmarks in ultrasound image data are labeled. The skeletal anatomy reference database includes a set of reference anatomical feature points, and skeletal anatomical landmarks are mapped to reference anatomical feature points in the set of reference anatomical feature points. Based on the mapping relationship, the corresponding skeletal anatomy reference data is obtained from the skeletal anatomy reference database. A topological graph is constructed based on geometric feature vectors, mapping relationships, and ultrasound image data.

3. The ultrasound-assisted system for automatic identification and diagnosis of emergency fractures according to claim 1, characterized in that: The feedback mechanism includes: The micro-damage areas detected by the advanced localization layer are fed back to the primary detection layer. The primary detection layer performs a second scan of the micro-damage areas at the sub-pixel level to obtain a sub-pixel-level fracture sign dataset of the micro-damage areas. The sub-pixel-level fracture sign dataset replaces the fracture sign dataset. The size of the initial fracture region in the primary detection layer is adjusted based on the quantified periosteal elevation height in the intermediate classification layer.

4. The ultrasound-assisted system for automatic identification and diagnosis of emergency fractures according to claim 1, characterized in that: The diagnostic decision module includes processing the identified data using diagnostic methods to obtain a diagnostic report; The diagnostic method includes: linking the AO classification database with the trauma scoring unit, identifying data input into the AO classification database and the trauma scoring unit and linking them, and generating a diagnostic report containing fracture type and displacement risk level.

5. The ultrasound-assisted system for automatic identification and diagnosis of emergency fractures according to claim 1, characterized in that: The identification method includes performing cortical continuity analysis on the topology map, scanning the geometric features of the cortical surface, detecting linear fracture zones with lengths exceeding a length threshold, identifying the fragmented edge features of comminuted fractures based on curvature mutation points, and obtaining cortical continuity data, strong echo data of bone fragments, and fragmented edge feature data of fractures. The establishment of the mapping relationship between the ultrasound coordinate system and the anatomical coordinate system includes: A three-dimensional view of the overlapping bone structure in ultrasound image data is unfolded using a surface reconstruction algorithm, and an ultrasound coordinate system is established. Based on skeletal analysis methods, anatomical reference data correlated with ultrasound image data is obtained. A surface reconstruction algorithm is used to unfold a three-dimensional view of the anatomical reference data and establish an anatomical coordinate system. The mapping relationship between the ultrasound coordinate system and the anatomical coordinate system is established based on skeletal anatomical landmarks and reference anatomical feature points.

6. The ultrasound-assisted system for automatic identification and diagnosis of emergency fractures according to claim 2, characterized in that: The geometric feature vector of the skeletal region includes the radius of curvature R and the cortical thickness gradient ΔT, and the anatomical landmarks include the epiphyseal plate growth line L and the articular surface boundary B.

7. The ultrasound-assisted system for automatic identification and diagnosis of emergency fractures according to claim 1, characterized in that: The quantization method includes: Measure the continuous value ΔH of the periosteal elevation height, quantify the hematoma area A, and mark the area where the soft tissue swelling is located; Generate a tissue response feature vector with continuous values ​​ΔH and hematoma area A. When ΔH is greater than the continuous value threshold, the periosteal region where ΔH is greater than the continuous value threshold is marked. The extent of soft tissue injury is determined by combining the periosteal region and the area of ​​soft tissue swelling. The extent of soft tissue injury, combined with Doppler blood flow signals, allows for the identification of indirect periosteal signs and reactions. A soft tissue response dataset is obtained by combining the types of indirect periosteal signs and the extent of soft tissue injury.

8. The ultrasound-assisted system for automatic identification and diagnosis of emergency fractures according to claim 1, characterized in that: The sub-pixel-level analysis includes: Analyze the spatial arrangement of trabecular bone structure, and detect and calculate the deviation of the normal vector angle of microfracture lines and fracture surfaces; Output micro-damage maps, establish spatial association between the coordinate set of the micro-damage core area and anatomical landmarks, and obtain micro-damage datasets by combining micro-fracture lines, fracture surface normal vector angle differences and spatial associations.

Images