Orthopedic trauma image recognition method, device, equipment, medium and computer program product

Abstract

The invention relates to the technical field of image processing, in particular to an orthopedic trauma image recognition method and device, equipment, a medium and a computer program product, and the method comprises the steps: carrying out the denoising of an initial bone image through a first filtering function, obtaining a first reference image, determining a second filtering parameter based on image processing, and carrying out the denoising again to generate a target bone image, n successive bone trauma features are integrated to obtain an image recognition result of the target object, and it can be known that the first filtering function of specific parameters is matched according to labels of different parts, and noise can be removed more accurately and details can be reserved according to noise distribution features of bone images of different parts; the filtering parameters are obtained according to the characteristics of each image, the filtering effect can be flexibly adjusted, noise suppression and feature retention are better balanced, bone features are effectively enhanced, wound details are highlighted, bone wound feature extraction is more accurate, comprehensive recognition of the recovery condition of a target object is facilitated, and the orthopedic wound image recognition accuracy is improved.

Description

Technical Field

[0001] The present invention generally relates to the technical field of image processing, and in particular, to an orthopedic trauma image recognition method, device, equipment, medium and computer program product. Background Art

[0002] Due to factors such as limited equipment accuracy and signal interference in existing image acquisition devices, various noises exist in the collected bone images, making key information such as fracture lines and bone injuries become blurred, and it is impossible to accurately extract the fine structures and texture information of bones. This not only affects the data accuracy of researchers' studies on bone trauma, resulting in possible deviations in research conclusions; in the teaching scenario, it also interferes with medical students' learning and cognition of normal and trauma bone images, hindering the progress of their professional skill cultivation; in the field of medical device research and development, it also misleads the judgment of R & D personnel on the improvement direction of image processing technology, making the R & D resources unable to be accurately invested. Therefore, accurate trauma image recognition plays an important role in multiple related fields.

[0003] In the task of orthopedic trauma image recognition, there are many limitations in the current image recognition technology. For example, while traditional mean filtering removes noise, it will blur the image indiscriminately, smoothing both the fine structures of bones and the noise together, resulting in a lack of accurate data basis for subsequent tasks such as bone mechanical property analysis and structural modeling based on images; median filtering, although having a certain role in protecting edges, is difficult to effectively remove noise for complex mixed noises, especially when the noise density is high, and in areas with rich textures, it is prone to problems of over-smoothing or texture distortion, causing the loss of the true information of bone textures and affecting the analysis of the structural integrity of bones; and various denoising techniques rely on the selection of parameters. However, the orthopedic image features are complex and variable, and it is difficult to find uniformly applicable parameters for images under different parts and different imaging conditions, resulting in unstable denoising effects. When extracting key features of bones, it is easy to have feature omissions or misjudgments, and reliable feature data cannot be provided for in-depth research on bones.

[0004] Therefore, in the task of orthopedic trauma image recognition, how to improve the recognition accuracy has become an urgent problem to be solved. Summary of the Invention

[0005] In view of the above-mentioned defects or deficiencies in the prior art, it is desired to provide an orthopedic trauma image recognition method, device, equipment, medium and computer program product.

[0006] In a first aspect, an embodiment of the present application provides an orthopedic trauma image recognition method, and the orthopedic trauma image recognition method includes the following steps: S1. Obtain N initial bone images of a target object, part labels corresponding to the N initial bone images, a first filtering function including first filtering parameters corresponding to the part labels, and the acquisition time corresponding to each initial bone image, where N is an integer greater than 1.

[0007] S2. Perform image processing on any one of the initial bone images to obtain a target bone image corresponding to the current initial bone image.

[0008] S3. Extract bone trauma features corresponding to the target bone image.

[0009] S4. According to the bone trauma features corresponding to the N target bone images and the acquisition time corresponding to each initial bone image, obtain an image recognition result corresponding to the target object, where the image recognition result includes the degree of recovery and the recovery trend.

[0010] In a second aspect, an orthopedic trauma image recognition device is provided in an embodiment of the present application. The orthopedic trauma image recognition device includes: A data acquisition module, configured to obtain N initial bone images of a target object, part labels corresponding to the N initial bone images, a first filtering function including first filtering parameters corresponding to the part labels, and the acquisition time corresponding to each initial bone image, where N is an integer greater than 1.

[0011] An image processing module, configured to perform image processing on any one of the initial bone images to obtain a target bone image corresponding to the current initial bone image.

[0012] A feature extraction module, configured to extract bone trauma features corresponding to the target bone image.

[0013] An image recognition module, configured to obtain an image recognition result corresponding to the target object according to the bone trauma features corresponding to the N target bone images and the acquisition time corresponding to each initial bone image, where the image recognition result includes the degree of recovery and the recovery trend.

[0014] In a third aspect, an electronic device is provided in an embodiment of the present application, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method described in the embodiment of the present application is implemented.

[0015] In a fourth aspect, a computer-readable storage medium is provided in an embodiment of the present application, on which a computer program is stored. When the program is executed by a processor, the method described in the embodiment of the present application is implemented.

[0016] In a fifth aspect, a computer program product is provided in an embodiment of the present application, including a computer program. When the computer program is executed by a processor, the method described in the embodiment of the present application is implemented.

[0017] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Other features, objectives, and advantages of this application will become more apparent by reading the detailed description of the non-limiting embodiments with reference to the following drawings: Figure 1 FIG. shows the implementation environment architecture diagram of the orthopedic trauma image recognition method provided by an embodiment of this application; Figure 2 FIG. shows the schematic flowchart of the orthopedic trauma image recognition method provided by an embodiment of this application; Figure 3 FIG. shows the schematic flowchart of step S1 in the orthopedic trauma image recognition method provided by an embodiment of this application; Figure 4 FIG. shows the schematic flowchart of step S2 in the orthopedic trauma image recognition method provided by an embodiment of this application; Figure 5 FIG. shows the schematic flowchart of step S23 in the orthopedic trauma image recognition method provided by an embodiment of this application; Figure 6 FIG. shows the schematic flowchart of step S233 in the orthopedic trauma image recognition method provided by an embodiment of this application; Figure 7 FIG. shows the schematic flowchart of step S4 in the orthopedic trauma image recognition method provided by an embodiment of this application; Figure 8 FIG. shows the exemplary structural block diagram of the orthopedic trauma image recognition device provided by an embodiment of this application; Figure 9 FIG. shows the schematic structural diagram of a computer system of an electronic device or server suitable for implementing the embodiments of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The following further elaborates on this application in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only for explaining the relevant invention and not for limiting the invention. Additionally, it should be noted that for the sake of description, only the parts related to the invention are shown in the drawings.

[0020] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will elaborate on this application in detail with reference to the drawings and embodiments.

[0021] For the specific implementation environment of the orthopedic trauma image recognition method proposed in this application, refer to Figure 1 .Figure 1 The figure shows the implementation environment architecture diagram of the orthopedic trauma image recognition method provided by the embodiments of the present application.

[0022] As Figure 1 shown, the implementation environment architecture includes: a terminal device 101 and a server 102.

[0023] The terminal device 101 can be a desktop computer, a laptop computer, a smart phone, a tablet computer, an e-book reader, smart glasses, a smart watch, etc., but is not limited thereto.

[0024] The server 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. Among them, the server 102 is used to provide replacement entries to the terminal device 101 and execute the recognition strategy for target entry anomalies.

[0025] The terminal device 101 and the server 102 are directly or indirectly connected through a wired or wireless communication method. Optionally, the above-mentioned wireless network or wired network uses standard communication technologies and / or protocols. The network is usually the Internet, or can be any network, including but not limited to any combination of a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network, or a virtual private network.

[0026] The orthopedic trauma image recognition method proposed in the present application can be implemented by an orthopedic trauma image recognition device, and the orthopedic trauma image recognition device can be installed on the terminal device or the server.

[0027] To further illustrate the technical solutions provided by the embodiments of the present application, the following will be described in detail in conjunction with the accompanying drawings and specific implementation manners. Although the embodiments of the present application provide the method operation instruction steps shown in the following embodiments or drawings, based on routine or non-creative labor, the method may include more or fewer operation instruction steps. In steps where there is no necessary causal relationship logically, the execution order of these steps is not limited to the execution order provided by the embodiments of the present application. When the method is actually processed or the device is executed, it can be executed in the order shown in the embodiments or drawings or executed in parallel.

[0028] It should be noted that the acquisition or use of the data in the embodiments of the present application requires the consent of the user. Relevant data can only be obtained after the user authorizes and permits it, and the acquisition or use of the data complies with relevant laws and regulations.

[0029] Please refer to Figure 2 , Figure 2 which shows a schematic flowchart of an orthopedic trauma image recognition method provided by an embodiment of the present application. As Figure 2 shown, the orthopedic trauma image recognition method includes: S1, obtaining N initial bone images of a target object, part labels corresponding to the N initial bone images, a first filtering function including first filtering parameters corresponding to the part labels, and the acquisition time corresponding to each initial bone image, where N is an integer greater than 1.

[0030] S2, performing image processing on any one of the initial bone images to obtain a target bone image corresponding to the current initial bone image.

[0031] S3, extracting bone trauma features corresponding to the target bone image.

[0032] S4, obtaining an image recognition result corresponding to the target object according to the bone trauma features corresponding to the N target bone images and the acquisition time corresponding to each initial bone image, where the image recognition result includes the degree of recovery and the recovery trend.

[0033] Among them, the target object refers to the user object who needs to perform orthopedic trauma image recognition. The initial bone images of the target object are collected at different time points through devices such as X-rays and CTs, which may contain various noise interferences that affect the accuracy of the recognition result.

[0034] The part label marks the corresponding body part for each initial bone image, such as body parts like the head, arm, leg, trunk, etc. Due to factors such as the physiological structure and tissue composition of different parts of the human body, when collecting trauma images for different parts, the noise distribution characteristics on the images are different. For example, the diaphysis is mainly composed of dense cortical bone, with a relatively simple and uniform structure. In imaging examinations, the noise distribution is often relatively uniform, and the overall noise level is relatively low. The metaphysis has more cancellous bone, with a complex trabecular structure. Noise is more likely to appear in the trabecular spaces, and the noise distribution shows characteristics related to the trabecular structure, presenting as scattered dots or patches, and the density will change with the density of the trabeculae. In the skull, air-containing cavities such as the paranasal sinuses are prone to generating artifacts and relatively high levels of noise in imaging examinations due to the large density difference between gas and surrounding bone and soft tissue, and the noise distribution is usually irregularly distributed around the edge of the air cavity. In parts rich in soft tissues such as muscles and fat, the noise distribution is related to the uniformity of the soft tissues, blood vessel distribution, etc. Generally, the noise is relatively uniform, but there may be slightly higher noise around blood vessels.

[0035] Therefore, in this embodiment, the part label corresponding to the initial bone image and the first filtering function including the first filtering parameter corresponding to the part label are obtained, which facilitates subsequent different filtering processes according to the bone characteristics of different parts, thereby improving the noise filtering effect and further enhancing the accuracy of subsequent image recognition.

[0036] The first filtering function can perform specific mathematical operations on each pixel and its neighboring pixels in the initial bone image, and according to the setting of the first filtering parameter, specifically remove noise and retain the important features of the image.

[0037] Perform image processing on any initial bone image, remove the noise in the current initial bone image, and obtain the target bone image corresponding to the current initial bone image, improve the image quality, and further enhance the accuracy of subsequent recognition of the recovery of bone trauma.

[0038] Extract the features related to bone trauma in the target bone image, for example, information such as the position, length, and shape of the fracture line, the number, size, and position of bone fragments, and the area of bone density change, as the feature basis for subsequent recognition of the recovery of bone trauma.

[0039] Comprehensively consider the bone trauma features extracted from N target bone images and the acquisition time corresponding to each initial bone image. Combining the acquisition time with the bone trauma features can analyze the change of bone trauma over time. For example, if the fracture line gradually narrows and the bone fragments gradually decrease over time, it indicates that the bone is in a recovery state. From this, a quantitative assessment of the recovery degree and a qualitative description of the overall recovery situation, such as good recovery or slow recovery, can be obtained and output as the image recognition result.

[0040] As described above, perform image processing on the initial bone image to generate a target bone image, extract bone trauma features from the target bone image, and comprehensively consider the trauma features of N target bone images and the corresponding acquisition time to obtain the image recognition result of the target object. It can be seen that by matching specific first filtering functions and parameters according to different part labels, the noise distribution characteristics of bone images of different parts can be more accurately removed, details can be retained, the extraction of bone trauma features can be more accurate, which helps to more accurately recognize the recovery situation of the target object, thereby improving the recognition accuracy of orthopedic trauma images.

[0041] Please refer to Figure 3 , Figure 3 which shows a schematic flowchart of step S1 provided by an embodiment of the present application. S1 includes the following steps: S11, obtain a plurality of bone image samples corresponding to the part label, the noise image sample corresponding to each bone image sample, and the first preset filtering model including the preset filtering parameter.

[0042] S12. Filter the noisy image sample according to the first preset filtering model to obtain the reference image sample corresponding to the noisy image sample.

[0043] S13. Obtain the image error according to the reference image sample corresponding to the noisy image sample and the bone image sample, and construct an objective function according to the image error.

[0044] S14. Update the preset filtering parameters by the gradient descent method until the objective function converges, and determine the updated preset filtering parameters corresponding to the convergence of the objective function as the first filtering parameters corresponding to the part label.

[0045] S15. Construct a first filtering function according to the first filtering parameters.

[0046] Among them, the bone image sample represents the image of the corresponding part in the normal or ideal state, while the noisy image sample is the image situation with noise in the actual acquisition process. By methods such as computer reconstruction of the noisy image sample, the original data obtained from different angles or levels is reconstructed into an intuitive two-dimensional or three-dimensional image, the noise interference is removed, the resolution and clarity of the image are enhanced, the detailed information is highlighted, and the bone image sample close to the ideal state is obtained as the basis for obtaining the first filtering parameters.

[0047] Since computer image reconstruction requires a large amount of computing resources and time, which to a certain extent limits its application in scenarios with high real-time requirements and a large number of images. Therefore, in this embodiment, the computer reconstructed image is used as the sample basis, and the first filtering parameters corresponding to each part label are corrected through the samples to improve the denoising effect corresponding to the first filtering parameters. In the subsequent analysis, the initial bone image is preliminarily denoised based on the first filtering parameters, the denoising efficiency is improved on the basis of ensuring the denoising effect, and the denoising cost is reduced.

[0048] The first preset filtering model containing the preset filtering parameters is used to filter the noisy image sample to obtain the corresponding reference image sample, so as to generate a reference image close to the ideal bone image. And the filtered reference image sample is compared with the original bone image sample, and the image error between the two is calculated, which reflects the degree of difference between the reference image and the ideal bone image.

[0049] An objective function is constructed based on the image error. By calculating the gradient of the objective function with respect to the parameters, the parameters are gradually adjusted along the opposite direction of the gradient so that the value of the objective function continuously decreases until the objective function converges, that is, the value of the objective function no longer changes significantly. At this time, the updated preset filtering parameters corresponding to it are determined as the first filtering parameters corresponding to the part label, thus ensuring that the filtering parameters can be optimized according to the actual sample data to achieve the best denoising effect.

[0050] As described above, by analyzing the bone image samples and noise image samples of specific parts and optimizing the filtering parameters based on the objective function, the finally obtained first filtering function can be highly adaptive to the noise characteristics of this part, can more effectively remove the noise unique to different parts, while retaining the detailed information of the bones to the greatest extent, improving the accuracy and pertinence of filtering, and enabling the entire orthopedic trauma image recognition method to have stronger versatility and stability when facing diverse actual situations.

[0051] Please refer to Figure 4 , Figure 4 shows a schematic flowchart of step S2 provided by an embodiment of the present application. S2 includes the following steps: S21. For any initial bone image, filter the current initial bone image according to the first filtering function including the first filtering parameter to obtain the first reference image corresponding to the current initial bone image.

[0052] S22. Perform a preset process on the first reference image to obtain a second reference image, where the preset process includes binarization, erosion, and dilation.

[0053] S23. Obtain the second filtering parameter according to the first reference image and the second reference image, where the second filtering parameter corresponds to the second filtering model.

[0054] S24. Filter the first reference image according to the second filtering model to obtain the target bone image corresponding to the current initial bone image.

[0055] Among them, the first filtering function can perform specific mathematical operations on each pixel and its neighboring pixels in the initial bone image. According to the setting of the first filtering parameter, it can specifically remove noise and retain the important features of the image to obtain the first reference image corresponding to the current initial bone image. Compared with the initial image, the noise of the first reference image is initially suppressed, and the image quality is improved, laying a foundation for subsequent processing.

[0056] The binarization process can convert the first reference image into a binary image, that is, there are only two pixel values in the image, usually 0 (black) and 255 (white). By setting a gray value threshold, the pixels with gray values greater than the gray value threshold in the image are set to one value, and those less than the threshold are set to another value, which helps to highlight the main structure in the first reference image, initially separate the bones from the background such as surrounding tissues, and facilitate subsequent processing. For example, in a bone image, the bone part with a higher gray value can be set to white, and other backgrounds can be set to black.

[0057] The erosion process can perform an erosion operation on the binarized image, causing the boundaries of the objects in the image to contract inward, removing some isolated noise points or small burrs, and making the outline of the bone clearer and more regular.

[0058] The dilation process is the opposite of the erosion process. The dilation process will expand the boundaries of the objects in the image, fill the noise gaps, and obtain a second reference image.

[0059] According to the comparative analysis between the reference images before and after the preset processing, it can provide guidance on the denoising degree for the secondary filtering, obtain appropriate second filtering parameters, and further filter and denoise the first reference image based on the second filtering model to obtain a target bone image with higher quality.

[0060] In a specific embodiment, the first filtering parameter is (a 0 , a 1 , a 2 , a 3 , a 4 , a 5 ). Correspondingly, the first filtering model including the first filtering parameter is f(x, y)=a 0 + a 1 ×x + a 2 ×y + a 3 ×x 2 + a 4 ×x×y + a 5 ×y 2 . The pixel value at the position (x, y) in the initial bone image is h(x, y). Then, after filtering the current initial bone image according to the first filtering function including the first filtering parameter, the pixel value at the position (x, y) in the first reference image corresponding to the current initial bone image is r(x, y)= h(x, y)×f(x, y).

[0061] The second filtering parameter is (Z, Q). According to Q, a convolution kernel of size (L, L) is obtained, where L=(Q - 1) / 2. Correspondingly, the second filtering model including the second filtering parameter is g(x, y)=(1 / (2×π×Z 2 ))×e^(-((x 2 +y 2 ) / (2×Z 2 ))), representing the coefficient of the convolution kernel at the position (x, y). Then, filtering the first reference image according to the second filtering model, the pixel value at the position (x, y) in the target bone image corresponding to the current initial bone image is: c(x, y)=Σ L u=-L (Σ L v=-L(r(x+u, y+v) × g(u, v))). Where r(x+u, y+v) is the value of the pixel at position (x+u, y+v) in the first reference image.

[0062] As described above, each initial bone image is denoised using the first filtering function with the first filtering parameter to obtain the first reference image. Then, the first reference image is sequentially subjected to preset image processing operations of binarization, erosion, and dilation to obtain the second reference image, so as to highlight the bone structure, remove noise, and fill gaps. Based on the first reference image and the second reference image, the second filtering parameter is determined, and the first reference image is denoised again to generate the target bone image, further balancing noise removal and detail retention. The bone trauma features are extracted from the target bone image, and the trauma features of N target bone images and the corresponding acquisition times are integrated to obtain the image recognition result of the target object. It can be seen that by matching a specific first filtering function and parameter according to different part labels, the noise distribution characteristics of bone images of different parts can be more accurately removed, details can be retained, and a high-quality image basis for subsequent processing can be provided; by combining the preset processing steps with the second filtering parameter to obtain the second filtering model, the filtering effect can be flexibly adjusted according to the characteristics of each image, better balancing noise suppression and feature retention, effectively enhancing the bone features, highlighting the trauma details, making the extraction of bone trauma features more accurate, and contributing to a comprehensive and dynamic recognition of the recovery situation of the target object, thereby improving the accuracy of orthopedic trauma image recognition.

[0063] Please refer to Figure 5 , Figure 5 FIG. shows a schematic flowchart of step S23 provided by an embodiment of the present application. S23 includes the following steps: S231, perform edge detection on the first reference image to obtain the first edge image.

[0064] S232, perform edge detection on the second reference image to obtain the second edge image.

[0065] S233, according to the first edge image and the second edge image, obtain the edge difference degree between the first reference image and the second reference image.

[0066] S234, according to the edge difference degree and a preset difference degree threshold, obtain the second filtering parameter.

[0067] Among them, those skilled in the art know that any edge detection method in the prior art falls within the protection scope of the present invention and will not be elaborated here. For example, edge detection methods such as Sobel operator, Laplacian operator, and Canny operator.

[0068] The preset difference degree threshold is a preset standard used to determine whether the edge difference between the first reference image and the second reference image is within an acceptable range. According to the comparison result of the edge difference degree and the threshold, the second filtering parameter is determined.

[0069] As described above, according to the edge difference situation between the first reference image and the second reference image, the parameters of the second filtering model are adaptively determined. While removing noise, important features of the image are retained as much as possible, improving the accuracy of subsequent bone trauma feature extraction, and ultimately enhancing the accuracy of the entire orthopedic trauma image recognition.

[0070] Please refer to Figure 6 , Figure 6 FIG. shows a schematic flowchart of step S233 provided by an embodiment of the present application. S233 includes the following steps: S2331, match all the first edge pixel points and all the second edge pixel points to obtain a number of edge pixel point pairs. Among them, each edge pixel point pair includes a first edge pixel point and a matching second edge pixel point.

[0071] S2332, for any edge pixel point pair, determine the distance between the first edge pixel point and the second edge pixel point in the current edge pixel point pair as the edge difference value of the current edge pixel point pair.

[0072] S2333, determine the variance of the edge difference values of all edge pixel point pairs as the edge difference degree between the first reference image and the second reference image.

[0073] Among them, all the first edge pixel points and all the second edge pixel points can be matched based on the features of the pixel points. For example, local feature descriptors of each first edge pixel point and second edge pixel point are extracted through the Scale-Invariant Feature Transform (SIFT) algorithm or the Speeded Up Robust Feature (SURF) algorithm. For the feature descriptor of each first edge pixel point, the most matching feature descriptor is searched in the set of feature descriptors of the second edge pixel points to obtain the corresponding edge pixel point pair. Common matching methods include nearest neighbor matching, k-nearest neighbor matching, etc.

[0074] Taking the Euclidean distance as an example for the distance calculation method, it intuitively reflects the difference in the spatial positions of two pixel points in the image. The larger the distance, the greater the position difference between the two corresponding edge pixel points. And the variance of the edge difference values of all edge pixel point pairs is used to reflect the edge difference degree between the first reference image and the second reference image.

[0075] As described above, by matching edge pixel points, calculating the difference value, and statistically analyzing the variance to determine the edge difference degree, starting from the microscopic pixel level, the information of the entire edge image is comprehensively considered, which can more comprehensively and accurately reflect the edge difference between two reference images, providing a quantitative and reliable basis for determining a suitable second filtering model in the subsequent process.

[0076] In a specific embodiment, the second filtering parameter includes a first filtering coefficient and a second filtering coefficient, and S234 includes the following steps: S2341: Compare the edge difference degree with a preset difference degree threshold, and determine the minimum value between the edge difference degree and the preset difference degree threshold as the first filtering coefficient.

[0077] S2342: According to the first filtering coefficient and a preset parameter mapping model, obtain the second filtering coefficient corresponding to the first filtering coefficient, where the parameter mapping model is used to represent the mapping relationship between the first filtering coefficient and the second filtering coefficient.

[0078] Among them, if the edge difference degree is less than the preset threshold, it indicates that the edge difference between the two reference images is relatively small. At this time, using the edge difference degree as the first filtering coefficient, a relatively small filtering intensity can be adopted to avoid the loss of useful information in the image caused by excessive filtering. If the edge difference degree is greater than the preset threshold, it indicates that the edge difference is large, and there may be many interference factors such as noise. At this time, selecting the preset difference degree threshold as the first filtering coefficient can limit the filtering intensity from being too large, ensuring that while removing noise, key information such as the edges of the image is retained as much as possible. The specific value of the preset difference degree threshold can be set by the implementer according to the actual situation. For example, the preset difference degree threshold can be set to 2.

[0079] The preset parameter mapping model is a pre-established model that describes the corresponding relationship between the first filtering coefficient and the second filtering coefficient, and can be set by the implementer according to the actual situation. For example, the preset parameter mapping model can be Q = r×Z + 1, where Z refers to the first filtering coefficient, r refers to the preset weight, and Q refers to the second filtering coefficient. The specific value of r can be set by the implementer according to the actual situation. For example, r can be set to 6.

[0080] In a specific embodiment, the first filtering coefficient can be used to represent the standard deviation size, and the second filtering coefficient can be used to represent the Gaussian kernel size.

[0081] As described above, adaptively determining the first filtering coefficient and the second filtering coefficient of the second filtering model according to the edge difference between the first reference image and the second reference image can adjust the filtering intensity according to the actual situation of the image, and while removing noise, retain the edge and detail information of the image as much as possible, laying a foundation for obtaining a high-quality target bone image and accurately extracting bone trauma features subsequently.

[0082] In another embodiment, S2 includes the following steps: S201, for any initial bone image, perform a preset process on the current initial bone image to obtain a third reference image corresponding to the current initial bone image, where the preset process includes binarization, erosion, and dilation; S202, according to the current initial bone image and the corresponding third reference image, obtain a third filtering parameter, where the third filtering parameter corresponds to the second filtering model.

[0083] S203, filter the current initial bone image according to the first filtering model, the first preset weight corresponding to the first filtering model, the second filtering model, and the second preset weight corresponding to the second filtering model, to obtain a target bone image corresponding to the current initial bone image.

[0084] Among them, the method for obtaining the third filtering parameter can refer to S231 - S234, that is, refer to the method for obtaining the second filtering parameter.

[0085] Then, the current initial bone image can be filtered according to the first filtering model and the second filtering model at the same time, and the corresponding weights are respectively configured, so as to respectively configure the importance and influence degree of the two filtering models during filtering, improve the filtering effect, and further improve the recognition accuracy of subsequent orthopedic images. And compared with the method of steps S21 - S24, the parallel filtering method reduces the resource occupancy rate and improves the processing efficiency of the initial bone image.

[0086] In a specific embodiment, the obtained first filtering parameter is (a 0 , a 1 , a 2 , a 3 , a 4 , a 5 ), the first preset weight is q 1 , correspondingly, the first filtering model is f(x, y) = a 0 + a 1 × x + a 2 × y + a 3 × x 2 + a 4 × x × y + a 5 × y 2。The third filtering parameter is (T, P). A convolution kernel of size (B, B) is obtained according to P, where B = (P - 1) / 2. Correspondingly, the second filtering model is g(x, y) = (1 / (2×π×T 2 ))×e^(-((x 2 +y 2 ) / (2×T 2 ))), which represents the coefficient of the convolution kernel at the position (x, y). The second preset weight is q 2 。

[0087] The pixel value at the position (x, y) in the initial bone image is h(x, y). Then, the initial bone image is filtered according to the first filtering model, the first preset weight corresponding to the first filtering model, the second filtering model, and the second preset weight corresponding to the second filtering model, and the pixel value at the position (x, y) in the target bone image is obtained as follows: c(x, y) = q 1 ×(h(x, y)×f(x, y)) + q 2 ×(Σ B u=-B (Σ B v=-B (h(x + u, y + v)×g(u, v)))). Where h(x + u, y + v) is the pixel value at the position (x + u, y + v) in the initial bone image.

[0088] In a specific embodiment, several target bone images obtained through the above steps S21 - S24 in the historical operation can be used as key image samples. Referring to steps S11 - S14, the method for obtaining the first weight and the second weight includes the following steps: S204, obtaining the first initial weight corresponding to the first filtering model and the second initial weight corresponding to the second filtering model; S205, for any key image sample, filtering the initial bone image corresponding to the current key image sample according to the first filtering model, the first initial weight corresponding to the first filtering model, the second filtering model, and the second initial weight corresponding to the second filtering model, and obtaining the target image sample corresponding to the current key image sample; S206, obtaining the image error sample according to the current key image sample and the corresponding target image sample, and constructing a key function according to the image error sample; S207, updating the first initial weight and the second initial weight through the gradient descent method until the key function converges, and determining the updated first initial weight and the second initial weight corresponding to the convergence of the key function as the first initial weight and the second initial weight.

[0089] Among them, the key image samples represent images with good denoising effects. By comparing the target image samples filtered according to the first filtering model, the first initial weight, the second filtering model, and the second initial weight with the corresponding key image samples, the image error between the two can be calculated, which can reflect the quality of the filtering effect according to the first filtering model, the first initial weight, the second filtering model, and the second initial weight.

[0090] Therefore, a key function is constructed based on the image error samples. By calculating the gradient of the key function with respect to the weights and gradually adjusting the weights along the opposite direction of the gradient, the value of the key function can be continuously reduced until the key function converges, that is, the value of the key function no longer changes significantly. At this time, the updated first initial weight and second initial weight are determined as the first initial weight and the second initial weight.

[0091] Therefore, the optimized first initial weight and second initial weight can be directly used to directly filter and denoise the initial bone image in combination with the first filtering model and the second filtering model, thereby ensuring a good denoising effect while reducing the resource occupancy rate and improving the processing efficiency of the initial bone image.

[0092] Please refer to Figure 7 , Figure 7 FIG. shows a schematic flow chart of step S4 provided by an embodiment of the present application. S4 includes the following steps: S41, obtaining a preset large language model and a preset task description text.

[0093] S42, sorting the bone trauma features corresponding to N target bone images in the order of acquisition time from earliest to latest to obtain a bone trauma feature sequence.

[0094] S43, sorting the N acquisition times in the order of acquisition time from earliest to latest to obtain an acquisition time sequence.

[0095] S44, inputting the bone trauma feature sequence, the acquisition time sequence, and the preset task description text into the preset large language model to obtain an image recognition result corresponding to the target object.

[0096] Among them, the large language model (LLM) is an artificial intelligence model based on deep learning. After being trained with a large amount of text data, it has powerful language understanding and generation capabilities. The preset large language model in this embodiment is a model that has been specifically screened or trained and is suitable for processing tasks related to orthopedic image analysis.

[0097] The task description text is a pre-set text used to clearly inform the large language model of the task to be completed. For example, "Based on the provided sequence of bone trauma characteristics and the corresponding acquisition time sequence, analyze and give the recovery degree and recovery trend of the target object."

[0098] Input the sorted sequence of bone trauma characteristics, the acquisition time sequence, and the pre-set task description text into the pre-set large language model. Based on its language understanding and analysis capabilities, the large language model deeply processes the input information, mines the variation law of bone trauma characteristics in the time sequence, and thus infers the recovery degree and recovery trend of the target object. For example, the recovery degree may be presented in the form of a percentage, such as "The recovery degree is 60%". The recovery trend may be described as "The recovery speed is accelerating" or "The recovery speed tends to be stable", etc.

[0099] In another specific embodiment, the pre-set large language model can be a large prediction model fine-tuned based on Chat GPT-4. Input the sorted sequence of bone trauma characteristics, the acquisition time sequence, and the pre-set task description text into the large prediction model. The large prediction model can encode the acquisition time sequence as a natural language description of time series features, such as encoding it as "Area of the fractured arm region: 120 mm² on the 1st day to 80 mm² on the 30th day", and then, based on its language understanding and analysis capabilities, deeply process the input information to generate an interpretable report that meets the requirements of the implementer, including the prediction of the recovery degree and recovery trend of the target object.

[0100] It should be noted that although the operations of the method of the present invention are described in a specific order in the drawings, this does not require or imply that these operations must be performed in that specific order, or that all the shown operations must be performed to achieve the desired result.

[0101] Please refer to Figure 8 , Figure 8 which shows an exemplary structural block diagram of an orthopedic trauma image recognition device 800 provided by an embodiment of the present application. As Figure 8 shown, the orthopedic trauma image recognition device 800 includes: A data acquisition module 81, configured to acquire N initial bone images of a target object, part labels corresponding to the N initial bone images, a first filtering function including a first filtering parameter corresponding to the part label, and the acquisition time corresponding to each initial bone image, where N is an integer greater than 1.

[0102] An image processing module 82, configured to perform image processing on any one of the initial bone images to obtain a target bone image corresponding to the current initial bone image.

[0103] A feature extraction module 83, configured to extract bone trauma features corresponding to a target bone image.

[0104] An image recognition module 84, configured to obtain an image recognition result corresponding to a target object according to the bone trauma features corresponding to N target bone images and the acquisition time corresponding to each initial bone image, where the image recognition result includes a recovery degree and a recovery trend.

[0105] In a specific embodiment, the data acquisition module 81 includes: A data acquisition sub-module, configured to obtain a plurality of bone image samples corresponding to a part label, a noise image sample corresponding to each bone image sample, and a first preset filtering model including preset filtering parameters.

[0106] A sample filtering sub-module, configured to filter the noise image sample according to the first preset filtering model to obtain a reference image sample corresponding to the noise image sample.

[0107] A target function construction sub-module, configured to obtain an image error according to the reference image sample corresponding to the noise image sample and the bone image sample, and construct a target function according to the image error.

[0108] A parameter update sub-module, configured to update the preset filtering parameters by the gradient descent method until the target function converges, and determine the updated preset filtering parameters corresponding to the convergence of the target function as the first filtering parameters corresponding to the part label.

[0109] A first filtering function construction sub-module, configured to construct a first filtering function according to the first filtering parameters.

[0110] In a specific embodiment, the image processing module 82 includes: A first filtering sub-module, configured to filter any current initial bone image according to the first filtering function including the first filtering parameters to obtain a first reference image corresponding to the current initial bone image; An image processing sub-module, configured to perform preset processing on the first reference image to obtain a second reference image, where the preset processing includes binarization processing, corrosion processing, and dilation processing; A second filtering parameter acquisition sub-module, configured to obtain second filtering parameters according to the first reference image and the second reference image, where the second filtering parameters correspond to a second filtering model; A second filtering sub-module, configured to filter the first reference image according to the second filtering model to obtain a target bone image corresponding to the current initial bone image.

[0111] In a specific embodiment, the second filtering parameter acquisition sub-module includes: A first edge detection unit for performing edge detection on a first reference image to obtain a first edge image.

[0112] A second edge detection unit for performing edge detection on a second reference image to obtain a second edge image.

[0113] A difference degree acquisition unit for obtaining the edge difference degree between the first reference image and the second reference image according to the first edge image and the second edge image.

[0114] A second filtering parameter acquisition unit for obtaining a second filtering parameter according to the edge difference degree and a preset difference degree threshold.

[0115] In a specific embodiment, the first edge image includes a plurality of first edge pixel points, the second edge image includes a plurality of second edge pixel points, and the difference degree acquisition unit includes: A pixel point matching sub-unit for matching all the first edge pixel points and all the second edge pixel points to obtain a plurality of edge pixel point pairs, where each edge pixel point pair includes a first edge pixel point and a matched second edge pixel point.

[0116] A difference value acquisition sub-unit for determining, for any edge pixel point pair, the distance between the first edge pixel point and the second edge pixel point in the current edge pixel point pair as the edge difference value of the current edge pixel point pair.

[0117] A difference degree acquisition sub-unit for determining the variance of the edge difference values of all edge pixel point pairs as the edge difference degree between the first reference image and the second reference image.

[0118] In a specific embodiment, the second filtering parameter includes a first filtering coefficient and a second filtering coefficient, and the second filtering parameter acquisition unit includes: A first filtering coefficient acquisition sub-unit for comparing the edge difference degree with a preset difference degree threshold and determining the minimum value of the edge difference degree and the preset difference degree threshold as the first filtering coefficient.

[0119] A second filtering coefficient acquisition sub-unit for obtaining the second filtering coefficient corresponding to the first filtering coefficient according to the first filtering coefficient and a preset parameter mapping model, where the parameter mapping model is used to represent the mapping relationship between the first filtering coefficient and the second filtering coefficient.

[0120] In a specific embodiment, the image recognition module 84 includes: A model acquisition sub-module for obtaining a preset large language model and a preset task description text.

[0121] The first sequence acquisition sub-module is configured to sort the bone trauma features corresponding to N target bone images in the order of acquisition time from the earliest to the latest, and obtain a bone trauma feature sequence.

[0122] The second sequence acquisition sub-module is configured to sort the N acquisition times in the order of acquisition time from the earliest to the latest, and obtain an acquisition time sequence.

[0123] The image recognition sub-module is configured to input the bone trauma feature sequence, the acquisition time sequence, and a preset task description text into a preset large language model, and obtain an image recognition result corresponding to the target object, where the image recognition result includes the degree of recovery and the recovery trend.

[0124] It should be understood that the various modules or units described in apparatus 800 correspond to the respective steps in the method described with reference Figure 2 Accordingly, the operations and features described above for the method are equally applicable to apparatus 800 and the modules included therein, and will not be repeated here. Apparatus 800 may be pre-implemented in a browser or other secure application of an electronic device, or may be loaded into the browser or its secure application of the electronic device by means of downloading or the like. The corresponding modules in apparatus 800 may cooperate with the modules in the electronic device to implement the solutions of the embodiments of the present application.

[0125] The division of the several modules or units mentioned in the above detailed description is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more of the above-described modules or units may be embodied in one module or unit. Conversely, the features and functions of one module or unit described above may be further divided and embodied by a plurality of modules or units.

[0126] The following refers to Figure 9 , Figure 9 which shows a schematic structural diagram of a computer system of an electronic device or a server suitable for implementing the embodiments of the present application.

[0127] As Figure 9 shown, the computer system includes a central processing unit (CPU) 901, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 902 or the program loaded from the storage section 908 into the random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation instructions of the system are also stored. The CPU 901, the ROM 902, and the RAM 903 are connected to each other through a bus 904. The input / output (I / O) interface 905 is also connected to the bus 904.

[0128] The following components are connected to the I / O interface 905: an input section 906 including a keyboard, a mouse, etc.; an output section 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, a modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed so that a computer program read therefrom is installed into the storage section 908 as needed.

[0129] Specifically, according to an embodiment of the present application, the process described above with reference to the flowchart Figure 2 can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product that includes a computer program carried on a computer-readable medium, and the computer program includes program code for performing the method shown in the flowchart. In such an embodiment, the computer program includes program code for performing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by a central processing unit (CPU) 901, the above-described functions defined in the system of the present application are performed.

[0130] It should be noted that the computer-readable medium shown in this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this application, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination of the above.

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operation instructions of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram can represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two connected blocks can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that executes the specified functions or operation instructions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0132] The units or modules involved in the embodiments described in this application can be implemented in software or in hardware. The described units or modules can also be provided in a processor. Among them, the names of these units or modules do not, in some cases, constitute a limitation on the units or modules themselves.

[0133] As another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments, or may exist alone without being assembled into the electronic device. The above computer-readable storage medium stores one or more programs, and when the above programs are executed by one or more processors, they are used to implement the orthopedic trauma image recognition method described in this application.

[0134] The above description is only a preferred embodiment of this application and an explanation of the technical principles applied. Those skilled in the art should understand that the scope of disclosure involved in this application is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the foregoing disclosure concept. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features having similar functions disclosed in this application.

Claims

1. A method for orthopedic trauma image recognition, characterized in that: The orthopedic trauma image recognition method comprises: S1, obtaining N initial bone images of the target object, part labels corresponding to the N initial bone images, a first filter function including a first filter parameter corresponding to the part labels, and an acquisition time corresponding to each initial bone image, wherein N is an integer greater than 1; S2, performing image processing on any initial bone image to obtain a target bone image corresponding to the current initial bone image; S3, extracting bone trauma features corresponding to the target bone image; S4, obtaining an image recognition result corresponding to the target object according to the bone trauma characteristics corresponding to the N target bone images and the acquisition time corresponding to each initial bone image, wherein the image recognition result includes a recovery degree and a recovery trend.

2. The orthopedic trauma image recognition method according to claim 1, characterized in that: S1 includes the following steps: S11, obtaining a plurality of bone image samples corresponding to the part label, a noise image sample corresponding to each bone image sample, and a first preset filtering model including preset filtering parameters; S12, filtering the noise image sample according to the first preset filtering model to obtain a reference image sample corresponding to the noise image sample; S13, obtaining an image error according to the reference image sample and the bone image sample corresponding to the noise image sample, and constructing an objective function according to the image error; S14, updating the preset filtering parameters by a gradient descent method until the objective function converges, and determining the updated preset filtering parameters corresponding to the convergence of the objective function as the first filtering parameters corresponding to the part label; S15: constructing the first filter function according to the first filter parameters.

3. The orthopedic trauma image recognition method according to claim 1, characterized in that: S2 includes the following steps: S21, for any initial bone image, filtering the current initial bone image according to the first filtering function including the first filtering parameter, to obtain a first reference image corresponding to the current initial bone image; S22, performing a preset process on the first reference image to obtain a second reference image, wherein the preset process includes a binarization process, an erosion process, and a dilation process; S23, acquiring a second filtering parameter according to the first reference image and the second reference image, wherein the second filtering parameter corresponds to a second filtering model; S24: Filter the first reference image according to the second filtering model to obtain a target bone image corresponding to the current initial bone image.

4. The orthopedic trauma image recognition method according to claim 3, characterized in that: S23 includes the following steps: S231, performing edge detection on the first reference image to obtain a first edge image; S232, performing edge detection on the second reference image to obtain a second edge image; S233, acquiring an edge difference degree between the first reference image and the second reference image according to the first edge image and the second edge image; S234: Acquire a second filtering parameter according to the edge difference degree and a preset difference degree threshold.

5. The orthopedic trauma image recognition method according to claim 4, characterized in that: The first edge image includes a plurality of first edge pixel points, and the second edge image includes a plurality of second edge pixel points. S233 includes the following steps: S2331, matching all first edge pixel points with all second edge pixel points to obtain a plurality of edge pixel pairs, wherein each edge pixel pair includes a first edge pixel point and a matching second edge pixel point; S2332: for any edge pixel pair, determine the distance between the first edge pixel and the second edge pixel in the current edge pixel pair as the edge difference value of the current edge pixel pair; S2333: Determine the variance of the edge difference values ​​of all edge pixel pairs as the edge difference degree between the first reference image and the second reference image.

6. The orthopedic trauma image recognition method according to claim 1, characterized in that: S4 includes the following steps: S41, obtaining a preset large language model and a preset task description text; S42, sorting the bone trauma features corresponding to the N target bone images in descending order of acquisition time to obtain a bone trauma feature sequence; S43, sorting the N collection times in descending order of collection time to obtain a collection time sequence; S44, inputting the bone trauma feature sequence, the acquisition time sequence and the preset task description text into the preset large language model to obtain the image recognition result corresponding to the target object.

7. An orthopedic trauma image recognition device, characterized in that: The orthopedic trauma image recognition device comprises: A data acquisition module, used to acquire N initial bone images of the target object, part labels corresponding to the N initial bone images, a first filter function including a first filter parameter corresponding to the part labels, and an acquisition time corresponding to each initial bone image, wherein N is an integer greater than 1; An image processing module is used to perform image processing on any initial bone image to obtain a target bone image corresponding to the current initial bone image; A feature extraction module, used to extract bone trauma features corresponding to the target bone image; The image recognition module is used to obtain the image recognition result corresponding to the target object according to the bone trauma characteristics corresponding to N target bone images and the acquisition time corresponding to each initial bone image, wherein the image recognition result includes the recovery degree and recovery trend.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the orthopedic trauma image recognition method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the orthopedic trauma image recognition method as described in any one of claims 1-6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the orthopedic trauma image recognition method described in any one of claims 1 to 6 are implemented.

Images