Osteoporosis magnetic resonance image analysis system and method based on edge calculation

Abstract

The invention discloses an osteoporosis magnetic resonance image analysis system and method based on edge calculation, and relates to the technical field of electrical digital data processing. According to the system, node suitability verification is carried out on a knee joint magnetic resonance image of a target object locally at an edge computing node, and matching analysis of bone state data and corresponding bone structure feature data is carried out based on a node suitability verification result, so that bone structure areas in the knee joint magnetic resonance image are screened; performing edge calculation and real-time positioning on the bone structure area based on a preset bone structure feature template, judging whether edge calculation parameters are analyzed and processed or not at an edge end, and finally judging whether to perform bone state grading evaluation or not according to a judgment result of the edge end. The problem that the timeliness of bone state analysis is not high due to the fact that the tracking precision of the inter-frame displacement of the osteoporosis magnetic resonance image is lagged in the edge calculation process is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric digital data processing, and in particular to an osteoporosis magnetic resonance image analysis system and method based on edge computing. BACKGROUND

[0002] Magnetic resonance imaging (MRI) has become one of the core technologies for diagnosing osteoporosis due to its high-resolution imaging capability of bone microstructure. The existing technology usually realizes osteoporosis MRI image analysis by combining a cloud server with a traditional machine learning algorithm. The specific process is as follows: first, the collected original MRI image data is compressed and uploaded to the cloud; image noise is removed by Gaussian filtering; image registration is completed based on the gray threshold method; the image deviation collected by different devices is corrected; then, the region growing algorithm is used to segment the bone regions such as femur and tibia; the support vector machine (SVM) and random forest algorithm are used to extract feature parameters such as bone density mean and trabecular spacing from the segmented regions; finally, the classification model is inputted, and the three-level classification of normal, osteopenia, and osteoporosis is realized through feature matching.

[0003] For example, the Chinese invention patent with the publication number CN120339267B discloses a light-weight 3D medical image real-time inference method and system based on edge computing, which includes: predicting sample image data and globally guiding a light-weight child model; monitoring anisotropy of input medical images to determine the extreme ratio of any axis resolution; checking whether the original medical image and the target 3D medical image shape are consistent; performing 2D resampling on each XY screen perpendicular to the low-resolution axis; automatically determining overlap compensation and weight allocation to generate a weight map; using the light-weight child model for prediction, and performing sliding window inference and image post-processing.

[0004] For example, the Chinese invention patent with the publication number CN111062043B discloses a medical image recognition method and system based on edge computing, which includes: a data using end sends a first request for edge computing in the domain, finds a data owning end that meets the first request in the cloud center according to the first request, establishes a sandbox and encrypts the algorithm using a private key, and sends it to the cloud center; sends data request information to the data owning end that meets the first request to query the data request information, and sends the data set corresponding to the query result to the cloud center after encryption with a private key, decrypts the algorithm and the data set in the sandbox, executes the algorithm on the data set to obtain an image recognition result set, and sends it to the data owner and the data using end after encryption.

[0005] The above-mentioned technology at least has the following technical problems: In the process of image analysis, the edge computing node needs to analyze high frame rate bone quality images in real time in a localized environment, both to maintain the spatial resolution corresponding to the fine features of bone quality (bone trabecular texture on the bone surface) and to ensure the continuous temporal coherence of consecutive frames (tracking of bone structure displacement between adjacent frames). This double demand superimposition may cause the local edge computing node to process a large amount of data, which may cause image frame loss or lag due to uneven memory allocation. In the case of resource constraints, the edge node will reduce the frequency to ensure real-time performance, and the frequency reduction will blur the analysis of bone trabecular texture details and prolong the single-frame analysis time, so that the data volume and analysis delay cause the bone structure to jump between consecutive frames, thereby reducing the cooperative adaptability between the temporal continuity of the image and the resource scheduling. In the process of edge computing, the bone quality magnetic resonance image frame displacement tracking accuracy is lagging behind, and the bone quality state analysis timeliness is not high. SUMMARY

[0006] To solve the technical problem of low timeliness of bone quality state analysis in the prior art, the embodiments of the present application provide a bone quality magnetic resonance image analysis system and method based on edge computing. The technical solution is as follows: On the one hand, a bone quality magnetic resonance image analysis system based on edge computing is provided, which includes the following modules: an edge bone quality data matching analysis module, an edge computing real-time positioning module, and a bone quality state classification evaluation module. The edge bone quality data matching analysis module is used to verify the node adaptability of the knee joint magnetic resonance image of the target object locally on the edge computing node, to improve the adaptability between the analysis process of the knee joint magnetic resonance image and the processing capability of the edge computing node, and to perform matching analysis of bone quality state data and corresponding bone structure feature data based on the results of node adaptability verification, to screen the bone structure region in the knee joint magnetic resonance image. The edge computing real-time positioning module is used to perform edge computing real-time positioning of the bone structure region based on a preset bone structure feature template, to determine a target analysis region that can reflect the change of bone structure, and to determine whether to perform analysis and processing of edge computing parameters at the edge. The target analysis region represents a bone tissue region that can intuitively reflect the degree of osteoporosis through the features of the magnetic resonance image. The bone quality state classification evaluation module is used to determine whether to perform bone quality state classification evaluation according to the results of edge determination, and the bone quality state classification evaluation is used to generate a structured analysis result of the target analysis region.

[0007] In another aspect, an edge computing-based osteoporosis magnetic resonance image analysis method is provided, which comprises the following steps: step one, performing node adaptability verification on the knee joint magnetic resonance image of a target object locally at an edge computing node, and performing matching analysis of bone quality state data and corresponding bone structure feature data based on the result of node adaptability verification; step two, performing edge computing real-time positioning on the bone structure region based on a preset bone structure feature template, and determining whether to perform analysis and processing of edge computing parameters at the edge; and step three, determining whether to perform bone quality state classification and evaluation according to the result of edge determination.

[0008] The technical scheme provided by the embodiment of the application has at least the following beneficial effects: 1. By constructing the node adaptability verification of abnormal source allocation processing, the precise adaptation of the edge node and the image data is realized. First, based on the node memory available capacity and the image single frame data amount, the node adaptability analysis value is obtained, the adaptability is comprehensively evaluated from two core dimensions of data amount and transmission capacity, and when there is an adaptation abnormality, the abnormal source of the dimension is further distinguished. This mechanism can dynamically select a processing strategy according to the abnormal source, avoid resource waste, guarantee the integrity and availability of the image data, effectively solve the problem of analysis interruption or precision reduction caused by insufficient adaptability of the edge node and the image data, and lay a stable data foundation for subsequent bone quality state matching analysis.

[0009] 2. By the combined cooperative processing of partition lossless compression and format cooperative conversion, the bone feature details in the image are maximally preserved while solving the node adaptation abnormality. When the node adaptation abnormality is caused by non-single dimension, first, the data amount proportion of different structure regions in the knee joint magnetic resonance image is analyzed by the regional lossless compression algorithm, a regional data amount compression ratio adjustment value is generated, a high compression ratio is used for the background area without key information, the overall data amount is efficiently reduced, the image data amount is reduced to the preset node memory load range, and this combined strategy organically combines data amount control and format matching. It can not only guarantee the integrity of the key features of the bone by lossless compression, but also eliminate data analysis obstacles by format cooperative conversion, realize the dual goals of load reduction and precision preservation, and provide high-quality image data support for subsequent bone structure feature comparison.

[0010] 3. By constructing a real-time positioning mechanism for evaluating the shielding proportion and adjusting the dynamic parameters, the accurate demarcation of the target area is realized. In the prior art, when determining the target analysis area of the osteoporosis magnetic resonance image, a fixed bone structure feature template is usually used for contour alignment to directly demarcate the area, without considering the shielding problems such as soft tissue overlap and artifact interference that may exist in the bone trabecula. First, the obtained bone structure area is aligned with the preset bone structure feature template, and the bone tissue boundary coordinates are extracted to define the preliminary distribution range of the bone structure area. Then, the non-shielding proportion of the bone trabecula in the distribution range is calculated and compared with the preset non-shielding proportion threshold of the bone trabecula. By evaluating the non-shielding condition of the bone trabecula in real time and dynamically adjusting the detection parameters according to the deviation degree, the limitations of the existing fixed parameter positioning are broken, and it is ensured that the target analysis area can contain the non-shielding bone trabecula area to the greatest extent, so as to accurately reflect the changes of the bone structure and provide an accurate area basis for subsequent osteoporosis grading evaluation.

[0011] 4. In the prior art, when performing osteoporosis grading, a single bone feature parameter is usually used for determination, and the synergistic consideration of the two core parameters of bone trabecula density and cortical bone thickness is lacking, which leads to the determination result being easily affected by the fluctuation of a single parameter and the accuracy being insufficient. The flow process retrieves the preset osteoporosis grading and corresponding reference bone structure feature parameters from the edge-end local database, obtains the to-be-analyzed bone trabecula density and to-be-analyzed cortical bone thickness in the knee magnetic resonance image, and compares them with the reference parameters layer by layer. This double-parameter layered comparison mechanism includes the two core bone feature parameters in the synergistic determination, compared with the single parameter determination or fuzzy interval determination of the prior art, it can more comprehensively reflect the osteoporosis state, avoid the misjudgment of osteopenia caused by only focusing on the bone trabecula density and ignoring the normal cortical bone thickness, or the missed judgment of osteoporosis caused by only focusing on the cortical bone thickness and ignoring the too low bone trabecula density. The clear determination rule further reduces the interference of subjective factors, so that the osteoporosis determination result is more objective and accurate. BRIEF DESCRIPTION OF DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0013] Figure 1 The structural schematic diagram of the osteoporosis magnetic resonance image analysis system based on edge computing provided by the embodiments of the present application is shown in the figure. Figure 2 The flowchart corresponding to the edge bone data matching analysis module provided by the embodiments of the present application is shown in the figure. Figure 3A flowchart corresponding to the edge computing real-time positioning module provided in the embodiment of the present application is shown in the figure. Figure 4 A flowchart corresponding to the bone state grading evaluation module provided in the embodiment of the present application is shown in the figure. Figure 5 A flowchart of the osteoporosis magnetic resonance image analysis method based on edge computing provided in the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0014] The technical solutions in the present application will be described below with reference to the accompanying drawings.

[0015] In the embodiments of the present application, the words such as "example", "for example" and the like are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be either one of the two.

[0016] In the embodiments of the present application, sometimes the subscript such as W1 may be written in the form of non-subscript such as W1, and when the difference is not emphasized, the meanings expressed are consistent.

[0017] In order to make the technical problems, technical solutions and advantages to be solved by the present application more clear, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0018] The embodiment of the present application provides an osteoporosis magnetic resonance image analysis system based on edge computing. As shown in the structural schematic diagram of the osteoporosis magnetic resonance image analysis system based on edge computing, the processing flow of the system can include the following modules: Figure 1 ​The edge bone data matching analysis module is used for verifying the node adaptability of the knee joint magnetic resonance image of a target object locally at the edge computing node, so as to improve the adaptability between the analysis process of the knee joint magnetic resonance image and the processing capability of the edge computing node, and meanwhile, based on the result of the node adaptability verification, the matching analysis of the bone state data and the corresponding bone structure feature data is performed, so as to screen the bone structure region in the knee joint magnetic resonance image; the edge computing real-time positioning module is used for performing edge computing real-time positioning on the bone structure region based on a preset bone structure feature template, so as to determine a target analysis region capable of reflecting the change of the bone structure, and meanwhile, it is determined whether to perform analysis and processing of the edge computing parameters at the edge, the target analysis region representing a bone tissue region in which the degree of osteoporosis (such as the sparsity of bone trabecula and the reduction of bone density) can be directly reflected through the features of the magnetic resonance image; and the bone state grading evaluation module is used for determining whether to perform bone state grading evaluation according to the result of the determination at the edge, and the bone state grading evaluation is used for generating a structured analysis result of the target analysis region.

[0019] In the embodiment, the three modules form a progressive cooperative relationship, the edge bone data matching analysis module serves as the basis, the adaptability of the knee joint magnetic resonance image and the processing capability of the node is verified locally at the edge computing node first, the bone structure region is screened out, and a reliable analysis object is provided for subsequent processing; the edge computing real-time positioning module takes the bone structure region as the input, performs real-time positioning in combination with the preset bone structure feature template, determines the target analysis region capable of reflecting the change of the bone structure, and simultaneously determines whether the edge computing parameters need to be adjusted, so as to realize accurate focusing from the structure region to the key analysis region; and the bone state grading evaluation module performs grading evaluation based on the determination result of the positioning module, generates the structured analysis result of the target analysis region, and forms a complete closed loop from data preprocessing to accurate analysis.

[0020] After the system is applied, through the cooperative linkage of the three modules, the adaptability of the image analysis and the edge node processing capability can be improved, the analysis interruption caused by the mismatch of resources can be avoided, the accurate positioning and grading evaluation can be performed, the effective extraction and state judgment of the osteoporosis related features can be ensured, the analysis process is more in line with the needs of real-time and accuracy, and more reliable image analysis support can be provided for the diagnosis and treatment of osteoporosis.

[0021] As shown in Figure 2 The flow chart corresponding to the edge bone data matching analysis module provided by the embodiment of the application is shown in FIG. 1. The node adaptability analysis value is obtained through the node adaptability verification, and is compared with the preset value. When the obtained node adaptability analysis value is not less than the preset node adaptability analysis value, the matching analysis is performed, otherwise the node adaptability abnormality analysis is performed. Through analyzing whether the source of the node adaptability abnormality is a single dimension or a non-single dimension, it is determined whether to perform data compression processing or compression ratio adjustment and current data format adjustment.

[0022] It is further understood that the node adaptability verification specifically includes the following steps: based on the available capacity of the node memory and the single-frame image data volume, an image data volume adaptation score S1 is obtained, wherein the greater the value, the higher the node memory adaptability, and the image data volume adaptation score represents the ratio of the available capacity of the node memory to the single-frame image data volume; based on the node interface bandwidth and the bandwidth required for image transmission, a bandwidth support score S2 is obtained, wherein the greater the value, the higher the node bandwidth adaptability, and the bandwidth support score represents the ratio of the node interface bandwidth to the bandwidth required for image transmission; based on the harmonic mean result of the normalized image data volume adaptation score and the bandwidth support score, a node adaptability analysis value S is obtained, which is compared with a preset adaptation degree to determine whether the node adaptability is qualified.

[0023] Specifically, the mathematical expression of the node adaptability analysis value S is S = 1 / (1 / S1 + 1 / S2). In the node adaptability verification, the image data volume adaptation score and the bandwidth support score are two key dimensions that are indispensable to each other, and the node needs to meet the requirements of both. The harmonic mean gives higher weight to the score with a lower value. For example, if S1 is significantly greater than S2, the harmonic mean result will be significantly biased towards S2, which truly reflects the adaptability defects caused by insufficient bandwidth. Only when S1 and S2 are at a high level and the difference is small, can a higher S value be obtained, ensuring that the final node adaptability analysis value can accurately measure the comprehensive adaptability of the node in terms of memory and bandwidth, and avoiding false judgment of the overall qualification of the node due to the compliance of a single dimension.

[0024] If the obtained node adaptability analysis value is greater than or equal to the preset adaptation degree, it is determined that the node adaptability is qualified, and the matching analysis of the bone state data and the corresponding bone structure feature data is directly performed. If the obtained node adaptability analysis value is less than the preset adaptation degree, it is determined that the node adaptability is abnormal, and the source of the node adaptability abnormality is determined. If the node adaptability abnormality is caused by a single dimension (such as only the data volume exceeding the standard), compression processing feedback is performed to prompt the edge end to compress the image data of the corresponding dimension based on the edge end lightweight algorithm. If the node adaptability abnormality is caused by non-single dimension (such as the data volume exceeding the standard and the format being incompatible), combined strategy processing feedback is performed to prompt the edge end to perform format type conversion based on the regional lossless compression algorithm. After the node adaptability abnormality is calibrated, the node adaptability analysis value is re-obtained. If the re-obtained node adaptability analysis value is greater than or equal to the preset adaptation degree, the node adaptability abnormality calibration is completed, and the matching analysis of the bone state data and the corresponding bone structure feature data is performed. Otherwise, a node resource shortage warning is sent to prompt the preset personnel to intervene.

[0025] The combination strategy processes feedback, specifically: through the sub-region lossless compression algorithm, the proportion of data quantity of non-single dimension is obtained, a sub-region data quantity compression ratio adjustment value is generated to reduce the corresponding image data quantity to the preset node memory load; during the adjustment process of the image data quantity compression, if the data format of the current image data quantity is different from the node compatible data format, the format conversion tool of the edge end is called to convert the data format of the corresponding image data quantity to the node compatible format type; if the data format of the current image data quantity is the same as the node compatible data format, the image data quantity compression ratio adjustment is completed, and the matching analysis of the bone state data and the corresponding bone structure feature data is performed.

[0026] In the embodiment, the data quantity proportion is obtained by the number of data storage bytes of a single region / the total number of image data storage bytes, the preset adaptation degree is set according to the load capacity (hardware parameter library) of the edge node to set the node adaptation degree interval, the edge end inputs the current data quantity into the edge end lightweight algorithm to obtain the data quantity of the image data of the corresponding dimension; based on the sub-region lossless compression algorithm, the historical data quantity proportion and the compression ratio of the data quantity are taken as training samples, the mapping relationship between the data quantity proportion input and the data quantity compression output is learned through the data compression algorithm, the data quantity proportion is taken as the input, and the corresponding sub-region data quantity compression ratio is obtained as the output, and the format conversion tool stores all conversion forms of the data format, which can convert the current incompatible data format into the data format type compatible with the edge node.

[0027] The node adaptation verification step improves the matching accuracy of the edge computing node and the osteoporosis magnetic resonance image analysis demand through the adaptation analysis logic and the hierarchical exception handling strategy, the node adaptation analysis value is coupled through the cooperative gain item and the imbalance item, and the balance of the memory bearing capacity and the bandwidth support capacity can be accurately captured, when the two are synchronously adapted, the analysis value remains at a reasonable level; if there is obvious imbalance, even if a single index performs well, the analysis value will faithfully reflect the overall adaptation defects, and misjudgment is avoided from the source. This accurate judgment ensures that only the truly adapted node enters the subsequent analysis, reduces the analysis interruption or feature loss caused by resource mismatch, makes the image processing process of the edge end more stable, and lays a foundation for accurate extraction of bone structure features.

[0028] The hierarchical processing strategy for adapting to the exception balances the demand of resource optimization and data integrity, the lightweight compression in single dimension exception, the appropriate adjustment for multi-dimension to avoid invalid processing of the dimension, the sub-regional lossless compression and format conversion in non-single dimension exception, which preserves the details of the key feature area in the differential compression ratio, eliminates the parsing obstacles through format conversion, ensures the reduction of data volume without destroying the key image features. This targeted processing makes the exception calibration more efficient, which can adjust the image data to the node bearing range, and maximally preserves the core information for bone status evaluation, avoids the problem of sacrificing precision for adaptation, makes the subsequent bone structure area screening and state classification more reliable, and improves the practical value of the whole edge analysis system.

[0029] As shown in Figure 3 The flow chart corresponding to the edge computing real-time positioning module provided by the embodiment of the application is shown in the figure. The bone status classification state obtained through matching analysis is compared with the reference classification state layer by layer, at least one parameter matching the reference interval of bone reduction, matching the bone qualified state and matching the osteoporosis state are screened out, and the bone structure area is extracted based on the matching analysis result and the image segmentation algorithm.

[0030] It is further understood that the matching analysis of the bone status data and the corresponding bone structure feature data is performed, and the specific process is as follows: the preset bone status classification state and the reference bone structure feature parameters corresponding to each state are called from the edge local database, the preset bone status classification state includes bone qualified, bone reduction and osteoporosis, the reference bone structure feature parameters include the upper limit of trabecular bone density (usually 1.2 g / cm³), the lower limit of cortical bone thickness (usually 0.3 cm), the lower limit of trabecular bone density (usually 0.9 g / cm³) and the upper limit of cortical bone thickness (usually 0.5 cm); the bone structure feature to be analyzed in the knee joint magnetic resonance image is obtained and standardized, the parameter unit and value range are unified, the bone structure feature to be analyzed includes the bone trabecular density to be analyzed and the cortical bone thickness to be analyzed; the standardized bone structure feature to be analyzed is compared with the reference bone structure feature parameter layer by layer to obtain the layer-by-layer comparison result, the layer-by-layer comparison result includes the matching bone qualified state, the matching bone reduction state and the matching osteoporosis state.

[0031] If the trabecular bone density to be analyzed is not less than the upper limit of the trabecular bone density and the cortical bone thickness to be analyzed is not less than the upper limit of the cortical bone thickness, it is determined that the layer-by-layer comparison result is a matched bone quality qualified state; if the trabecular bone density to be analyzed is between the lower limit and the upper limit of the trabecular bone density, or the cortical bone thickness to be analyzed is between the lower limit and the upper limit of the cortical bone thickness, and at least one parameter of the layer-by-layer comparison result is in the matched bone quality reduction reference interval, it is determined that the layer-by-layer comparison result is a matched bone quality reduction state; if the trabecular bone density to be analyzed is less than the lower limit of the trabecular bone density and the cortical bone thickness to be analyzed is less than the lower limit of the cortical bone thickness, it is determined that the layer-by-layer comparison result is a matched bone quality osteoporosis state; based on the layer-by-layer comparison result determined, the bone quality structure region of the corresponding state is extracted from the knee joint magnetic resonance image based on the image segmentation algorithm, and the matching analysis is completed.

[0032] In the embodiment, the preset bone quality state classification state is set according to the reference standard of the bone quality structure characteristics of the historical bone quality state, and the values of the upper limit of the trabecular bone density, the lower limit of the cortical bone thickness, the lower limit of the trabecular bone density and the upper limit of the cortical bone thickness are not fixed, and in the actual application process, the preset personnel can adjust the values according to the application purpose of the current scene; by clearly defining the image characteristics of the estimated value structure in this state, the characteristics are converted into the constraint conditions of the image segmentation algorithm, and the segmentation algorithm parameters are dynamically adjusted according to the state characteristics.

[0033] By constructing a standardized comparison framework and a double-parameter collaborative determination logic, the accuracy and consistency of the matching analysis of the bone quality state and the structure characteristics are improved. The preset classification state and reference parameters are called from the edge local database, which not only avoids the influence of cloud transmission delay on real-time analysis, but also ensures the stability of the comparison benchmark through the standardized parameters stored locally. The standardized processing of the trabecular bone density and the cortical bone thickness eliminates the differences in parameter units and value ranges under different image acquisition conditions, so that the characteristics to be analyzed and the reference parameters are in the same comparable dimension, and the comparison deviation caused by the non-uniform benchmark is avoided.

[0034] The layer-by-layer comparison of the double parameters takes into account the collaborative significance of the trabecular bone and the cortical bone in osteoporosis evaluation, and reduces the determination error through clear interval division. This design allows the matching analysis to accurately capture the qualified performance of the double characteristics in the bone quality qualified state, and accurately identify the transition characteristics in bone quality reduction and the characteristic data in bone quality osteoporosis, ensuring the pertinence of the extraction of the bone quality structure region in different states. Overall, through the standardized and collaborative design, the matching analysis result is more in line with the actual state of the bone, providing a reliable basis for subsequent target region positioning and classification evaluation.

[0035] As Figure 4As shown, the bone state classification evaluation module provided by the embodiment of the application corresponds to a flowchart. The trabecular bone unobstructed proportion obtained is compared with a preset value. When the obtained value is greater than the preset value, the current region is taken as a target analysis region, and bone state classification evaluation is performed on this basis. When the obtained value is not greater than the preset value, edge calculation parameter analysis processing is performed. The deviation obtained is judged. The results of deviation judgment are used for edge detection gradient adjustment and detection window adjustment. Whether the obtained deviation value is qualified is re-judged. The results of re-judgment of the deviation are used for bone state classification evaluation and positioning abnormality early warning. The bone state classification evaluation outputs a classification report through the obtained feature matching parameters. The report includes the osteoporosis grade determined according to the bone state, and outputs the corresponding suggestion decision.

[0036] It is further understood that the edge calculation real-time positioning is performed on the bone structure region. The specific process is as follows: the obtained bone structure region is aligned with a preset bone structure feature template to extract bone tissue boundary coordinates. The preset bone structure feature template stored in the edge node is called first. The contour feature points of the obtained bone structure region image to be analyzed are obtained through a feature point matching algorithm to obtain the feature point correspondence between the bone structure region to be analyzed and the template. Based on these feature points, the contour coordinates of the region to be analyzed are mapped to the coordinate system of the template to realize the alignment of the contour. The bone tissue boundary coordinates are used to define the distribution range of the bone structure region containing the matching osteoporosis state. The trabecular bone unobstructed proportion in the distribution range is obtained and compared with a preset trabecular bone unobstructed proportion. If the obtained trabecular bone unobstructed proportion is greater than the preset trabecular bone unobstructed proportion, the distribution range corresponding to the current trabecular bone unobstructed proportion is determined as the target analysis region, and bone state classification evaluation is performed. If the obtained trabecular bone unobstructed proportion is not greater than the preset trabecular bone unobstructed proportion, the edge calculation parameter analysis processing is performed to update the detection window width and the edge detection gradient.

[0037] Specifically, the analysis and processing of the edge computing parameter, the specific process is: based on the current obtained trabecular bone unobstructed proportion and the preset trabecular bone unobstructed proportion, the proportion deviation is obtained and compared with the set maximum deviation reference value: if the proportion deviation is greater than the set maximum deviation reference value, the arithmetic mean of the original detection window width and the dynamic expansion width is taken as the new detection window width, and the arithmetic mean of the edge detection gradient and the gradient reduction threshold is taken as the new edge detection gradient; if the proportion deviation is not greater than the set maximum deviation reference value, the dynamic expansion width is taken as the new detection window width, and the gradient reduction value is taken as the new edge detection gradient; based on the updated detection window width or edge detection gradient, the distribution range is redefined and the bone tissue boundary coordinates are extracted, and the trabecular bone unobstructed proportion is reacquired; if the reacquired trabecular bone unobstructed proportion is still less than the preset trabecular bone unobstructed proportion, the edge computing positioning abnormal early warning is carried out, otherwise the corresponding bone structure region is determined as the target analysis region, and the bone quality state classification evaluation is carried out, specifically: If the feature parameter matching degree of the target analysis region is not greater than the preset feature parameter matching degree, the bone quality state is directly determined according to the matching result, and a classification report is output, otherwise a supplementary feature verification is carried out to improve the matching degree between the feature parameters and the osteoporosis state, and the feature parameter matching degree is used to quantify the similarity between the bone quality characteristics of the analysis region and the actual bone quality state characteristics; the specific process of the supplementary feature verification is: if the trabecular bone spacing deviation value is greater than the set maximum spacing deviation, the trabecular bone spacing parameter is input into the edge feature correction algorithm to output a correction coefficient, otherwise, the arithmetic mean of the trabecular bone spacing and the cortical bone thickness deviation value is input into the edge feature correction algorithm to output a correction coefficient; the classification is re-determined after adjusting the feature parameters according to the correction coefficient.

[0038] The classification report is output as the result of edge determination, specifically: the preset osteoporosis evaluation level is called from the edge database, and the cortical bone thickness detection result in the target analysis region is synchronously acquired, the preset evaluation level includes osteoporosis level one, osteoporosis level two and osteoporosis level three, and the specified target osteoporosis degree corresponding to osteoporosis level one, osteoporosis level two and osteoporosis level three increases in turn; if the cortical bone thickness detection result meets the osteoporosis level one, a level one instruction is sent to prompt regular review, if the cortical bone thickness detection result meets the osteoporosis level two, a level two instruction is sent to mark calcium intervention, if the cortical bone thickness detection result meets the osteoporosis level three, a level three instruction is sent to prompt clinical diagnosis and treatment; the marking results of each target analysis region are summarized to generate a classification report of the target analysis region osteoporosis level and intervention scheme.

[0039] In the embodiment, the preset bone structure feature template is a preset template stored in the local edge node, aggregated based on preset labeled features, individual differences removed to form an initial template, and set in the edge node, combined with the bone status images in the historical analysis process to obtain dynamic iteration; the trabecular bone unshielded ratio is obtained by dividing the number (or area) of unshielded trabecular bone regions by the total number (or area) of segmented trabecular bone regions, and the preset trabecular bone unshielded ratio is stored in the bone structure feature template, the sample contains various types / degrees of trabecular bone shielding scenarios, and the distribution characteristics of the unshielded ratio under different bone conditions are extracted by statistical analysis to select the ratio interval that can stably reflect the effective observation range of the trabecular bone under the condition; the ratio deviation is the difference between the current obtained trabecular bone unshielded ratio and the preset trabecular bone unshielded ratio, and the set maximum deviation reference value is also stored in the bone structure feature template, which is the maximum value of the trabecular bone unshielded ratio deviation in the historical image analysis process, the set maximum spacing deviation is the maximum spacing deviation set in the historical grading evaluation process, the trabecular spacing deviation value is the difference between the preset reference template and the obtained trabecular spacing, and the cortical bone thickness deviation is the difference between the cortical bone thickness set in the preset reference template and the obtained cortical bone thickness. The labeled results of each target analysis region are summarized to generate a grading report of the target analysis region osteoporosis grade and intervention scheme, and the grading report is taken as a structured analysis result; the edge feature correction algorithm is constructed and dynamically optimized based on historical bone feature data and the lightweight demand of edge computing, the correction data of the historical trabecular spacing, cortical bone thickness deviation value and corresponding correction scheme are taken as training samples, the mapping relationship between the deviation value input and the correction coefficient output is learned through a lightweight neural network, an edge correction model is obtained, the model is lightweighted after training to reduce the calculation complexity to adapt to the edge node, and the adjustment of the feature parameters is based on the multiplication of the obtained correction coefficient to obtain the corrected feature parameters, wherein the expression of the feature parameter matching degree M is: ; In the formula, n represents the number of bone feature parameters, i represents the order of the number of bone feature parameters, x i represents the standardized value of the i-th to-be-analyzed feature parameter, y i represents the standard value of the i-th reference feature parameter, r x represents the correlation coefficient between the to-be-analyzed feature parameters, and r yThe formula ensures that the matching degree of each core bone quality characteristic parameter is taken into account through the multi-characteristic average similarity, avoids the excessive influence of the deviation of a single parameter on the matching result, and uses the synergistic correlation r to obtain the correlation rule between the to-be-analyzed characteristics and the reference characteristics, so that the quantification of the matching degree is more in line with the actual correlation of the bone quality characteristics, and the overall evaluation from the single parameter matching to the consistent overall characteristic trend is realized. The multi-characteristic average similarity and the synergistic similarity are coupled, which avoids the one-sidedness of single-dimensional analysis and accurately reflects the internal correlation rule between the bone quality characteristics.

[0040] Through dynamic parameter adjustment and occlusion adaptive processing, the accuracy and stability of target analysis area positioning are improved, and the real-time evaluation of the trabecular bone unoccluded ratio becomes the core basis for positioning. By comparing the actual ratio with the preset standard ratio, the occlusion interference in the region can be accurately identified. For the case of insufficient ratio, the width of the detection window and the edge detection gradient are adjusted in stages. When the deviation is small, the window is moderately expanded and the gradient is reduced to capture more effective regions. When the deviation is large, the mean value parameter is used to balance the expansion and the accuracy to avoid region distortion. This dynamic adjustment mechanism enables the positioning process to adapt to the occlusion differences in the image, ensures that the finally determined target analysis area excludes interference to the greatest extent, and reflects the real state of the bone structure, providing a high-quality analysis object for subsequent hierarchical evaluation and reducing the feature misjudgment caused by region deviation.

[0041] The hierarchical evaluation link outputs the results of supplementary verification and clinical adaptation, enhances the reliability of bone quality state analysis, generates a correction coefficient through the synergistic analysis of the trabecular bone spacing and bone cortex thickness deviation when the characteristic parameter matching degree is insufficient, dynamically adjusts the characteristic parameters to improve the matching accuracy, and makes the hierarchical result more in line with the actual bone quality state. The output hierarchical report not only includes the osteoporosis grade, but also corresponds to different recommended schemes combined with the bone cortex thickness detection result, reduces the hierarchical error, and enables the edge-end image analysis to not only accurately judge the bone quality state, but also enhance the actual application efficiency.

[0042] As Figure 5As shown, a flowchart of an edge computing-based osteoporosis magnetic resonance image analysis provided by an embodiment of the present application, the edge computing-based osteoporosis magnetic resonance image analysis method provided by an embodiment of the present application comprises the following steps: step one, performing node adaptability verification on the knee joint magnetic resonance image of the target object locally at the edge computing node, and simultaneously performing matching analysis of the bone mass state data and the corresponding bone structure feature data based on the results of the node adaptability verification; step two, performing edge computing real-time positioning on the bone structure region based on a preset bone structure feature template, and simultaneously determining whether to perform analysis and processing of edge computing parameters at the edge; step three, determining whether to perform bone mass state classification evaluation according to the results of the edge determination.

[0043] In the present embodiment, the three steps are cooperatively linked to realize efficient adaptation and accurate evaluation of osteoporosis magnetic resonance images in the edge computing scenario. The node adaptability verification and matching analysis are combined to ensure that the image data is adapted to the processing capability of the edge node from the source, avoid analysis interruption or data loss caused by resource mismatch, ensure stable image processing at the edge, and reduce cloud delay and transmission delay. Real-time positioning adjusts the target analysis region adaptively to the bone trabecula occlusion ratio through dynamic parameter adjustment, ensures that the region reflects the real changes of the bone structure, provides a high-quality basis for evaluation, and avoids feature misjudgment caused by region definition deviation. The classification evaluation reduces the influence of single parameter deviation by means of determination logic and supplementary verification, and directly associates the analysis results with the actual demand through the output of the targeted intervention scheme.

[0044] The above embodiments can be implemented in whole or in part by software, hardware (such as a circuit), firmware, or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid state disk.

[0045] It should be understood that the term "and / or" in this document is merely used to describe associated objects, and it is possible that there are three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, B exists alone, and A, B can be singular or plural. In addition, the character " / " in this document generally represents an "or" relationship between the front and rear associated objects, but it can also represent an "and / or" relationship, which can be understood according to the context before and after.

[0046] In this application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or the like means any combination of the items, including single item or any combination of multiple items. For example, at least one of a, b, or c can mean a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be singular or plural.

[0047] It should be understood that the size of the sequence number of the above-mentioned processes in various embodiments of the application does not mean the order of execution, and the execution order of the processes should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the application.

[0048] Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0049] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working process of the above-mentioned devices, apparatuses and units can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0050] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are only schematic, and the division of units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed objects can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0051] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e., may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0052] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit.

[0053] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method of each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0054] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An osteoporosis magnetic resonance imaging analysis system based on edge computing, characterized in that, It includes the following modules: edge bone data matching and analysis module, edge computing real-time positioning module, and bone status grading assessment module; The edge bone data matching and analysis module is used to perform node adaptability verification on the knee joint magnetic resonance image of the target object locally on the edge computing node, so as to improve the adaptability between the analysis process of the knee joint magnetic resonance image and the processing capability of the edge computing node. At the same time, based on the results of the node adaptability verification, the matching analysis of bone status data and corresponding bone structure feature data is performed to screen bone structure regions in the knee joint magnetic resonance image. The edge computing real-time positioning module is used to perform edge computing real-time positioning of the bone structure region based on a preset bone structure feature template to determine the target analysis region reflecting changes in bone structure. At the same time, it determines whether to perform edge computing parameter analysis at the edge end. The target analysis region represents the bone tissue region that intuitively reflects the degree of osteoporosis through magnetic resonance imaging features. The bone quality grading assessment module is used to determine whether to perform bone quality grading assessment based on the results of edge end determination. The bone quality grading assessment is used to generate structured analysis results for the target analysis area.

2. The edge computing-based osteoporosis magnetic resonance imaging analysis system as described in claim 1, characterized in that, The node compatibility verification process is as follows: Based on the available memory capacity of nodes and the data volume of a single image frame, obtain the image data volume adaptation score; Based on the node interface bandwidth and the bandwidth required for image transmission, obtain the bandwidth support score; Based on the normalized image data volume adaptation score and bandwidth support score, the node adaptability analysis value is obtained and compared with the preset adaptability to determine whether the node adaptability is qualified. If the obtained node adaptability analysis value is greater than or equal to the preset adaptability, the node is deemed to be adaptable and the matching analysis between the bone status data and the corresponding bone structure feature data is directly performed. If the obtained node adaptability analysis value is less than the preset adaptability, the node adaptability is determined to be abnormal, and the source of the corresponding node adaptability abnormality is determined.

3. The edge computing-based osteoporosis magnetic resonance imaging analysis system as described in claim 2, characterized in that, The specific process for determining the source of the corresponding node adaptation anomaly is as follows: If the node adaptation anomaly originates from a single dimension, compression processing is performed as feedback to prompt the edge end to compress the image data of the corresponding dimension based on the edge end lightweight algorithm. If the node adaptation anomaly originates from a non-single dimension, a combined strategy is used to process the feedback, prompting the edge end to perform format type conversion based on a regional lossless compression algorithm; After node adaptation anomaly calibration, if the newly acquired node adaptation analysis value is greater than or equal to the preset adaptation degree, the node adaptation anomaly calibration is completed, and a matching analysis of bone status data and corresponding bone structure feature data is performed; otherwise, a node resource shortage warning is sent to prompt the preset personnel to intervene.

4. The osteoporosis magnetic resonance imaging analysis system based on edge computing as described in claim 3, characterized in that, The combined strategy for processing feedback specifically includes: By using a regional lossless compression algorithm, the proportion of data volume in non-single-dimensional data is obtained, and regional data volume compression ratio adjustment values ​​are generated to reduce the corresponding image data volume to the preset node memory load. During the adjustment of image data compression, if the data format of the current image data is different from the node-compatible data format, the format conversion tool at the edge is called to change the data format of the corresponding image data to the node-compatible format type. If the current image data format is the same as the node-compatible data format, then the image data compression ratio is adjusted, and a matching analysis of bone status data and corresponding bone structure feature data is performed.

5. The edge computing-based osteoporosis magnetic resonance imaging analysis system as described in claim 1, characterized in that, The specific process of matching and analyzing bone quality status data with corresponding bone structure feature data is as follows: The preset bone quality status grading status and the corresponding reference bone structure feature parameters are retrieved from the local database at the edge. The preset bone quality status grading status includes qualified bone quality, reduced bone mass, and osteoporosis. The reference bone structure feature parameters include the upper limit of trabecular density, the lower limit of cortical thickness, the lower limit of trabecular density, and the upper limit of cortical thickness. The bone structure features to be analyzed in the magnetic resonance imaging of the knee joint are obtained and standardized to unify the parameter units and numerical ranges. The bone structure features to be analyzed include the trabecular bone density and the cortical bone thickness. The standardized bone structure features to be analyzed are compared layer by layer with the reference bone structure feature parameters to obtain the layer by layer comparison results. The layer by layer comparison results include matching bone quality status, matching bone loss status, and matching osteoporosis status.

6. The edge computing-based osteoporosis magnetic resonance imaging analysis system as described in claim 5, characterized in that, The specific steps for obtaining the layer-by-layer comparison results are as follows: If the density of the trabecular bone to be analyzed is not less than the upper limit of the trabecular bone density and the thickness of the cortical bone to be analyzed is not less than the upper limit of the cortical bone thickness, then the layer-by-layer comparison result is determined to be a qualified state of matched bone quality. If the trabecular bone density to be analyzed is between the lower limit and the upper limit of trabecular bone density, or the cortical bone thickness to be analyzed is between the lower limit and the upper limit of cortical bone thickness, and at least one parameter matches the reference range for bone loss through layer-by-layer comparison, then the layer-by-layer comparison result is determined to be a matched state of bone loss. If the density of the trabecular bone to be analyzed is less than the lower limit of trabecular bone density and the thickness of the cortical bone to be analyzed is less than the lower limit of cortical bone thickness, then the layer-by-layer comparison result is determined to be a matched osteoporosis state. Based on the layer-by-layer comparison results obtained from the determination, the corresponding bone structure regions are extracted from the knee joint magnetic resonance images using an image segmentation algorithm to complete the matching analysis.

7. The edge computing-based osteoporosis magnetic resonance imaging analysis system as described in claim 1, characterized in that, The specific process for real-time edge calculation and localization of the bone structure region is as follows: The obtained bone structure region is aligned with the preset bone structure feature template to extract the bone tissue boundary coordinates, which are used to define the distribution range of bone structure regions that match the osteoporosis state. Obtain the percentage of unobstructed trabeculae within the distribution range and compare it with a preset percentage of unobstructed trabeculae: If the obtained percentage of unobstructed trabeculae is greater than the preset percentage of unobstructed trabeculae, then the corresponding bone structure region is determined as the target analysis region based on the distribution range corresponding to the current percentage of unobstructed trabeculae, and a bone status grading assessment is performed. If the percentage of unobstructed trabeculae obtained is not greater than the preset percentage of unobstructed trabeculae, then edge calculation parameters are analyzed and processed to update the detection window width and edge detection gradient.

8. The osteoporosis magnetic resonance imaging analysis system based on edge computing as described in claim 7, characterized in that, The analysis and processing of the edge computing parameters are as follows: Based on the currently obtained percentage of unobstructed trabecular bone, and comparing it with the preset percentage of unobstructed trabecular bone, the percentage deviation is obtained and compared with the set maximum deviation reference value: If the percentage deviation is greater than the set maximum deviation reference value, the arithmetic mean of the original detection window width and the dynamic expansion width will be used as the new detection window width, and the arithmetic mean of the edge detection gradient and the gradient reduction threshold will be used as the new edge detection gradient. If the percentage deviation is not greater than the set maximum deviation reference value, the dynamically expanded width will be used as the new detection window width, and the gradient reduction value will be used as the new edge detection gradient. Based on the updated detection window width and edge detection gradient, the distribution range is redefined and the coordinates of the bone tissue boundary are extracted, and the proportion of unobstructed trabecular bone is re-acquired. If the percentage of unobstructed trabeculae acquired again is still less than the percentage of unobstructed trabeculae, an edge calculation and localization anomaly warning will be issued; otherwise, the corresponding bone structure region will be identified as the target analysis region, and a bone status grading assessment will be performed.

9. The edge computing-based osteoporosis magnetic resonance imaging analysis system as described in claim 8, characterized in that, The specific process for conducting bone status grading assessment is as follows: If the feature parameter matching degree of the target analysis area is not greater than the preset feature parameter matching degree, the bone status is directly determined based on the matching result and a grading report is output. Otherwise, supplementary feature verification is performed to improve the matching degree between bone structure feature data and osteoporosis status. The feature parameter matching degree is used to quantify the similarity between the bone features of the analysis area and the actual bone status features. The supplementary feature verification process is as follows: if the deviation value of the trabecular spacing is greater than the set maximum spacing deviation, the trabecular spacing parameter is input into the edge feature correction algorithm to output the correction coefficient; otherwise, the arithmetic mean of the deviation values ​​of the trabecular spacing and the cortical thickness is input into the edge feature correction algorithm to output the correction coefficient, and the feature parameters are adjusted according to the correction coefficient before the grading is re-determined. The output grading report specifically involves: calling a preset osteoporosis assessment level from the edge database and simultaneously obtaining the bone cortical thickness detection results in the target analysis area. The preset assessment levels include osteoporosis level 1, osteoporosis level 2, and osteoporosis level 3. The degree of osteoporosis of the specified target corresponding to osteoporosis level 1, osteoporosis level 2, and osteoporosis level 3 increases sequentially. The labeling results of each target analysis area are summarized to generate a graded report on the osteoporosis level and intervention plan for the target analysis area.

10. An edge computing-based method for osteoporosis magnetic resonance imaging analysis, applied to the edge computing-based osteoporosis magnetic resonance imaging analysis system according to any one of claims 1-9, comprising the following steps: Step 1: Perform node adaptability verification on the knee joint MRI image of the target object locally on the edge computing node, and at the same time perform matching analysis between bone status data and corresponding bone structure feature data based on the results of node adaptability verification. Step 2: Based on the preset bone structure feature template, perform edge calculation and real-time positioning of the bone structure region, and at the same time determine whether to perform edge calculation parameter analysis at the edge end; Step 3: Based on the results of the edge assessment, determine whether to conduct a bone condition grading assessment.

Images