Method for identifying multiple orthopedic implants

Abstract

The invention relates to the technical field of orthopedic implant identification and tracing, in particular to a method for identifying multiple orthopedic implants, which comprises the following steps of: S1, constructing an adaptive intelligent identification model of annular code adaptive expansion and multi-dimensional feature fusion, and outputting an identification result; s2, constructing a closed-loop comparison system of a multi-brand coding feature library and a preoperative and postoperative full link, and forming a closed-loop process according to an identification result; s3, through a preliminary screening and secondary verification mechanism, automatically rejecting edge invalid orthopedic implants; according to the method, polar coordinate transformation is performed on the annular coding area of the orthopedic implant through the adaptive intelligent identification model, and the annular coding is converted into linear characters, so that the method has the effect of adaptive expansion of the annular coding, manual expansion operation is avoided conveniently, and a multi-dimensional feature fusion model is combined to realize adaptive expansion of the annular coding. At least nine implants can be synchronously recognized through a single frame of image, the recognition accuracy can be improved, and the recognition time can be shortened.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of orthopedic implant identification and traceability, in particular to a method for identifying multiple orthopedic implants. BACKGROUND

[0002] An orthopedic implant is a medical device used to repair or enhance the human skeleton or joints. Orthopedic implants are usually made of metal, ceramic, polymer materials or other biocompatible materials. Trauma-related orthopedic implants include bone plates, hollow screws, intramedullary nails, etc. During surgery, orthopedic implants (such as nails) implanted in the human body need to be tracked throughout their life cycle. During surgery, different companies' implants are placed in orthopedic implant carrying boxes. Each implant has a unique microcode for long-term tracking of postoperative implant usage, and it is necessary to accurately record which implants are used.

[0003] In the prior art, for example, Chinese Patent No. CN202411117094.0 discloses a method and system for identifying and detecting orthopedic instruments. The method for identifying and detecting orthopedic instruments includes: obtaining image information, which includes an initial image and a final image; identifying the category of the image information and assigning a corresponding identification and detection model to the image information based on the category, which can be used to identify the target features within the image information; establishing a data set A including the target features of the initial image and a data set B including the target features of the final image, comparing the data set A and the data set B and generating a difference set C, and generating orthopedic instrument usage data based on the difference set C. The present application can effectively improve the problems of poor instrument identification effect and difficulty in accurately obtaining specific orthopedic instrument data in the identification process of existing orthopedic implant instruments.

[0004] Although the prior art has the above advantages, the prior art has the following disadvantages: the coding rules of orthopedic implants from different manufacturers are not unified, the batch number, character format and brand logo differ, the existing identification technology relies on manual coding style differentiation by region, which is prone to identification deviation, the batch number, quantity and type of orthopedic implants need to be captured by a camera and manually checked and confirmed, postoperative images need to be manually compared with preoperative images, the carrying box images need to be screened for qualified certificates of the used materials and matched with the State Drug Charge Code, resulting in low efficiency and human error, in addition, the orthopedic implant coding, qualified certificate data and State Drug Charge Code do not form an automatic association link, and a traceability report that meets the policy requirements cannot be quickly generated.

[0005] In summary, developing a method for identifying multiple orthopedic implants is still a key problem that needs to be solved in the field of orthopedic implant identification and traceability. SUMMARY

[0006] The purpose of the present application is to solve the problem that the coding rules of orthopedic implants of different manufacturers are not unified, the batch number, character format and brand logo are different, the existing recognition technology relies on manual coding style differentiation, and recognition deviation is easy to occur, the batch number, quantity and type of orthopedic implants need to be confirmed by camera shooting and manual checking, postoperative manual comparison of preoperative and postoperative bearing box images, screening of qualified certificates of materials used and matching of China medicine charging codes, resulting in low efficiency and human error, in addition, the coding of orthopedic implants, qualified certificate data and China medicine charging code are not automatically associated, and a traceability report meeting the policy requirements cannot be quickly generated.

[0007] To achieve the above purpose, the present application provides the following technical solutions: The present application provides a method for recognizing a plurality of orthopedic implants, comprising the following steps: S1, constructing a self-adaptive intelligent recognition model of ring coding self-adaptive expansion and multi-dimensional feature fusion, and outputting a recognition result; S2, constructing a multi-brand coding feature library and a preoperative and postoperative closed-loop comparison system, and forming a closed-loop process according to the recognition result; S3, automatically removing edge invalid orthopedic implants through a preliminary screening + secondary verification mechanism.

[0008] Further, in step S1, the method for constructing a self-adaptive intelligent recognition model of ring coding self-adaptive expansion and multi-dimensional feature fusion and outputting a recognition result is: Wherein, the ring coding self-adaptive expansion adopts a polar coordinate transformation algorithm, performs geometric correction on the orthopedic implant image collected by the camera, maps the small coding area distributed in a ring in the image to a polar coordinate space, and the pixel point coordinates in the Cartesian coordinate system are , and the image center is , and the polar coordinate transformation formula is: In the formula, is the radial distance, is the polar angle, is an angle compensation factor obtained by fitting the center offset of the ring coding by the least square method, and the ring text is converted into a linearly recognizable area by angle compensation and pixel rearrangement, and linear coding text information is obtained; the threshold value of coding character definition is set to be gray value variance ≥ 50, and when the coding character definition in the linear area after expansion is lower than the threshold value, adaptive histogram equalization is triggered to optimize the coding character definition.

[0009] Further, in step S1, the method for constructing a self-adaptive intelligent recognition model of ring coding self-adaptive expansion and multi-dimensional feature fusion and outputting a recognition result is: In addition, multi-dimensional feature fusion is performed based on the improved SwinTransformer as a basis network to construct an adaptive intelligent recognition model fusing three types of features, including appearance features, coded character features and barcode features; the appearance features include the size of the orthopedic implant, i.e., diameter and length, the shape, i.e., head morphology and thread structure, and the batch number position, i.e., side, top, single side and multiple sides; the coded character features include the linear coded character information; and the barcode features include the product barcode data of the orthopedic implant in the outer package extracted through image recognition.

[0010] Further, in step S1, an adaptive intelligent recognition model of ring-shaped coding adaptive unfolding and multi-dimensional feature fusion is constructed, and the method for outputting the recognition result is as follows: It should be noted that the feature fusion layer of the adaptive intelligent recognition model inputs the three types of features, and the attention mechanism is used to assign the weights of the features, i.e., the coded character feature weight is 0.5, the appearance feature weight is 0.3, and the barcode feature weight is 0.2, and the expression is as follows: In the formula, is the fused feature vector, and the subscript fusion represents fusion, is the weight coefficient of the i-th type of feature, is the original feature vector of the i-th type of feature, is the attention score function, is the dot product operation of the feature vector and the transpose of the query vector, is the transpose symbol, is the query vector, is the feature dimension, and the initial weight is fixed during training: is the coded character feature weight, is the appearance feature weight, is the barcode feature weight, and the weight is updated through gradient descent optimization: In the formula, represents the updated weight of the i-th type of feature after the j-th iteration, and represents the current weight of the i-th type of feature during the j-th iteration, is the partial derivative symbol, is the learning rate, is the number of categories, is the loss function, is the true label, To predict the probability, the adaptive intelligent recognition model is trained to achieve the recognition result extraction of multiple different types and different brands of orthopedic implants in a single frame of image, and the recognition result includes the model number, batch number, brand and size of the orthopedic implant.

[0011] Further, in step S2, a multi-brand coding feature library and a preoperative and postoperative full-link closed-loop comparison system are constructed, and the method for forming a closed-loop process according to the recognition result is: It should be noted that the multi-brand coding feature library is constructed to record the coding rule data of at least 20 major orthopedic implant manufacturers, including: the batch number and character format of the spinal nail, trauma nail, plate and rod type orthopedic implants of each manufacturer; the brand coding prefix of each manufacturer; the orthopedic implant specifications and qualification standards corresponding to each brand coding; the multi-brand coding feature library supports online updating, and when a new manufacturer orthopedic implant is added, the recognition result is extracted by the adaptive intelligent recognition model , the Mahalanobis distance with the existing records in the library is calculated: In the formula, is the Mahalanobis distance for measuring the distance between two data points, represents the feature record of the new orthopedic implant, represents the standard orthopedic implant feature record already in the multi-brand coding feature library, represents the mean vector of the same type of implant feature record in the multi-brand coding feature library, is the mean value, and when, it is determined as a new record and automatically entered into the multi-brand coding feature library, is the threshold value of 95% confidence interval critical value.

[0012] Further, in step S2, a multi-brand coding feature library and a preoperative and postoperative full-link closed-loop comparison system are constructed, and the method for forming a closed-loop process according to the recognition result is: In addition, the closed-loop comparison system, in the preoperative stage, captures the image of the orthopedic implant carrying box through the camera, inputs the adaptive intelligent recognition model to extract the orthopedic implant information , wherein represents a set of preoperative orthopedic implant information, and the subscript pre represents preoperative, is the model number of the orthopedic implant, is the batch number of the orthopedic implant, is the brand of the orthopedic implant, is the size of the orthopedic implant, and is matched with the multi-brand coding feature library to confirm the brand and size, and is associated with the silicone template parameters of the customized round nail box and the data of the accompanying certificate of conformity, to generate a preoperative data account containing orthopedic implant information-certificate-silicone template parameters-national drug charging code; In the intraoperative stage, the bearing box image is photographed in real time, and the current implant information is extracted , and the preoperative image is compared by dynamic time warping, and the expression is: In the formula, indicates a dynamic time warping function, is the total length of the aligned sequence, indicates the intraoperative orthopedic implant information sequence, and the subscript intra indicates intraoperative, indicates the alignment path , the minimum value is found, is the time alignment path, and when , the extracted orthopedic implant information is marked, is a threshold value.

[0013] Further, in step S2, a multi-brand coding feature library and a preoperative and postoperative full-link closed-loop comparison system are constructed, and the method for forming a closed-loop process according to the identification result is: It should be noted that in the postoperative stage, the postoperative bearing box image is photographed, and the remaining orthopedic implant information is extracted by the adaptive intelligent recognition model , and the difference set with the preoperative account is calculated: In the formula, indicates a set of used orthopedic implant information, and the subscript used indicates that it has been used, indicates a set of postoperative remaining orthopedic implant information, and the subscript post indicates postoperative, the used orthopedic implant is automatically screened out, the corresponding qualified certificate and the Sinopharm charge code are matched, and the orthopedic implant use traceability report form meeting the national traceability requirements is generated.

[0014] Further, in step S3, the method for automatically removing edge invalid orthopedic implants by a preliminary screening + secondary verification mechanism is: Wherein, the preliminary screening, when the intelligent recognition model extracts features, a feature missing rate threshold value ≤ 30% is set, when the orthopedic implant is located at the edge of the image, resulting in missing appearance features or coding features, the orthopedic implant feature missing rate : In the formula, is a counting function, is a missing feature set, and the subscript miss indicates missing, is a total feature set, and the subscript total indicates overview, and when , it is marked as an edge invalid orthopedic implant.

[0015] Further, in step S3, the method for automatically removing invalid orthopedic implants on the edge through the preliminary screening + secondary verification mechanism is: It should be noted that when the orthopedic implant structure is damaged or the code is worn, the similarity of the structure corresponding to the standard form of the type in the multi-brand code feature library is compared, and the expression is: In the formula, indicates the structure similarity index, is a constant term of brightness comparison, is a constant term of covariance comparison, is the mean of the standard image to be detected, is the gray variance of the detected implant image and the standard implant image, is the gray covariance of the detected image and the standard image, is a constant, and when , it is marked as a defective orthopedic implant.

[0016] Further, in step S3, the method for automatically removing invalid orthopedic implants on the edge through the preliminary screening + secondary verification mechanism is: In addition, the secondary verification, the closed-loop comparison system verifies the parameter consistency of the orthopedic implant information after the preliminary screening with the silica gel template and the qualified certificate through the chi-square test, and the expression is: In the formula, indicates the chi-square statistic, indicates the measured value of the th parameter, indicates the standard value of the th parameter, is the total number of parameters participating in verification; when , it is determined that the parameters do not match, and the invalid orthopedic implants with parameter mismatch are removed, and only the effective orthopedic implant information with complete features and parameter compliance is retained for tracing, is the significance level.

[0017] Advantages Compared with the known prior art, the technical scheme provided by the present application has the following advantages: The adaptive intelligent recognition model of the present application first performs polar coordinate transformation on the annular code area of the orthopedic implant, converts the annular code into linear text, so that the present application has the effect of annular code adaptive expansion, which is convenient for avoiding manual expansion operation, and the multi-dimensional feature fusion model is combined, and at least 9 implants can be recognized synchronously in a single frame image, which is beneficial to improve the recognition accuracy and shorten the recognition time.

[0018] The application improves the multi-brand coding feature library to adapt to the coding rules of different manufacturers, without manually distinguishing the coding styles, solves the problem of identification confusion of multi-brand implants, automatically generates a traceability report, saves postoperative data processing time, and meets the policy requirements of the state for long-term tracking of orthopedic implants. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flowchart of a method for identifying multiple orthopedic implants. DETAILED DESCRIPTION

[0020] In order for those skilled in the art to better understand the application scheme, the technical solutions in the embodiments of the application will be clearly and completely described below in combination with the drawings in the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, not all. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the application.

[0021] It should be noted that the terms "first", "second", etc. in the specification and claims of the application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not necessarily limit to those steps or units clearly listed, but includes other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0022] The application will be further described in detail below in combination with the drawings: Embodiments: As shown in the drawings, the application provides a method for identifying multiple orthopedic implants, comprising the following steps: Figure 1 S1, constructing a ring coding adaptive expansion and multi-dimensional feature fusion adaptive intelligent identification model, and outputting an identification result; Further, in step S1, the method for constructing a ring coding adaptive expansion and multi-dimensional feature fusion adaptive intelligent identification model and outputting an identification result is: Wherein, the ring coding adaptive expansion adopts a polar coordinate transformation algorithm, performs geometric correction on the orthopedic implant images collected by the camera, maps the ring-distributed micro coding area in the image to the polar coordinate space, and the pixel point coordinates in the Cartesian coordinate system are ​, the image center is The polar coordinate transformation formula is: In the formula, is the radial distance, is the polar angle, is the angle compensation factor, which is obtained by fitting the center offset of the annular code by the least square method, and the annular text is converted into a linearly recognizable area by angle compensation and pixel rearrangement to obtain linearly encoded text information; the threshold of the encoding character definition is set to be a gray value variance ≥ 50, and when the encoding character definition of the unfolded linear area is lower than the threshold, the adaptive histogram equalization is triggered to optimize the encoding character definition.

[0023] Further, in step S1, an adaptive intelligent recognition model of annular code adaptive unfolding and multi-dimensional feature fusion is constructed, and the method for outputting the recognition result is: In addition, multi-dimensional feature fusion is used to construct an adaptive intelligent recognition model based on an improved SwinTransformer as a basic network, which fuses three types of features, including appearance features, encoding character features and barcode features; the appearance features: extract the size of the orthopedic implant: diameter, length, shape: head shape, thread structure, batch number position: side, top, single side and multiple sides; the encoding character features: extract the linearly encoded text information; the barcode features: extract the product barcode data of the orthopedic implant by image recognition.

[0024] Further, in step S1, an adaptive intelligent recognition model of annular code adaptive unfolding and multi-dimensional feature fusion is constructed, and the method for outputting the recognition result is: It should be noted that the three types of features are input into the feature fusion layer of the adaptive intelligent recognition model, and an attention mechanism is used to allocate the weights of each feature: the encoding character feature weight is 0.5, the appearance feature weight is 0.3, and the barcode feature weight is 0.2, and the expression is: In the formula, is the fusion feature vector, and the subscript fusion represents fusion, is the weight coefficient of the type feature, is the original feature vector of the type feature, is the attention score function, is the dot product operation of the feature vector and the transpose of the query vector, is the transpose symbol, is the query vector, is the feature dimension, and the initial weight is fixed during training: for encoding character feature weight, for appearance feature weight, for barcode feature weight, the weight update is optimized by gradient descent: wherein, indicates the class feature updated weight after the class feature current weight at the is the partial derivative symbol, is the learning rate, is the number of categories, is the loss function, is the true label, is the predicted probability, the adaptive intelligent recognition model is trained to achieve the recognition result extraction of multiple different types and different brands of orthopedic implants in a single frame of image, and the recognition result includes the model number, batch number, brand and size of the orthopedic implant.

[0025] In this embodiment, in the hospital orthopedic operating room, the implant carrying box is photographed in real time by the high-definition camera beside the operating table. When a certain brand of spinal nail is photographed, the adaptive intelligent recognition model first performs polar coordinate transformation on the head annular batch number area of the spinal nail, converts the annular batch number to linear text, and automatically starts the histogram equalization processing because the initial gray variance is 42 (lower than 50), so that the batch number characters are clear and identifiable. Then, the appearance features (diameter 6mm, length 30mm, thread structure) of the spinal nail, the linearized batch number characters (“20240615-02”), and the outer packaging barcode data (“6931234567890”) are extracted. Through the attention mechanism, the encoding character feature is given the highest weight, and after fusion, the model number is accurately identified as “SS-302”, the brand is “Johnson & Johnson”, and the batch number and size are consistent with the extracted information. It is convenient to solve the recognition problem caused by the geometric shape of the annular code, and to improve the success rate of micro-code recognition through polar coordinate transformation and adaptive clarity optimization. Through dimension feature fusion combined with dynamic weight optimization, the core role of the encoding character and the auxiliary verification of the appearance and barcode are taken into account, which is convenient to improve the single-frame recognition accuracy and recognition speed of multiple brands and multiple types of orthopedic implants, and is conducive to reducing the artificial checking time in the operation.

[0026] S2, a multi-brand encoding feature library and a preoperative and postoperative full-link closed-loop comparison system are constructed, and a closed-loop process is formed according to the recognition result; Further, in step S2, a multi-brand encoding feature library and a preoperative and postoperative full-link closed-loop comparison system are constructed, and a method for forming a closed-loop process according to the recognition result is: It should be noted that the multi-brand coding feature library is constructed, and coding rule data of at least 20 mainstream orthopedic implant manufacturers is collected, including: the batch number digit and character format of each manufacturer's spinal nail, trauma nail, plate and rod type orthopedic implant; the brand coding prefix of each manufacturer; the orthopedic implant specification and qualification standard corresponding to each brand code; the multi-brand coding feature library supports online updating, and when a new manufacturer's orthopedic implant is added, the adaptive intelligent recognition model is used to extract the recognition result , calculate the Mahalanobis distance with the existing record in the library: In the formula, is the Mahalanobis distance used to measure the distance between two data points, represents the feature record of the newly added orthopedic implant, represents the existing standard orthopedic implant feature record in the multi-brand coding feature library, represents the mean vector of the same type of implant feature record in the multi-brand coding feature library, is the mean value, and when, it is determined as a new record and automatically entered into the multi-brand coding feature library, is the threshold value, which is the 95% confidence interval critical value.

[0027] Further, in step S2, a multi-brand coding feature library and a preoperative and postoperative full-link closed-loop comparison system are constructed, and the method for forming a closed-loop process according to the recognition result is: In addition, the closed-loop comparison system, in the preoperative stage, captures the image of the orthopedic implant carrying box through the camera, inputs the adaptive intelligent recognition model to extract the orthopedic implant information , wherein represents a set of preoperative orthopedic implant information, and the subscript pre represents preoperative, is the model of the orthopedic implant, is the batch number of the orthopedic implant, is the brand of the orthopedic implant, is the size of the orthopedic implant, and the brand and specification are matched with the multi-brand coding feature library, and the silicone template parameters of the customized round nail box and the data of the accompanying certificate of conformity are associated to generate a preoperative data account containing orthopedic implant information-certificate-silicone template parameters-national drug charging code; In the intraoperative stage, the carrying box image is captured in real time to extract the current implant information , and the preoperative image is compared by dynamic time warping, and the expression is: In the formula, represents the dynamic time warping function, is the total length of the aligned sequence, represents the intraoperative orthopedic implant information sequence, and subscript intra represents intraoperative, represents the alignment path minimizing, is the time alignment path, and is the extracted orthopedic implant information, is a threshold value.

[0028] Further, in step S2, a multi-brand coding feature library and a preoperative and postoperative full-link closed-loop comparison system are constructed, and a method for forming a closed-loop process according to the identification result is: It should be noted that in the postoperative stage, a postoperative carrying box image is shot, and the remaining orthopedic implant information is extracted by the adaptive intelligent recognition model , and the difference set with the preoperative account is calculated: In the formula, represents the used orthopedic implant information set, and subscript used represents used, represents the postoperative remaining orthopedic implant information set, and subscript post represents postoperative. The used orthopedic implant is automatically screened out, the corresponding qualified certificate and the national drug charge code are matched, and the orthopedic implant use traceability report form meeting the national traceability requirements is generated.

[0029] In this embodiment, before the hospital orthopedic department carries out a spine surgery, a nurse uses a special camera to shoot an implant carrying box, automatically extracts three types of implant information (such as “Johnson & Johnson SS-302 spinal nail, batch number 20240615-02, diameter 6 mm”), matches and confirms the specifications with the multi-brand feature library, and associates the silica gel template parameters (pore diameter 6.2 mm), the certificate number and the national drug charge code to generate a preoperative account; during the operation, the doctor takes out one spinal nail at a time, shoots the remaining implant in real time, compares the preoperative sequence through the DTW algorithm, and quickly marks “2 SS-302 have been taken out”; after the operation, the camera shoots the carrying box of the remaining one spinal nail, confirms “2 have been used” through the difference set operation, automatically matches the corresponding qualified certificate information, generates a traceability report containing the batch number, brand and use quantity, and synchronously uploads it to the hospital management system. Through the multi-brand feature library covering the mainstream manufacturers and supporting automatic updating, the problems of low efficiency and incomplete brand coverage of traditional manual input are solved, which is convenient for shortening the response time of new manufacturer orthopedic implants in the warehouse, realizing full-process automation of the closed-loop comparison of preoperative-intraoperative-postoperative, is conducive to improving the intraoperative marking accuracy, shortening the postoperative traceability report generation time, and greatly reducing the non-surgical working time of medical staff.

[0030] S3, through the preliminary screening + secondary verification mechanism, automatically remove the edge invalid orthopedic implants; Further, in step S3, the method for automatically removing edge invalid orthopedic implants through a preliminary screening + secondary verification mechanism is: Wherein, in the preliminary screening, the intelligent recognition model sets a feature missing rate threshold of ≤ 30% when extracting features, and calculates the orthopedic implant feature missing rate when the orthopedic implant is located at the edge of the image, resulting in missing appearance features or coding features : In the formula, is a counting function, is a missing feature set, and the subscript miss represents missing, is a total feature set, and the subscript total represents a total overview, and when, it is marked as an edge invalid orthopedic implant.

[0031] Further, in step S3, the method for automatically removing edge invalid orthopedic implants through a preliminary screening + secondary verification mechanism is: It should be noted that when the orthopedic implant structure is damaged or the code is worn, the structural similarity of the corresponding type standard form in the multi-brand coding feature library is expressed as: In the formula, represents a structural similarity index, is a constant term for brightness comparison, is a constant term for covariance comparison, is the mean of the standard image to be detected, is the gray variance of the detected implant image and the standard implant image, is the gray covariance of the detected image and the standard image, is a constant, and when, it is marked as a defective orthopedic implant.

[0032] Further, in step S3, the method for automatically removing edge invalid orthopedic implants through a preliminary screening + secondary verification mechanism is: In addition, in the secondary verification, the closed-loop comparison system verifies the parameter consistency of the orthopedic implant information after the preliminary screening with the silica gel template and the certificate of conformity through a chi-square test, and the expression is: In the formula, represents the chi-square statistic, represents the measured value of the th parameter, represents the standard value of the th parameter, is the total number of parameters participating in the verification; and When the parameters are determined to be mismatched, the invalid orthopedic implants with mismatched parameters are removed, and only the valid orthopedic implant information with complete features and compliant parameters is reserved for tracing, is a significance level.

[0033] In the embodiment, when the hospital orthopedics department sorts the orthopedic implant information after the spine surgery, the automatic screening of the bearing box image is performed: first, one trauma nail is found to be located at the edge of the image, only two features (total features 7) of diameter and brand are extracted, the feature missing rate is about 71% (more than 30%), and it is marked as invalid edge; second, one plate rod type orthopedic implant has serious code wear, and the SSIM value with the standard image of the same type in the feature library is 0.6 (lower than the threshold value 0.7), which is marked as invalid defect; third, one spine nail passes the preliminary screening, but the actual diameter 5.8mm (standard value 5.0mm) is found in the secondary verification, and the chi-square value calculated by chi-square test exceeds the critical value corresponding to the significance level 0.05, which is determined to be parameter mismatch and removed. Finally, among the original 10 orthopedic implants to be traced, only 7 orthopedic implant information with complete features and compliant parameters enters the tracing process; The double-layer screening mechanism realizes the full automation of removing invalid implants, avoids the subjective error of manual screening, is conducive to improving the accuracy of removing invalid information, accurately retains valid information, reduces the processing amount of redundant data in subsequent tracing reports, facilitates improving the efficiency of the tracing process, and reduces the data sorting burden of medical staff.

[0034] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements will not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method of identifying a plurality of orthopedic implants, the method comprising: The method comprises the following steps: S1, constructing a ring code adaptive expansion and multi-dimensional feature fusion adaptive intelligent recognition model, and outputting a recognition result; S2, constructing a multi-brand code feature library and a preoperative and postoperative full-link closed-loop comparison system, and forming a closed-loop process according to the recognition result; S3, automatically removing edge invalid orthopedic implants through a preliminary screening + secondary verification mechanism.

2. The method of identifying a plurality of orthopedic implants of claim 1, wherein, In step S1, the method for constructing a ring code adaptive expansion and multi-dimensional feature fusion adaptive intelligent recognition model and outputting a recognition result is: Wherein, the annular coding is self-adaptive expansion, and a polar coordinate transformation algorithm is adopted to perform geometric correction on the orthopedic implant image collected by the camera, so as to map the annular distributed micro coding area in the image into a polar coordinate space, and the pixel point coordinates in the Cartesian coordinate system are , the image center is , and the polar coordinate transformation formula is: In the formula, is a radial distance, is a polar angle, is an angle compensation factor obtained by least square fitting of the ring-encoding circle center offset, the ring text is converted into a linearly recognizable area by angle compensation and pixel rearrangement, and linear encoding text information is obtained; the threshold of encoding character definition is set as a gray value variance ≥ 50, and when the encoding character definition of the linear area after expansion is lower than the threshold, adaptive histogram equalization is triggered to optimize the encoding character definition.

3. The method of identifying a plurality of orthopedic implants of claim 2, wherein, In step S1, the method for constructing a ring code adaptive expansion and multi-dimensional feature fusion adaptive intelligent recognition model and outputting a recognition result is: In addition, multi-dimensional feature fusion is used to construct an adaptive intelligent recognition model based on an improved SwinTransformer as a basic network, and three types of features including appearance features, code character features and barcode features are fused. The appearance features include the size of the orthopedic implant, the diameter and length, the shape, the head morphology and the thread structure, and the batch number position, the side, the top, the single side and the multiple sides.

4. The method of identifying a plurality of orthopedic implants of claim 3, wherein, The code character features include the linear code character information. The barcode features include the product barcode data of the orthopedic implant on the outer package. wherein, is the fusion feature vector, subscript fusion represents fusion, is the weight coefficient of the class feature, is the original feature vector of the class feature, is the attention score function, is the dot product operation of the feature vector and the query vector transpose, is the transpose symbol, is the query vector, is the feature dimension, the initial weight is fixed during training: is the encoding character feature weight, is the appearance feature weight, is the barcode feature weight, and the weight update is optimized by gradient descent: wherein, denotes the class feature updated weight after the class feature current weight at the is the partial derivative symbol, is the learning rate, is the number of classes, is the loss function, is the true label, is the predicted probability, the adaptive intelligent recognition model is trained to achieve the recognition result extraction of multiple different types and different brands of orthopedic implants in a single frame of image as the target, and the recognition result includes the orthopedic implant model, batch number, brand and size.

5. The method of identifying a plurality of orthopedic implants of claim 4, wherein, In step S1, the method for constructing a ring code adaptive expansion and multi-dimensional feature fusion adaptive intelligent recognition model and outputting a recognition result is: It should be noted that the multi-brand coding feature library is constructed, and coding rule data of at least 20 mainstream orthopedic implant manufacturers is collected, including: the batch number digit number and character format of the spinal nail, trauma nail, plate and rod type orthopedic implant of each manufacturer; the brand coding prefix of each manufacturer; the orthopedic implant specification and qualified standard corresponding to each brand coding; the multi-brand coding feature library supports online updating, and when a new manufacturer orthopedic implant is added, the adaptive intelligent recognition model is used to extract and recognize the result , the Mahalanobis distance with the existing record in the library is calculated: wherein, is the Mahalanobis distance for measuring the distance between two data points, represents the feature record of a new orthopedic implant, represents the existing standard orthopedic implant feature record in the multi-brand coded feature library, represents the mean vector of the same type of implant feature records in the multi-brand coded feature library, is the mean value, and when is the threshold value, which is the 95% confidence interval critical value.

6. The method of identifying a plurality of orthopedic implants of claim 5, wherein, It should be noted that the feature fusion layer of the adaptive intelligent recognition model inputs the three types of features, and uses an attention mechanism to assign the weights of the features, that is, the code character feature weight is 0.5, the appearance feature weight is 0.3, and the barcode feature weight is 0.2, and the expression is: Further, the closed-loop comparison system captures an image of an orthopedic implant carrying box through a camera at a preoperative stage, inputs the adaptive intelligent recognition model to extract orthopedic implant information wherein, represents a preoperative orthopedic implant information set, and the subscript pre represents preoperative, is a model of the orthopedic implant, is a batch number of the orthopedic implant, is a brand of the orthopedic implant, is a size of the orthopedic implant, and the brand and the specification are matched and confirmed with the multi-brand coding feature library, and the silicone template parameters of the customized round nail box and the data of the cargo certificate of conformity are associated to generate a preoperative data account containing orthopedic implant information-certificate-silicone template parameters-national drug charging code; At the intraoperative stage, the bearing box image is taken in real time, and the current implant information is extracted By comparing the preoperative image through dynamic time warping, expression: wherein denotes a dynamic time warping function, is the total length of the aligned sequence, denotes the intraoperative orthopaedic implant information sequence, the subscript intra denotes intraoperative, denotes the alignment path is minimized, is the time alignment path, and in the orthopaedic implant information is marked as extracted, is a threshold value.

7. The method of identifying a plurality of orthopedic implants of claim 6, wherein, In step S2, the method for constructing a multi-brand code feature library and a preoperative and postoperative full-link closed-loop comparison system and forming a closed-loop process according to the recognition result is: It should be noted that in the postoperative phase, the postoperative carrying box image is shot, and the remaining orthopedic implant information is extracted through the adaptive intelligent recognition model , and the difference set with the preoperative account is calculated: In the formula, represents a used orthopedic implant information set, and the subscript used represents used, represents a postoperative remaining orthopedic implant information set, the subscript post represents postoperative, the used orthopedic implants are automatically screened out, the corresponding qualified certificates and the sino-pharm charge codes are matched, and the orthopedic implant use traceability report form meeting the national traceability requirements is generated.

8. The method of identifying a plurality of orthopedic implants of claim 7, wherein, In step S2, the method for constructing a multi-brand code feature library and a preoperative and postoperative full-link closed-loop comparison system and forming a closed-loop process according to the recognition result is: In the preliminary screening, the intelligent recognition model sets a feature missing rate threshold of ≤ 30% when extracting features, and calculates the orthopedic implant feature missing rate when the orthopedic implant is located at the edge of the image, resulting in missing appearance features or coding features : wherein is a counting function, is a set of missing features, subscript miss denotes missing, is a set of total features, subscript total denotes total, at the marker is an edge invalid orthopedic implant.

9. The method of identifying a plurality of orthopedic implants of claim 8, wherein, In step S2, the method for constructing a multi-brand code feature library and a preoperative and postoperative full-link closed-loop comparison system and forming a closed-loop process according to the recognition result is: In step S3, the method for automatically removing edge invalid orthopedic implants through a preliminary screening + secondary verification mechanism is: wherein, represents a structural similarity index, is a constant term of the luminance ratio, is a constant term of the covariance ratio, is a mean value of the standard image, is a gray scale variance of the standard implant image, is a gray scale covariance of the standard image, is a constant, and when is marked as a defective orthopedic implant.

10. The method of identifying a plurality of orthopedic implants of claim 8, wherein, In step S3, the method for automatically removing edge invalid orthopedic implants through a preliminary screening + secondary verification mechanism is: It should be noted that when the structure of the orthopedic implant is damaged or the code is worn, the structural similarity of the corresponding model standard form in the multi-brand code feature library is compared, and the expression is: In step S3, the method for automatically removing edge invalid orthopedic implants through a preliminary screening + secondary verification mechanism is: In addition, the secondary verification is performed, and the closed-loop comparison system verifies the parameter consistency of the orthopedic implant information after the preliminary screening with the silica gel template and the certificate of conformity through a chi-square test, and the expression is: In the formula, χ2represents a chi-square statistic, represents the measured value of the th parameter, represents the standard value of the th parameter, is the total number of parameters participating in the verification; when , it is determined that the parameters do not match, the invalid orthopedic implant with parameter mismatch is removed, and only the valid orthopedic implant information with complete features and compliant parameters is retained for tracing, is the significance level.

Images