Intelligent intensive blood sampling management system based on GPU (Graphics Processing Unit)

Abstract

The invention discloses an intelligent intensive blood sampling management system based on a GPU, and belongs to the technical field of blood sampling management. The intelligent intensive blood sampling management system based on the GPU comprises a data acquisition unit and a data analysis unit. The problem that efficient management of the intensive blood sampling process is difficult to achieve in the prior art is solved, bar code and two-dimensional code information in the sample image can be rapidly and accurately recognized through the deep learning algorithm based on GPU acceleration, the sample information extraction speed and accuracy are improved, and the efficiency of blood sampling is improved. The trained deep learning model is used for analyzing the sample images, blood types and detection items can be automatically identified, errors caused by manual judgment are effectively avoided, samples are accurately classified according to the analysis result, the sample processing efficiency is improved, an automatic sorting device is accurately controlled according to the classification result, and the sorting efficiency is improved. And the sample test tubes are conveyed to designated positions, so that manual intervention is reduced, and meanwhile, the sorting accuracy and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of blood collection management, and in particular to a GPU-based intelligent intensive blood collection management system. Background Art

[0002] In modern healthcare, blood collection requires not only high efficiency but also accurate sample information and refined management. In traditional blood collection management processes, post-blood collection tube verification, classification, and sorting are largely manual tasks. This traditional blood collection management process, coupled with the large number of blood tubes and slow manual verification and sorting, can lead to long patient wait times and make efficient management of the intensive blood collection process difficult. Therefore, this process does not meet existing needs. Therefore, we propose a GPU-based intelligent intensive blood collection management system. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent intensive blood collection management system based on GPU. By collecting sample images through a high-resolution camera and utilizing a deep learning algorithm accelerated by GPU, it can accurately identify the barcode and QR code information in the sample image, automatically identify the blood type and test items, and accurately classify the samples according to the analysis results, and then control the automated sorting device to transfer the sample tubes to the designated location, thereby improving the efficiency and accuracy of blood collection management and solving the problems raised in the above-mentioned background technology.

[0004] To achieve the above objectives, the present invention provides the following technical solutions: a GPU-based intelligent intensive blood sampling management system, the system comprising a data acquisition unit and a data analysis unit, the data acquisition unit being configured to use a high-resolution camera to capture images of test tubes after blood sampling to obtain sample images, the data analysis unit comprising an information recognition module, a sample analysis module, a sample classification module, and a sorting control module;

[0005] The information recognition module is configured as a deep learning algorithm based on GPU acceleration to recognize the barcode and QR code information in the acquired sample image and extract the sample information;

[0006] The sample analysis module is configured to analyze the sample image using a trained deep learning model to automatically identify the blood type and test items;

[0007] The sample classification module is configured to classify the samples according to the analysis results;

[0008] The sorting control module is configured to control the automated sorting device according to the classification results to transfer the sample tubes to the designated locations.

[0009] Furthermore, the data acquisition unit includes:

[0010] An image acquisition module is configured to arrange a high-resolution camera in an image acquisition area, and to acquire an image of the test tube after blood collection through the high-resolution camera to obtain a sample image;

[0011] The image processing module is configured to perform denoising, sharpening, and motion compensation on the acquired sample images, specifically:

[0012] Denoising: using a denoising algorithm to remove noise from the sample image, where the denoising algorithm includes but is not limited to bilateral filtering;

[0013] Sharpening: Improving the detail clarity of the sample image through sharpening methods, where sharpening methods include but are not limited to gradient sharpening or Sobel operator;

[0014] Motion compensation: Use histogram equalization to adjust the brightness and contrast of the sample image.

[0015] Furthermore, the information identification module includes:

[0016] a data collection module configured to collect historical image data containing barcodes and QR codes, and annotate the barcodes and QR codes in the historical image data, wherein the annotation content includes the position and category information of the bounding box;

[0017] The model building module is configured to select a deep learning model suitable for barcode and QR code detection, including YOLO, and train the deep learning model using annotated historical image data. During the training process, GPU support for the deep learning framework is configured to utilize GPU accelerated computing to improve training efficiency.

[0018] The target recognition module is configured to input the acquired sample images into the trained deep learning model. The GPU-accelerated deep learning model is used to locate and decode the barcodes and QR codes in the sample images. Specifically:

[0019] Perform forward propagation on the input sample image, and the deep learning model outputs the location information of the barcode and QR code in the sample image;

[0020] Based on the position information of the barcode and the QR code, the integrity coefficient of the current sample image is obtained, thereby judging the integrity of the position information of the barcode and the QR code;

[0021] The barcode and QR code area with complete detected position information is decoded using a decoding library. The decoding library converts the information in the barcode or QR code into text format for extracting sample information.

[0022] The result extraction module is configured to extract sample information, including sample number and test items, based on the decoded barcode and QR code information.

[0023] Furthermore, the sample analysis module includes:

[0024] A feature extraction module is configured to load a trained deep learning model and input the acquired sample image into the deep learning model. The deep learning model outputs a high-dimensional feature vector, and the output feature vector is used to represent the feature information of the sample image;

[0025] a feature analysis module configured to further analyze the extracted feature vectors through cluster analysis and classification analysis to identify blood types and test items;

[0026] The result output module is configured to integrate the identification results of blood type and test items, generate a complete analysis report and provide it to the user.

[0027] Furthermore, the feature analysis module includes:

[0028] A result verification module is configured to verify the result by comparing it with the manual annotation results and counting the number of matches and mismatches through comparison;

[0029] A result display module is configured to display the comparison between the model recognition results and the manual annotation results in a table format, including matches and mismatches;

[0030] The feedback optimization module is configured to provide feedback on the displayed table and fine-tune or retrain the deep learning model based on the comparison and verification results.

[0031] Furthermore, the sample classification module includes:

[0032] a rule setting module configured to preset rules for sample classification based on blood type and test items, wherein the samples are classified into type A, type B, type AB, and type O based on blood type, and the samples are classified into routine testing and special testing based on test items;

[0033] The rule application module is configured to match the analysis results of each sample according to the preset classification rules and supports multi-level classification, with the first level classification based on blood type and the second level classification based on the test items;

[0034] The result management module is configured to store the classified sample information in a database, including the sample number, classification category and classification time, and allows users to query the sample classification results based on the sample number and classification category.

[0035] Furthermore, the rule application module includes:

[0036] The initial classification module is configured to preset a set of sample classification rules based on the clinical characteristics and treatment direction of the hospital to obtain the initial sample classification rules;

[0037] An information extraction module is configured to extract key features based on user information and doctor diagnosis information in the sample information to obtain key extracted information;

[0038] A comparison auxiliary module is configured to compare the key extraction information with the preset information in the preset information-rule matching table one by one, thereby determining the multi-dimensional sample classification rule that matches the key extraction information and obtaining the auxiliary classification rule;

[0039] A rule optimization module is configured to optimize the initial sample classification rule based on the auxiliary classification rule to obtain the sample classification rule;

[0040] A model building module is configured to mine the item frequency between historical test items based on a preset algorithm, and generate item association rules based on the item frequency, thereby establishing a test association model based on blood test items;

[0041] a correlation synthesis module configured to extract corresponding user information based on the sample information, thereby extracting item information of each blood test item of the user, and then combining the sample information with the input of the test correlation model to determine the item correlation;

[0042] A rule judgment module is configured to judge whether the sample classification rule can satisfy the project relevance;

[0043] If the sample classification rule can satisfy the item relevance, the sample classification rule is determined to be the final detection item classification rule of the sample;

[0044] Otherwise, the rules are optimized in combination with the item correlation to obtain the detection item classification rules.

[0045] Furthermore, the rule application module executes the following process:

[0046] Primary classification: Classify the sample into type A, type B, type AB or type O according to the blood type classification rules;

[0047] Secondary classification: Classify samples into routine testing or special testing according to the classification rules of test items;

[0048] Comprehensive classification: Combine the results of the first-level classification and the second-level classification to form the final classification category of the sample;

[0049] If the result of the first-level classification or second-level classification is an unknown type or unknown test item, the abnormality shall be clearly marked in the final classification category.

[0050] Furthermore, the sorting control module includes:

[0051] a result parsing module configured to parse the classification results into specific sorting instructions and determine the target location to which each sample tube should be transferred;

[0052] The sample sorting module is configured to control the automated sorting device to grab the sample tube according to the sorting instruction, transfer the sample tube from the starting position to the designated position, and monitor the sorting process in real time. If the sample tube breaks or the automated sorting device fails, an alarm is triggered and the sorting operation is suspended;

[0053] The sorting recording module is configured to record the sorting process, including sample information, target location and sorting time, and feed back the recorded sorting information to the user after the sorting is completed.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention uses a GPU-accelerated deep learning algorithm to quickly and accurately identify barcode and QR code information in sample images, greatly improving the speed and accuracy of sample information extraction. It uses a trained deep learning model to analyze sample images, automatically identifying blood types and test items, effectively avoiding errors in human judgment and improving the accuracy of sample analysis. It accurately controls the automated sorting device based on the classification results to transfer sample tubes to designated locations, reducing manual intervention and labor intensity while improving sorting accuracy and efficiency. The above design uses a GPU-accelerated deep learning algorithm to achieve efficient information recognition, sample analysis, classification, and sorting control, thereby improving the efficiency and accuracy of blood collection management. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is the overall structure diagram of the GPU-based intelligent intensive blood sampling management system of the present invention;

[0057] Figure 2 Schematic diagram of the structure of each module of the data analysis unit of the present invention. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0059] In order to solve the technical problem that the traditional blood collection management process, which relies mostly on manual work such as checking, classifying and sorting test tube information after blood collection, is faced with the problem that in the case of intensive blood collection, the number of blood tubes collected is huge, manual checking and sorting are slow, and patients may have to wait too long, making it difficult to achieve efficient management of the intensive blood collection process, please refer to Figure 1-Figure 2 , this embodiment provides the following technical solutions:

[0060] An intelligent intensive blood sampling management system based on a GPU, the system includes a data acquisition unit and a data analysis unit. The data acquisition unit is configured to use a high-resolution camera to capture images of test tubes after blood sampling to obtain sample images. The data analysis unit includes an information recognition module, a sample analysis module, a sample classification module, and a sorting control module.

[0061] The information recognition module is configured as a deep learning algorithm based on GPU acceleration to recognize the barcode and QR code information in the acquired sample image and extract the sample information;

[0062] The sample analysis module is configured to analyze the sample image using a trained deep learning model to automatically identify the blood type and test items;

[0063] The sample classification module is configured to classify the samples according to the analysis results;

[0064] The sorting control module is configured to control the automated sorting device according to the classification results to transfer the sample tubes to the designated locations.

[0065] The technical effects of the above content are as follows: the data acquisition unit acquires sample images through a high-resolution camera, providing a richer data foundation for subsequent information recognition and sample analysis. The information recognition module, based on a GPU-accelerated deep learning algorithm, can quickly identify the barcode and QR code information in the test tube image after blood collection and extract sample information. GPU acceleration significantly improves the image processing speed and can complete the recognition task of a large number of samples in a short time. The sample analysis module uses a trained deep learning model to analyze the sample image to automatically identify the blood type and test items, thereby avoiding the subjectivity and errors of manual judgment. The sample classification module can classify the samples according to the analysis results and can quickly distinguish different types of samples, facilitating subsequent testing and processing, thereby improving the operating efficiency of the entire blood collection management system. Finally, the sorting control module controls the automated sorting device according to the sample classification results to transfer the sample tubes to the designated location, thereby avoiding errors that may be caused by human operation and being able to quickly handle the sorting tasks of a large number of samples.

[0066] Data acquisition unit, including:

[0067] An image acquisition module is configured to arrange a high-resolution camera in the image acquisition area, and to capture an image of the test tube after blood collection through the high-resolution camera to obtain a sample image; the high-resolution camera is capable of capturing high-quality images under different lighting conditions (such as natural light, indoor light, etc.). Even in the case of insufficient ambient light or interference, the high-resolution image can still provide sufficient details to ensure the normal operation of the system;

[0068] The image processing module is configured to perform denoising, sharpening, and motion compensation on the acquired sample images, specifically:

[0069] Denoising: Denoising algorithms are used to remove noise from sample images. These algorithms include, but are not limited to, bilateral filtering. Denoising algorithms can remove noise while preserving image details, making the image clearer and reducing the interference of noise on subsequent information recognition and analysis. This improves the accuracy of the information recognition module in identifying sample information and reduces misjudgments.

[0070] Sharpening: Improving the detail clarity of the sample image through sharpening methods. Sharpening methods include, but are not limited to, gradient sharpening or the Sobel operator. Sharpening can enhance image edges and details, facilitating subsequent recognition and analysis.

[0071] Motion compensation: The histogram equalization method is used to adjust the brightness and contrast of the sample image. The histogram equalization method can automatically adjust the brightness and contrast of the image so that the image can maintain a good visual effect under different lighting conditions.

[0072] The technical effect of the above content is: the high-resolution camera used by the image acquisition module can capture clearer and more delicate image details on the test tube and its label. Compared with the low-resolution camera, the high-resolution image can more accurately display the texture of the barcode and QR code, thereby providing a richer data basis for subsequent information recognition and sample analysis. The sample image processed by the image processing module is clearer and more consistent, so that the information recognition module can recognize barcodes, QR codes and other information more quickly, reducing the recognition time and improving the overall operation efficiency of the system.

[0073] Information identification module, including:

[0074] a data collection module configured to collect historical image data containing barcodes and QR codes, and annotate the barcodes and QR codes in the historical image data, wherein the annotation content includes the position and category information of the bounding box;

[0075] The model building module is configured to select a deep learning model suitable for barcode and QR code detection, including YOLO, and train the deep learning model using annotated historical image data. During the training process, GPU support for the deep learning framework is configured to utilize GPU accelerated computing to improve training efficiency.

[0076] Among them, the deep learning model is able to learn the characteristics of barcodes and QR codes under different image conditions through training with a large amount of labeled data;

[0077] GPUs have powerful parallel computing capabilities and can handle multiple computing tasks simultaneously. GPU acceleration can significantly improve the model's inference speed, enabling the system to complete image forward propagation in a short period of time and quickly locate and decode barcodes and QR codes. This is crucial for improving the system's real-time performance and response speed, especially in intensive blood collection scenarios.

[0078] The target recognition module is configured to input the acquired sample images into the trained deep learning model. The GPU-accelerated deep learning model is used to locate and decode the barcodes and QR codes in the sample images. Specifically:

[0079] Perform forward propagation on the input sample image, and the deep learning model outputs the location information of the barcode and QR code in the sample image;

[0080] Based on the position information of the barcode and the QR code, the integrity coefficient of the current sample image is obtained, thereby judging the integrity of the position information of the barcode and the QR code;

[0081] The barcode and QR code area with complete detected position information is decoded using a decoding library. The decoding library converts the information in the barcode or QR code into text format for extracting sample information.

[0082] The result extraction module is configured to extract sample information, including sample number and test items, based on the decoded barcode and QR code information.

[0083] The technical effects of the above content are: the data collection module collects and labels historical image data containing barcodes and QR codes, which can provide high-quality training data for the deep learning model, so that the model can learn the features of barcodes and QR codes more accurately, thereby improving the recognition accuracy. The model construction module selects a deep learning model suitable for barcode and QR code detection and uses the labeled historical image data for training, which can quickly and accurately locate and identify targets in the image. During the model training process, the parallel computing capability of the GPU can significantly improve the training and inference speed of the deep learning model, so that the system can process a large number of sample images in a short time, thereby improving the operating efficiency of the entire blood collection management system. The target recognition module uses the GPU-accelerated deep learning model to locate and decode the barcodes and QR codes in the sample image, so that the result extraction module can extract sample information, thereby ensuring that the sample information can be extracted quickly and accurately, reducing the patient's waiting time during the blood collection process.

[0084] In this embodiment, the integrity factor is T;

[0085]

[0086] Among them, T is the integrity coefficient of the current sample image, is the pixel integrity indicator, is the structural integrity index, is the content integrity index, α1 is the pixel integrity index weight, α2 is the structure integrity index weight, α3 is the content integrity index weight, M is the total number of pixels in the current sample image, m is the number of abnormal pixels, μ x is the image pixel mean of the current sample image, is the image variance of the current sample image, σ xy is the covariance between the current sample image and the standard image of the corresponding image type, μ y is the standard image pixel mean of the current sample image corresponding to the image type, is the standard image variance of the image type corresponding to the current sample image, C1 and C2 are constants used to avoid the denominator being 0, n is the number of valid semantic elements for semantic extraction from the current sample image, and N is the expected number of semantic elements for the sample image. The sum of the weights of α1, α2, and α3 is 1.

[0087] The technical effect of the above content is: by calculating the integrity coefficient to determine the integrity of the position information in the sample image, it can ensure that only the valid area is processed during processing and analysis, thereby improving processing accuracy and efficiency, thereby improving decoding efficiency, ensuring that the sample information can be extracted quickly and accurately, and reducing the patient's waiting time during the blood collection process.

[0088] In summary, the sample analysis module significantly improves the accuracy of identifying blood types and test items through the dual guarantees of efficient feature extraction, cluster analysis and classification analysis of the deep learning model, as well as fast and accurate result output.

[0089] Sample analysis module, including:

[0090] A feature extraction module is configured to load a trained deep learning model and input the acquired sample image into the deep learning model. The deep learning model outputs a high-dimensional feature vector. The output feature vector is used to represent the characteristic information of the sample image, including details such as blood color, liquid level, and test tube label;

[0091] a feature analysis module configured to further analyze the extracted feature vectors through cluster analysis and classification analysis to identify blood types and test items;

[0092] The result output module is configured to integrate the identification results of blood type and test items, generate a complete analysis report and provide it to the user.

[0093] The technical effects of the above content are: the feature extraction module can extract high-dimensional feature vectors from the sample image by loading the trained deep learning model. The extracted feature vectors can comprehensively and accurately represent the feature information of the sample image. The feature analysis module can classify similar feature vectors into one category through cluster analysis, thereby discovering potential similarities in the sample image. The classification analysis can classify the feature vectors based on the existing category labels and identify specific blood types and test items. The dual analysis method can analyze the sample image from different angles to improve the accuracy and reliability of the recognition results. The result output module can quickly generate an analysis report and provide it to the user, reducing user waiting time.

[0094] Feature analysis module, including:

[0095] The result verification module is configured to verify the model by comparing it with the manual annotation results. By comparing and counting the number of matches and mismatches, the accuracy and error rate of the model can be quantitatively evaluated;

[0096] A result display module is configured to display the comparison between the model recognition results and the manual annotation results in a table format, including matches and mismatches;

[0097] A feedback optimization module is configured to provide feedback on the displayed table and fine-tune or retrain the deep learning model based on the comparison and verification results;

[0098] Among them, fine-tuning can adjust certain parameters of the model to improve the model's ability to recognize specific samples;

[0099] Retraining can use more training data to retrain the model when major problems are found in the model, thereby improving the overall performance of the model.

[0100] The technical effects of the above content are: the result verification module can directly evaluate whether the recognition results of the deep learning model are accurate by comparing and verifying with the manually labeled results, so that it can quickly discover the errors and shortcomings of the model in actual applications, and provide a clear direction for model optimization. The result display module displays the results in tabular form, allowing users to easily analyze the recognition results of the model, so that they can intuitively understand the performance of the model, and the feedback optimization module can continuously optimize according to the data in actual applications, gradually improve the accuracy and reliability of the model, and make it better adapt to the complex situations in actual blood collection management scenarios.

[0101] Sample classification module, including:

[0102] a rule setting module configured to preset rules for sample classification based on blood type and test items, wherein the samples are classified into type A, type B, type AB, and type O based on blood type, and the samples are classified into routine testing and special testing based on test items;

[0103] The rule application module is configured to match the analysis results of each sample according to the preset classification rules and supports multi-level classification, with the first level classification based on blood type and the second level classification based on the test items;

[0104] The result management module is configured to store the classified sample information in a database, including the sample number, classification category and classification time, and allows users to query the sample classification results based on the sample number and classification category.

[0105] The technical effects of the above content are: the rule setting module reduces the ambiguity and uncertainty in the classification process and improves the accuracy of classification by pre-setting sample classification rules. The rule application module can classify samples more finely according to the preset classification rules and supports multi-level classification, thereby improving the degree of refinement of classification. The sample information classified by the result management module is stored in the database, so that the sample classification information is effectively managed and preserved, which is convenient for subsequent query and analysis, and allows users to query the sample classification results according to the sample number and classification category, thereby enhancing the traceability of the system. Users can quickly locate the classification information of specific samples, which is convenient for sample management and tracking, and also provides a traceability basis for possible errors or abnormal situations.

[0106] Rule application module, including:

[0107] The initial classification module is configured to preset a set of sample classification rules based on the clinical characteristics and treatment direction of the hospital to obtain the initial sample classification rules;

[0108] An information extraction module is configured to extract key features based on user information and doctor diagnosis information in the sample information to obtain key extracted information;

[0109] A comparison auxiliary module is configured to compare the key extraction information with the preset information in the preset information-rule matching table one by one, thereby determining the multi-dimensional sample classification rule that matches the key extraction information and obtaining the auxiliary classification rule;

[0110] A rule optimization module is configured to optimize the initial sample classification rule based on the auxiliary classification rule to obtain the sample classification rule;

[0111] A model building module is configured to mine the item frequency between historical test items based on a preset algorithm, and generate item association rules based on the item frequency, thereby establishing a test association model based on blood test items;

[0112] a correlation synthesis module configured to extract corresponding user information based on the sample information, thereby extracting item information of each blood test item of the user, and then combining the sample information with the input of the test correlation model to determine the item correlation;

[0113] A rule judgment module is configured to judge whether the sample classification rule can satisfy the project relevance;

[0114] If the sample classification rule can satisfy the item relevance, the sample classification rule is determined to be the final detection item classification rule of the sample;

[0115] Otherwise, the rules are optimized in combination with the item correlation to obtain the detection item classification rules.

[0116] In this embodiment, the sample classification rules are a series of standards and conditions for classifying samples, helping to divide the samples into different categories.

[0117] In this embodiment, the sample information is various detailed information about the sample, covering multiple aspects such as user information and doctor's diagnosis information, which can be obtained by association after decoding the barcode and QR code. For example, in a blood test sample, the sample information may include the patient's (ie, user's) name, age, gender, contact information, and the doctor's preliminary diagnosis based on the patient's symptoms, such as "the patient has obvious symptoms of polydipsia and polyuria recently, and it is suspected that blood sugar control is poor. It is recommended to conduct blood sugar and glycosylated hemoglobin tests".

[0118] In this embodiment, user information refers to basic personal information related to the sample provider (usually a patient), such as name, age, gender, contact information, etc.

[0119] In this embodiment, the doctor's diagnosis information refers to the doctor's judgment and diagnostic direction on the patient's condition based on the patient's symptoms, physical signs, preliminary examination results, etc. For example, the doctor's diagnosis information may be "The patient reported that he has blurred vision, polydipsia, polyuria, and weight loss recently. The fasting blood glucose test value is 8.5mmol / L (normal range 3.9-6.1mmol / L). The preliminary diagnosis is diabetes, and a comprehensive blood test is recommended."

[0120] In this embodiment, key information extraction refers to screening out key information that has an important impact on sample classification and analysis from the sample information, and removing irrelevant or redundant information. For example, the key features extracted from the sample information may be "50 years old, male, polydipsia and polyuria, blurred vision, fasting blood glucose 8.5mmol / L, diabetes."

[0121] In this embodiment, the preset information-rule matching table is a pre-set table, which contains various preset information and corresponding sample classification rules, and is used to match the extracted key information with the preset information to determine the appropriate sample classification rules. For example, age > 60 years old, family history of diabetes | special test sample (high risk of diabetes), the former is the preset information, and the latter is the corresponding sample classification rule.

[0122] In this embodiment, the auxiliary classification rule is a multidimensional sample classification rule that matches the key extraction information after comparing the key extraction information with the preset information in the preset information-rule matching table one by one. For example, the classification rule of "routine test sample (sample with poor blood sugar control)" corresponding to "male, fasting blood sugar >7.0mmol / L and symptoms of polydipsia and polyuria" is matched in the preset information-rule matching table. This rule is the auxiliary classification rule.

[0123] In this embodiment, rule optimization refers to adjusting and improving existing sample classification rules to make them more consistent with actual needs and sample characteristics, thereby improving the accuracy and effectiveness of classification.

[0124] In this embodiment, the preset algorithm is a specific algorithm for mining the frequency of items between historical test items, such as the Apriori algorithm. The Apriori algorithm can scan historical diabetes-related blood test data, count the frequencies of different test item combinations, and find frequently occurring item combinations.

[0125] In this embodiment, the frequency of items is the frequency of occurrence of each detection item combination in historical detection items, which is usually measured by indicators such as support. The higher the support, the higher the frequency of occurrence of the item combination. Determining the frequency of items includes: setting a minimum support threshold according to the data volume and business needs; judging that the detection items below the minimum support threshold in the comparison results of each detection item do not have item correlation, and eliminating the detection items below the minimum support threshold in the comparison results of each detection item, thereby starting from a single detection item, gradually generating a frequent item set containing more detection items; generating a corresponding association rule for each frequent item set, and measuring the frequency of occurrence of the association rule. The higher the support, the higher the frequency between items, which indicates that the rule is more universal.

[0126] In this embodiment, the item association rule is a rule generated based on the frequency of items, which describes the association relationship between test items. For example, the appearance of a certain test item may imply the necessity of another test item. The generated association rule may be "If the patient undergoes a fasting blood glucose test and the result shows a high blood glucose value, then there is a high probability that a glycated hemoglobin test and a blood lipid test are required", which indicates that there is an association between the fasting blood glucose test results and the glycated hemoglobin test and the blood lipid test.

[0127] In this embodiment, the detection association model is a model established based on the item association rules, which is used to analyze and determine the correlation between different blood test items. By inputting the user's item information and sample information into the detection association model, the item association between the current sample information and the user's remaining item information can be obtained, and a preliminary inference can be made as to whether the sample information belongs to a routine test or a special test.

[0128] In this embodiment, the item information refers to the item information of the blood test items recorded by the user in the current test cycle, including the name of the test item, the test value, the normal reference range, etc.

[0129] In this embodiment, item correlation refers to the degree of correlation between different blood test items, reflecting the relationship and dependency between the test items. For example, there is a correlation between fasting blood glucose test, glycated hemoglobin test and blood lipid test items in the diagnosis and treatment of diabetes. When a user performs fasting blood glucose test and glycated hemoglobin test at the same time, it can be initially inferred that the user has diabetes, and the sample classification rules can be combined to classify the current sample as a special test.

[0130] In this embodiment, the test item classification rule refers to the final rule for classifying blood test items obtained after rule optimization. The test item classification rule comprehensively considers the sample classification rule and item correlation. For example, the final test item classification rule may be "For samples suspected of diabetes and fasting blood glucose >7.0mmol / L, classify them as routine tests (samples with poor blood glucose control)."

[0131] The working principle of the above content is as follows: First, the initial sample classification rules are preset based on the hospital's characteristics. Key information such as user and doctor diagnosis is extracted from the sample information. This information is compared with the preset information-rule matching table to determine auxiliary classification rules, which are used to optimize the initial sample classification rules and obtain the sample classification rules. At the same time, a preset algorithm is used to mine the frequency of historical test items, generate item association rules, and establish a test association model to determine item relevance. Finally, it is determined whether the sample classification rules meet the item relevance. If so, they are used as the final rules. Otherwise, the rules are optimized again based on the relevance, resulting in a more scientific and reasonable test item classification rule.

[0132] The technical effect of the above content is: by presetting initial classification rules and combining them with key information matching optimization, sample classification can be made more accurate and more consistent with the current hospital treatment direction. At the same time, by using algorithms to mine project correlations and build models to obtain the relationship between test items, and then combining the test item correlation judgment and optimization of classification rules, the determined test item classification rules can be made more scientific and accurate to a greater extent, thereby improving diagnostic efficiency and accuracy, and enhancing the overall level of medical care.

[0133] The rule application module executes the following process:

[0134] Primary classification: Classify the sample into type A, type B, type AB or type O according to the blood type classification rules;

[0135] Secondary classification: Classify samples into routine testing or special testing according to the classification rules of test items;

[0136] Comprehensive classification: Combine the results of the first-level classification and the second-level classification to form the final classification category of the sample;

[0137] If the result of the first-level classification or second-level classification is an unknown type or unknown test item, the abnormality shall be clearly marked in the final classification category.

[0138] The technical effects of the above content are as follows: the hierarchical management method of primary classification and secondary classification makes the sample classification logic clearer and easier for system managers to understand and operate. The comprehensive classification method can fully reflect the characteristics of the sample and provide more detailed information for subsequent sample processing and analysis. By marking abnormalities in the final classification category, abnormalities in the blood type or test items of the sample can be discovered in time, allowing the system to respond quickly when faced with unknown types or unknown test items, avoiding subsequent problems caused by incorrect classification.

[0139] Sorting control module, including:

[0140] a result parsing module configured to parse the classification results into specific sorting instructions and determine the target location to which each sample tube should be transferred;

[0141] The sample sorting module is configured to control the automated sorting device to grab the sample tube according to the sorting instruction, transfer the sample tube from the starting position to the designated position, and monitor the sorting process in real time. If the sample tube breaks or the automated sorting device fails, an alarm is triggered and the sorting operation is suspended;

[0142] The sorting recording module is configured to record the sorting process, including sample information, target location and sorting time, and feed back the recorded sorting information to the user after the sorting is completed.

[0143] The technical effects of the above content are as follows: the result analysis module can quickly convert classification results into sorting instructions, can accurately determine the target location to which each sample tube should be transferred, ensuring that the automated sorting device can accurately perform the sorting task, the sample sorting module can quickly perform the sorting task, and automated sorting reduces the tediousness and error rate of manual operation, thereby improving the accuracy and efficiency of sorting. Real-time monitoring of the sorting process can promptly detect sample tube rupture or automated sorting device failure. Once a problem is detected, an alarm is immediately triggered and the sorting operation is suspended, avoiding subsequent problems caused by equipment failure or sample damage, thereby improving the reliability of the system. The sorting record module feeds back sorting information to the user, so that the user can understand the sorting status and results in real time. The sorting control module significantly improves the accuracy and efficiency of sorting through the precise instruction generation of the result analysis module, the automated sorting and real-time monitoring of the sample sorting module, and the data recording and feedback functions of the sorting record module, thereby enhancing the reliability and security of the system.

[0144] Working principle: The deep learning algorithm based on GPU acceleration can quickly identify the barcode and QR code information in the sample images captured by the high-resolution camera to extract the sample information. GPU acceleration significantly improves the image processing speed and can complete the recognition task of a large number of samples in a short time. By analyzing the sample images using the trained deep learning model, it can automatically identify the blood type and test items, thus avoiding the subjectivity and errors of manual judgment. The samples are then classified according to the analysis results, and different types of samples can be quickly distinguished, which facilitates subsequent testing and processing. The automatic sorting device is controlled according to the sample classification results to transfer the sample tubes to the designated location, thereby avoiding the errors that may be caused by human operation and being able to quickly handle the sorting tasks of a large number of samples.

[0145] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0146] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. GPU-based intelligent intensive blood collection management system, characterized by: The system includes a data acquisition unit and a data analysis unit. The data acquisition unit is configured to use a high-resolution camera to capture images of the test tube after blood collection to obtain sample images. The data analysis unit includes an information recognition module, a sample analysis module, a sample classification module and a sorting control module. The information recognition module is configured as a deep learning algorithm based on GPU acceleration to recognize the barcode and QR code information in the acquired sample image and extract the sample information; The sample analysis module is configured to analyze the sample image using a trained deep learning model to automatically identify the blood type and test items; The sample classification module is configured to classify the samples according to the analysis results; The sorting control module is configured to control the automated sorting device according to the classification results to transfer the sample tubes to the designated locations.

2. The GPU-based intelligent intensive blood sampling management system according to claim 1, characterized in that: The data acquisition unit includes: An image acquisition module is configured to arrange a high-resolution camera in an image acquisition area, and to acquire an image of the test tube after blood collection through the high-resolution camera to obtain a sample image; The image processing module is configured to perform denoising, sharpening, and motion compensation on the acquired sample images, specifically: Denoising: using a denoising algorithm to remove noise from the sample image, where the denoising algorithm includes but is not limited to bilateral filtering; Sharpening: Improving the detail clarity of the sample image through sharpening methods, where sharpening methods include but are not limited to gradient sharpening or Sobel operator; Motion compensation: Use histogram equalization to adjust the brightness and contrast of the sample image.

3. The GPU-based intelligent intensive blood sampling management system according to claim 1, characterized in that: The information identification module includes: a data collection module configured to collect historical image data containing barcodes and QR codes, and annotate the barcodes and QR codes in the historical image data, wherein the annotation content includes the position and category information of the bounding box; The model building module is configured to select a deep learning model suitable for barcode and QR code detection, including YOLO, and train the deep learning model using annotated historical image data. During training, GPU support is configured for the deep learning framework to utilize GPU-accelerated computations to improve training efficiency. The target recognition module is configured to input the acquired sample images into the trained deep learning model. The GPU-accelerated deep learning model is used to locate and decode the barcodes and QR codes in the sample images. Specifically: Perform forward propagation on the input sample image, and the deep learning model outputs the location information of the barcode and QR code in the sample image; Based on the position information of the barcode and the QR code, the integrity coefficient of the current sample image is obtained, thereby judging the integrity of the position information of the barcode and the QR code; The barcode and QR code area with complete detected position information is decoded using a decoding library. The decoding library converts the information in the barcode or QR code into text format for extracting sample information. The result extraction module is configured to extract sample information, including sample number and test items, based on the decoded barcode and QR code information.

4. The GPU-based intelligent intensive blood sampling management system according to claim 1, characterized in that: The sample analysis module includes: A feature extraction module is configured to load a trained deep learning model and input the acquired sample image into the deep learning model. The deep learning model outputs a high-dimensional feature vector, and the output feature vector is used to represent the feature information of the sample image; a feature analysis module configured to further analyze the extracted feature vectors through cluster analysis and classification analysis to identify blood types and test items; The result output module is configured to integrate the identification results of blood type and test items, generate a complete analysis report and provide it to the user.

5. The GPU-based intelligent intensive blood sampling management system according to claim 4, characterized in that: The feature analysis module includes: A result verification module is configured to verify the result by comparing it with the manual annotation results and counting the number of matches and mismatches through comparison; A result display module is configured to display the comparison between the model recognition results and the manual annotation results in a table format, including matches and mismatches; The feedback optimization module is configured to provide feedback on the displayed table and fine-tune or retrain the deep learning model based on the comparison and verification results.

6. The GPU-based intelligent intensive blood sampling management system according to claim 1, characterized in that: The sample classification module includes: a rule setting module configured to preset rules for sample classification based on blood type and test items, wherein the samples are classified into type A, type B, type AB, and type O based on blood type, and the samples are classified into routine testing and special testing based on test items; The rule application module is configured to match the analysis results of each sample according to the preset classification rules and supports multi-level classification, with the first level classification based on blood type and the second level classification based on the test items; The result management module is configured to store the classified sample information in a database, including the sample number, classification category and classification time, and allows users to query the sample classification results based on the sample number and classification category.

7. The GPU-based intelligent intensive blood sampling management system according to claim 6, characterized in that: The rule application module includes: The initial classification module is configured to preset a set of sample classification rules based on the clinical characteristics and treatment direction of the hospital to obtain the initial sample classification rules; An information extraction module is configured to extract key features based on user information and doctor diagnosis information in the sample information to obtain key extracted information; A comparison auxiliary module is configured to compare the key extraction information with the preset information in the preset information-rule matching table one by one, thereby determining the multi-dimensional sample classification rule that matches the key extraction information and obtaining the auxiliary classification rule; A rule optimization module is configured to optimize the initial sample classification rule based on the auxiliary classification rule to obtain the sample classification rule; A model building module is configured to mine the item frequency between historical test items based on a preset algorithm, and generate item association rules based on the item frequency, thereby establishing a test association model based on blood test items; a correlation synthesis module configured to extract corresponding user information based on the sample information, thereby extracting item information of each blood test item of the user, and then combining the sample information with the input of the test correlation model to determine the item correlation; A rule judgment module is configured to judge whether the sample classification rule can satisfy the project relevance; If the sample classification rule can satisfy the item relevance, the sample classification rule is determined to be the final detection item classification rule of the sample; Otherwise, the rules are optimized in combination with the item correlation to obtain the detection item classification rules.

8. The GPU-based intelligent intensive blood sampling management system according to claim 6, characterized in that: The rule application module executes the following process: Primary classification: Classify the sample into type A, type B, type AB or type O according to the blood type classification rules; Secondary classification: Classify samples into routine testing or special testing according to the classification rules of test items; Comprehensive classification: Combine the results of the first-level classification and the second-level classification to form the final classification category of the sample; If the result of the first-level classification or second-level classification is an unknown type or unknown test item, the abnormality shall be clearly marked in the final classification category.

9. The GPU-based intelligent intensive blood sampling management system according to claim 1, characterized in that: The sorting control module includes: a result parsing module configured to parse the classification results into specific sorting instructions and determine the target location to which each sample tube should be transferred; The sample sorting module is configured to control the automated sorting device to grab the sample tube according to the sorting instruction, transfer the sample tube from the starting position to the designated position, and monitor the sorting process in real time. If the sample tube breaks or the automated sorting device fails, an alarm is triggered and the sorting operation is suspended; The sorting recording module is configured to record the sorting process, including sample information, target location and sorting time, and feed back the recorded sorting information to the user after the sorting is completed.

Images