Uric acid monitoring method based on deep learning

Abstract

The present invention discloses a uric acid monitoring method based on deep learning, which relates to the technical field of uric acid monitoring, and includes the following steps: obtaining a pre-processed uric acid detection ultrasonic image to provide a high-quality data basis for subsequent deep learning analysis; in the pre-processed ultrasonic image, applying a deep learning algorithm to preliminarily identify the uric acid pathological characteristics and detect the patient's lesion area; after identifying the pathological characteristics, obtaining image information at the pathological characteristic position, and further extracting features from the image information. The present invention analyzes the pre-processed ultrasonic image through deep learning, identifies the pathological characteristics and further extracts features and abnormal analysis, evaluates the image quality of the pathological characteristic position, ensures that the image quality is normal before identification, improves the recognition accuracy of uric acid crystals and other lesions, reduces missed diagnosis and misdiagnosis, and provides strong support for clinical diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of uric acid monitoring, and in particular to a uric acid monitoring method based on deep learning. Background Art

[0002] Uric acid monitoring based on deep learning refers to the use of deep learning technology to analyze and process biomedical data related to uric acid levels, thereby achieving accurate measurement and prediction of uric acid concentrations. Deep learning is an advanced artificial intelligence technology that can automatically extract features, identify patterns, and make accurate predictions from large amounts of complex, nonlinear medical data by building and training multi-layer neural networks. In uric acid monitoring, this data can come from chemical analysis of blood or urine, image processing, and measurement of other physiological indicators.

[0003] Deep learning models can be trained to analyze the chemical composition of a patient's urine or blood samples. By learning the complex relationships between these data, the model is able to predict uric acid levels. Compared with traditional methods, deep learning-based monitoring methods have higher accuracy and flexibility, and can detect abnormal uric acid levels earlier, thereby facilitating early diagnosis and treatment of related diseases such as gout. In addition, this method can also achieve continuous monitoring. Through portable or wearable devices, patients can obtain their own uric acid level data in real time in their daily lives, facilitating timely intervention measures.

[0004] When using deep learning technology to analyze and process image data, uric acid crystals and other related pathological features are usually captured through ultrasonic imaging technology. By analyzing these ultrasonic images, deep learning algorithms can identify and quantify features related to high uric acid, such as the location, size, and morphology of uric acid crystals, thereby achieving accurate monitoring of uric acid levels and early warning of the disease, and improving the efficiency of clinical diagnosis and treatment.

[0005] The prior art has the following deficiencies:

[0006] When the existing technology uses deep learning technology to analyze and process ultrasound images, it usually directly analyzes the preprocessed ultrasound images. When an ultrasound image has abnormalities at the pathological feature location that cannot be removed by preprocessing, if a direct analysis is performed, the image abnormalities at the pathological feature location may cause the deep learning algorithm to be unable to accurately identify uric acid crystals or other related lesions. For example, abnormalities in the image may mask the presence of uric acid crystals, causing the algorithm to mistakenly believe that the lesion area is normal, resulting in a missed diagnosis. This may result in the patient not receiving the necessary treatment in a timely manner, and the disease further deteriorating.

[0007] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0008] The purpose of the present invention is to provide a uric acid monitoring method based on deep learning. Through deep learning analysis of preprocessed ultrasonic images, pathological features are identified and features and abnormal analysis are further extracted. The image quality of the pathological feature position is evaluated to ensure that the image quality is normal before identification, thereby improving the recognition accuracy of uric acid crystals and other lesions, reducing missed diagnoses and misdiagnoses. For abnormal image quality, the abnormal range is determined as local or overall through analysis. When the overall abnormality occurs, medical personnel are notified to check the equipment. When the local abnormality occurs, rescanning is recommended, thereby improving the image quality and the accuracy of deep learning analysis, ensuring that the pathological feature recognition results are reliable, and providing strong support for clinical diagnosis to solve the problems in the above-mentioned background technology.

[0009] In order to achieve the above object, the present invention provides the following technical solution: a uric acid monitoring method based on deep learning, comprising the following steps:

[0010] Obtain pre-processed uric acid detection ultrasonic images to provide a high-quality data foundation for subsequent deep learning analysis;

[0011] In the preprocessed ultrasound images, a deep learning algorithm is applied to preliminarily identify the pathological features of uric acid and detect the patient's lesion area;

[0012] After the pathological features are identified, image information at the pathological feature locations is obtained, and further features are extracted from the image information. After abnormality analysis is performed on the extracted features, a data analysis model is established based on the features after abnormality analysis, and the image quality at the pathological feature locations is evaluated to identify image quality abnormalities;

[0013] When the image quality is abnormal at the position of the pathological feature, a circle with an area of ​​S is drawn with the center of the pathological feature as the origin, and several circles with an area of ​​S are randomly selected at other positions of the ultrasound image for comprehensive analysis, and the image quality abnormality is divided into local abnormality and overall abnormality;

[0014] For overall abnormal ultrasound images, an early warning notification is issued to remind relevant medical personnel that there is an abnormality in the ultrasound image capture equipment. For locally abnormal ultrasound images, medical personnel are advised to rescan the local area with abnormal image quality at the location of the pathological features to obtain clear and accurate local images.

[0015] Preferably, the specific steps of applying a deep learning algorithm to preliminarily identify the uric acid pathological characteristics and detecting the patient's lesion area are as follows:

[0016] Collect a large amount of uric acid detection ultrasound image data, including normal samples and lesion samples with clear uric acid pathological characteristics;

[0017] Annotate the collected images and mark the lesion area and normal area;

[0018] Use the preprocessed labeled data to build a deep learning model, divide the labeled data into a training set and a validation set, and use the training set to train the deep learning model. During the training process, repeatedly adjust the parameters of the deep learning model to learn the uric acid pathological characteristics in the image;

[0019] After the model training is completed, use an independent test data set to evaluate the model. Based on the evaluation results, optimize the model to improve the model performance.

[0020] The trained and optimized model is deployed to the ultrasound image recognition system to perform preliminary recognition of the pathological features of the acquired ultrasound images and detect the lesion area.

[0021] Preferably, further feature extraction is performed on the image information, and the extracted features include motion artifact information and depth attenuation information. Motion artifact refers to image blur and distortion caused by movement of the patient during ultrasonic imaging. Depth attenuation refers to the phenomenon that when ultrasonic waves propagate in tissues, their signal strength gradually weakens with increasing depth.

[0022] Preferably, after performing abnormal analysis on the motion artifact information and depth attenuation information, a motion artifact complexity index and a depth attenuation abnormality index are generated, a data analysis model is established based on the motion artifact complexity index and the depth attenuation abnormality index after the abnormal analysis, and an image quality assessment coefficient is generated. The image quality at the pathological feature position is evaluated by the image quality assessment coefficient to identify image quality abnormalities.

[0023] Preferably, after the pathological feature is identified, the image information at the pathological feature position is obtained, and the image quality assessment coefficient generated by analyzing the image information at the pathological feature position is compared and analyzed with a preset image quality reference threshold to identify abnormalities in the image quality. The comparison and analysis process is as follows:

[0024] If the image quality assessment coefficient is greater than or equal to the image quality reference threshold, a normal image quality signal is generated, and the image quality at the pathological feature position is marked as normal;

[0025] If the image quality assessment coefficient is less than the image quality reference threshold, an image quality abnormality signal is generated, marking the image quality at the pathological feature position as abnormal.

[0026] Preferably, the specific steps of dividing the abnormal image quality into local abnormality and overall abnormality are as follows:

[0027] Determine the center position of the pathological feature, recorded as point , whose coordinates are , calculate the image quality assessment coefficient for the image area at the pathological feature location;

[0028] by With as the center, draw a circle with a radius of R, and record the area of ​​the circle as S;

[0029] Randomly select L points at other locations in the ultrasound image As the center of the circle, where f=1, 2, 3, 4, ..., L, L is a positive integer, and the coordinates of each point are recorded as , for each point Draw circles with radius R as the center, making sure that the area of ​​each circle is S;

[0030] The image quality assessment coefficient is calculated for each randomly selected circle, and the calculated image quality assessment coefficient is calibrated as ,in, represents the image quality assessment coefficient corresponding to the f-th point;

[0031] The image quality assessment coefficients of all random circles After comprehensive analysis, a comprehensive image quality assessment coefficient is calculated. The calculation expression is: , where represents the comprehensive image quality assessment coefficient;

[0032] The comprehensive image quality assessment coefficient is compared and analyzed with the pre-set comprehensive image quality reference threshold. If the comprehensive image quality assessment coefficient is less than or equal to the comprehensive image quality reference threshold, the image quality abnormality is judged as a local abnormality. If the comprehensive image quality assessment coefficient is greater than the comprehensive image quality reference threshold, the image quality abnormality is judged as an overall abnormality.

[0033] Preferably, after the pathological feature is identified, the motion artifact information at the pathological feature position is obtained, and after the motion artifact information is subjected to abnormal analysis processing, the specific steps of generating the motion artifact complexity index are as follows:

[0034] Divide the ultrasound image I into multiple small blocks , the size of each small block is n×n pixels, then: , i represents the block index of the image in the vertical direction (i.e., row direction), j represents the block index of the image in the horizontal direction (i.e., column direction), and n represents the side length of the small block, which is usually a fixed value, such as 8 or 16 pixels;

[0035] For each small block , use the optical flow method to detect motion artifacts, set and Represents small pieces Medium Pixels The horizontal and vertical motion components of the motion vector field are calculated as follows: , where Indicates Pixels in a small block The motion vector field of

[0036] Calculate the motion complexity of each small block. The motion complexity is defined as the variance of all motion vectors in the small block, reflecting the uniformity of the motion within the block. The calculation formula of motion complexity is: , where Indicates small pieces The average value of the inner motion vector field is calculated as: , where Indicates The movement complexity of each small block;

[0037] Combine the motion complexity of all small blocks and calculate the motion artifact complexity index of the image. The calculation expression is: , where represents the motion artifact complexity index, and N represents the total number of small blocks in the image.

[0038] Preferably, after the pathological feature is identified, the depth attenuation information at the position of the pathological feature is obtained, and after the depth attenuation information is subjected to abnormal analysis processing, the specific steps of generating the depth attenuation abnormality index are as follows:

[0039] The depth attenuation information at the pathological feature position is extracted from the ultrasonic image, and the extracted depth attenuation data is represented as a vector A. ,in represents the signal strength at depth k;

[0040] Calculate the attenuation gradient at each depth position to reflect the rate at which the signal strength changes with depth. Let the attenuation gradient vector be G. ,in Represents the attenuation gradient at the kth depth position. The calculation expression of the attenuation gradient is: , where represents the attenuation gradient at the kth depth position, represents the kth depth position;

[0041] A nonlinear regression model is used to fit the depth attenuation data. The signal attenuation model is assumed to be an exponential attenuation model. The expression of the signal attenuation model is: , where represents the initial signal strength, represents the attenuation coefficient;

[0042] Using the fitted attenuation coefficient , calculate the residual between the actual signal strength and the fitted model to form the residual vector R, ,in, , represents the residual at the kth depth position;

[0043] The depth attenuation anomaly index is calculated based on the residual vector R, and the calculation expression is: , where represents the depth attenuation anomaly index, , represents the normalized absolute value of the residual.

[0044] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0045] The present invention performs deep learning analysis on preprocessed ultrasonic images, and after identifying the pathological features, further performs feature extraction and abnormality analysis on the image information, evaluates the image quality at the pathological feature location, identifies image quality abnormalities, and only performs pathological feature identification when the image quality is normal, thereby avoiding image abnormalities interfering with the judgment of the deep learning algorithm, greatly improving the recognition accuracy of uric acid crystals and other related lesions, reducing the possibility of missed diagnosis and misdiagnosis, and enabling patients to obtain correct diagnosis and treatment in a timely manner.

[0046] The present invention aims at identifying abnormal image quality. By analyzing the image information at the position of pathological features, it is determined whether the scope of the abnormal image quality is local or overall. In the case of overall abnormality, an early warning notification is issued to prompt relevant medical personnel to check the ultrasonic imaging equipment to solve the equipment problem at the source. In the case of local abnormality, it is recommended that medical personnel rescan the local area to obtain clear and accurate images. This not only improves the overall quality of the image, but also ensures the accuracy of deep learning analysis, ensures that the results of pathological feature recognition are reliable, and provides more powerful support for clinical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0048] Figure 1 This is a flow chart of the method for uric acid monitoring based on deep learning of the present invention. DETAILED DESCRIPTION

[0049] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0050] The present invention provides Figure 1 The deep learning-based uric acid monitoring method shown comprises the following steps:

[0051] Obtain pre-processed uric acid detection ultrasonic images to provide a high-quality data foundation for subsequent deep learning analysis;

[0052] Before using deep learning technology to analyze uric acid detection ultrasound images, the images need to be preprocessed first. The preprocessing steps include removing noise, correcting artifacts, enhancing contrast and edges, etc., with the aim of improving the clarity and recognizability of the image. Through preprocessing, most of the interference factors are eliminated, making the image cleaner and more accurate, thereby providing high-quality data input for the deep learning algorithm. A high-quality data foundation is crucial for the training and prediction of deep learning models, because the performance of the model is highly dependent on the quality of the input data. The preprocessed images can better retain pathological features, reduce the interference of erroneous information, improve the accuracy of model recognition and analysis of lesions, and thus improve the effectiveness of disease diagnosis and treatment. In short, the preprocessing process is to ensure the high quality of the input data so that deep learning analysis can be carried out on a more reliable and accurate data basis.

[0053] In the preprocessed ultrasound images, a deep learning algorithm is applied to preliminarily identify the pathological features of uric acid and detect the patient's lesion area;

[0054] The specific steps for applying deep learning algorithms to preliminarily identify uric acid pathological features and detect the patient's lesion area are as follows:

[0055] Collect a large amount of uric acid detection ultrasound image data, including normal samples and lesion samples with clear uric acid pathological characteristics;

[0056] Annotate the collected images and mark the lesion area and normal area;

[0057] Data collection and labeling are the basis of the entire process. By collecting a large amount of real ultrasound image data and having experts label the uric acid lesion areas in detail, a rich and accurate data source can be provided for the subsequent model training. The labeling process requires professional knowledge to ensure that the lesion areas are accurately labeled so that the model can learn effective features.

[0058] Use the preprocessed labeled data to build a deep learning model (commonly used models include convolutional neural networks (CNNs)), divide the labeled data into a training set and a validation set, and use the training set to train the deep learning model. During the training process, repeatedly adjust the parameters of the deep learning model to learn the uric acid pathological characteristics in the image;

[0059] Model construction and training are the core steps. By selecting a suitable deep learning model architecture, such as a convolutional neural network (CNN), and using a large amount of preprocessed labeled data for training, the model can gradually learn pathological features. The validation set is used to evaluate model performance to ensure that the model not only performs well on the training data, but can also accurately identify lesion areas on new data.

[0060] After the model training is completed, use an independent test data set to evaluate the model. Based on the evaluation results, optimize the model to improve the model performance.

[0061] Model evaluation and optimization are important steps to ensure model performance. Evaluation indicators include accuracy, recall, F1 score, etc. The accuracy and robustness of the model are evaluated by using the test data set, the weaknesses of the model are identified, and the model architecture and hyperparameters are adjusted to optimize it. This process is iterative, and the model is continuously improved until it reaches the expected performance level to ensure accurate identification of lesion areas.

[0062] The trained and optimized model is deployed to the ultrasound image recognition system to perform preliminary recognition of pathological features on the acquired ultrasound images and detect the lesion area;

[0063] Deployment and application are key steps to put the model into practical use. By integrating the optimized model into the medical system and providing corresponding training to medical staff, we can ensure that the model is effectively applied in the clinical environment. The lesion recognition results of the model on new images will assist doctors in diagnosis, improve the accuracy and efficiency of diagnosis, and thus realize the actual value of the model.

[0064] After the pathological features are identified, image information at the pathological feature locations is obtained, and further features are extracted from the image information. After abnormality analysis is performed on the extracted features, a data analysis model is established based on the features after abnormality analysis, and the image quality at the pathological feature locations is evaluated to identify image quality abnormalities;

[0065] The image information is further subjected to feature extraction, and the extracted features include motion artifact information and depth attenuation information. Motion artifact refers to image blur and distortion caused by patient movement during ultrasound imaging (such as breathing, heartbeat, or sudden body movement). Depth attenuation refers to the phenomenon that when ultrasound waves propagate in tissues, their signal strength gradually weakens with increasing depth. After abnormal analysis and processing of the motion artifact information and depth attenuation information, a motion artifact complexity index and a depth attenuation abnormality index are generated. A data analysis model is established based on the motion artifact complexity index and the depth attenuation abnormality index after abnormal analysis to generate an image quality assessment coefficient. The image quality at the pathological feature location is evaluated by the image quality assessment coefficient to identify image quality abnormalities.

[0066] After the pathological features are identified, when the image information at the location of the pathological features is obtained, if the complexity of the image motion artifacts corresponding to the location of the pathological features is high, it will have a serious impact on the image analysis accuracy of the uric acid pathological features. The high complexity of motion artifacts means that the image is blurred and deformed to a greater extent due to patient movement or natural organ movement (such as breathing, heartbeat). This blurring and deformation will cause the static structures in the image to appear distorted, and pathological features such as the edges, shapes, and sizes of uric acid crystals will be difficult to accurately identify. These distortions will interfere with the deep learning algorithm's accurate positioning and analysis of the lesion area, and may lead to erroneous feature extraction and classification, thereby affecting the accurate diagnosis of pathological features.

[0067] In addition, high-complexity motion artifacts may lead to the loss of detailed information of the lesion area, causing the pathological features in the image to be obscured or misjudged. Deep learning algorithms rely on high-quality and high-resolution image data for accurate feature extraction and analysis, while motion artifacts can reduce the overall quality of the image, increase the noise level, and destroy the detailed structure of the image. Even after preprocessing steps such as motion correction and denoising, it is difficult to fully restore the detailed information destroyed by motion artifacts. Therefore, high-complexity motion artifacts can significantly affect the image analysis accuracy of uric acid pathological features, which may lead to missed diagnosis or misdiagnosis, thereby affecting the treatment and prognosis of patients.

[0068] After the pathological features are identified, the motion artifact information at the pathological feature position is obtained, and after abnormal analysis and processing of the motion artifact information, the specific steps of generating the motion artifact complexity index are as follows:

[0069] Divide the ultrasound image I into multiple small blocks , the size of each small block is n×n pixels (this can help refine the analysis of motion artifacts and improve the accuracy of local detection), then: , i represents the block index of the image in the vertical direction (i.e., row direction), j represents the block index of the image in the horizontal direction (i.e., column direction), and n represents the side length of the small block, which is usually a fixed value, such as 8 or 16 pixels;

[0070] For each small block , use the optical flow method to detect motion artifacts, set and Represents small pieces Medium Pixels The horizontal and vertical motion components of the motion vector field are calculated as follows: , where Indicates Pixels in a small block The motion vector field of

[0071] Calculate the motion complexity of each small block. The motion complexity is defined as the variance of all motion vectors in the small block, reflecting the uniformity of the motion within the block. The calculation formula of motion complexity is: , where Indicates small pieces The average value of the inner motion vector field is calculated as: , where Indicates The movement complexity of each small block;

[0072] Combine the motion complexity of all small blocks and calculate the motion artifact complexity index of the image. The calculation expression is: , where represents the motion artifact complexity index, and N represents the total number of small blocks in the image.

[0073] It can be known from the motion artifact complexity index that after the pathological features are identified, the motion artifact information at the pathological feature position is obtained and analyzed. The larger the value of the motion artifact complexity index generated after the analysis, the greater the negative impact of the motion artifact on the accuracy of the uric acid pathological feature image analysis; conversely, the smaller the value of the generated motion artifact complexity index, the smaller the impact of the motion artifact on the accuracy of the uric acid pathological feature image analysis. This means that the higher the value of the motion artifact complexity index, the more serious the motion artifact in the image, resulting in lower accuracy in pathological feature recognition and analysis, while the lower the value of the motion artifact complexity index, the better the image quality, which is conducive to accurate pathological feature analysis and diagnosis.

[0074] After the pathological features are identified, when obtaining the image information at the location of the pathological features, if the image depth attenuation corresponding to the location of the pathological features is relatively serious, it will have a serious impact on the image analysis accuracy of the uric acid pathological features. The deep attenuation phenomenon will cause the energy of the ultrasonic signal to gradually weaken during the propagation process, especially in the deeper tissue layers, this phenomenon will be more obvious. Due to the weakening of the signal, the intensity of the echo signal decreases, causing the details in the image to become blurred and unclear. For tiny pathological features such as uric acid crystals, this blur will cover up their subtle features, making it difficult for deep learning algorithms to accurately identify and quantify these lesions. This not only affects the accurate positioning of the pathological features, but may also lead to misjudgment or missed judgment, affecting the reliability of the diagnostic results.

[0075] In addition, depth attenuation also affects the accurate analysis of the morphology, size, and boundaries of pathological features. Lesions such as uric acid crystals usually need to be differentially diagnosed through their morphological features. If the image depth attenuation is severe, the contrast of the lesion area will drop significantly, the boundaries will become blurred, and the morphological features will be difficult to distinguish. In this case, even with advanced image enhancement and signal processing techniques, it is difficult to fully restore the original details, which affects the deep learning algorithm to extract effective features. Ultimately, the degradation of image quality caused by depth attenuation not only limits the recognition of pathological features, but may also lead to delayed diagnosis, and patients cannot obtain accurate treatment plans in time. Therefore, depth attenuation has a very serious impact on the accuracy of image analysis of uric acid pathological features.

[0076] After the pathological features are identified, the depth attenuation information at the pathological feature position is obtained, and after abnormal analysis and processing of the depth attenuation information, the specific steps of generating the depth attenuation abnormality index are as follows:

[0077] The depth attenuation information at the pathological feature position is extracted from the ultrasonic image, and the extracted depth attenuation data is represented as a vector A. ,in represents the signal strength at depth k;

[0078] Vector A contains the signal intensity data at different positions from the surface to the deep layer, which reflects the attenuation of ultrasound in the tissue.

[0079] Calculate the attenuation gradient at each depth position to reflect the rate at which the signal strength changes with depth. Let the attenuation gradient vector be G. ,in Represents the attenuation gradient at the kth depth position. The calculation expression of the attenuation gradient is: , where represents the attenuation gradient at the kth depth position, represents the kth depth position;

[0080] Attenuation gradients represent the rate of change of signal intensity between adjacent depths, and these gradients can reveal the local characteristics of signal intensity changes.

[0081] A nonlinear regression model is used to fit the depth attenuation data. The signal attenuation model is assumed to be an exponential attenuation model. The expression of the signal attenuation model is: , where represents the initial signal strength, represents the attenuation coefficient;

[0082] Through nonlinear regression fitting, the attenuation coefficient can be obtained , this parameter reflects the overall signal attenuation characteristics.

[0083] Using the fitted attenuation coefficient , calculate the residual between the actual signal strength and the fitted model to form the residual vector R, ,in, , represents the residual at the kth depth position;

[0084] Residual It reflects the difference between the actual signal strength and the fitted model, which can reveal anomalies in deep attenuation.

[0085] The depth attenuation anomaly index is calculated based on the residual vector R, and the calculation expression is: , where represents the depth attenuation anomaly index, , represents the normalized absolute value of the residual.

[0086] After identifying the pathological features, the depth attenuation information at the pathological feature position is obtained, and the depth attenuation abnormality index generated after analysis reflects the degree of attenuation abnormality of the image. The larger the performance value of the depth attenuation abnormality index, the more severe the depth attenuation at the pathological feature position, the greater the abnormal difference in signal intensity, and thus the greater the negative impact on the image analysis accuracy of the uric acid pathological feature. On the contrary, the smaller the performance value of the generated depth attenuation abnormality index, the more uniform the depth attenuation, the more stable the signal intensity, and the less impact on the image analysis accuracy of the uric acid pathological feature.

[0087] The data analysis model can be a machine learning model, etc., which is not specifically limited here, and can realize the motion artifact complexity index and depth attenuation anomaly index Perform comprehensive analysis to generate image quality assessment coefficients In order to realize the technical solution of the present invention, the present invention provides a specific implementation method;

[0088] Image quality assessment coefficient The resulting calculation formula is: , where , The motion artifact complexity index and depth attenuation anomaly index The preset scaling factor of , Both are greater than 0.

[0089] It can be seen from the image quality assessment coefficient that after the pathological feature is identified, the motion artifact information and depth attenuation information at the pathological feature position are obtained. The larger the performance value of the motion artifact complexity index generated after the motion artifact information is subjected to abnormal analysis processing, the larger the performance value of the depth attenuation abnormality index generated after the depth attenuation information is subjected to abnormal analysis processing, that is, the smaller the performance value of the generated image quality assessment coefficient, indicating that the image quality at the pathological feature position is worse, and the negative impact on the image analysis accuracy of the uric acid pathological feature is greater. Conversely, it indicates that the negative impact on the image analysis accuracy of the uric acid pathological feature is smaller.

[0090] After the pathological features are identified, the image information at the pathological feature location is obtained, and the image quality assessment coefficient generated by analyzing the image information at the pathological feature location is compared and analyzed with a preset image quality reference threshold to identify abnormalities in the image quality. The comparison and analysis process is as follows:

[0091] If the image quality assessment coefficient is greater than or equal to the image quality reference threshold, a normal image quality signal is generated, and the image quality at the pathological feature position is marked as normal. The deep learning model can be used to efficiently identify the uric acid pathological features.

[0092] If the image quality assessment coefficient is less than the image quality reference threshold, an abnormal image quality signal is generated, and the image quality at the pathological feature location is marked as abnormal. When the image analysis of uric acid pathological features is performed through a deep learning model, the image analysis accuracy will be greatly affected.

[0093] When the image quality is abnormal at the position of the pathological feature, a circle with an area of ​​S is drawn with the center of the pathological feature as the origin, and several circles with an area of ​​S are randomly selected at other positions of the ultrasound image for comprehensive analysis, and the image quality abnormality is divided into local abnormality and overall abnormality;

[0094] The specific steps for classifying image quality abnormalities into local abnormalities and overall abnormalities are as follows:

[0095] Determine the center position of the pathological feature, recorded as point , whose coordinates are , calculate the image quality assessment coefficient for the image area at the pathological feature location;

[0096] In ultrasound images, existing technologies usually use image segmentation and feature extraction algorithms to determine the center position of pathological features. The images are processed through deep learning models (such as convolutional neural networks, CNNs) to identify the boundaries of pathological features and calculate the centroid of these boundaries through geometric methods. The centroid is the center position of the pathological feature. This method combines automated image analysis technology with precise mathematical calculations to effectively determine the center position of pathological features.

[0097] by With as the center, draw a circle with a radius of R, and record the area of ​​the circle as S;

[0098] Randomly select L points at other locations in the ultrasound image As the center of the circle, where f=1, 2, 3, 4, ..., L, L is a positive integer, and the coordinates of each point are recorded as , for each point Draw circles with radius R as the center, making sure that the area of ​​each circle is S;

[0099] The image quality assessment coefficient is calculated for each randomly selected circle, and the calculated image quality assessment coefficient is calibrated as ,in, represents the image quality assessment coefficient corresponding to the f-th point;

[0100] The image quality assessment coefficients of all random circles After comprehensive analysis, a comprehensive image quality assessment coefficient is calculated. The calculation expression is: , where represents the comprehensive image quality assessment coefficient;

[0101] The comprehensive image quality assessment coefficient is compared and analyzed with the pre-set comprehensive image quality reference threshold. If the comprehensive image quality assessment coefficient is less than or equal to the comprehensive image quality reference threshold, the image quality abnormality is judged as a local abnormality. If the comprehensive image quality assessment coefficient is greater than the comprehensive image quality reference threshold, the image quality abnormality is judged as an overall abnormality.

[0102] For overall abnormal ultrasound images, an early warning notification is issued to remind relevant medical personnel that there is an abnormality in the ultrasound imaging equipment. For local abnormal ultrasound images, medical personnel are advised to rescan the local area with abnormal image quality at the pathological feature location to obtain a clear and accurate local image.

[0103] Since overall abnormalities are often caused by systemic problems, such as improper equipment calibration, probe failure, or poor imaging conditions, early warning notifications should be issued to alert relevant medical personnel to possible abnormalities in ultrasound imaging equipment. Specific measures include checking the calibration status of the equipment, detecting and replacing faulty probes, and optimizing the imaging environment and conditions. In this way, abnormal problems with ultrasound imaging equipment can be fed back and resolved in a timely manner, ensuring the quality of subsequent images and the accuracy of diagnosis, and improving the overall efficiency and reliability of medical services.

[0104] It is recommended that medical personnel rescan local areas where abnormal image quality exists at the location of pathological features to obtain clearer and more accurate local images. This measure will help to more accurately identify and analyze pathological features, improve the accuracy and reliability of diagnosis, and ensure that patients receive timely and correct treatment.

[0105] The present invention performs deep learning analysis on preprocessed ultrasonic images, and after identifying the pathological features, further performs feature extraction and abnormality analysis on the image information, evaluates the image quality at the pathological feature location, identifies image quality abnormalities, and only performs pathological feature identification when the image quality is normal, thereby avoiding image abnormalities interfering with the judgment of the deep learning algorithm, greatly improving the recognition accuracy of uric acid crystals and other related lesions, reducing the possibility of missed diagnosis and misdiagnosis, and enabling patients to obtain correct diagnosis and treatment in a timely manner.

[0106] The present invention aims at identifying abnormal image quality. By analyzing the image information at the position of pathological features, it is determined whether the scope of the abnormal image quality is local or overall. In the case of overall abnormality, an early warning notification is issued to prompt relevant medical personnel to check the ultrasonic imaging equipment to solve the equipment problem at the source. In the case of local abnormality, it is recommended that medical personnel rescan the local area to obtain clear and accurate images. This not only improves the overall quality of the image, but also ensures the accuracy of deep learning analysis, ensures that the results of pathological feature recognition are reliable, and provides more powerful support for clinical diagnosis.

[0107] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0108] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0109] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0110] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A uric acid monitoring method based on deep learning, characterized in that: The following steps are involved: Obtain pre-processed uric acid detection ultrasonic images to provide a high-quality data foundation for subsequent deep learning analysis; In the preprocessed ultrasound images, a deep learning algorithm is applied to preliminarily identify the pathological features of uric acid and detect the patient's lesion area; After the pathological features are identified, image information at the pathological feature locations is obtained, and further features are extracted from the image information. After abnormality analysis is performed on the extracted features, a data analysis model is established based on the features after abnormality analysis, and the image quality at the pathological feature locations is evaluated to identify image quality abnormalities; When the image quality is abnormal at the pathological feature position, a circle with an area of S , and continue to randomly select several circles with an area of S The circles are analyzed comprehensively to classify the abnormal image quality into local abnormality and overall abnormality. For overall abnormal ultrasound images, an early warning notification is issued to remind relevant medical personnel that there is an abnormality in the ultrasound imaging equipment. For local abnormal ultrasound images, medical personnel are advised to rescan the local area with abnormal image quality at the pathological feature location to obtain a clear and accurate local image; The specific steps for classifying image quality abnormalities into local abnormalities and overall abnormalities are as follows: Determine the center position of the pathological feature, recorded as point , whose coordinates are , calculate the image quality assessment coefficient for the image area at the pathological feature location; by With centered at R The area of ​​the circle is S ; Randomly select other positions in the ultrasound image L Points As the center of the circle, f =1, 2, 3, 4, ..., L , L is a positive integer, and the coordinates of each point are , for each point With centered at R circles, ensuring that the area of ​​each circle is S ; The image quality assessment coefficient is calculated for each randomly selected circle, and the calculated image quality assessment coefficient is calibrated as ,in, Indicates f The image quality assessment coefficient corresponding to each point; The image quality assessment coefficients of all random circles After comprehensive analysis, a comprehensive image quality assessment coefficient is calculated. The calculation expression is: , where represents the comprehensive image quality assessment coefficient; The comprehensive image quality assessment coefficient is compared and analyzed with the pre-set comprehensive image quality reference threshold. If the comprehensive image quality assessment coefficient is less than or equal to the comprehensive image quality reference threshold, the image quality abnormality is judged as a local abnormality. If the comprehensive image quality assessment coefficient is greater than the comprehensive image quality reference threshold, the image quality abnormality is judged as an overall abnormality.

2. The uric acid monitoring method based on deep learning according to claim 1, characterized in that: The specific steps for applying deep learning algorithms to preliminarily identify uric acid pathological features and detect the patient's lesion area are as follows: Collect a large amount of uric acid detection ultrasound image data, including normal samples and lesion samples with clear uric acid pathological characteristics; Annotate the collected images and mark the lesion area and normal area; Use the preprocessed labeled data to build a deep learning model, divide the labeled data into a training set and a validation set, and use the training set to train the deep learning model. During the training process, repeatedly adjust the parameters of the deep learning model to learn the uric acid pathological characteristics in the image; After the model training is completed, use an independent test data set to evaluate the model. Based on the evaluation results, optimize the model to improve the model performance. The trained and optimized model is deployed to the ultrasound image recognition system to perform preliminary recognition of the pathological features of the acquired ultrasound images and detect the lesion area.

3. The uric acid monitoring method based on deep learning according to claim 1, characterized in that: The image information is further extracted for features, including motion artifact information and depth attenuation information. Motion artifact refers to image blur and distortion caused by the movement of the patient during ultrasound imaging. Depth attenuation refers to the phenomenon that when ultrasound waves propagate in tissues, their signal strength gradually weakens with increasing depth.

4. The uric acid monitoring method based on deep learning according to claim 3, characterized in that: After abnormal analysis of the motion artifact information and depth attenuation information, the motion artifact complexity index and the depth attenuation abnormality index are generated. A data analysis model is established based on the motion artifact complexity index and the depth attenuation abnormality index after the abnormal analysis, and an image quality assessment coefficient is generated. The image quality at the pathological feature position is evaluated by the image quality assessment coefficient to identify image quality abnormalities.

5. The uric acid monitoring method based on deep learning according to claim 4, characterized in that: After the pathological features are identified, the image information at the pathological feature location is obtained, and the image quality assessment coefficient generated by analyzing the image information at the pathological feature location is compared and analyzed with a preset image quality reference threshold to identify abnormalities in the image quality. The comparison and analysis process is as follows: If the image quality assessment coefficient is greater than or equal to the image quality reference threshold, a normal image quality signal is generated, and the image quality at the pathological feature position is marked as normal; If the image quality assessment coefficient is less than the image quality reference threshold, an image quality abnormality signal is generated, marking the image quality at the pathological feature position as abnormal.

6. The uric acid monitoring method based on deep learning according to claim 4, characterized in that: After the pathological features are identified, the motion artifact information at the pathological feature position is obtained, and after abnormal analysis and processing of the motion artifact information, the specific steps of generating the motion artifact complexity index are as follows: Divide the ultrasound image I into multiple small blocks , the size of each small block is n × n Pixels, then: i Indicates the vertical block index of the image. j Indicates the horizontal block index of the image. n Indicates the side length of the small block; For each small block , use the optical flow method to detect motion artifacts, set and Represents small pieces Medium Pixels The horizontal and vertical motion components of the motion vector field are calculated as follows: , where Indicates Pixels in a small block The motion vector field of Calculate the motion complexity of each small block. The motion complexity is defined as the variance of all motion vectors in the small block, reflecting the uniformity of the motion within the block. The calculation formula of motion complexity is: , where Indicates small pieces The average value of the inner motion vector field is calculated as: , where Indicates The movement complexity of each small block; Combine the motion complexity of all small blocks and calculate the motion artifact complexity index of the image. The calculation expression is: , where represents the motion artifact complexity index, N Represents the total number of patches in the image.

7. The uric acid monitoring method based on deep learning according to claim 4, characterized in that: After the pathological features are identified, the depth attenuation information at the pathological feature position is obtained, and after abnormal analysis and processing of the depth attenuation information, the specific steps of generating the depth attenuation abnormality index are as follows: Extract the depth attenuation information at the pathological feature location from the ultrasound image and represent the extracted depth attenuation data as a vector A , ,in Indicates depth k Signal strength at Calculate the attenuation gradient at each depth position to reflect the rate at which the signal strength changes with depth. Suppose the attenuation gradient vector is G , ,in Indicates k The attenuation gradient at the depth position, the calculation expression of the attenuation gradient is: , where No. k The attenuation gradient at the depth position, Indicates k depth positions; A nonlinear regression model is used to fit the depth attenuation data. The signal attenuation model is assumed to be an exponential attenuation model. The expression of the signal attenuation model is: , where represents the initial signal strength, represents the attenuation coefficient; Using the fitted attenuation coefficient Calculate the residual between the actual signal strength and the fitted model to form a residual vector R , ,in, , Indicates k The residual of the depth position; Based on the residual vector R Calculate the depth attenuation anomaly index. The calculation expression is: , where represents the depth attenuation anomaly index, , represents the normalized absolute value of the residual.

Images