Intelligent analysis system based on instantaneous elastography technology

Abstract

The invention relates to the field of application of an instantaneous elasticity imaging technology to liver sclerosis detection, in particular to an intelligent analysis system based on the instantaneous elasticity imaging technology. Comprising a shear wave imaging module, a medical history detection calling module and an intelligent analysis module, and the shear wave imaging module is used for generating shear waves and longitudinal waves through energy pulses and conducting imaging processing on real-time data of the shear waves obtained through patient detection; the medical history detection calling module is used for acquiring detailed patient conditions through the patient medical history and detection records; and the intelligent analysis module is used for jointly analyzing through the shear wave imaging module and the medical history detection calling module so as to determine disease causes and give suggestions. According to the image drawn by the shear wave, the condition of the liver of a patient can be reflected more visually and accurately, the judgment of the liver is more accurate, inaccurate factors in illness state judgment are reduced, and the system can reduce errors caused by the influence of factors such as the technical level of an operator, equipment performance and the like when the liver is detected by an instantaneous elastography technology.

Description

Technical Field

[0001] The present invention relates to the field of liver cirrhosis detection using transient elastic imaging technology, and in particular to an intelligent analysis system based on transient elastic imaging technology. Background Art

[0002] Transient elastography technology can be used for liver stiffness detection to detect and diagnose liver fibrosis at an early stage, reduce the risk of liver biopsy, and reduce complications such as chronic liver disease and cirrhosis. Traditional transient elastography technology is a non-invasive detection technology that uses a dedicated probe to emit a fixed low-frequency shear wave to instantaneously actively excite the liver tissue, while emitting a high-frequency signal to track the propagation speed of the shear wave in the liver, thereby obtaining the liver stiffness value.

[0003] However, since this technology only derives the liver stiffness value, the test results will be affected by many factors, such as obesity or the size of the intercostal space, which may lead to inaccurate test results. Liver inflammation, intrahepatic cholestasis and other conditions will also cause local changes in the liver stiffness value of the liver, which will cause misdiagnosis or inaccurate test results. Therefore, it is still necessary to combine invasive examination methods such as liver puncture biopsy to make specific location judgments and other comprehensive diagnoses, which goes against the original intention of designing transient elastography technology for liver testing. Summary of the invention

[0004] To solve the above problems, the present invention provides an intelligent analysis system based on transient elastic imaging technology, which can more intuitively and accurately reflect the patient's liver condition, more accurately judge the liver condition, reduce inaccurate factors in disease judgment, and the system can avoid errors in transient elastic imaging technology when detecting the liver due to factors such as the operator's technical level and equipment performance.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] An intelligent analysis system based on transient elastic imaging technology comprises: a shear wave imaging module, a medical history detection and retrieval module and an intelligent analysis module which are wirelessly connected in sequence, wherein:

[0007] The shear wave imaging module is used to generate shear waves and longitudinal waves through energy pulses, and to perform imaging processing on the real-time data of shear waves obtained from patient testing;

[0008] The medical history test retrieval module is used to obtain detailed patient conditions through patient medical history and test records;

[0009] The intelligent analysis module is used to jointly analyze through the shear wave imaging module and the medical history detection and retrieval module to determine the cause of the disease and give suggestions.

[0010] Furthermore, the method of generating shear waves and longitudinal waves by energy pulses and performing imaging processing on real-time shear wave data acquired by patient detection comprises the following steps:

[0011] S1: Acquire ultrasonic signals through signal transmission, signal reception and signal quality processing;

[0012] S2: Focus the received signal to the target area to improve the resolution and contrast of the image;

[0013] S3: converting the processed signal into an image;

[0014] S4: Feature extraction for image readability and diagnostic value.

[0015] Furthermore, the acquisition of ultrasonic signals through signal transmission, signal reception and signal quality processing comprises the following steps:

[0016] S11: Generate low-frequency elastic waves through an external mechanical vibration device to act on liver tissue;

[0017] S12: Low-frequency elastic waves propagate in liver tissue, exciting shear waves;

[0018] S13: the ultrasonic transducer array transmits an ultrasonic beam to the shear wave region;

[0019] S14: The transducer receives the reflected ultrasonic signal and performs amplification, filtering and digitization operations.

[0020] Furthermore, focusing the received signal to the target area to improve the resolution and contrast of the image includes the following steps:

[0021] S21: Calculate the delay time required for each array element to receive the signal according to the relative position of the transducer array element and the target point;

[0022] S22: adjusting the time of the signal received by each array element according to the calculated delay time;

[0023] S23: Assign a certain weight to the signal of each array element to control the directivity and shape characteristics of the beam;

[0024] S24: summing up the signals of each array element after delay adjustment and weighting processing to obtain a beamformed signal.

[0025] Furthermore, converting the processed signal into an image comprises the following steps:

[0026] S31: obtaining a series of signal data received by the ultrasonic transducer after beamforming, and performing necessary preprocessing to divide the imaging area into a number of small spatial grids;

[0027] S32: for each grid point, estimating relevant parameter values ​​of the grid point according to information of the strength and propagation time of surrounding received signals;

[0028] S33: Perform multiple iterative calculations to continuously optimize the values ​​of grid points and perform image verification based on longitudinal wave positioning performance;

[0029] S34: constructing a final liver image according to the calculated values ​​of each grid point.

[0030] Furthermore, the feature extraction of the readability and diagnostic value of the image includes the following steps:

[0031] S41: Calculate the histogram of the image constructed in step S3, calculate the cumulative histogram, determine the new grayscale value corresponding to each grayscale level through the cumulative histogram, and perform contrast stretching;

[0032] S42: Determine the size of the filter window, calculate the average value of the pixels in its neighborhood for each pixel and replace the corresponding pixel, calculate the median value of the pixels in the neighborhood and replace the central pixel;

[0033] S43: using the maximum inter-class variance method to classify pixels in the image whose grayscale values ​​are greater than or equal to the threshold into one category, and pixels whose grayscale values ​​are less than the threshold into another category, selecting a seed pixel, setting a grayscale difference threshold, and starting from the seed pixel, merging adjacent pixels that meet the similarity criterion into the same area, and continuously growing the area;

[0034] S44: Perform Gaussian filtering to smooth the image, calculate the gradient amplitude and direction, perform non-maximum suppression, refine the edge, use double thresholds for edge connection and judgment, and extract liver shape features.

[0035] Furthermore, an evaluation and error correction module is included, which is used to perform the following steps:

[0036] A1: Collect a large amount of raw image data of the liver detected by transient elastography technology, as well as the corresponding actual liver condition information annotated by doctors, and use a convolutional neural network. The model includes multiple convolutional layers, pooling layers, and fully connected layers;

[0037] A2: Input the prepared image data and annotation information into the model, forward propagate and calculate the prediction results, define the loss function, calculate the gradient through the back propagation algorithm, and update the model parameters;

[0038] A3: Input the questionable or erroneous image into the model trained in step A2, and the model outputs the corrected prediction result of liver problems;

[0039] A4: Randomly select an area on the image to crop and retain, randomly rotate the image at a certain angle, and make slight random changes to the color channels of the image; try different learning rates, adjust the depth and width structure parameters of the model, and try different regularization strengths;

[0040] A5: Use precision, recall, and F1 value metrics to evaluate the accuracy of the model in predicting liver problems, and use the confusion matrix to visually view the correct and incorrect predictions for each category.

[0041] Furthermore, the medical history detection and retrieval module uses the medical information platform to retrieve the information, and marks the patient's liver disease history, medication and test items in the retrieved information;

[0042] The medical information platform is based on the medical personnel's use of standardized terminology in their daily work to accurately enter the patient's medical history, medication records and test records into the medical information platform and update them in real time; the medical information platform is equipped with a permission management mechanism.

[0043] Furthermore, the shear wave imaging module and the medical history detection and retrieval module are used to jointly analyze the cause of the disease and give suggestions, including the following steps:

[0044] B1: Through the shear wave imaging module and the medical history detection and retrieval module, the liver stiffness value, echo characteristics, liver function index, virus markers, medication records and detailed medical history are extracted as key features;

[0045] B2: Using the improved clustering algorithm and based on the expert database, a diagnostic model is constructed. The detected liver images and information of the patient are input into the diagnostic model to output the most likely cause of the disease.

[0046] B3: Generate personalized treatment recommendations based on the most likely cause of the disease output and expert database matching, and feed back the patient's treatment effects and expert recommendations into the diagnostic model and treatment recommendations for continuous optimization.

[0047] Furthermore, the improved clustering algorithm comprises the following steps:

[0048] B21: Preprocess and standardize the key features collected in step B1, divide the data into multiple clusters, and randomly select K data points as the initial cluster centers;

[0049] B22: Calculate the distance from each data point to each cluster center, and assign the data point to the cluster to which the nearest cluster center belongs; recalculate the center of each cluster, that is, the average value of the features of all data points in the cluster;

[0050] B23: Repeat step B22 until the location of the cluster center no longer changes significantly or the set number of iterations is reached, analyze the clustering results, observe the characteristic distribution of patient data in different clusters, and discover patterns related to the cause or condition.

[0051] The beneficial effects of the present invention are:

[0052] 1. The shear wave imaging module uses real-time shear wave data to draw images, which can intuitively understand the liver condition and avoid inaccurate judgments caused by obesity or the size of the intercostal space. The shear wave imaging module processes the image in high definition and extracts image features to clearly reflect the liver condition, providing more reference information for misdiagnosis or inaccurate test results caused by liver inflammation, intrahepatic cholestasis, etc., thereby improving the accuracy of disease diagnosis.

[0053] 2. The evaluation and error correction module reduces the error of transient elastography technology in detecting the liver due to factors such as the operator's technical level and equipment performance. By comparing the test results with the liver data in the system for error correction and training, it can provide more accurate images and improve the accuracy of disease diagnosis.

[0054] 3. The medical history detection and retrieval module obtains the patient's real condition through the medical information platform. Before using transient elastic imaging technology to judge the liver condition, it combines the patient's historical condition, medication and previous tests to understand the patient's condition, and makes a preliminary judgment on possible or existing liver inflammation, intrahepatic cholestasis and other conditions, thereby reducing other risks such as misdiagnosis caused by inaccurate condition judgment due to the above conditions.

[0055] 4. The intelligent analysis module uses an improved clustering algorithm combined with an expert database to make a preliminary judgment on the advancedness of the image and give corresponding suggestions. It provides reference suggestions for doctors when judging patients' liver images and reduces the possibility of doctors misjudgment of images due to subjective reasons. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the structure of the intelligent analysis system based on transient elastic imaging technology;

[0057] Figure 2 is a schematic diagram of the shear wave imaging module process;

[0058] Figure 3 It is a flowchart of the evaluation and error correction module. DETAILED DESCRIPTION

[0059] See also Figure 1-3 As shown, the present invention relates to an intelligent analysis system based on transient elastic imaging technology.

[0060] Example:

[0061] An intelligent analysis system based on transient elastic imaging technology comprises: a shear wave imaging module, a medical history detection and retrieval module and an intelligent analysis module which are wirelessly connected in sequence, wherein:

[0062] The shear wave imaging module is used to generate shear waves and longitudinal waves through energy pulses, and to perform imaging processing on the real-time data of shear waves obtained from patient testing;

[0063] The medical history test retrieval module is used to obtain detailed patient conditions through patient medical history and test records;

[0064] The intelligent analysis module is used to jointly analyze through the shear wave imaging module and the medical history detection and retrieval module to determine the cause of the disease and give suggestions.

[0065] Furthermore, the method of generating shear waves and longitudinal waves by energy pulses and performing imaging processing on real-time shear wave data acquired by patient detection comprises the following steps:

[0066] S1: Acquire ultrasonic signals through signal transmission, signal reception and signal quality processing;

[0067] S2: Focus the received signal to the target area to improve the resolution and contrast of the image;

[0068] S3: converting the processed signal into an image;

[0069] S4: Feature extraction for image readability and diagnostic value.

[0070] Furthermore, the acquisition of ultrasonic signals through signal transmission, signal reception and signal quality processing comprises the following steps:

[0071] S11: Generate low-frequency elastic waves through an external mechanical vibration device to act on liver tissue;

[0072] S12: Low-frequency elastic waves propagate in liver tissue, exciting shear waves;

[0073] S13: the ultrasonic transducer array transmits an ultrasonic beam to the shear wave region;

[0074] S14: The transducer receives the reflected ultrasonic signal and performs amplification, filtering and digitization operations.

[0075] Specifically, when shear waves propagate in the liver, if the liver is cirrhotic, the propagation speed will be fast, and if the liver is in normal condition, the propagation speed will be slow. An ultrasonic transducer is required to obtain the shear wave state. The ultrasonic transducer emits an ultrasonic beam that hits the shear wave and then reflects, thereby indirectly obtaining the speed state of the shear wave. These ultrasonic signals will produce reflection and scattering when encountering liver tissue and shear waves. The transducer continuously captures these echo signals and accurately records key information such as the arrival time, intensity, and frequency of each echo signal. At the same time, the system will synchronously record the time point when the elastic wave is generated and other related trigger information to ensure accurate association and analysis later.

[0076] The large amount of echo signal data collected is preprocessed, amplified, filtered and digitized to extract the effective signal segments related to shear wave propagation.

[0077] Furthermore, focusing the received signal to the target area to improve the resolution and contrast of the image includes the following steps:

[0078] S21: Calculate the delay time required for each array element to receive the signal according to the relative position of the transducer array element and the target point;

[0079] S22: adjusting the time of the signal received by each array element according to the calculated delay time;

[0080] S23: Assign a certain weight to the signal of each array element to control the directivity and shape characteristics of the beam;

[0081] S24: summing up the signals of each array element after delay adjustment and weighting processing to obtain a beamformed signal.

[0082] Specifically, in step S2, based on an improved beamforming algorithm, the received signal is focused on the target area to improve the resolution and contrast of the image.

[0083] In the delay calculation in step S21, the position coordinates of the transducer array element are assumed to be (x i ,y i ), the target point coordinates are (x0, y0), and the speed of sound is c. Then the distance d from array element i to the target point is i It can be calculated by the formula:

[0084]

[0085] The corresponding delay time t i for:

[0086]

[0087] In step S22, the signal adjustment is performed for the signal s received by the array element i. i (t), and delay it on the time axis to obtain a new signal s′ i (tt i );

[0088] In the weighted processing in step S23, the signal of each array element is multiplied by the corresponding weight coefficient w i , and get the weighted signal w i s′ i (tt i );

[0089] The signals in step S24 are summed to obtain the beamformed signal S(t):

[0090]

[0091] Where N is the total number of array elements.

[0092] Furthermore, converting the processed signal into an image comprises the following steps:

[0093] S31: obtaining a series of signal data received by the ultrasonic transducer after beamforming, and performing necessary preprocessing to divide the imaging area into a number of small spatial grids;

[0094] S32: for each grid point, estimating relevant parameter values ​​of the grid point according to information of the strength and propagation time of surrounding received signals;

[0095] S33: Perform multiple iterative calculations to continuously optimize the values ​​of grid points and perform image verification based on longitudinal wave positioning performance;

[0096] S34: constructing a final liver image according to the calculated values ​​of each grid point.

[0097] Specifically, step S3 is to convert the processed signal into an image based on the improved reconstruction algorithm.

[0098] The data collection and preprocessing and spatial grid division in step S31 are to obtain a series of signal data S from the ultrasonic transducer. i (t), where i represents the array element number and t represents time. Perform preprocessing operations such as denoising and filtering on these data to obtain purer data. Divide the imaging area into a uniform three-dimensional space grid, and the coordinates of each grid point are (x m ,y m , z m ), m represents the grid point number.

[0099] The calculation of the grid point values ​​in step S32 is to calculate the gray value (or other relevant parameter values) of each grid point by the following formula:

[0100]

[0101] Among them, S i (t i，m ) represents the propagation time t of array element i at the corresponding grid point i，m The signal strength at time K i，m is a weight coefficient related to factors such as the distance from the array element to the grid point, D i，m Represents the distance from array element i to the grid point, which can be calculated through spatial coordinates as follows:

[0102]

[0103] The iterative optimization in step S33 can be performed by iteratively adjusting the weight coefficients and other parameters to optimize the values ​​of the grid points and improve the accuracy and quality of the image. Each iteration can use the following update formula:

[0104]

[0105] Where α is an adjustment step size, is the desired grid point value.

[0106] The image generation in step S34 is to generate a corresponding two-dimensional image according to the grayscale value or other parameter value of each grid point finally calculated.

[0107] More specifically, when the ultrasonic probe emits an ultrasonic pulse, it is decomposed into shear waves and longitudinal waves. For the longitudinal waves, image verification can be performed based on the positioning performance of the reflected longitudinal wave signal combined with the corresponding two-dimensional image that has been generated to determine the accuracy of the image.

[0108] Furthermore, the feature extraction of the readability and diagnostic value of the image includes the following steps:

[0109] S41: Calculate the histogram of the image constructed in step S3, calculate the cumulative histogram, determine the new grayscale value corresponding to each grayscale level through the cumulative histogram, and perform contrast stretching;

[0110] S42: Determine the size of the filter window, calculate the average value of the pixels in its neighborhood for each pixel and replace the corresponding pixel, calculate the median value of the pixels in the neighborhood and replace the central pixel;

[0111] S43: using the maximum inter-class variance method to classify pixels in the image whose grayscale values ​​are greater than or equal to the threshold into one category, and pixels whose grayscale values ​​are less than the threshold into another category, selecting a seed pixel, setting a grayscale difference threshold, and starting from the seed pixel, merging adjacent pixels that meet the similarity criterion into the same area, and continuously growing the area;

[0112] S44: Perform Gaussian filtering to smooth the image, calculate the gradient amplitude and direction, perform non-maximum suppression, refine the edge, use double thresholds for edge connection and judgment, and extract liver shape features.

[0113] Specifically, step S4 is to further process the reconstructed image to improve the readability and diagnostic value of the image.

[0114] Image enhancement in step S41, to equalize the histogram, requires calculating the histogram of the original image: for gray level i, its frequency of occurrence:

[0115] where n i is the number of grayscale pixels, and N is the total number of pixels in the image. Calculate the cumulative histogram

[0116]

[0117] Determine the new grayscale value corresponding to each grayscale level through the cumulative histogram: g(k)=int[(L-1)s(k)+0.5]

[0118] Then perform contrast stretching. Assuming the stretching interval is [a, b] to [c, d], for the pixel value x, the new value y is:

[0119]

[0120] In step S42, the smoothing process uses mean filtering and median filtering. For example, the mean filtering is based on a 3x3 window. For the center pixel (x, y), the new value is the average value of the surrounding 9 pixel values, that is,

[0121]

[0122] The median filter also uses a 3x3 window to sort the 9 pixel values ​​and take the middle value as the new value of the center pixel.

[0123] The image segmentation in step S43 is performed by threshold segmentation and region growing method;

[0124] More specifically, threshold segmentation uses the maximum inter-class variance method to calculate the inter-class variance The threshold T when the variance is the largest is the optimal threshold, where the inter-class variance formula is: w0 and w1 are the ratios of the two types of pixels, and u0 and u1 are the average grayscales of the two types of pixels.

[0125] The region growing method is to start from the seed pixel (x s ,y s ) and calculate the grayscale difference between the neighboring pixel (x, y) and the seed pixel |I(x, y)-I(x s ,y s )|, if it is less than the set threshold, they will be merged into the same area.

[0126] The feature extraction in step S44 involves edge detection and shape feature extraction;

[0127] The specific steps of edge detection are:

[0128] 1. Gaussian filtering: The filtered image I′=I*G;

[0129] 2. Gradient calculation: G x (x,y)=i(x+1,y)-I(x-1,y),G y (x,y)=I(x,y+1)-I(x,y-1)

[0130] Gradient Amplitude direction

[0131] 3. Non-maximum suppression: retain the local maximum value in the gradient direction;

[0132] 4. Double threshold: connect strong edges and weak edges.

[0133] The area of ​​shape feature extraction is calculated by counting the number of pixels in the area, the perimeter can be calculated by the sum of the distances between edge pixels, and the circularity is usually defined as 4πA / P 2 , where A is the area and P is the perimeter.

[0134] Furthermore, an evaluation and error correction module is included, which is used to perform the following steps:

[0135] A1: Collect a large amount of raw image data of the liver detected by transient elastography technology, as well as the corresponding actual liver condition information annotated by doctors, and use a convolutional neural network. The model includes multiple convolutional layers, pooling layers, and fully connected layers;

[0136] A2: Input the prepared image data and annotation information into the model, forward propagate and calculate the prediction results, define the loss function, calculate the gradient through the back propagation algorithm, and update the model parameters;

[0137] A3: Input the questionable or erroneous image into the model trained in step A2, and the model outputs the corrected prediction result of liver problems;

[0138] A4: Randomly select an area on the image to crop and retain, randomly rotate the image at a certain angle, and make slight random changes to the color channels of the image; try different learning rates, adjust the depth and width structure parameters of the model, and try different regularization strengths;

[0139] A5: Use precision, recall, and F1 value metrics to evaluate the accuracy of the model in predicting liver problems, and use the confusion matrix to visually view the correct and incorrect predictions for each category.

[0140] Specifically, in the step of evaluating the implementation of the error correction module, step A1 is data preparation and building a deep learning model;

[0141] More specifically, we will collect a wide range of original images of the liver detected by transient elastography from different patients and under different testing conditions. These images should cover various liver conditions, including healthy liver, different degrees of fatty liver, cirrhosis, etc. Liver disease experts will carefully annotate each image, clearly indicate the actual liver problem corresponding to the image, such as liver hardness level, whether there is a lesion, etc., and accurately associate this annotation information with the image.

[0142] ResNet50 is selected as the convolutional neural network architecture. This architecture contains multiple convolutional layers. The first layer can have 64 convolution kernels of size 7×7, which are used to extract the primary features of the image; followed by a pooling layer, such as maximum pooling, to reduce the size of the feature map and reduce the amount of calculation; followed by multiple residual modules, each of which consists of multiple convolutional layers to achieve deep feature extraction and transmission; and finally a fully connected layer is connected to map the extracted features to the category space of liver problems.

[0143] Step A2 is the training process: the prepared image data and annotation information are input into the model in batches. Each batch can contain, for example, 32 images; the model performs forward propagation and calculates the prediction results of liver problems through operations such as convolution and pooling. For each category, a predicted probability distribution is obtained; the loss function is defined, the cross entropy loss, for a sample, if the true category is c, the predicted probability is p c , then the loss is -log(p c ), sum the losses of all samples; calculate the gradient of each model parameter relative to the loss using the back-propagation algorithm using the automatic derivation mechanism; use stochastic gradient descent (SGD) or its variants, such as the Adam optimizer, to update the model parameters according to the learning rate and gradient. For example, for a parameter w, the update formula is:

[0144]

[0145] Where α is the learning rate.

[0146] Step A3 is image error correction. When encountering an image with doubts or obvious errors, it is input into the trained model; the model outputs the corrected prediction results for liver problems based on the learned features and patterns; and the results are compared and analyzed with the original results to determine the effectiveness of the error correction.

[0147] Step A4 is data enhancement and hyperparameter tuning, where the data enhancement involves random cropping, random rotation, and color jittering;

[0148] More specifically, random cropping is to randomly select a rectangular area on the image for cropping, and the position and size of the cropping are random. For example, the coordinates of the upper left corner of the cropping area can be randomly generated within the image range, and the side length of the cropping area can be randomly selected within a certain range, such as [32,64] pixels.

[0149] Random rotation is to perform random rotation within a certain angle range, such as [-30 degrees, 30 degrees]. The pixel coordinates after rotation are calculated by trigonometric functions to achieve the image rotation effect.

[0150] Color jitter is a small random change to each color channel of the image (such as R, G, B). You can add a small random value to each channel, such as between [-0.1, 0.1], to simulate different lighting conditions.

[0151] Hyperparameter tuning is to try a series of different values ​​for the learning rate, such as [0.0001, 0.001, 0.01, 0.1], observe the performance of the model on the validation set, and select the optimal learning rate; adjust the depth of the model, such as increasing or decreasing the number of residual modules; adjust the width of the model, such as increasing or decreasing the number of convolution kernels in each convolutional layer; try different regularization parameters, such as the weight coefficient of L2 regularization, and observe the inhibitory effect on model overfitting.

[0152] Step A5 is model evaluation, involving accuracy, recall, and F1 value;

[0153] More specifically, accuracy is calculated as the ratio of correctly predicted samples to the total number of samples; accuracy is equal to the number of correctly predicted samples divided by the total number of samples.

[0154] The recall rate is the ratio of the number of samples correctly predicted to be in a certain category to the actual number of samples in that category for a specific liver problem category; the recall rate is equal to the number of samples correctly predicted to be in that category divided by the actual number of samples in that category.

[0155] The F1 value comprehensively considers the accuracy and recall rate, and measures the model performance through the harmonic mean of the two; the F1 value is equal to 2 multiplied by the product of the accuracy and recall rate divided by the sum of the accuracy and recall rate.

[0156] You can also use the confusion matrix to show the correct and incorrect predictions for each category in detail, and intuitively understand the performance of the model on different categories.

[0157] Furthermore, the medical history detection and retrieval module uses the medical information platform to retrieve the information, and marks the patient's liver disease history, medication and test items in the retrieved information;

[0158] The medical information platform is based on the medical personnel's use of standardized terminology in their daily work to accurately enter the patient's medical history, medication records and test records into the medical information platform and update them in real time; the medical information platform is equipped with a permission management mechanism.

[0159] Specifically, the medical information platform is already in use in actual applications. This system needs to establish a connection to apply the medical information platform. When medical personnel enter data, they must ensure that the data format is unified and the terminology is standardized to facilitate subsequent retrieval and analysis.

[0160] In order to protect the privacy and personal information of patients, the system needs to take strict security measures, such as encryption technology, firewalls, etc., to prevent data leakage and unauthorized access; whenever there is new medical history, test results or medication, the records in the system should be updated in time to keep the data timely and accurate.

[0161] Furthermore, the shear wave imaging module and the medical history detection and retrieval module are used to jointly analyze the cause of the disease and give suggestions, including the following steps:

[0162] B1: Through the shear wave imaging module and the medical history detection and retrieval module, the liver stiffness value, echo characteristics, liver function index, virus markers, medication records and detailed medical history are extracted as key features;

[0163] B2: Using the improved clustering algorithm and based on the expert database, a diagnostic model is constructed. The detected liver images and information of the patient are input into the diagnostic model to output the most likely cause of the disease.

[0164] B3: Generate personalized treatment recommendations based on the most likely cause of the disease output and expert database matching, and feed back the patient's treatment effects and expert recommendations into the diagnostic model and treatment recommendations for continuous optimization.

[0165] Specifically, the images generated by transient elastography technology are associated and integrated with the corresponding patient test records (such as blood tests, imaging tests, etc.), medication records and detailed medical history to ensure the integrity and accuracy of the information; special image analysis software or algorithms are used to process transient elastography images to extract key features, such as liver stiffness values, echo characteristics, etc.; combined with the patient's medical history information (including previous medical history, family medical history, lifestyle habits, etc.), various test results (such as liver function indicators, viral markers, etc.) and medication history, a comprehensive analysis and evaluation is conducted.

[0166] Using machine learning or deep learning technology, a diagnostic model is constructed based on a large amount of known case data. The model can predict possible causes of the disease based on the input images and related information; the current patient's images and other information are input into the diagnostic model, and the most likely cause of the disease is determined through the model's output and the doctor's professional judgment; based on the determined cause, with reference to clinical guidelines, expert consensus, and past treatment experience, and combined with the patient's individual situation (such as age, physical condition, etc.), personalized treatment recommendations are generated. The treatment recommendations may include drug treatment plans, treatment cycles, and subsequent follow-up plans.

[0167] If necessary, for complex or difficult cases, multidisciplinary experts should be organized to consult, integrate opinions from different professional fields, and further improve the cause diagnosis and treatment recommendations.

[0168] During the treatment process, we continuously monitor changes in the patient's condition, including new test results, image changes, etc., and adjust treatment recommendations in a timely manner based on actual conditions; we feed back treatment effects into the diagnostic model and treatment recommendation generation system, and continuously optimize the model and recommendations to improve accuracy and effectiveness.

[0169] Furthermore, the improved clustering algorithm comprises the following steps:

[0170] B21: Preprocess and standardize the key features collected in step B1, divide the data into multiple clusters, and randomly select K data points as the initial cluster centers;

[0171] B22: Calculate the distance from each data point to each cluster center, and assign the data point to the cluster to which the nearest cluster center belongs; recalculate the center of each cluster, that is, the average value of the features of all data points in the cluster;

[0172] B23: Repeat step B22 until the location of the cluster center no longer changes significantly or the set number of iterations is reached, analyze the clustering results, observe the characteristic distribution of patient data in different clusters, and discover patterns related to the cause or condition.

[0173] Specifically, suppose there is a set of data points X = {x1, x2, ..., x n}, each data point x i Contains multiple features; determines the number of clusters K, randomly selects K data points as the initial cluster centers, denoted by C = {c1, c2, ..., c k}.

[0174] Assign data points: For each data point x i , calculate it to each cluster center c j The Euclidean distance

[0175]

[0176] Where m is the number of features, x i，k Represents data point x i The kth eigenvalue of j，k Represents the cluster center c j，k Then x i Assigned to the cluster with the closest cluster center.

[0177] Update cluster center: For each cluster, recalculate the cluster center. That is, the new cluster center c j Each eigenvalue of is the average of the corresponding eigenvalues ​​of all data points in the cluster. The formula is:

[0178]

[0179] Where N j is the number of data points in cluster j.

[0180] The data points are assigned and the cluster centers are updated repeatedly until the positions of the cluster centers no longer change significantly or the set number of iterations is reached, and the final clustering results are analyzed and interpreted.

[0181] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. An intelligent analysis system based on transient elastic imaging technology, characterized in that: include: The shear wave imaging module, the medical history detection and retrieval module and the intelligent analysis module are connected in sequence by wireless communication, wherein: The shear wave imaging module is used to generate shear waves and longitudinal waves through energy pulses, and to perform imaging processing on the real-time data of shear waves obtained from patient testing; The medical history test retrieval module is used to obtain detailed patient conditions through patient medical history and test records; The intelligent analysis module is used to jointly analyze through the shear wave imaging module and the medical history detection and retrieval module to determine the cause of the disease and give suggestions.

2. The intelligent analysis system based on transient elastic imaging technology according to claim 1, characterized in that: The method of generating shear waves and longitudinal waves by energy pulses and performing imaging processing on the real-time shear wave data obtained by patient detection includes the following steps: S1: Acquire ultrasonic signals through signal transmission, signal reception and signal quality processing; S2: Focus the received signal to the target area to improve the resolution and contrast of the image; S3: converting the processed signal into an image; S4: Feature extraction for image readability and diagnostic value.

3. The intelligent analysis system based on transient elastic imaging technology according to claim 2, characterized in that: The method of acquiring ultrasonic signals by signal transmission, signal reception and signal quality processing comprises the following steps: S11: Generate low-frequency elastic waves through an external mechanical vibration device to act on liver tissue; S12: Low-frequency elastic waves propagate in liver tissue, exciting shear waves; S13: the ultrasonic transducer array transmits an ultrasonic beam to the shear wave region; S14: The transducer receives the reflected ultrasonic signal and performs amplification, filtering and digitization operations.

4. The intelligent analysis system based on transient elastic imaging technology according to claim 2, characterized in that: The step of focusing the received signal to the target area and improving the resolution and contrast of the image comprises the following steps: S21: Calculate the delay time required for each array element to receive the signal according to the relative position of the transducer array element and the target point; S22: adjusting the time of the signal received by each array element according to the calculated delay time; S23: Assign a certain weight to the signal of each array element to control the directivity and shape characteristics of the beam; S24: summing up the signals of each array element after delay adjustment and weighting processing to obtain a beamformed signal.

5. The intelligent analysis system based on transient elastic imaging technology according to claim 2, characterized in that: The process of converting the processed signal into an image comprises the following steps: S31: obtaining a series of signal data received by the ultrasonic transducer after beamforming, and performing necessary preprocessing to divide the imaging area into a number of small spatial grids; S32: for each grid point, estimating relevant parameter values ​​of the grid point according to information of the strength and propagation time of surrounding received signals; S33: Perform multiple iterative calculations to continuously optimize the values ​​of grid points and perform image verification based on longitudinal wave positioning performance; S34: constructing a final liver image according to the calculated values ​​of each grid point.

6. The intelligent analysis system based on transient elastic imaging technology according to claim 2, characterized in that: The feature extraction of the readability and diagnostic value of the image includes the following steps: S41: Calculate the histogram of the image constructed in step S3, calculate the cumulative histogram, determine the new grayscale value corresponding to each grayscale level through the cumulative histogram, and perform contrast stretching; S42: Determine the size of the filter window, calculate the average value of the pixels in its neighborhood for each pixel and replace the corresponding pixel, calculate the median value of the pixels in the neighborhood and replace the central pixel; S43: using the maximum inter-class variance method to classify pixels in the image whose grayscale values ​​are greater than or equal to the threshold into one category, and pixels whose grayscale values ​​are less than the threshold into another category, selecting a seed pixel, setting a grayscale difference threshold, and starting from the seed pixel, merging adjacent pixels that meet the similarity criterion into the same area, and continuously growing the area; S44: Perform Gaussian filtering to smooth the image, calculate the gradient amplitude and direction, perform non-maximum suppression, refine the edge, use double thresholds for edge connection and judgment, and extract liver shape features.

7. The intelligent analysis system based on transient elastic imaging technology according to claim 1, characterized in that: Also included is an evaluation and error correction module for performing the following steps: A1: Collect a large amount of raw image data of the liver detected by transient elastography technology, as well as the corresponding actual liver condition information annotated by doctors, and use a convolutional neural network. The model includes multiple convolutional layers, pooling layers, and fully connected layers; A2: Input the prepared image data and annotation information into the model, forward propagate and calculate the prediction results, define the loss function, calculate the gradient through the back propagation algorithm, and update the model parameters; A3: Input the questionable or erroneous image into the model trained in step A2, and the model outputs the corrected prediction result of liver problems; A4: Randomly select an area on the image to crop and retain, randomly rotate the image at a certain angle, and make slight random changes to the color channels of the image; try different learning rates, adjust the depth and width structure parameters of the model, and try different regularization strengths; A5: Use precision, recall, and F1 value metrics to evaluate the accuracy of the model in predicting liver problems, and use the confusion matrix to visually view the correct and incorrect predictions for each category.

8. The intelligent analysis system based on transient elastic imaging technology according to claim 1, characterized in that: The medical history detection and retrieval module uses the medical information platform to retrieve the patient's liver disease history, medication and test items in the retrieved information; The medical information platform is based on the medical personnel's use of standardized terminology in their daily work to accurately enter the patient's medical history, medication records and test records into the medical information platform and update them in real time; the medical information platform is equipped with a permission management mechanism.

9. The intelligent analysis system based on transient elastic imaging technology according to claim 1, characterized in that: The shear wave imaging module and the medical history detection and retrieval module jointly analyze to determine the cause and give suggestions, including the following steps: B1: Through the shear wave imaging module and the medical history detection and retrieval module, the liver stiffness value, echo characteristics, liver function index, virus markers, medication records and detailed medical history are extracted as key features; B2: Using the improved clustering algorithm and based on the expert database, a diagnostic model is constructed. The detected liver images and information of the patient are input into the diagnostic model to output the most likely cause of the disease. B3: Generate personalized treatment recommendations based on the most likely cause of the disease output and expert database matching, and feed back the patient's treatment effects and expert recommendations into the diagnostic model and treatment recommendations for continuous optimization.

10. The intelligent analysis system based on transient elastic imaging technology according to claim 9, characterized in that: The improved clustering algorithm comprises the following steps: B21: Preprocess and standardize the key features collected in step B1, divide the data into multiple clusters, and randomly select K data points as the initial cluster centers; B22: Calculate the distance from each data point to each cluster center, and assign the data point to the cluster to which the nearest cluster center belongs; recalculate the center of each cluster, that is, the average value of the features of all data points in the cluster; B23: Repeat step B22 until the location of the cluster center no longer changes significantly or the set number of iterations is reached, analyze the clustering results, observe the characteristic distribution of patient data in different clusters, and discover patterns related to the cause or condition.

Images