Microvascular imaging system based on ultrasonic waves
Through ultrasonic-based microvascular imaging system, piezoelectric crystal and Doppler effect detection technology, combined with machine learning and deep learning, the problem of insufficient resolution and sensitivity of traditional ultrasonic imaging technology in microvascular imaging is solved, and high resolution is achieved. Deep penetration and real-time imaging, suitable for clinical applications.
Patent Information
- Application Number
- CN202411086710.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-08
- Publication Date
- 2024-11-15
AI Technical Summary
Traditional ultrasonic imaging technology has insufficient resolution and sensitivity in microvascular imaging, making it difficult to clearly display the microvascular structure. The prior art such as optical imaging and magnetic resonance imaging also lacks penetration depth and real-time.
Using ultrasonic-based microvascular imaging system, including an ultrasonic generator, a microvascular imaging module and an image analysis module, ultrasonic waves are generated through piezoelectric crystal vibration, and the blood flow velocity and direction are detected using the Doppler effect. Combined with machine learning and deep learning technology, we extracted The structural characteristics and blood flow parameters of microvascular, and the quantitative analysis results are output.
It realizes high resolution, deep penetration and real-time imaging, can automatically identify and analyze microvascular structures, provide quantitative analysis results, is suitable for microvascular imaging of deep tissues, and is non-invasive and safe, suitable for clinical applications.
Smart Images

Figure CN118948330A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of microvascular imaging, and in particular to a microvascular imaging system based on ultrasound. Background Art
[0002] Microvascular imaging is of great significance in medical diagnosis and treatment. Structural and functional changes in microvessels are often early signs of many diseases, such as tumors, diabetes, and cardiovascular diseases. Therefore, the development of a high-resolution, non-invasive microvascular imaging technology is of great value for the early diagnosis and treatment of diseases. Ultrasonic imaging technology has become a hot topic in research and application due to its advantages such as non-invasiveness, real-time, and high resolution. However, the resolution and sensitivity of traditional ultrasonic imaging technology in microvascular imaging are still insufficient and need further improvement and optimization.
[0003] Current microvascular imaging technologies mainly include the following:
[0004] Optical imaging techniques, such as optical coherence tomography (OCT) and laser Doppler imaging (LDI), have high resolution but limited penetration depth and are suitable for superficial tissues.
[0005] Magnetic resonance imaging (MRI): It has good soft tissue contrast, but the equipment is expensive, the imaging time is long, and it is not suitable for real-time monitoring.
[0006] Traditional ultrasonic imaging: Although it has good penetration depth and real-time imaging capabilities, its resolution and sensitivity in microvascular imaging are insufficient, making it difficult to clearly display the microvascular structure. Summary of the invention
[0007] The purpose of the present invention is to provide an ultrasound-based microvascular imaging system to solve the above technical problems;
[0008] The technical problem solved by the present invention can be achieved by adopting the following technical solutions:
[0009] Ultrasound-based microvascular imaging system, including,
[0010] Ultrasonic generator, which generates ultrasonic waves for microvascular imaging through the vibration of piezoelectric crystals;
[0011] A microvascular imaging module, connected to the ultrasonic generator, for detecting blood flow velocity and direction based on the Doppler effect and generating microvascular network ultrasonic images;
[0012] An image analysis module is connected to the microvascular imaging module to receive the microvascular network ultrasound image. The image analysis module is used to identify and analyze the structural and functional changes of microvessels, detect and obtain the structural characteristics and blood flow parameters of microvessels, and output quantitative analysis results. The image analysis module includes:
[0013] A multi-scale analysis unit, used for extracting the structural features from the microvascular network ultrasound image at multiple scales;
[0014] The image analysis model is connected to the multi-scale analysis unit, extracts the structural features and the blood flow parameters based on machine learning, and outputs the quantitative analysis results.
[0015] Preferably, the piezoelectric crystals are distributed in an array, and the ultrasonic generator controls the excitation time of each piezoelectric crystal in a phased array manner to control the focusing position and direction of the ultrasonic wave.
[0016] Preferably, the microvascular network ultrasound image is an ultrasound image obtained by imaging the ultrasonic echo signals of the microvessels after using a microbubble contrast agent.
[0017] Preferably, the image analysis module further comprises a filtering unit connected to the multi-scale analysis unit for preprocessing the microvascular network ultrasound image, wherein the filtering unit comprises:
[0018] An image segmentation subunit, used for segmenting the microvascular network ultrasound image into a plurality of local blocks, each of which contains a set number of pixels;
[0019] A local statistical feature calculation subunit, connected to the image segmentation subunit, and configured to calculate a local statistical feature for each of the local blocks;
[0020] The adaptive filter adjustment subunit is connected to the local statistical feature calculation subunit and is used to adjust the filter parameters of the filter according to the local statistical feature. The formula is:
[0021]
[0022] Wherein, (x, y) is the relative position of the pixel in the local block with the center of the local block as the origin; h(x, y) is the weight of the filter at the position (x, y), and σ is the filter parameter adaptively adjusted according to the local statistical features;
[0023] The filter processing subunit includes the filter connected to the adaptive filter adjustment subunit, and is used to perform filter processing on the microvascular network ultrasound image according to the filter parameters.
[0024] Preferably, the multi-scale analysis unit uses Gaussian pyramid and wavelet transform to extract the structural features from the microvascular network ultrasound image at multiple scales;
[0025] The Gaussian pyramid performs Gaussian blur processing on the original microvascular network ultrasound image to obtain a blurred image, and downsamples the blurred image to obtain an image of the next scale. After multiple downsampling, a predetermined number of scale layers is obtained. The formula is:
[0026] I I+1 (x,y)=(I I *G σ )↓2
[0027] Among them I I represents the I-th layer image, G σ represents a Gaussian filter, ↓2 represents a downsampling operation;
[0028] The wavelet transform uses a wavelet basis function to perform wavelet decomposition on the microvascular network ultrasound image to obtain subband coefficients in different directions. The formula is:
[0029]
[0030] Where W j (x, y) represents the wavelet coefficient at the jth scale, I(x, y) represents the pixel value of the original microvascular network ultrasound image at (x, y), ∑ m,n The summation symbol in I(x,y) represents the summation of all pixel positions (m,n) in the microvascular network ultrasound image, ψ j,m,n (x, y) represents the spatial position of the wavelet basis function at scale j and position (m, n).
[0031] Preferably, the image analysis model uses a convolutional neural network to identify and analyze microvessels;
[0032] The convolutional neural network includes:
[0033] An input layer, used for inputting the microvascular network ultrasound image;
[0034] The convolution layer is connected to the input layer, and convolutions the input microvascular network ultrasound image through multiple convolution kernels to extract the feature map. The formula is:
[0035]
[0036] Wherein, f(x,y) represents the feature map, I(x,y) represents the pixel value of the microvascular network ultrasound image at (x,y), and k(i,j) represents the convolution kernel;
[0037] The pooling layer is connected to the convolutional layer and is used to downsample the feature map. The formula is:
[0038] P(x,y)=maxi,j f(x+i,y+j)
[0039] Wherein, P(x, y) represents the output result of the pooling layer, corresponding to the value after pooling at position (x, y) in the feature map, and max i,j It means that at the given position of f(x,y), the largest value is selected as the value of P(x,y) in the i×j region;
[0040] A fully connected layer, connected to the pooling layer, used to expand the output result of the pooling layer into a one-dimensional vector and perform classification;
[0041] The output layer is connected to the fully connected layer and is used to output the recognition result of the microvessels.
[0042] Preferably, the image analysis model further uses a generative adversarial network to generate microvascular images, and the generative adversarial network includes:
[0043] A generator is used to generate the microvascular image, and the generating formula is:
[0044] G(z)=σ1(W g ·z+b g )
[0045] Where z represents the random noise vector, W g and b g represents the weight and bias of the generator, σ1 represents the activation function;
[0046] The discriminator is connected to the generator and is used to receive input image data and determine whether the input image data is a real image or the microvascular image generated by the generator. The formula is:
[0047] D(x)=σ2(W d ·x+b d )
[0048] Where x represents the input image data, W d and b d represents the weight and bias of the discriminator, and σ2 represents the activation function.
[0049] Preferably, the microvascular imaging module comprises
[0050] A high-speed data acquisition card, used for acquiring the ultrasonic echo signal;
[0051] An image processing unit, connected to the high-speed data acquisition card, receives the ultrasonic echo signal, detects the blood flow velocity and direction based on the Doppler effect, and generates the microvascular network ultrasonic image;
[0052] The real-time image display unit is connected to the image processing unit and is used to display the microvascular network ultrasound image in real time.
[0053] Preferably, the quantitative analysis results at least include the density, diameter, and number of branches of microvessels, and the image analysis module further includes a report generation unit for generating an analysis report including the quantitative analysis results.
[0054] Preferably, the microvascular imaging module and the image analysis module are arranged on a cloud platform, and a distributed computing framework and a stream processing framework are used to perform real-time data processing. The cloud platform also includes a data storage module, which connects the microvascular imaging module and the image analysis module, and uses object storage and distributed storage to store the microvascular network ultrasound image and the quantitative analysis results.
[0055] Beneficial effects of the present invention: Due to the adoption of the above technical solution, the present invention can achieve high resolution, deep penetration and real-time imaging through an ultrasound-based microvascular imaging system, and automatically identify and analyze microvascular structures and output quantitative analysis results. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 is a structural diagram of an ultrasound-based microvascular imaging system in an embodiment of the present invention;
[0057] Figure 2 A schematic diagram of the filter unit architecture in an embodiment of the present invention;
[0058] Figure 3 A schematic diagram of a convolutional neural network architecture in an embodiment of the present invention;
[0059] Figure 4 A schematic diagram of a generative adversarial network architecture in an embodiment of the present invention. DETAILED DESCRIPTION
[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0061] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.
[0062] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.
[0063] Ultrasound-based microvascular imaging systems, such as Figures 1 to 4As shown, including,
[0064] An ultrasonic generator 1 generates ultrasonic waves for microvascular imaging by vibrating a piezoelectric crystal 11;
[0065] A microvascular imaging module 2, connected to the ultrasonic generator 1, is used to detect blood flow velocity and direction based on the Doppler effect and generate an ultrasonic image of a microvascular network;
[0066] The image analysis module 3 is connected to the microvascular imaging module 2 to receive the microvascular network ultrasound image. The image analysis module 3 is used to identify and analyze the structural and functional changes of the microvessels, detect the structural characteristics and blood flow parameters of the microvessels, and output the quantitative analysis results. The image analysis module 3 includes:
[0067] A multi-scale analysis unit 32, used for extracting structural features from microvascular network ultrasound images at multiple scales;
[0068] The image analysis model 33 is connected to the multi-scale analysis unit 32, extracts structural features and blood flow parameters based on machine learning, and outputs quantitative analysis results.
[0069] Specifically, the present invention provides an ultrasound-based microvascular imaging system that can achieve high resolution, deep penetration and real-time imaging, addressing the deficiencies of the prior art in microvascular imaging and assisting in subsequent clinical diagnosis and treatment needs.
[0070] In a preferred embodiment, the piezoelectric crystals 11 are distributed in an array, and the ultrasonic generator 1 controls the excitation time of each piezoelectric crystal 11 in a phased array manner to control the focusing position and direction of the ultrasonic wave.
[0071] Specifically, the present invention generates high-frequency ultrasound by vibrating the piezoelectric crystal 11. The piezoelectric crystal 11 vibrates mechanically under the action of the electric field, and the frequency range is set to 10MHz to 30MHz. High-frequency ultrasound can provide higher spatial resolution, which helps to clearly display the microvascular structure.
[0072] Ultrasound has good tissue penetration ability and can penetrate thicker tissue layers to image deeper microvascular networks. Therefore, the present invention is not only suitable for microvascular imaging of superficial tissues, but also for microvascular imaging of deep tissues.
[0073] In addition, ultrasound imaging technology is non-invasive, safe, and does not cause radiation damage to patients. Compared with other imaging technologies, such as X-rays and CT, ultrasound imaging does not cause ionizing radiation hazards to the human body and is suitable for repeated detection and long-term monitoring.
[0074] Due to its non-invasiveness and safety, the ultrasonic microvascular imaging system is suitable for a wide range of clinical applications, including but not limited to tumor detection, cardiovascular disease diagnosis, diabetic complications monitoring, etc. The system provided by the present invention can provide reliable imaging support in different clinical scenarios to meet the diagnosis and treatment needs of different diseases.
[0075] Therefore, the ultrasound-based microvascular imaging system provided by the present invention overcomes the shortcomings of the prior art, has the advantages of high resolution, deep penetration, real-time imaging and non-invasiveness, and has broad clinical application prospects. In particular, its application in tumor vascular imaging and its significant effects provide strong technical support for clinical diagnosis and treatment.
[0076] More specifically, phased array technology is used to focus the ultrasonic energy on the target area by controlling the excitation time of each piezoelectric crystal 11. Phased array technology can dynamically adjust the focus position to improve the signal-to-noise ratio and resolution of the imaging area. Phased array technology can also achieve multi-angle imaging and provide richer vascular structure information.
[0077] Preferably, the present invention uses high-frequency ultrasound and phased array technology to achieve high-resolution, multi-angle microvascular imaging and provide more comprehensive vascular information.
[0078] In a preferred embodiment, the microvascular network ultrasound image is an ultrasound image obtained by imaging the ultrasonic echo signals of the microvessels after using a microbubble contrast agent.
[0079] Specifically, the present invention uses the Doppler effect to detect the blood flow velocity and direction. Doppler imaging technology generates dynamic images of microvascular networks by analyzing changes in ultrasonic frequency, and can display the blood flow velocity and direction in real time, thereby evaluating the functional state of blood vessels.
[0080] The use of microbubble contrast agents can enhance the reflection of ultrasonic signals. Microbubble contrast agents form a strong reflective interface in blood vessels, which can significantly improve the contrast and resolution of microvascular imaging. The use of microbubble contrast agents can more clearly show the structure and blood flow of microvessels.
[0081] By using microbubble contrast agents to enhance the reflection of ultrasound signals, microvascular structures can be clearly displayed even in deep tissues. Contrast agents can significantly increase the contrast of imaging, allowing the details of deep microvascular networks to be presented.
[0082] Preferably, the present invention uses Doppler imaging and microbubble contrast agents to enhance blood flow detection, provide high-contrast imaging results, and ensure clear visibility of blood flow and vascular structure.
[0083] In a preferred embodiment, the image analysis module 3 further includes a filtering unit 31 connected to the multi-scale analysis unit 32 for preprocessing the microvascular network ultrasound image. The filtering unit 31 includes:
[0084] The image segmentation subunit 311 is used to segment the microvascular network ultrasound image into a plurality of local blocks, each local block including a set number of pixels;
[0085] A local statistical feature calculation subunit 312, connected to the image segmentation subunit 311, for calculating local statistical features for each local block;
[0086] The adaptive filter adjustment subunit 313 is connected to the local statistical feature calculation subunit 312 and is used to adjust the filter parameters of the filter according to the local statistical features. The formula is:
[0087]
[0088] Where (x, y) is the relative position of the pixel in the local block with the center of the local block as the origin; h(x, y) is the weight of the filter at the position (x, y), and σ is the filter parameter adaptively adjusted according to the local statistical characteristics;
[0089] The filter processing subunit 314 includes a filter connected to the adaptive filter adjustment subunit 313, and is used to perform filter processing on the microvascular network ultrasound image according to the filter parameters.
[0090] Specifically, the purpose of the filtering unit 31 is to remove noise from the image while retaining the detailed structure of the microvessels. Commonly used filtering algorithms include Gaussian filtering, median filtering, and adaptive filtering. Existing filtering algorithms usually use fixed filtering parameters and cannot be adaptively adjusted according to the local features of the image. This technical solution can more effectively remove noise while retaining the detailed structure of the microvessels by introducing an adaptive filter.
[0091] In the present invention, an improved adaptive filtering algorithm is adopted. The adaptive filtering adjustment subunit 313 adaptively adjusts the filtering parameters according to the local features of the image to achieve the best denoising effect. The formula is:
[0092]
[0093] Among them, (x, y) is the relative position of the pixel in the local block with the center of the local block as the origin; h(x, y) is the weight of the filter at the position (x, y), and σ is the filter parameter adaptively adjusted according to the local statistical characteristics.
[0094] The above formula is a two-dimensional Gaussian function, which is used as the reference of adaptive filtering parameters. h(x,y) represents the weight of the filter at position (x,y). is the normalization constant of the Gaussian distribution, ensuring that the function integrates to 1 in the entire domain, where σ is the filter parameter adaptively adjusted according to local statistical characteristics, is the exponential part of the Gaussian function, where x 2 +y 2 Represents the square of the distance from the position (x, y) to the center point. The negative sign of the exponential part indicates that the weight decreases with the increase of distance. The farther the distance, the smaller the weight. (x, y) is the relative position of the pixel in the local block with the center of the local block as the origin. σ is the filter parameter adaptively adjusted according to the local statistical characteristics, and is adjusted according to the variance of the local block. If the variance of the local block is large, σ is large, and the Gaussian function will have a significant response in a larger area, and vice versa. The weight is applied to the filter, and the pixels of each local block are processed to remove noise and retain the detailed structure.
[0095] In a preferred embodiment, the multi-scale analysis unit 32 uses Gaussian pyramid and wavelet transform to extract structural features from the microvascular network ultrasound image at multiple scales;
[0096] The Gaussian pyramid performs Gaussian blur processing on the original microvascular network ultrasound image to obtain a blurred image, and then downsamples the blurred image to obtain an image of the next scale. After multiple downsampling, the predetermined scale layer is obtained. The formula is:
[0097] I I+1 (x,y)=(I I *G σ )↓2
[0098] Among them I I represents the I-th layer image, G σ represents a Gaussian filter, ↓2 represents a downsampling operation;
[0099] Wavelet transform uses wavelet basis functions to perform wavelet decomposition on microvascular network ultrasound images to obtain subband coefficients in different directions. The formula is:
[0100]
[0101] Where W j (x, y) represents the wavelet coefficient at the jth scale, I(x, y) represents the pixel value of the original microvascular network ultrasound image at (x, y), ∑ m,n The summation symbol in I(x,y) represents the summation of all pixel positions (m,n) in the microvascular network ultrasound image, ψ j,m,n (x,y) represents the spatial position of the wavelet basis function at scale j and position (m,n).
[0102] Specifically, the multi-scale analysis unit 32 processes the image at different scales using a multi-scale analysis method, so as to more comprehensively capture the structural features of the microvessels. The multi-scale analysis method of the present invention includes Gaussian pyramid and wavelet transform.
[0103] Gaussian Pyramid constructs a Gaussian pyramid of the image, each layer represents an image at a different scale.
[0104] The specific steps are as follows:
[0105] Step 1: Perform Gaussian blur processing on the original image to obtain a blurred image.
[0106] Step 2: downsample the blurred image to obtain an image of the next scale.
[0107] Step 3: Repeat the above steps until the predetermined number of scale layers is reached.
[0108] The formula is as follows,
[0109] I I+1 (x,y)=(I I *G σ )↓2
[0110] Among them I I represents the I-th layer image, G σ represents a Gaussian filter, ↓2 represents a downsampling operation;
[0111] First, for the I-th layer image I I Perform Gaussian blur processing. Gaussian filter G σ The original image is smoothed and blurred by reducing the high frequency between pixels in the image. The result of Gaussian blurring is a blurred image that retains the overall structural features of the original image but reduces the detail information. Next, the blurred image is downsampled.
[0112] The purpose of downsampling is to reduce the resolution of the image, that is, to reduce the number of pixels in the image, thereby reducing the size of the image. This step usually uses a sampling interval to reduce the size of the image. A common way is to take samples of every other pixel, or use methods such as pixel averaging to reduce the resolution of the image.
[0113] The above steps are iterated continuously, that is, the I+1th layer image is obtained by using the Ith layer image, until the predetermined number of scale layers is reached. The image of each layer has a greater blur and lower resolution than the previous layer.
[0114] The constructed Gaussian pyramid can provide image representations at different scales, thereby allowing the analysis and processing of image features at different spatial scales.
[0115] Wavelet transform can decompose an image into components of different scales and directions. Specifically in image processing, wavelet transform decomposes the image by selecting appropriate wavelet basis functions, such as Haar wavelet or Daubechies wavelet, to obtain its subband coefficients in different scales and directions.
[0116] The formula is,
[0117]
[0118] Where W j (x, y) represents the wavelet coefficient at the jth scale, I(x, y) represents the pixel value of the original microvascular network ultrasound image at (x, y), ∑ m,n The summation symbol in I(x,y) represents the summation of all pixel positions (m,n) in the microvascular network ultrasound image, ψ j,m,n (x,y) represents the spatial position of the wavelet basis function at scale j and position (m,n).
[0119] The wavelet basis function determines the characteristics of the wavelet transform, including the properties of the transform, such as orthogonality, compact support, and frequency response. Common wavelet basis functions include Haar wavelet and Daubechies wavelet. Each basis function is suitable for different types of signal or image processing tasks. In the above formula, I(x, y) represents the pixel value of the original microvascular network ultrasound image at (x, y), that is, the original image to be processed, and W j (x, y) represents the wavelet coefficient at the jth scale, that is, the image subband coefficient at the jth level obtained by wavelet transform, which describes the local characteristics of the image at different spatial scales. j,m,n (x, y) represents the spatial position of the wavelet basis function at scale j and position (m, n). The key to wavelet transform is that it can provide both spatial and scale information, so that the structural features of microvascular ultrasound images at different scales can be analyzed and processed.
[0120] In a preferred embodiment, the image analysis model 33 uses a convolutional neural network 4 to identify and analyze microvessels;
[0121] Convolutional neural network 4 includes,
[0122] An input layer 41 is used to input microvascular network ultrasound images;
[0123] The convolution layer 42 is connected to the input layer 41, and convolves the input microvascular network ultrasound image through multiple convolution kernels to extract the feature map. The formula is:
[0124]
[0125] Among them, f(x,y) represents the feature map, I(x,y) represents the pixel value of the microvascular network ultrasound image at (x,y), and k(i,j) represents the convolution kernel;
[0126] The pooling layer 43 is connected to the convolution layer 42 and is used to downsample the feature map. The formula is:
[0127] P(x,y)=max i,j f(x+i,y+j)
[0128] Among them, P(x,y) represents the output result of the pooling layer 43, which corresponds to the value after pooling at position (x,y) in the feature map, and max i,j It means that at the given position of f(x,y), the largest value is selected as the value of P(x,y) in the i×j region;
[0129] The fully connected layer 44 is connected to the pooling layer 43 and is used to expand the output result of the pooling layer 43 into a one-dimensional vector and perform classification;
[0130] The output layer 45 is connected to the fully connected layer 44 and is used to output the recognition result of the microvessels.
[0131] Specifically, the image analysis model 33 of the present invention adopts a deep learning model, which can automatically learn the features in the image and perform recognition and classification. The present invention adopts a convolutional neural network 4 to automatically recognize and analyze microvessels;
[0132] The input layer 41 receives the microvascular network ultrasound image as input data.
[0133] The convolution layer 42 uses multiple convolution kernels to perform convolution operations on the input image. Each convolution kernel can be regarded as a small window, and local features are extracted on the input image by sliding the window.
[0134] The formula is,
[0135]
[0136] Among them, f(x,y) represents the feature map, I(x,y) represents the pixel value of the microvascular network ultrasound image at (x,y), and k(i,j) represents the convolution kernel;
[0137] The above formula represents the basic form of the convolution operation, where f(x,y) is a position in the feature map, obtained by weighted summing the input image I(x,y) and the convolution kernel k(i,j); each element of the convolution kernel k(i,j) is multiplied by the corresponding element of the input image I(x+i,y+j), and then these products are accumulated to obtain a pixel value of the feature map. This process is repeated by sliding the convolution kernel on the input image until the entire area of the input image is covered.
[0138] The pooling layer 43 receives the feature map of the convolution layer 42 as input and performs a downsampling operation on it, which helps to reduce the size of the feature map while retaining the main features, thereby reducing the computational complexity of subsequent processing;
[0139] The formula is,
[0140] P(x,y)=max i,j f(x+i,y+j)
[0141] Among them, P(x,y) represents the output result of the pooling layer 43, which corresponds to the value after pooling at position (x,y) in the feature map, and max i,j It means that at the given position of f(x,y), the largest value in the i×j region is selected as the value of P(x,y); for example, the i×j region is a 2x2 or 3x3 region, and the largest value in this region is selected as the value of P(x,y);
[0142] In the above formula, P(x, y) represents the operation of the pooling layer 43. The present invention adopts maximum pooling, that is, the maximum value is selected as the output in the pooling window. The pooling layer 43 reduces the spatial size of the feature map by taking the maximum value in the local area. If it is a 2x2 area, then the size of the output feature map will be half of the input feature map. The maximum pooling operation provides position invariance, that is, a slight translation of the input feature map will not significantly affect the output after pooling, which is helpful for feature learning. In addition, the dimensionality reduction effect of the pooling layer 43 also helps to reduce the number of network parameters, thereby reducing the risk of overfitting and improving the generalization ability of the network.
[0143] The fully connected layer 44 flattens the output of the pooling layer 43 into a one-dimensional vector, and then classifies it through a neural network to integrate the features extracted by the pooling layer 43;
[0144] The output layer 45 receives the output of the fully connected layer 44 and generates the final microvessel recognition result.
[0145] Preferably, the convolutional neural network 4 can effectively learn and extract features from the input image by stacking convolutional layers 42, pooling layers 43 and fully connected layers 44, thereby realizing automatic recognition and analysis of details such as microvessels.
[0146] In a preferred embodiment, the image analysis model 33 further uses a generative adversarial network 5 to generate a microvascular image. The generative adversarial network 5 includes:
[0147] Generator 51 is used to generate microvascular images. The generating formula is:
[0148] G(z)=σ1(W g ·z+b g )
[0149] Where z represents the random noise vector, W g and b g represents the weight and bias of generator 51, σ1 represents the activation function;
[0150] The discriminator 52 is connected to the generator 51 and is used to receive input image data and determine whether the input image data is a real image or a microvascular image generated by the generator 51. The formula is:
[0151] D(x)=σ2(W d ·x+b d )
[0152] Where x represents the input image data, W d and b d represents the weight and bias of the discriminator 52, and σ2 represents the activation function.
[0153] Preferably, the image analysis module 3 of the present invention adopts advanced image analysis algorithms to improve image quality, automatically identify and extract microvascular structures, reduce human intervention, and assist in improving diagnostic efficiency.
[0154] Specifically, the role of the Generative Adversarial Network 5 (GAN) is mainly to generate high-resolution microvascular images and to determine the authenticity of the input images.
[0155] Generator 51 (Generator): Generator 51 receives a random noise vector z, which is usually sampled from a specific distribution, such as a normal distribution.
[0156] Generator 51 uses the learned parameters W g and b d , converting random noise z into a high-resolution microvascular image. This process is expressed as,
[0157] G(z)=σ1(W g ·z+b g )
[0158] Where z represents the random noise vector, W g and b grepresents the weight and bias of the generator 51, and σ1 represents the activation function, such as ReLU or sigmoid activation function.
[0159] The goal of the generator 51 is to generate microvascular images as realistic as possible, so that these generated images can deceive the discriminator 52 and make it be considered by the discriminator 52 as real images.
[0160] Discriminator 52: The discriminator 52 receives an input image x, which can be a real microvascular image or an image generated by the generator 51. The discriminator 52 uses the learned parameters W d and b d , calculate the probability that the input image x is a real image.
[0161] The expression of this process is,
[0162] D(x)=σ2(W d ·x+b d )
[0163] Where x represents the input image data, W d and b d represents the weight and bias of the discriminator 52, and σ2 represents the activation function.
[0164] The goal of the discriminator 52 is to be able to accurately distinguish between real microvascular images and generated microvascular images. The discriminator 52 is trained as a binary classifier to distinguish whether the image output by the generator 51 looks like a real microvascular image.
[0165] The Generative Adversarial Network 5 (GAN) drives the generator 51 and the discriminator 52 to continuously improve their performance by making them confront and compete with each other. Competition drives the generator 51 to generate more realistic and detailed microvascular images, while the discriminator 52 also becomes more accurate. Therefore, the microvascular images generated by the generator 51 can be used to enhance the performance of the convolutional neural network 4 (CNN) for more accurate analysis and identification of microvascular features.
[0166] Preferably, the present invention has significant advantages in image processing and image analysis by introducing efficient filtering algorithms, multi-scale analysis and deep learning models, and can more effectively remove noise, extract multi-scale features, and automatically identify and analyze microvascular structures, significantly improving the performance and practicality of microvascular imaging systems.
[0167] In a preferred embodiment, the microvascular imaging module 2 includes
[0168] High-speed data acquisition card, used to obtain ultrasonic echo signals;
[0169] The image processing unit is connected to a high-speed data acquisition card, receives ultrasonic echo signals, detects blood flow velocity and direction based on the Doppler effect, and generates microvascular network ultrasonic images;
[0170] The real-time image display unit is connected to the image processing unit and is used to display the microvascular network ultrasound image in real time.
[0171] Specifically, the present invention adopts a high-speed data acquisition card to acquire ultrasonic echo signals in real time. The high sampling rate of the acquisition card ensures the integrity and accuracy of the signal, providing a high-quality data basis for subsequent image processing.
[0172] The image processing unit processes the collected ultrasonic signals in real time and generates microvascular imaging results. The image processing unit uses parallel processing technology to ensure the real-time nature of the imaging process. The image processing unit uses the Doppler effect to detect the blood flow velocity and direction. Doppler imaging technology generates dynamic images of the microvascular network by analyzing the changes in ultrasonic frequency. It can display the blood flow velocity and direction in real time and help evaluate the functional status of microvessels.
[0173] The high-speed image processing unit processes the collected ultrasonic signals in real time and generates high-quality imaging results. The application of parallel processing technology ensures the real-time nature of the imaging process and can dynamically monitor the blood flow changes in microvessels; combined with the real-time image display unit, the microvascular imaging results are displayed in real time. Doctors can instantly observe the structure and blood flow changes of microvessels.
[0174] Specifically, the image processing unit introduces an image processing method when generating microvascular network ultrasound images, which at least includes
[0175] Image denoising methods, such as median filtering and Gaussian filtering, remove the noise generated during the imaging process and improve image quality.
[0176] Image enhancement methods, such as histogram equalization and contrast enhancement, are used to improve the visual effect of the image and make the microvascular structure clearer.
[0177] Image segmentation methods, through image segmentation methods such as threshold segmentation, region growing, etc., automatically identify and extract microvascular structures to provide more accurate vascular network images.
[0178] Specifically, the real-time image display unit integrates a high-resolution display to display the microvascular imaging results in real time, providing an intuitive visual reference, and assisting the user to instantly observe the microvascular structure and blood flow changes through the display for subsequent diagnosis.
[0179] Preferably, the real-time imaging and display functions of the microvascular imaging module 2 ensure the real-time nature and high-quality display of the imaging process, thereby assisting doctors in making rapid diagnoses and decisions.
[0180] In a preferred embodiment, the quantitative analysis result at least includes the density, diameter, and number of branches of microvessels, and the image analysis module 3 further includes a report generation unit for generating an analysis report including the quantitative analysis result.
[0181] Specifically, the image analysis module 3 integrates intelligent analysis software to automatically identify and analyze microvascular structure and functional changes. The software uses machine learning and artificial intelligence technology to automatically detect the morphological characteristics and blood flow parameters of microvessels and provide quantitative analysis results.
[0182] The image analysis module 3 can quantify the density, diameter, number of branches and other parameters of microvessels, and provide doctors with detailed diagnostic information.
[0183] In a preferred embodiment, the microvascular imaging module 2 and the image analysis module 3 are arranged on a cloud platform, and a distributed computing framework and a stream processing framework are used to perform real-time data processing. It also includes a data storage module 6, which connects the microvascular imaging module 2 and the image analysis module 3, and uses object storage and distributed storage to store microvascular network ultrasound images and quantitative analysis results.
[0184] Specifically, in order to realize the real-time processing and cloud computing functions of the microvascular imaging system, it is necessary to comprehensively utilize high-performance computing resources, distributed computing architecture and high-speed network transmission technology.
[0185] The present invention selects a suitable cloud computing platform, such as Amazon Web Services (AWS), Google Cloud Platform (GCP) and Microsoft Azure.
[0186] AWS: Provides a variety of computing services (such as EC2, Lambda), storage services (such as S3), database services (such as RDS, DynamoDB), etc.
[0187] GCP: Provides powerful computing engines (Compute Engine), container services (Kubernetes Engine), data processing services (BigQuery), etc.
[0188] Azure: Provides comprehensive computing resources (Virtual Machines), AI services (Azure Machine Learning), data storage (Blob Storage), etc.
[0189] The traditional local server processing capacity used in the prior art is limited, and it is difficult to cope with large-scale data processing and real-time computing requirements. The present invention provides elastically scalable computing resources and storage capabilities through a cloud platform, which can meet the needs of high-performance computing and big data processing.
[0190] More specifically, the present invention uses an efficient data transmission method combined with a cloud platform to achieve high-speed data transmission to the cloud for processing, including:
[0191] Use efficient data transmission protocols, such as HTTP / 2 and gRPC, to increase transmission speed;
[0192] Compress the transmitted data, such as using GZIP or Snappy, to reduce the amount of data transmitted;
[0193] Use fragmented transmission to transmit large data in fragments, and use multi-threading or multi-connection for parallel transmission to increase transmission speed;
[0194] And use encryption protocols such as TLS / SSL to ensure the security of data transmission.
[0195] Traditional data transmission methods are slow and less secure. The present invention can significantly improve data transmission speed and security by optimizing network protocols, data compression and fragmented transmission to meet real-time processing requirements.
[0196] The present invention also uses distributed computing to distribute tasks to multiple computing nodes for parallel processing, thereby improving computing efficiency and processing capacity.
[0197] Use distributed computing frameworks such as Apache Hadoop and Apache Spark for large-scale data processing. Hadoop is suitable for batch processing tasks and processes data through the MapReduce programming model. Spark supports batch processing and stream processing, has in-memory computing capabilities, and has faster processing speed.
[0198] Use container technologies such as Docker and Kubernetes for application deployment and management to improve the portability and scalability of the system.
[0199] Use task scheduling systems, such as Apache Airflow and Kubernetes CronJob, to schedule and manage tasks to ensure that tasks are executed on time.
[0200] The traditional single-machine processing capability is limited and it is difficult to cope with large-scale data processing needs. Preferably, the present invention improves processing efficiency and system scalability through distributed computing, parallel processing and task scheduling, and can cope with large-scale data processing needs.
[0201] Since real-time processing requires high-performance computing resources and efficient processing algorithms to ensure that data can be quickly processed and analyzed after collection, the present invention uses a stream processing framework, such as Apache Flink and Apache Storm, for real-time data processing; Flink has the ability to provide high-throughput and low-latency stream processing and supports complex event processing, while Storm provides real-time data processing capabilities and is suitable for low-latency real-time analysis scenarios.
[0202] Traditional batch processing methods are difficult to meet real-time processing requirements. The present invention can achieve low-latency, high-throughput data processing by introducing a stream processing framework and a real-time database to meet real-time processing requirements.
[0203] Specifically, the data storage module 6 supports the data storage of microvascular network ultrasound images, and uses high-capacity storage devices to ensure the safe preservation of imaging data. The storage system is designed as an expandable structure that can meet the storage requirements of large amounts of data.
[0204] The data storage module 6 also provides data management functions to facilitate the retrieval and analysis of imaging data of microvascular network ultrasound images. The data storage module 6 supports the classification, labeling and archiving of data to facilitate subsequent analysis and comparison. Doctors can retrieve historical imaging data through the data management system for longitudinal comparison and trend analysis.
[0205] The data storage module 6 of the present invention can ensure the rapid reading, writing and management of a large amount of data and support subsequent data analysis and processing. The data storage module 6 includes:
[0206] Use object storage services such as AWS S3 and Google Cloud Storage for large-scale data storage, supporting high-concurrency access and data persistence.
[0207] Use distributed file storage such as HDFS and Ceph for data storage and management, supporting large-scale data storage and efficient access.
[0208] Use distributed databases such as Cassandra and HBase for structured data storage, supporting high concurrent access and fast query.
[0209] By introducing cloud computing platforms, efficient data transmission, distributed computing, real-time processing, and efficient data storage, the real-time processing capability and scalability of the microvascular imaging system can be significantly improved. Compared with the prior art, the present invention has significant advantages in processing power, data transmission speed, real-time performance, and storage capacity, and can meet the needs of high-performance computing and large-scale data processing, and improve the overall performance of the system and user experience.
[0210] Preferably, the automated analysis of the present invention improves diagnostic accuracy, and efficient data storage and management supports long-term research and follow-up.
[0211] In a specific embodiment, taking tumor vascular imaging as an example,
[0212] Tumor growth is accompanied by the formation of new blood vessels (angiogenesis), and microvascular imaging can be used to detect tumors at an early stage. The new blood vessel network inside and around the tumor is an important sign of tumor growth and metastasis. Therefore, clear imaging and analysis of these microvascular networks is of great significance for the early diagnosis and treatment of tumors.
[0213] In the preliminary preparation step, the patient is placed in front of the ultrasound imaging device to ensure that the target area, such as the location of the tumor, is exposed and prepared for imaging. The patient has been injected with an ultrasound contrast agent, such as a microbubble contrast agent. The contrast agent is injected into the blood circulation through an intravenous injection to enhance the ultrasound reflection signal of the microvessels.
[0214] The real-time process of the present invention:
[0215] Step 1, start the high-frequency ultrasonic generator 1, and set the frequency at 20 MHz to ensure high-resolution imaging. Phased array technology is used to focus ultrasonic energy to the target area to ensure signal strength and resolution.
[0216] Step 2, the ultrasonic echo signal is acquired through the microvascular imaging module 2, and a real-time microvascular network ultrasonic image is generated. The tumor and the surrounding new blood vessel network are dynamically displayed using Doppler imaging technology.
[0217] The high-speed data acquisition card acquires the ultrasonic echo signal in real time and generates the microvascular network ultrasonic image through the high-speed image processing unit. The image quality is optimized by using the denoising, enhancement and segmentation algorithms to ensure the clear visibility of the microvascular structure.
[0218] Step 3, the image analysis module 3 intelligently analyzes and automatically identifies and analyzes the changes in microvascular structure and function, quantifies the density and distribution of new blood vessels, provides quantitative analysis results, and generates detailed reports, including microvascular density, diameter, number of branches and other parameters, to help doctors make accurate diagnoses.
[0219] Doctors can subsequently assess the malignancy and development trend of the tumor based on the imaging results and develop treatment plans, such as surgical resection, radiotherapy or chemotherapy.
[0220] Imaging data and analysis results are stored and managed for subsequent tracking and comparison, helping doctors evaluate treatment effectiveness and adjust treatment plans.
[0221] This case demonstrates the application of the present invention in tumor vascular imaging and its significant effects, providing strong technical support for clinical diagnosis and treatment.
[0222] The above description is only a preferred embodiment of the present invention, and does not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.
Claims
1. Ultrasound-based microvascular imaging system, It is characterized in that include, An ultrasonic generator (1) generates ultrasonic waves for microvascular imaging by vibrating a piezoelectric crystal (1); A microvascular imaging module (2) is connected to the ultrasonic generator (1) and is used to detect blood flow velocity and direction based on the Doppler effect and generate a microvascular network ultrasonic image; An image analysis module (3) is connected to the microvascular imaging module (2) to receive the microvascular network ultrasound image. The image analysis module (3) is used to identify and analyze the structural and functional changes of the microvessels, detect and obtain the structural characteristics and blood flow parameters of the microvessels, and output quantitative analysis results. The image analysis module (3) includes: A multi-scale analysis unit (32), used for extracting the structural features from the microvascular network ultrasound image at multiple scales; The image analysis model (33) is connected to the multi-scale analysis unit (32), extracts the structural features and the blood flow parameters based on machine learning, and outputs the quantitative analysis results.
2. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The piezoelectric crystals (1) are distributed in an array, and the ultrasonic generator (1) controls the excitation time of each piezoelectric crystal (1) in a phased array manner, so as to control the focusing position and direction of the ultrasonic wave.
3. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The microvascular network ultrasound image is an ultrasound image obtained by imaging the ultrasonic echo signals of the microvessels after using a microbubble contrast agent.
4. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The image analysis module (3) further comprises a filtering unit (31) connected to the multi-scale analysis unit (32) for preprocessing the microvascular network ultrasound image, wherein the filtering unit (31) comprises: An image segmentation subunit (311), used for segmenting the microvascular network ultrasound image into a plurality of local blocks, each of which contains a set number of pixels; A local statistical feature calculation subunit (312), connected to the image segmentation subunit (311), for calculating a local statistical feature for each local block; The adaptive filter adjustment subunit (313) is connected to the local statistical feature calculation subunit (312) and is used to adjust the filter parameters of the filter according to the local statistical feature. The formula is: Wherein, (x, y) is the relative position of the pixel in the local block with the center of the local block as the origin; h(x, y) is the weight of the filter at the position (x, y), and σ is the filter parameter adaptively adjusted according to the local statistical features; The filter processing subunit (314) includes the filter connected to the adaptive filter adjustment subunit (313), and is used to perform filter processing on the microvascular network ultrasound image according to the filter parameters.
5. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The multi-scale analysis unit (32) uses Gaussian pyramid and wavelet transform to extract the structural features from the microvascular network ultrasound image at multiple scales; The Gaussian pyramid performs Gaussian blur processing on the original microvascular network ultrasound image to obtain a blurred image, and downsamples the blurred image to obtain an image of the next scale. After multiple downsampling, a predetermined number of scale layers is obtained. The formula is: I I+1 (x,y)=(I I *G σ )↓2 Among them I I represents the I-th layer image, G σ represents a Gaussian filter, ↓2 represents a downsampling operation; The wavelet transform uses a wavelet basis function to perform wavelet decomposition on the microvascular network ultrasound image to obtain subband coefficients in different directions. The formula is: Where W j (x, y) represents the wavelet coefficient at the jth scale, I(x, y) represents the pixel value of the original microvascular network ultrasound image at (x, y), ∑ m,n The summation symbol in I(x,y) represents the summation of all pixel positions (m,n) in the microvascular network ultrasound image, ψ j,m,n (x, y) represents the spatial position of the wavelet basis function at scale j and position (m, n).
6. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The image analysis model (33) uses a convolutional neural network (4) to identify and analyze microvessels; The convolutional neural network (4) comprises: An input layer (41), used for inputting the microvascular network ultrasound image; The convolution layer (42) is connected to the input layer (41), and convolves the input microvascular network ultrasound image through multiple convolution kernels to extract a feature map. The formula is: Wherein, f(x,u) represents the feature map, I(x,y) represents the pixel value of the microvascular network ultrasound image at (x,y), and k(i,j) represents the convolution kernel; The pooling layer (43) is connected to the convolution layer (42) and is used to downsample the feature map. The formula is: P(x,y)=max i,j f(x+i,y+j) Wherein, P(x, y) represents the output result of the pooling layer (43), corresponding to the value after pooling at the position (x, y) in the feature map, and max i,j It means that at the given position of f(x,y), the largest value is selected as the value of P(x,y) in the i×j region; A fully connected layer (44), connected to the pooling layer (43), used to expand the output result of the pooling layer (43) into a one-dimensional vector and perform classification; The output layer (45) is connected to the fully connected layer (44) and is used to output the recognition result of the microvessels.
7. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The image analysis model (33) also uses a generative adversarial network (5) to generate a microvascular image. The generative adversarial network (5) includes: The generator (51) is used to generate the microvascular image, and the generating formula is: G(z)=σ 1 (IN g ·z+b g ) Where z represents the random noise vector, W g and b g represents the weights and biases of the generator (51), σ 1 represents the activation function; The discriminator (52) is connected to the generator (51) and is used to receive input image data and determine whether the input image data is a real image or the microvascular image generated by the generator (51). The formula is: D(x)=σ 2 (W d ·x+b d ) Where x represents the input image data, W d and b d represents the weight and bias of the discriminator (52), σ 2 Represents the activation function.
8. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The microvascular imaging module (2) comprises A high-speed data acquisition card (21) for acquiring the ultrasonic echo signal; An image processing unit (22) is connected to the high-speed data acquisition card (21), receives the ultrasonic echo signal, detects the blood flow velocity and direction based on the Doppler effect, and generates the microvascular network ultrasonic image; A real-time image display unit (23) is connected to the image processing unit (22) and is used to display the microvascular network ultrasound image in real time.
9. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The quantitative analysis result at least includes the density, diameter and number of branches of microvessels. The image analysis module (3) also includes a report generation unit for generating an analysis report including the quantitative analysis result.
10. The ultrasound-based microvascular imaging system according to claim 1, It is characterized in that The microvascular imaging module (2) and the image analysis module (3) are arranged on a cloud platform and use a distributed computing framework and a stream processing framework to perform real-time data processing. The cloud platform also includes a data storage module (6) which is connected to the microvascular imaging module (2) and the image analysis module (3) and uses object storage and distributed storage to store the microvascular network ultrasound image and the quantitative analysis result.
