Wearable deep learning-based ultrasound image quality enhancement method and system

Abstract

The application belongs to the technical field of image processing. A wearable ultrasonic image quality enhancement method and system based on deep learning are provided. The single-angle echo data is IQ demodulated to obtain complex IQ data containing amplitude and phase information. The complex IQ data is time delay compensated to obtain time delay IQ data. The time delay IQ data is input into an encoder to obtain low-dimensional feature representation. The low-dimensional feature representation is input into a time-frequency feature fusion module to obtain a reconstructed enhanced ultrasonic image. The application models the echo data of single-angle plane waves through a deep learning model, improves the resolution, contrast and image quality of the ultrasonic image, retains the high frame rate characteristics of single-angle imaging, ensures that it is suitable for portable and low-power wearable ultrasonic equipment, and meets the needs of actual monitoring and clinical application.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a wearable ultrasound image quality enhancement method and system based on deep learning. Background Art

[0002] The statements in this section merely provide background art related to the present invention and do not necessarily constitute prior art.

[0003] Wearable ultrasound devices are gaining attention due to their portability and real-time monitoring capabilities. However, these devices still face numerous challenges in practical medical applications. Limited by device size and power consumption, the number and performance of sensors are often inferior to those of traditional ultrasound devices, resulting in insufficient clarity and contrast in acquired ultrasound images. Furthermore, factors such as patient movement, ambient noise, and poor probe contact can cause noise, blurring, and artifacts in ultrasound images, compromising diagnostic accuracy.

[0004] Conventional coherent plane wave compounding (CPWC) significantly improves image quality through multi-angle transmission and coherent summation. However, this method requires multiple plane wave transmissions, resulting in a significant decrease in frame rate and making it unsuitable for real-time wearable ultrasound devices. Summary of the Invention

[0005] To address the shortcomings of the existing technology, the present invention provides a wearable ultrasound image quality enhancement method and system based on deep learning. By modeling the echo data of a single-angle plane wave through a deep learning model, the method improves the resolution, contrast, and image quality of ultrasound images while retaining the high frame rate characteristics of single-angle imaging, ensuring its suitability for portable, low-power wearable ultrasound devices and meeting the needs of actual monitoring and clinical applications.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a wearable ultrasound image quality enhancement method based on deep learning.

[0008] A wearable ultrasound image quality enhancement method based on deep learning includes the following processes:

[0009] Perform IQ demodulation on the single-angle echo data to obtain complex IQ data containing amplitude and phase information;

[0010] Performing time delay compensation on the complex IQ data to obtain time delayed IQ data;

[0011] Input the time-delayed IQ data into the encoder to obtain a low-dimensional feature representation;

[0012] inputting the low-dimensional feature representation into a time-frequency feature fusion module to obtain a reconstructed enhanced ultrasound image.

[0013] As a further limitation of the first aspect of the application, the complex IQ data is time delay compensated to obtain time delayed IQ data, comprising:

[0014] wherein, is the time delayed IQ data, wherein, , , and denotes the coordinates of the imaging point, and denotes the position coordinates of the th probe, denotes the single angle, is the complex IQ data, represents the wave speed, represents the propagation time of the transmitted signal at the angle , represents the propagation time of the received signal at the angle , represents the total time of flight.

[0015] As a further limitation of the first aspect of the application, the time-frequency feature fusion module comprises a time domain branch and a frequency domain branch, the time domain branch obtains a time domain feature representation through time domain feature extraction;

[0016] The frequency domain branch comprises a low frequency branch and a high frequency branch in parallel, the low frequency branch extracts a low frequency feature representation through a low frequency feature extraction module, and the high frequency branch extracts a high frequency feature representation through a high frequency feature extraction module;

[0017] The feature fusion module performs weighted fusion on the time domain feature representation, the low frequency feature representation and the high frequency feature representation, and outputs a reconstructed enhanced ultrasound image.

[0018] As a further limitation of the first aspect of the application, the time domain branch obtains a time domain feature representation through time domain feature extraction, comprising:

[0019] The input feature is first normalized through a normalization layer, and the normalized feature is subjected to a convolution layer to capture the change pattern in the time dimension through a local receptive field. After the convolution operation of the convolution layer, the feature is subjected to a nonlinear transformation through an activation function, and a time domain feature representation is output.

[0020] As a further limitation of the first aspect of the application, the low-frequency branch extracts a low-frequency feature representation through a low-frequency feature extraction module, including:

[0021] The low-frequency branch converts the input feature from the time domain to the frequency domain through a two-dimensional Fourier transform, passes through a low-pass filter with learnable parameters, extracts the low-frequency component of the signal, and obtains the low-frequency feature representation after the low-frequency feature extraction module and the two-dimensional inverse Fourier transform;

[0022] The high-frequency branch extracts a high-frequency feature representation through a low-frequency feature extraction module, including:

[0023] The high-frequency branch converts the input feature from the time domain to the frequency domain through a two-dimensional Fourier transform, passes through a high-pass filter with learnable parameters, extracts the high-frequency component of the signal, and obtains the high-frequency feature representation after the high-frequency feature extraction module and the two-dimensional inverse Fourier transform.

[0024] As a further limitation of the first aspect of the application, the feature fusion module performs weighted fusion on the time domain feature representation, the low-frequency feature representation and the high-frequency feature representation, including:

[0025] An adaptive weight learning module is used to integrate the time domain feature representation, the low-frequency feature representation and the high-frequency feature representation, generate three feature weights, and normalize the three feature weights through a normalization function to ensure that the sum of the weights is 1;

[0026] The normalized weights are multiplied by the time domain feature, the low-frequency feature and the high-frequency feature respectively, and the features are weighted and processed. Finally, the weighted features are linearly superimposed, converted to the frequency domain through a two-dimensional Fourier transform, filtered through learnable parameters, and then transformed back to the time-space domain to obtain the reconstructed enhanced ultrasound image.

[0027] As a further limitation of the first aspect of the application, the loss function is a weighted sum of the pixel-level loss, the structural similarity loss and the wavelet loss.

[0028] In a second aspect, the application provides a wearable ultrasound image quality enhancement system based on deep learning.

[0029] A wearable ultrasound image quality enhancement system based on deep learning, including:

[0030] A data demodulation unit configured to IQ demodulate single-angle echo data to obtain complex IQ data containing amplitude and phase information;

[0031] A time delay compensation unit configured to perform time delay compensation on the complex IQ data to obtain time-delayed IQ data;

[0032] a data encoding unit configured to input the time-delayed IQ data into an encoder to obtain a low-dimensional feature representation;

[0033] a time-frequency feature fusion unit configured to input the low-dimensional feature representation into a time-frequency feature fusion module to obtain a reconstructed enhanced ultrasound image.

[0034] In a third aspect, the present application provides a computer device, comprising: a processor and a computer readable storage medium;

[0035] a processor adapted to execute a computer program;

[0036] a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the deep learning-based wearable ultrasound image quality enhancement method according to the first aspect of the present application.

[0037] In a fourth aspect, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by a processor to implement the deep learning-based wearable ultrasound image quality enhancement method according to the first aspect of the present application.

[0038] Compared with the prior art, the present application has the following beneficial effects:

[0039] 1. The present application significantly improves the imaging quality of a wearable ultrasound device by using a deep learning-based image reconstruction model. Under the constraints of low power consumption and single-angle plane wave transmission, the generated image is close to the effect of multi-angle compound imaging (CPWC), while retaining the high frame rate characteristic of single-angle plane wave imaging.

[0040] 2. The present application introduces a time-frequency feature fusion module (TFF), which effectively extracts and utilizes time-domain and frequency-domain information through a multi-branch design and a dynamic weight fusion mechanism. Compared with traditional models, the present application further improves the resolution and contrast of the image, and generates high-quality images suitable for dynamic monitoring and clinical diagnosis, thereby providing reliable support for the practical application of wearable ultrasound devices.

[0041] The advantages of the additional aspects of the present application will be partially given in the following description, partially become obvious from the following description, or be understood through the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0042] The accompanying drawings, which form a part of the present application, are used to provide further understanding of the present application, and the illustrative embodiments of the present application and their description are used to explain the present application, and do not constitute improper limitations on the present application.

[0043] Figure 1A model training process schematic diagram provided for the embodiment 1 of the present application;

[0044] Figure 2 A model application process schematic diagram provided for the embodiment 1 of the present application;

[0045] Figure 3 An effect schematic diagram before and after enhancement provided for the embodiment 1 of the present application, wherein, (A) is an ultrasound image before enhancement, and (B) is an ultrasound image after enhancement;

[0046] Figure 4 A schematic diagram of a wearable ultrasound image quality enhancement system based on deep learning provided for the embodiment 2 of the present application;

[0047] Figure 5 A schematic diagram of a computer device provided for the embodiment 3 of the present application. DETAILED DESCRIPTION

[0048] The present application will be further described below in conjunction with the accompanying drawings and embodiments.

[0049] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0050] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0051] Embodiment 1:

[0052] The present implementation proposes a wearable ultrasound image quality enhancement method based on deep learning, which effectively improves the imaging quality of wearable ultrasound equipment and reduces the influence caused by the limitations of device volume, power consumption and sensor performance. At the same time, the method uses single-angle transmitted echo data as input to generate high-quality images, which reduces the transmission times of the wearable ultrasound probe as much as possible under the premise of not sacrificing image quality, so as to meet the requirements of wearability and low power consumption.

[0053] The present application respectively uses software simulation and real human body acquisition method to obtain echo signal data. Specifically, for simulated data, first generate a scatter pattern with different acoustic impedance distribution, and then calculate the probe echo data through software simulation. For real human body data, first, the probe is attached to the skin, and the plane wave imaging method is used to transmit plane waves from multiple angles in turn, and receive echo data, and then the sampled echo data is obtained through ADC sampling, denoted as , wherein, is the transmission angle of the plane wave. The echo data of each angle contains echo signals from each array element, denoted as wherein, represents the i-th array element, it should be noted that in subsequent calculations, if not specifically marked , only is marked, it means that the echo data of each array element is calculated in turn in the echo data with the emission angle .

[0054] More specifically, if it is multi-angle, assuming there are 75 angles, each angle contains the echo signal of 128 array elements, the echo signal of each array element is discretely sampled, and the sampling point is assumed to be 4096. The shape of the final data is RF (75, 128, 4096) such as a size, corresponding to RF (alpha, i, t); if it is single angle, it is RF (1, 128, 4096), the number of imaging points is obtained according to the image size, assuming that the imaging size is , the number of imaging points to be calculated is 65536, that is, 65536 (x, y). The meaning of the per-array element calculation of the present implementation is that all angles are traversed first (here the angle is 1, so it does not need to be traversed), and then all array elements in each angle are traversed. For each array element [coordinate (x i , y i )], each imaging point [65536 (x, y)] is traversed, and each imaging point is calculated according to formulas (2)~(5). The shape of the final TDIQ is (1, 128, ), which can be directly input into the model.

[0055] For the echo data RF, any single angle is selected, that is , that is, the echo data of the plane wave of a single angle is used as input, and the method of CPWC is used for imaging for all angles , to obtain a high-quality image as a target, denoted as HQI, and HQI constitute an example in the data set.

[0056] The training framework of the model is shown in Figure 1 , which is divided into three parts, including echo data representation, image reconstruction model and loss function constraint.

[0057] In the present implementation, the echo data representation stage, specifically, includes:

[0058] First, the single-angle echo data is IQ demodulated to obtain complex IQ data containing amplitude and phase information, denoted as:

[0059] ​ (1);

[0060] wherein, represents the emission angle The echo signal received by the i-th array element, is the carrier frequency, represents time.

[0061] IQ demodulation is a demodulation technique for recovering the original baseband signal from the received modulated signal, which is widely used in wireless communication, radar, ultrasonic detection and other fields. The core idea is to separate the in-phase (I) and quadrature (Q) components of the signal using orthogonal carriers, so as to extract the amplitude and phase information of the signal.

[0062] Then, according to the coordinates of the imaging point and the position of the transmitting array element, the time of flight (TOF) is calculated, and then the time delay compensation is applied to the IQ data to obtain the time-delayed IQ data (TDIQ):

[0063] (2) ；

[0064] (3);

[0065] (4);

[0066] (5);

[0067] wherein, and represent the coordinates of the imaging point, and represent the position coordinates of the i-th probe, y is 0 by default, and c represents the wave speed, represents the propagation time of the transmitted signal at angle , represents the propagation time of the received signal at angle , represents the total time of flight.

[0068] Finally, is input to the autoencoder structure for data compression and dimension reduction. Specifically, Data is first processed by an encoder for feature extraction and compression. The purpose of the encoder design is to significantly reduce the amount of data and redundant information while retaining the key features of the imaging points to the greatest extent. The output of the encoder is a compact low-dimensional feature representation, which will serve as input for subsequent image reconstruction. At the same time, in order to ensure the effectiveness of the compressed features, the autoencoder structure also contains a decoder for restoring the low-dimensional feature representation generated by the encoder to the original data. The purpose of this process is to evaluate the degree of information retention during compression and restoration, ensuring that the low-dimensional features extracted by the encoder can cover the key information of the original data.

[0069] In this implementation, the image reconstruction model, specifically, includes:

[0070] In order to fully extract the information in the data compressed by the encoder, the present application proposes a time-frequency feature fusion module (Time-Frequency Feature Fusion Module, TFF) as a basic module in the network structure, which is applicable to network structures including but not limited to Unet and ResNet, etc. This module extracts and fuses time domain and frequency domain features of input features through multi-branch design, generating high-quality feature representation, providing key support for image reconstruction. The image reconstruction model of the present application takes the time-frequency feature fusion module as the core, which includes feature extraction module (TFE) and feature fusion module (FFM) two main parts. The feature extraction module is composed of time domain branch and frequency domain branch, which is used to extract comprehensive time domain and frequency domain information from input features, while the feature fusion module fuses the features output by the multi-branch, as the output of the time-frequency feature fusion module.

[0071] The feature extraction module first processes the input feature into the time domain branch and the frequency domain branch respectively. The time domain branch obtains the time domain feature representation through time domain feature extraction. Specifically, the input feature first passes through a normalization layer to adjust the numerical range of the feature and reduce the deviation of the data distribution. The normalized feature passes through a convolution layer to capture the change pattern in the time dimension through a local receptive field. After convolution operation, the feature is nonlinearly transformed through an activation function to further enhance the expression ability of the feature. Finally, the time domain branch outputs the time domain feature representation .

[0072] The frequency domain branch is further divided into a low frequency branch and a high frequency branch, which respectively process the low frequency and high frequency components of the signal. The low frequency branch first converts the input feature from the time domain to the frequency domain through a two-dimensional Fourier transform (2DFFT), and the result of the Fourier transform contains the amplitude and phase information of the signal. Then, the low frequency component is extracted through a low-pass filter with learnable parameters. Subsequently, the low frequency component is processed through a low frequency feature extraction module to obtain a low frequency feature representation . Specifically, the low frequency component enters a complex domain normalization layer to adjust the amplitude distribution of the frequency domain data, making it more suitable for subsequent convolution operations. The normalized feature is further captured through a complex convolution layer to capture local feature information in the low frequency component. Subsequently, the convolution feature is nonlinearly transformed through a complex activation function to enhance the expression ability of the feature. Finally, the low frequency feature is restored to the time-space domain through a two-dimensional inverse Fourier transform (2DIFFT) to generate a low frequency feature representation .

[0073] The processing process of the high frequency branch is similar to that of the low frequency branch. First, the input feature is converted to the frequency domain through a two-dimensional Fourier transform; then, a high-pass filter with learnable parameters is used to extract the high frequency component of the signal. Subsequently, the high frequency component is processed through a high frequency feature extraction module to obtain a high frequency feature representation . Specifically, the high frequency component is processed through a complex domain normalization layer to adjust the distribution of the feature; the normalized feature is captured through a complex convolution layer to capture the detailed characteristics of the high frequency information. Subsequently, the convolution feature is further enhanced through a complex activation function to enhance the expression ability of the high frequency feature. Finally, the high frequency feature is restored to the time-space domain through a two-dimensional inverse Fourier transform to generate a final high frequency feature representation .

[0074] The feature fusion module receives the feature representations output by the time domain branch, the low frequency branch and the high frequency branch, and performs weighted fusion on them. First, an adaptive weight learning module is used to integrate the three features to generate three feature weights. These weights are normalized through a normalization function to ensure that the sum of the weights is 1. The normalized weights are multiplied by the time domain feature, the low frequency feature and the high frequency feature respectively to perform weighted processing on the features. Finally, the weighted features are linearly superimposed, then converted to the frequency domain through a two-dimensional Fourier transform, filtered through learnable parameters, and then transformed back to the time-space domain to obtain the output of the time-frequency feature fusion module .

[0075] It should be noted that the feature extraction structure of the time domain branch, the feature extraction structure after filtering of the frequency domain branch, and the adaptive weight learning module are not limited to the combination of the normalization layer, the convolution layer and the nonlinear layer mentioned above.

[0076] Through the above structural design, the image reconstruction model combines and builds TFF as a basic module, and finally reconstructs a high-quality image, ensuring that the output of the model has good definition and contrast.

[0077] In the present embodiment, the loss function constraint includes:

[0078] For the autoencoder, the MSE loss is used to measure the difference between the input data (i.e. ) and the decoder output , so as to evaluate the information retention ability of the autoencoder in the compression process:

[0079] (6);

[0080] Wherein, represents the total number of elements in , represents the th element in X.

[0081] For the reconstruction model, in order to ensure that the model can generate high-quality reconstructed images, the present application designs a set of multi-objective loss function constraint mechanism to constrain and optimize the model from multiple levels of pixel level, structure level and frequency domain level. By combining different loss functions, the model improves the image resolution and contrast while considering the preservation of global information and local details.

[0082] The model uses pixel-level loss to constrain the pixel difference between the reconstructed image and the target image . Specifically, the pixel-level loss calculates the mean square error (MSE) between the predicted image and the real image to ensure the reconstruction accuracy at the pixel level:

[0083] (7);

[0084] Wherein, N represents the total number of elements in Y.

[0085] The present application introduces structure similarity (SSIM) loss to evaluate the structural consistency between the reconstructed image and the target image. The SSIM loss quantifies the similarity of image brightness, contrast and local structure, effectively improving the model's ability to preserve image details and visual quality:

[0086] (8);

[0087] represents the mean value of the target image Y, represents the mean value of the reconstructed image . denotes the square of denotes the square of denotes the square of (the standard deviation of the target image Y), that is, the variance of the target image Y, denotes the square of (the standard deviation of the reconstructed image ), that is, the variance of the reconstructed image ; denotes the covariance of the target image Y and the reconstructed image ; and are constants, which can be set as needed to prevent the denominator from being 0, or have a default calculation method, such as , , L is the dynamic range of the image (such as an 8-bit grayscale image L=255), 0.01 by default, 0.03 by default.

[0088] In addition, in order to further optimize the frequency domain information, the application designs a loss function based on wavelet transform. Specifically, the target image and the predicted image are respectively decomposed by wavelet, and the approximation coefficients and the detail coefficients are obtained. In the wavelet loss, the importance of global information and local details is balanced by weighted sum of the mean square error of the approximation coefficients and the detail coefficients, so as to enhance the sensitivity of the model to different frequency information.

[0089] (9);

[0090] wherein, and are weight coefficients, both of which are 0.5; A and D represent the approximation coefficients and the detail coefficients after wavelet transform of the target image Y; and represent the approximation coefficients and the detail coefficients after wavelet transform of the reconstructed image ; , , , are tensors, and the subscript represents the first element; represents the number of elements of (same as ), represents the number of elements of (same as ), and here is also used for traversal in the summation sign.​​

[0091] The total loss function of the model is composed of a weighted combination of multiple loss functions, which effectively improves the overall quality of image reconstruction through multi-objective constraints. Each loss function cooperates during the training process to guide the model to generate high-resolution, high-contrast ultrasound images that meet the requirements of wearable ultrasound devices for image quality:

[0092] (10).

[0093] wherein, , and are weight coefficients.

[0094] The model is first trained on a large number of simulation data sets to obtain pre-training weights, and then the autoencoder structure is fine-tuned on real human data sets to obtain the final model. The loss function described above is used for constraint during the training process.

[0095] The image reconstruction model of the present application can generate high-quality ultrasound images based on the plane wave echo data emitted by a wearable ultrasound device in a single shot. Specifically, the application of the model includes the following steps:

[0096] As shown in Figure 2 , first, a single-angle plane wave emission is performed using a wearable ultrasound probe, and the corresponding echo signal is received. After analog-to-digital conversion (ADC sampling) processing, the received echo signal generates the original data after sampling. Subsequently, according to the coordinate information of the imaging area, the time of flight (TOF) of each imaging point is calculated, and the original echo data is compensated for TOF to generate time-delayed IQ data (IQ ); then, the generated is compressed and reduced in dimension by the encoder to extract a compact low-dimensional feature representation; subsequently, the compressed feature representation is input into the image reconstruction model to finally generate an enhanced ultrasound image, and the quality of the output image approaches the effect of multi-angle compound imaging (CPWC), while retaining the high frame rate advantage of single-angle plane wave imaging. The generated high-quality image is suitable for clinical diagnosis or dynamic monitoring tasks, such as cardiovascular health assessment or real-time functional monitoring. As shown in Figure 3 , the effect diagram before and after enhancement is shown in Figure 3 , where (A) is the ultrasound image before enhancement, Figure 3 (B) is the ultrasound image after enhancement, and it can be seen that the image quality has been significantly enhanced after using the method of the present implementation.

[0097] Example 2:

[0098] As shown in Figure 4As shown, the present implementation provides a deep learning-based wearable ultrasound image quality enhancement system, comprising:

[0099] a data demodulation unit configured to IQ demodulate the single-angle echo data to obtain complex IQ data containing amplitude and phase information;

[0100] a time delay compensation unit configured to perform time delay compensation on the complex IQ data to obtain time-delayed IQ data;

[0101] a data encoding unit configured to input the time-delayed IQ data into an encoder to obtain a low-dimensional feature representation;

[0102] a time-frequency feature fusion unit configured to input the low-dimensional feature representation into a time-frequency feature fusion module to obtain a reconstructed enhanced ultrasound image.

[0103] The specific working methods of each unit are described in Embodiment 1 and will not be repeated here.

[0104] It can be understood that the above-mentioned units can be combined into one or several other units respectively or entirely, or some of the units can be further split into a plurality of units with smaller functions to constitute, which can achieve the same operation without affecting the realization of the technical effects of the embodiments of the present application. The above-mentioned units are divided based on logical functions, and in actual application, the functions of one unit can also be realized by multiple units, or the functions of multiple units can be realized by one unit. In other embodiments of the present application, the system can also include other units, and in actual application, these functions can also be realized by other units and can be realized by multiple units in cooperation.

[0105] According to another embodiment of the present application, the system described in the embodiment can be constructed and the method of Embodiment 1 can be realized by running a computer program (including program code) capable of performing each step involved in the corresponding method described in Embodiment 1 on a general computing device such as a computer including processing elements and storage elements such as a Central Processing Unit (CPU), a Random Access Memory (RAM), a Read Only Memory (ROM), etc., the computer program can be recorded on a computer readable recording medium such as a computer readable recording medium, and loaded into the above-mentioned computing device through the computer readable recording medium and run therein.

[0106] Embodiment 3:

[0107] As Figure 5As shown, the present implementation provides an electronic device including a processor 1001, a communication interface 1002, and a computer readable storage medium 1003. Among them, the processor 1001, the communication interface 1002, and the computer readable storage medium 1003 can be connected through a bus or other means.

[0108] Among them, the communication interface 1002 is used to receive and send data, the computer readable storage medium 1003 can be stored in the memory of the electronic device, the computer readable storage medium 1003 is used to store a computer program, the computer program includes program instructions, and the processor 1001 is used to execute the program instructions stored by the computer readable storage medium 1003.

[0109] The processor 1001 (also called CPU (Central Processing Unit, Central Processing Unit)) is the computing core and control core of the electronic device, which is suitable for implementing one or more instructions, and is particularly suitable for loading and executing one or more instructions to realize the corresponding method process or corresponding function.

[0110] The processor 1001 is configured to perform the following process:

[0111] IQ demodulation is performed on the single-angle echo data to obtain complex IQ data containing amplitude and phase information;

[0112] Time delay compensation is performed on the complex IQ data to obtain time-delayed IQ data;

[0113] The time-delayed IQ data is input into an encoder to obtain a low-dimensional feature representation;

[0114] The low-dimensional feature representation is input into a time-frequency feature fusion module to obtain a reconstructed enhanced ultrasound image.

[0115] The specific working method is described in Example 1, which will not be repeated here.

[0116] Example 4:

[0117] The present implementation provides a computer readable storage medium (Memory), which is a memory device in an electronic device, used to store programs and data. It can be understood that the computer readable storage medium here can include the built-in storage medium in the electronic device, and of course can also include the expansion storage medium supported by the electronic device. The computer readable storage medium provides a storage space, and the storage space stores the processing system of the electronic device.

[0118] Also, stored in the storage space are one or more instructions adapted to be loaded and executed by the processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0119] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to implement the following processes:

[0120] IQ demodulation is performed on the single-angle echo data to obtain complex IQ data containing amplitude and phase information;

[0121] Time delay compensation is performed on the complex IQ data to obtain time-delayed IQ data;

[0122] The time-delayed IQ data is input into an encoder to obtain a low-dimensional feature representation;

[0123] The low-dimensional feature representation is input into a time-frequency feature fusion module to obtain a reconstructed enhanced ultrasound image.

[0124] The specific working method is described in Embodiment 1 and will not be repeated here.

[0125] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present application can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those of ordinary skill in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0126] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of the present application are generated. The computer can be a general purpose computer, a special purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in or transmitted by a computer readable storage medium. The computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (for example, coaxial cable, optical fiber, digital line (DSL)) or wireless (for example, infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium that can be accessed by a computer or a data processing device such as a server, data center, etc. integrated with one or more available media. The available media can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD), or a semiconductor medium (for example, solid state disk (SSD)) and the like.

[0127] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A deep learning-based wearable ultrasound image quality enhancement method, characterized by, The method comprises the following processes: IQ demodulation is performed on single-angle echo data to obtain complex IQ data containing amplitude and phase information; time delay compensation is performed on the complex IQ data to obtain time-delayed IQ data; the time-delayed IQ data is input into an encoder to obtain a low-dimensional feature representation; the low-dimensional feature representation is input into a time-frequency feature fusion module to obtain a reconstructed enhanced ultrasound image; time delay compensation is performed on the complex IQ data to obtain time-delayed IQ data, comprising: wherein, is the time delayed IQ data, wherein, , , and denotes the coordinate of the imaging point, and denotes the position coordinate of the th probe, denotes the single angle, is the complex IQ data, represents the wave velocity, represents the propagation time of the transmitted signal at the angle , represents the propagation time of the received signal at the angle , represents the total time of flight; The time-frequency feature fusion module comprises a time domain branch and a frequency domain branch, and the time domain branch obtains a time domain feature representation through time domain feature extraction; The frequency domain branch comprises a low frequency branch and a high frequency branch in parallel, the low frequency branch extracts a low frequency feature representation through a low frequency feature extraction module, and the high frequency branch extracts a high frequency feature representation through a high frequency feature extraction module; The feature fusion module performs weighted fusion on the time domain feature representation, the low frequency feature representation and the high frequency feature representation, and outputs a reconstructed enhanced ultrasound image.

2. The deep learning-based wearable ultrasound image quality enhancement method of claim 1, wherein The time domain branch obtains a time domain feature representation through time domain feature extraction, comprising: The input feature is first normalized through a normalization layer, and the normalized feature is subjected to a convolution layer to capture the change pattern in the time dimension through a local receptive field, and then subjected to an activation function for nonlinear transformation of the feature, and finally outputting the time domain feature representation.

3. The deep learning-based wearable ultrasound image quality enhancement method of claim 1, wherein The low frequency branch extracts a low frequency feature representation through a low frequency feature extraction module, comprising: The low frequency branch converts the input feature from the time domain to the frequency domain through a two-dimensional Fourier transform, extracts the low frequency component of the signal through a low pass filter with learnable parameters, and obtains the low frequency feature representation through a low frequency feature extraction module and a two-dimensional inverse Fourier transform; The high frequency branch extracts a high frequency feature representation through a low frequency feature extraction module, comprising: The high frequency branch converts the input feature from the time domain to the frequency domain through a two-dimensional Fourier transform, extracts the high frequency component of the signal through a high pass filter with learnable parameters, and obtains the high frequency feature representation through a high frequency feature extraction module and a two-dimensional inverse Fourier transform.

4. The deep learning-based wearable ultrasound image quality enhancement method of claim 1, wherein The feature fusion module performs weighted fusion on the time domain feature representation, the low frequency feature representation and the high frequency feature representation, comprising: An adaptive weight learning module is used to integrate the time domain feature representation, the low frequency feature representation and the high frequency feature representation to generate three feature weights, and a normalization function is used to normalize the three feature weights to ensure that the sum of the weights is 1. The normalized weights are multiplied with the time domain feature, the low frequency feature and the high frequency feature respectively, and the features are weighted and processed. Finally, the weighted features are linearly superimposed, converted to the frequency domain through two-dimensional Fourier transform, filtered through learnable parameters, and then transformed back to the time-space domain to obtain the reconstructed enhanced ultrasound image.

5. The deep learning-based wearable ultrasound image quality enhancement method according to any one of claims 1-4, characterized in that, The loss function is a weighted sum of the pixel-level loss, the structural similarity loss, and the wavelet loss.

6. A deep learning based wearable ultrasound image quality enhancement system, characterized by, The deep learning-based wearable ultrasound image quality enhancement method according to any one of claims 1-5, comprising: a data demodulation unit configured to IQ demodulate the single-angle echo data to obtain complex IQ data containing amplitude and phase information; a time delay compensation unit configured to perform time delay compensation on the complex IQ data to obtain time-delayed IQ data; a data encoding unit configured to input the time-delayed IQ data into an encoder to obtain low-dimensional feature representation; a time-frequency feature fusion unit configured to input the low-dimensional feature representation into a time-frequency feature fusion module to obtain a reconstructed enhanced ultrasound image.

7. A computer device, comprising: comprising: a processor and a computer readable storage medium; a processor adapted to execute a computer program; a computer readable storage medium having a computer program stored therein, wherein the computer program is executed by the processor to implement the deep learning-based wearable ultrasound image quality enhancement method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by the processor to implement the deep learning-based wearable ultrasound image quality enhancement method according to any one of claims 1-5.

Images