Real-time dynamic MRI tumor imaging method and system based on compressed sensing

Abstract

The invention provides a real-time dynamic MRI (Magnetic Resonance Imaging) tumor imaging method and system based on compressed sensing, and relates to the technical field of medical imagines.The method comprises the following steps: acquiring two types of weighted image dynamic signal data generated by a detection device at different time points due to structural transformation through a compressed sensing technology; a multi-echo water-fat separation sequence is adopted for dynamic scanning, signals are separated according to water-fat chemical shift differences, and an initial water diagram and a non-water phase diagram are reconstructed and generated; performing time resolution analysis in combination with the dynamic signal data and the two types of images, generating a signal intensity curve of the detection device, and separating a specificity enhancement signal; and finally, in combination with the time dimension information, reconstructing the image by using a compressed sensing algorithm to obtain a dynamic enhanced image representing the target tissue, and according to the application, the artifacts are inhibited through the time evolution mode of the detection device, so that accurate real-time dynamic MRI tumor imaging can be realized.

Description

Technical Field

[0001] This application relates to the field of medical imaging technology, and in particular to a real-time dynamic MRI tumor imaging method and system based on compressed sensing. Background Technology

[0002] In scenarios such as accurately distinguishing tumor tissue from surrounding normal tissue and detecting small tumor lesions in the early stages, imaging technology needs to have high spatial resolution, high temporal dynamics, and strong specificity to clearly present tumor boundaries and capture signals of small lesions, providing accurate basis for clinical treatment planning and efficacy evaluation.

[0003] Currently, the most commonly used clinical approach is dynamic contrast-enhanced magnetic resonance imaging (MRI). This technique utilizes the differences in contrast agent perfusion between tissues to dynamically acquire MRI signals at different time points. After image reconstruction, blood perfusion information of the tumor tissue is obtained, thereby helping to determine the location and extent of the tumor.

[0004] However, existing methods have significant drawbacks: contrast agents lack tumor targeting, resulting in small enhancement differences between normal and tumor tissues, insufficient specificity, and a tendency to blur tumor boundaries; their ability to detect small tumor lesions is limited, constrained by spatial resolution and signal-to-noise ratio, making it difficult to capture signal changes in small lesions; the scanning time is long, the temporal resolution is low, making it impossible to monitor instantaneous signal changes in tumor tissue in real time, and they are susceptible to motion artifacts, affecting imaging accuracy. Summary of the Invention

[0005] The purpose of this application is to provide a real-time dynamic MRI tumor imaging method and system based on compressed sensing, so as to solve the problem of low temporal resolution in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a real-time dynamic MRI imaging method based on compressed sensing, comprising: Compressed sensing technology was used to collect dynamic signal data from the first and second type weighted images of the detection device at different time points caused by structural changes. A multi-echo water-fat separation sequence was used to dynamically scan the target area. Multi-echo data was acquired synchronously during the dynamic scanning process. The multi-echo data was separated according to the difference in chemical shift between water and fat to obtain water signals and fat signals. An initial water map was generated by MRI image reconstruction based on the separated water signals, and a non-water phase map was generated by MRI image reconstruction based on the separated fat signals. Based on the dynamic signal data, the initial water map, and the non-aqueous phase map, the multi-echo data is subjected to time-resolved analysis to generate a signal intensity curve of the detection device as a function of time. Based on the enhancement characteristics of the first type of weighted image signal and the second type of weighted image signal at different time points in the signal intensity curve, the specific first type of enhancement signal and the second type of enhancement signal are separated. Based on the specific first and second type enhancement signals, and combined with the time dimension information in the dynamic signal data, a compressed sensing algorithm is used to reconstruct MRI images to obtain dynamic enhanced images of the target tissue. The specific evolution pattern of the detection device in the time dimension is used to identify and suppress artifact interference in the dynamic enhanced images.

[0007] Optionally, based on the specific first-type and second-type enhancement signals, and combined with the temporal dimension information in the dynamic signal data, a compressed sensing algorithm is used to reconstruct the MRI image to obtain a dynamic enhanced image of the target tissue. Furthermore, the specific evolutionary pattern of the detection device in the temporal dimension is utilized to identify and suppress artifact interference in the dynamic enhanced image, including: The specific first type of enhancement signal and the specific second type of enhancement signal are combined into a joint enhancement signal, and the spatial location of the joint enhancement signal is consistent with the dynamic signal data. Based on the time dimension information of the dynamic signal data, a time evolution constraint function is constructed, which describes the prior pattern of the signal strength of the detection device changing with time. Using the joint enhanced signal as the initial input and the time evolution constraint function as the sparsity constraint, an objective optimization function is constructed. By iteratively solving the objective optimization function, the signal estimation value of each spatial location at continuous time nodes is updated. The updated signal estimate is compared with the dynamic signal data by residual. If the residual does not converge, the iteration is repeated. When the residual converges, the signal estimate is subjected to Fourier transform to reconstruct a dynamic enhanced image of the target tissue. During the reconstruction process, abnormal fluctuation points in the signal intensity curve that do not conform to the time evolution constraint function are detected. The spatial locations corresponding to the abnormal fluctuation points are marked as artifact interference regions, and the signal intensity of the artifact interference regions is set to zero in the dynamically enhanced image.

[0008] Optionally, using the joint enhanced signal as the initial input and the time evolution constraint function as the sparsity constraint, a target optimization function is constructed. The signal estimate for each spatial location at consecutive time nodes is updated by iteratively solving the target optimization function, including: The joint enhanced signal is assigned to the signal estimate as the initial value, and the objective optimization function is constructed using the time evolution constraint function as the basis for sparsity evaluation. The objective optimization function includes a data matching part between the signal estimate and the dynamic signal data, and a pattern conformance part of the signal estimate under the time evolution constraint function; The signal estimate is updated through a cyclic calculation process. In each iteration, the target optimization function value corresponding to the current signal estimate is calculated, and the signal estimate is adjusted based on the target optimization function value. The calculation process is repeated until the change in the target optimization function value is lower than the preset stopping condition, and then the updated signal estimate is output.

[0009] Optionally, based on the dynamic signal data, the initial water map, and the non-aqueous phase map, the multi-echo data is subjected to time-resolved analysis to generate a signal intensity curve of the detection device over time. Based on the enhancement characteristics of the first-type weighted image signal and the second-type weighted image signal at different time points in the signal intensity curve, specific first-type enhancement signals and second-type enhancement signals are separated, including: Using the initial water map as a spatial mask, the signal values ​​of the background region in the dynamic signal data are removed. The fat region is marked in the non-aqueous phase map, and the signal values ​​of the fat region in the dynamic signal data are filtered out. Based on the signal values ​​obtained after removing background and fat regions, a two-dimensional matrix of signal intensity at time nodes and spatial locations is constructed, wherein the rows of the two-dimensional matrix correspond to spatial locations and the columns correspond to time nodes. Curve fitting is performed on the signal intensity data representing the time change of each row in the two-dimensional matrix to generate the first type weighted image signal intensity curve and the second type weighted image signal intensity curve for each spatial location. Based on the rising slope and peak position of the first type of weighted image signal intensity curve, the region that meets the preset threshold condition is marked as a specific first type of enhanced signal; Based on the descent delay characteristics of the second type of weighted image signal intensity curve, regions that meet the preset time delay conditions are marked as specific second type enhanced signals.

[0010] Optionally, curve fitting is performed on the signal intensity data representing the time variation of each row in the two-dimensional matrix to generate the first-type weighted image signal intensity curve and the second-type weighted image signal intensity curve for each spatial location, including: Extract the time-series signal data corresponding to each spatial location in the two-dimensional matrix, establish the correspondence between time nodes and signal strength values, and form a set of discrete data points; Based on the physical characteristics of the detection device, the signal data is divided into a first signal channel and a second signal channel, wherein the first signal channel corresponds to the fast response characteristic and the second signal channel corresponds to the delayed response characteristic. For the first signal channel, a polynomial function approximation method is used, and the fitting parameters are determined by the least squares method to obtain the continuous function expression of the first signal channel; For the second signal channel, an exponential decay function model is used for fitting, and the decay constant and initial amplitude parameters are solved by nonlinear regression to obtain the continuous function expression of the second signal channel; Based on the reliability of the signal strength, different weights are assigned to the two continuous function expressions, and the continuous function expressions processed by the weighting factors are smoothed to eliminate high-frequency noise, thereby generating the first type of weighted image signal strength curve and the second type of weighted image signal strength curve.

[0011] Optionally, compressed sensing technology is used to acquire dynamic signal data from the first-type weighted image and the second-type weighted image at different time points caused by structural changes, including: At each time point, the first type of excitation pulse sequence is emitted using the radio frequency coil of the magnetic resonance device to excite the first type of signal response of the detection device with the frequency parameters corresponding to the first type of weighted image, and the first type of frequency domain data is acquired based on the random frequency sampling template. At the same time point, switch to the frequency parameters corresponding to the second type of weighted image, emit the second type of excitation pulse sequence to excite the second type of signal response of the detection device, and collect the second type of frequency domain data based on the random frequency sampling template; The first type of frequency domain data and the second type of frequency domain data at each time point are integrated in chronological order to form dynamic signal data containing spatial coordinates, time points, and signal type identifiers.

[0012] Optionally, a multi-echo water-fat separation sequence is used to dynamically scan the target region. Multi-echo data is acquired synchronously during the dynamic scan. The multi-echo data is separated based on the chemical shift differences between water and fat to obtain water and fat signals. An initial water map is generated based on the separated water signal through MRI image reconstruction, and a non-water phase map is generated based on the separated fat signal through MRI image reconstruction. This includes: The target area is continuously scanned using a multi-echo gradient echo sequence. The original frequency domain data is acquired synchronously at each echo time point, and the original frequency domain data at each echo time point is combined into a multi-channel multi-echo data block. Based on the preset chemical shift difference between water and fat, a phase difference evolution model is constructed that includes phase accumulation functions at different echo time points. The multi-echo data block is input into the phase difference evolution model, and the water signal and fat signal of each voxel are separated by solving a system of linear equations. The separated water signal is subjected to inverse Fourier transform to reconstruct a two-dimensional initial water map, and the separated fat signal is subjected to inverse Fourier transform to reconstruct a two-dimensional non-aqueous phase map.

[0013] Secondly, this application provides a real-time dynamic MRI tumor imaging system based on compressed sensing, comprising: The acquisition module uses compressed sensing technology to acquire dynamic signal data from the first and second type weighted images of the detection device at different time points caused by structural changes. The separation module is used to dynamically scan the target area using a multi-echo water-fat separation sequence. During the dynamic scanning process, multi-echo data is acquired synchronously. The multi-echo data is separated according to the chemical shift difference between water and fat to obtain water signals and fat signals. An initial water map is generated by MRI image reconstruction based on the separated water signals, and a non-water phase map is generated by MRI image reconstruction based on the separated fat signals. The generation module is used to perform time-resolved analysis on the multi-echo data based on the dynamic signal data, the initial water map and the non-aqueous phase map, generate the signal intensity curve of the detection device changing with time, and separate the specific first-type enhancement signal and the second-type enhancement signal according to the enhancement characteristics of the first-type weighted image signal and the second-type weighted image signal at different time points in the signal intensity curve. The reconstruction module is used to reconstruct MRI images using a compressed sensing algorithm based on the specific first-type enhancement signal and the second-type enhancement signal, combined with the time dimension information in the dynamic signal data, to obtain a dynamic enhanced image of the target tissue, and to identify and suppress artifact interference in the dynamic enhanced image by utilizing the specific evolution pattern of the detection device in the time dimension.

[0014] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of a real-time dynamic MRI tumor imaging method based on compressed sensing as described in the first aspect above.

[0015] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of a real-time dynamic MRI tumor imaging method based on compressed sensing as described in the first aspect above.

[0016] This application provides a real-time dynamic MRI tumor imaging method based on compressed sensing. By employing compressed sensing technology to acquire dynamic signal data of two types of weighted images generated by structural changes at different time points, it can capture specific dynamic signals related to the target tissue by leveraging the targeting and structural change characteristics of the detection device's microenvironment. Simultaneously, compressed sensing technology can efficiently acquire signals, laying the foundation for accurate target tissue identification. By using a multi-echo water-lipid separation sequence to dynamically scan and separate water and fat signals, generating initial water and non-water phase images, interference between water and fat signals can be eliminated, resulting in pure basic signals and images, providing high-quality data for subsequent analysis. By combining dynamic signal data with the two types of images for time-resolved analysis, signal intensity curves are generated and specific enhanced signals are separated. This accurately captures the signal change patterns of the detection device over time, and signals only related to the target tissue are selected based on enhancement features, effectively eliminating interference from normal tissue and improving the specificity of target tissue identification. Based on the specific enhanced signals and time dimension information, the image is reconstructed using a compressed sensing algorithm, and artifacts are suppressed by utilizing the temporal evolution mode of the detection device, ultimately obtaining a highly accurate dynamic enhanced image of the target tissue. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a real-time dynamic MRI tumor imaging method based on compressed sensing, provided for an embodiment of this application; Figure 2 A flowchart illustrating a real-time dynamic MRI tumor imaging method based on compressed sensing, provided in this application embodiment; Figure 3 A scene diagram illustrating a real-time dynamic MRI tumor imaging method based on compressed sensing, provided in an embodiment of this application. Figure 4 This is a schematic diagram of the structure of a real-time dynamic MRI tumor imaging system based on compressed sensing, provided in an embodiment of this application. Detailed Implementation

[0019] In tumor imaging, the commonly used dynamic contrast-enhanced magnetic resonance imaging (fMRI) technique has significant shortcomings, failing to meet the clinical demand for accurate detection. The gadolinium-based contrast agent used in this technique lacks tumor specificity, resulting in minimal difference in imaging enhancement between tumor tissue and surrounding normal tissue, making it difficult for physicians to clearly distinguish tumor boundaries. For small, early-stage tumor lesions, insufficient imaging clarity and signal sensitivity also hinder accurate detection. Furthermore, its long scan time prevents real-time capture of dynamic changes in the tumor and makes it susceptible to image interference from minor body movements, further affecting the accuracy of tumor imaging.

[0020] To address the aforementioned issues, this application proposes a real-time dynamic MRI tumor imaging method based on compressed sensing. This method first uses compressed sensing technology to capture imaging signals from the detection device at different times; then, it separates water and fat signals in the image using a specialized scanning sequence to generate a clearer baseline image; next, it combines these signals and images to analyze the signal variation patterns of the detection device and filter out specific signals belonging to the target tissue; finally, it reconstructs the image using a compressed sensing algorithm to obtain a dynamically enhanced image displaying only the target tissue, and can also eliminate image interference by utilizing the signal variation patterns of the detection device. This method improves the differentiation between target and normal tissues through targeted detection devices, enhances the detection capability of small lesions through signal separation and precise algorithms, while shortening scan time and reducing image interference.

[0021] This application primarily uses active tumor tissue as an example to illustrate the following embodiments. To enable those skilled in the art to better understand the present application, further detailed descriptions are provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0022] The core of this application is to provide a real-time dynamic MRI tumor imaging method based on compressed sensing, and a flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes: S101. Compressed sensing technology is used to collect dynamic signal data from the first type of weighted image and the second type of weighted image of the detection device at different time points caused by structural transformation. Optionally, step S101 may specifically include the following steps: S1011. At each time point, the first type of excitation pulse sequence is emitted using the radio frequency coil of the magnetic resonance device to excite the first type of signal response of the detection device with the frequency parameters corresponding to the first type of weighted image, and the first type of frequency domain data is collected based on the random frequency sampling template. S1012. At the same time point, switch to the frequency parameter corresponding to the second type of weighted image, transmit the second type of excitation pulse sequence to excite the second type of signal response of the detection device, and collect the second type of frequency domain data based on the random frequency sampling template. S1013. Integrate the first type of frequency domain data and the second type of frequency domain data at each time point in chronological order to form dynamic signal data containing spatial coordinates, time points and signal type identifiers.

[0023] In the above steps, the detection device in practical applications can be a pH-responsive bimodal nanoprobe, which is a nanoscale detection substance that is sensitive to environmental pH values ​​and undergoes conformational changes and generates two different signal responses under acidic conditions such as tumor microenvironments. The target area refers to a biological tissue sample or preclinical model under magnetic resonance scanning conditions. The set time nodes are multiple specific time points selected for signal acquisition, used to capture the detected signal characteristics that change over time. The random frequency sampling template is a template generated for each time node to guide the MRI equipment in acquiring frequency domain data. Its sampling position is randomly generated and meets the preset compression ratio condition. The compression ratio condition is the ratio requirement between the sampled data volume and the complete data volume, which can reduce the sampling volume while ensuring signal quality. The radio frequency coil of the MRI equipment is the component in the MRI equipment used to transmit excitation pulses and receive signals. The first type of excitation pulse sequence and the second type of excitation pulse sequence are two radio frequency pulse sequences with different parameters, used to excite the signal response of the detection device corresponding to the first type of weighted image and the second type of weighted image, respectively. The first type of weighted image and the second type of weighted image are MRI images with two different imaging weights, which can reflect the signal characteristics of different tissues. The first type of frequency domain data and the second type of frequency domain data are the raw signal data in the frequency domain acquired through the corresponding excitation pulse sequences. The dynamic signal data is a comprehensive signal dataset that integrates the two types of frequency domain data at each time node and includes spatial coordinates, time nodes, and signal type identifiers.

[0024] In this embodiment, firstly, in step S1011, the radio frequency coil of the magnetic resonance imaging (MRI) device emits a first type of excitation pulse sequence to excite the first type of signal response of the detection device with the frequency parameters corresponding to the first type of weighted image. First type of frequency domain data is acquired based on a random frequency sampling template. For example, at the first time point, 0.5 hours after the start of detection, medical personnel assist the target area D to lie down in the scanning chamber of the MRI device, adjusting the body position to a comfortable and stable state to avoid data acquisition being affected by changes in body position during the scan. Subsequently, the MRI device is started, and the radio frequency coil of the device is controlled to emit the first type of excitation pulse sequence. The frequency parameters of this sequence are set to a specific resonance frequency matching the first type of weighted image. Simultaneously, the MRI device, based on the random frequency sampling template corresponding to this time point, only acquires signal data at the frequency positions marked in the template. The acquisition process lasts approximately 5 minutes, ultimately obtaining the first type of frequency domain data for this time point. This data is saved through the device's built-in storage module and reflects the first type of signal characteristics generated at different positions of the detection device at this time.

[0025] Next, in step S1012, at the same time point, the frequency parameters corresponding to the second type of weighted image are switched, and a second type of excitation pulse sequence is emitted to excite the second type of signal response of the detection device. Second type of frequency domain data is acquired based on a random frequency sampling template. For example, after completing the first time point (0.5 hours of detection) and acquiring the first type of frequency domain data, without moving the target region D, the control terminal of the MRI equipment automatically switches the frequency parameters to values ​​matching the second type of weighted image, and then controls the radio frequency coil to emit a second type of excitation pulse sequence. This sequence generates a signal response corresponding to the second type of weighted image. Similarly, based on the random frequency sampling template at this time point, the device acquires signal data at the frequency positions marked on the template. The acquisition duration is the same as the first type of frequency domain data acquisition, approximately 5 minutes, ultimately obtaining the second type of frequency domain data for this time point. This data is also saved to the storage module, reflecting the characteristics of the second type of signal generated at different locations at the same time point, achieving continuous acquisition of the two signals at the same time point, and avoiding signal deviation caused by changes in the target region's position.

[0026] Finally, in step S1013, the first and second type frequency domain data at each time point are integrated in chronological order to form dynamic signal data containing spatial coordinates, time points, and signal type identifiers. For example, after acquiring the two types of frequency domain data at four time points, the computer retrieves all data from the storage module of the MRI equipment and arranges them in chronological order, i.e., first arranging the two types of data for 0.5 hours, then arranging the two types of data for 1 hour, 1.5 hours, and 2 hours in sequence. Subsequently, the computer obtains the spatial coordinates corresponding to each data point through the positioning function of the MRI equipment, that is, the specific location information of each region in the body, and labels each set of data with the corresponding time point and signal type identifier. After labeling, the computer integrates these data into a complete dynamic signal data file, which can be opened and viewed with professional software, clearly showing the changes in the two signals generated by the detection devices at different times and locations, providing basic data for subsequent image analysis and reconstruction.

[0027] In the overall scheme of step S101 above, by setting multiple time nodes and generating random frequency sampling templates, it is possible to track signal changes over time and reduce the amount of sampling data based on compressed sensing technology, thereby improving signal acquisition efficiency. By acquiring two types of frequency domain data at the same time node, dual signal information of the detection device can be obtained, enriching the signal dimensions. By integrating the data to form dynamic signal data containing multi-dimensional information, it can provide comprehensive and continuous signal support for subsequent image analysis and reconstruction. Overall, it realizes efficient and accurate acquisition of relevant dynamic signals of the target tissue, laying a data foundation for the subsequent generation of specific images.

[0028] S102. The target area is dynamically scanned using a multi-echo water-fat separation sequence. Multi-echo data is acquired synchronously during the dynamic scanning process. The multi-echo data is separated according to the chemical shift difference between water and fat to obtain water signals and fat signals. An initial water map is generated by MRI image reconstruction based on the separated water signals, and a non-water phase map is generated by MRI image reconstruction based on the separated fat signals. Optionally, step S102 may specifically include the following steps: S1021. The target area is continuously scanned using a multi-echo gradient echo sequence. The original frequency domain data is acquired synchronously at each echo time point, and the original frequency domain data at each echo time point is combined into a multi-channel multi-echo data block. S1022. Based on the preset chemical shift difference between water and fat, a phase difference evolution model containing phase accumulation functions at different echo time points is constructed. The multi-echo data block is input into the phase difference evolution model, and the water signal and fat signal of each voxel are separated by solving the linear equation system. S1023. Perform an inverse Fourier transform on the separated water signal to reconstruct a two-dimensional initial water map, and perform an inverse Fourier transform on the separated fat signal to reconstruct a two-dimensional non-aqueous phase map.

[0029] In the above steps, the multi-echo water-lipid separation sequence is a magnetic resonance imaging (MRI) scan sequence that can acquire multiple echo signals in a single scan and achieve signal separation based on the chemical differences between water and fat. Dynamic scanning refers to continuous, uninterrupted scanning of the target region to capture dynamic changes in the signal within the region. Multi-echo data refers to all signal data acquired at multiple echo time points during the dynamic scanning process. The chemical shift difference between water and fat refers to the difference in resonance frequencies generated in the magnetic resonance field due to the different chemical structures of water and fat molecules; this difference is the core basis for achieving water-lipid signal separation. The water signal is the MRI signal derived solely from water molecules separated from the multi-echo data, and the fat signal is the MRI signal derived solely from fat molecules. MRI image reconstruction is the process of converting the acquired frequency domain signal data into a visualized image. The initial water map is an MRI image reconstructed based on the water signal, mainly showing the distribution of water molecules in the body. The non-water phase map is an MRI image reconstructed based on the fat signal, mainly showing the distribution of adipose tissue in the body. The multi-echo gradient echo sequence is a specific type of multi-echo water-lipid separation sequence. The implementation involves several mechanisms: Gradient magnetic field control allows for the acquisition of multiple echo signals within a short time; the target region refers to the area within the body to be scanned, selected according to detection requirements, such as the location of suspected target tissue; echo time points refer to specific time nodes during the scanning process where echo signals are acquired, with signals at different echo time points carrying different information; raw frequency domain data is unprocessed signal data directly acquired from the MRI equipment in the frequency domain; multi-channel multi-echo data blocks are structured data sets formed by integrating raw frequency domain data from multiple echo time points according to channel and time order; preset chemical shift difference is a pre-set value based on the known resonance frequency difference between water and fat, used to construct a signal separation model; the phase difference evolution model is a mathematical model describing the phase change of water and fat signals at different echo time points, including phase accumulation functions at different echo time points; a voxel is the smallest pixel unit in an MRI image, with each voxel corresponding to a signal in a tiny region within the body; inverse Fourier transform is a signal processing technique that converts frequency domain signals into spatial domain signals, a key technology for MRI image reconstruction.

[0030] In this embodiment, firstly, the target region is continuously scanned using a multi-echo gradient echo sequence in step S1021. Frequency domain raw data is acquired synchronously at each echo time point. The frequency domain raw data at each echo time point is combined into a multi-channel multi-echo data block. For example, when detecting a target region D of suspected target tissue, the abdominal region where the suspected tumor is located is identified as the target region. The magnetic resonance imaging (MRI) device is started, and a multi-echo gradient echo sequence is selected. Scanning parameters are set so that the device generates 8 echo time points in one continuous scan, such as from 1ms to 8ms, with a time point set every 1ms. During the scan, the device's multi-channel radio frequency coil synchronously acquires the frequency domain raw data of the target region at each echo time point, with each channel corresponding to an independent set of raw data. After the scan, the frequency domain raw data of each channel at the 8 echo time points are integrated in the order of "channel-echo time point" to form a multi-channel multi-echo data block containing channel information, time information, and signal data, providing complete raw data support for subsequent water-lipid signal separation.

[0031] Secondly, based on the preset chemical shift difference between water and fat in step S1022, a phase difference evolution model containing phase accumulation functions at different echo time points is constructed. Multi-echo data blocks are input into the phase difference evolution model, and the water and fat signals of each voxel are separated by solving a system of linear equations. For example, a preset chemical shift difference between water and fat is first set based on the known difference in their resonance frequencies. Then, based on this difference, a phase difference evolution model is constructed: the model contains a phase accumulation function corresponding to each echo time point, which describes the change in the phase difference between water and fat signals as the echo time progresses. Then, the multi-channel multi-echo data blocks obtained in step S1021 are input into this model. For each voxel in the data block, the model establishes a system of linear equations regarding the water and fat signal intensities based on the signal intensity and phase information at different echo time points. By solving the linear equations corresponding to each voxel through matrix operations, the water and fat signals of that voxel can be extracted separately from the mixed signal, achieving accurate separation of water and fat signals and avoiding interference from the superposition of the two signals in subsequent imaging.

[0032] Finally, in step S1023, the separated water signals are subjected to inverse Fourier transform to reconstruct a two-dimensional initial water map, and the separated fat signals are subjected to inverse Fourier transform to reconstruct a two-dimensional non-aqueous phase map. For example, water signals from all voxels are extracted from the separation results in step S1022, and these water signals are arranged in order of spatial location of the voxels to form a complete water signal dataset. Inverse Fourier transform is applied to this dataset to convert the water signals, which were originally in the frequency domain, into image signals in the spatial domain. After post-processing such as image denoising and contrast adjustment, a two-dimensional initial water map is generated. The bright areas in this image correspond to areas with dense water molecules, which can clearly show the water distribution in the target area. At the same time, fat signals from all voxels are extracted and arranged in the same order of spatial location to form a fat signal dataset. Similarly, inverse Fourier transform is applied to convert it into a spatial domain image signal, and after post-processing, a two-dimensional non-aqueous phase map is generated. The bright areas in this image correspond to areas with dense adipose tissue, which can intuitively show the fat distribution in the target area and lay the foundation for subsequent analysis of relevant signals of the target tissue.

[0033] In practical applications, firstly, based on the preliminary imaging results of target region E, medical staff determine the liver as the target area for scanning. A B-brand MRI machine is selected and a multi-echo gradient echo sequence is loaded, setting 10 echo time points from 0.8ms to 8.8ms with 0.8ms intervals. Simultaneously, the machine's 16-channel radio frequency coil is activated to improve signal acquisition efficiency. After the scan begins, the machine continuously and dynamically scans the liver region of target region E. At each echo time point, the 16-channel coil synchronously acquires raw frequency domain data. After the scan is completed, the machine automatically integrates the raw data from the 16 channels at the 10 echo time points into a multi-channel multi-echo data block and stores it in the machine's image processing terminal. Next, the terminal automatically constructs a phase difference evolution model based on preset water and fat chemical shift differences. After inputting the multi-channel multi-echo data block into the model, the model uses a built-in linear equation solving algorithm to separate water and fat signals pixel by pixel. During the separation process, the terminal displays the separation progress in real time to ensure the stability of data processing. Finally, the terminal performs inverse Fourier transforms on the separated water and fat signals, and then uses a built-in image reconstruction algorithm for noise reduction and edge enhancement to generate a clear initial water map and non-water phase map. Medical staff can view the two images through the device's display interface and clearly distinguish the distribution of water molecules and fat in the liver region. Throughout the process, the device's operating software automatically records all parameter settings and processing steps, facilitating subsequent data traceability and review.

[0034] In the overall scheme of step S102 above, by using multi-echo gradient echo sequences for dynamic scanning and combining multi-channel multi-echo data blocks, rich echo signal data can be acquired at once, providing sufficient raw information for water-fat signal separation and avoiding separation errors caused by insufficient data. By constructing a phase difference evolution model based on the water-fat chemical shift difference and solving a system of linear equations, the essential differences between water and fat can be accurately utilized to achieve signal separation, effectively eliminating the interference of water and fat signals superimposed on each other, and ensuring high purity of the separated water and fat signals. By reconstructing the initial water map and non-water phase map through inverse Fourier transform, the abstract frequency domain signal can be converted into an intuitive visual image, clearly presenting the distribution of water molecules and fat tissue in the body. This step successfully solves the problem of water-fat signal mixing in traditional MRI imaging, laying a high-quality image foundation for subsequent combination of target tissue-specific signal analysis and accurate identification of target tissue and normal tissue. At the same time, the dynamic scanning method can also adapt to the needs of the entire real-time dynamic MRI tumor imaging method, ensuring the continuity of data acquisition.

[0035] S103. Based on the dynamic signal data, the initial water map and the non-aqueous phase map, perform time-resolved analysis on the multi-echo data to generate a signal intensity curve of the detection device changing with time, and separate the specific first-type enhancement signal and the second-type enhancement signal according to the enhancement characteristics of the first-type weighted image signal and the second-type weighted image signal at different time points in the signal intensity curve. Optionally, step S103 may specifically include the following steps: S1031. Using the initial water map as a spatial mask, remove the signal values ​​of the background region in the dynamic signal data, mark the fat region in the non-aqueous phase map, and filter out the signal values ​​of the fat region in the dynamic signal data. S1032. Based on the signal values ​​obtained after removing the background region and filtering out the fat region, construct a two-dimensional matrix of signal intensity with respect to time nodes and spatial locations, wherein the rows of the two-dimensional matrix correspond to spatial locations and the columns correspond to time nodes. S1033. Perform curve fitting on the data of signal intensity change over time represented by each row of the two-dimensional matrix to generate the first type weighted image signal intensity curve and the second type weighted image signal intensity curve for each spatial location. Specifically, step S1033 may include the following processes: extracting time-series signal data corresponding to each spatial location in the two-dimensional matrix, establishing the correspondence between time nodes and signal strength values, and forming a set of discrete data points; dividing the signal data into a first signal channel and a second signal channel according to the physical characteristics of the detection device, wherein the first signal channel corresponds to fast response characteristics and the second signal channel corresponds to delayed response characteristics; for the first signal channel, using a polynomial function approximation method, determining the fitting parameters through the least squares method to obtain the continuous function expression of the first signal channel; for the second signal channel, using an exponential decay function model for fitting, solving the decay constant and initial amplitude parameters through a nonlinear regression method to obtain the continuous function expression of the second signal channel; assigning different weights to the two continuous function expressions according to the reliability of the signal strength, and smoothing the continuous function expressions processed by the weighting factors to eliminate high-frequency noise, generating a first-type weighted image signal strength curve and a second-type weighted image signal strength curve.

[0036] S1034. Based on the rising slope and peak position of the first type of weighted image signal intensity curve, mark the region that meets the preset threshold condition as a specific first type of enhanced signal. S1035. Based on the descent delay characteristics of the second type of weighted image signal intensity curve, the region that meets the preset time delay condition is marked as a specific second type of enhanced signal.

[0037] In the above steps, time-resolved analysis is a processing method that combines time-dimensional information to analyze signal data in order to capture the signal's variation over time; specific type I / II enhanced signals are type I / II weighted image signals that are only related to the target tissue and have specific time-enhancing characteristics; spatial masking is a tool that uses water molecule distribution information in the initial water map to screen out the target area and exclude irrelevant areas; background area is an area in the initial water map with extremely low water molecule content and no clinical significance; two-dimensional matrix is ​​structured data formed by organizing the processed signal data according to the "spatial location-time node" dimension, with rows corresponding to different spatial locations in the body and columns corresponding to different signal acquisition time nodes; the first signal channel corresponds to the rapid response of the detection device. The signal channel corresponds to the characteristic signal channel, and the second signal channel corresponds to the signal channel with the delayed response characteristic of the detection device; the weighting factor is the importance coefficient assigned to different fitting functions based on the reliability of signal strength, used to optimize the accuracy of the curve; the rising slope is the degree of inclination of the rising phase of the first type of weighted image signal intensity curve, and the peak position is the time point when the curve reaches the highest signal intensity; the preset threshold condition is a pre-set standard for judging whether the first type of signal is a specific enhanced signal, such as the minimum rising slope and the range of peak positions; the falling delay characteristic is the characteristic that the falling phase of the second type of weighted image signal intensity curve lags behind the ordinary tissue signal, and the preset time delay condition is a pre-set delay time standard for judging whether the second type of signal is a specific enhanced signal.

[0038] In this embodiment, firstly, step S1031 uses the initial water map as a spatial mask to remove signal values ​​from background regions in the dynamic signal data. Then, in the non-aqueous phase image, fat regions are marked, and signal values ​​from these fat regions in the dynamic signal data are filtered out. For example, when processing the abdominal target region D, regions with brightness below a set value in the initial water map are identified as background regions, such as air regions at the abdominal edge or non-target organ regions. Using the initial water map as a mask, signal values ​​corresponding to these background regions are deleted from the dynamic signal data. Simultaneously, in the non-aqueous phase image, regions with brightness above a set value are marked as fat regions, such as the abdominal subcutaneous fat layer or intra-abdominal adipose tissue. Signal values ​​corresponding to these fat regions are then deleted from the dynamic signal data, retaining only signal data from non-background and non-fat regions within the target region, thus reducing interference from irrelevant signals in subsequent analysis.

[0039] Secondly, based on the signal values ​​obtained after removing the background region and filtering out the fat region in step S1032, a two-dimensional matrix of signal intensity at time nodes and spatial locations is constructed. The rows of the two-dimensional matrix correspond to spatial locations, and the columns correspond to time nodes. For example, for the target region D, the abdominal target region, the processed signal data is organized according to "spatial location - time node": each row of the matrix represents a spatial location, such as a voxel in the target region, corresponding to a small area in the body, and each column represents a signal acquisition time node, such as the previously set 0.5 hours, 1 hour, 1.5 hours, and 2 hours. Each element in the matrix is ​​the signal intensity value corresponding to "a certain spatial location - a certain time node". This matrix structures the scattered signal data, which is convenient for subsequent analysis of signal changes over time according to spatial location.

[0040] Next, in step S1033, curve fitting is performed on the signal intensity data representing the time variation of each row in the two-dimensional matrix to generate the first-type weighted image signal intensity curve and the second-type weighted image signal intensity curve for each spatial location. Specifically, the process is as follows: First, extract the time-series signal data corresponding to each spatial location in the two-dimensional matrix, i.e., the signal intensity values ​​at that location arranged in chronological order according to time nodes, and establish the correspondence between time nodes and signal intensity values ​​to form a set of discrete data points. Then, based on the physical characteristics of the detection device, the signal data is divided into a first signal channel and a second signal channel. The first signal channel corresponds to the rapid response characteristics of the detection device in the target area, such as a rapid rise in signal intensity within a short period. The second signal channel corresponds to the delayed response characteristics of the detection device, such as a slow decline in signal intensity after reaching a peak. For the first signal channel, a polynomial function approximation method is used, and the coefficients of the polynomial are calculated using the least squares method to minimize the sum of squared errors between the fitted curve and the discrete data points, thus obtaining the first signal channel. A continuous function expression is obtained for the second signal channel. An exponential decay function model is used for fitting, and the decay constant and initial amplitude parameter in the model are solved using a nonlinear regression method. The decay constant reflects the signal descent rate, and the initial amplitude parameter reflects the initial signal strength, thus obtaining a continuous function expression for the second signal channel. Finally, based on the signal strength reliability of the two signal channels (e.g., the first signal channel is more stable in the early time nodes and assigned a higher weight, while the second signal channel is more reliable in the later time nodes and assigned a corresponding weight), different weight factors are assigned to the two continuous function expressions. The weighted function expressions are then smoothed to eliminate curve fluctuations caused by high-frequency noise, ultimately generating the first-type weighted image signal intensity curve and the second-type weighted image signal intensity curve for each spatial location. For example, in target area D, at a certain spatial location within the target area, the first-type curve shows a rapid upward trend from 0.5 to 1 hour after the start of detection, while the second-type curve shows a slow downward trend after 1 hour.

[0041] Next, in step S1034, based on the rising slope and peak position of the first type of weighted image signal intensity curve, the region that meets the preset threshold conditions is marked as a specific first type of enhanced signal. For example, a threshold for the rising slope of the first type curve is preset, such as the minimum value of the rising slope, and a threshold for the peak position is preset, such as the range of the peak position. The first type curves of all spatial positions in the target area D are analyzed: if the rising slope of the curve at a certain spatial position reaches or exceeds the preset threshold, and the peak position is within the preset time range, it indicates that the signal at that position meets the fast response characteristics of the detection device for the target tissue area, and the first type signal at that position is marked as a specific first type of enhanced signal; otherwise, if the rising slope is insufficient or the peak position deviates, it is determined to be a signal related to non-target tissue.

[0042] Finally, in step S1035, based on the descent delay characteristics of the second type weighted image signal intensity curve, regions that meet the preset time delay conditions are marked as specific second type enhanced signals. For example, a descent delay threshold for the second type curve is preset. If the time for the signal to drop to half the peak intensity after reaching the peak is not less than a certain value, the second type curves of all spatial locations within the target area D are analyzed. If the time for the curve at a certain spatial location to drop to half the peak intensity reaches or exceeds the preset threshold, it indicates that the signal at that location conforms to the delay attenuation characteristics of the detection device for the target tissue area. The signal of normal tissue drops faster, and the second type signal at that location is marked as a specific second type enhanced signal. If the descent speed is too fast and the delay threshold is not reached, it is determined to be a signal related to non-target tissue.

[0043] In practical applications, firstly, staff import the previously acquired dynamic signal data of the target area E (liver region), initial aqueous phase map, and non-aqueous phase map into the A-brand medical image processing software. The software automatically uses the initial aqueous phase map as a spatial mask to remove the signal from the intestinal gas background region surrounding the liver. Simultaneously, it marks the fatty capsule region on the liver surface in the non-aqueous phase map, filters out the signal from this region, and retains only the signal data from the liver parenchyma. Next, based on the processed signal data, the software constructs a two-dimensional matrix: the matrix rows correspond to 1000 spatial locations within the liver parenchyma, divided by voxels, and the columns correspond to four time points: 0.4 hours, 0.8 hours, 1.2 hours, and 1.6 hours after the start of detection. Each matrix element is filled with the signal intensity value corresponding to the location and time. Subsequently, the software performs curve fitting on each row of the matrix: extracting time-series data for each spatial location, dividing the signal into a first signal channel and a second signal channel. The first signal channel corresponds to a fast response of 0.4-0.8 hours, and the second signal channel corresponds to a delayed response of 0.8-1.6 hours. The first channel is fitted using a cubic polynomial function with the least squares method, and the second channel is fitted using an exponential decay function with nonlinear regression. Weights are then assigned based on channel reliability: the first channel has a weight of 0.6, and the second channel has a weight of 0.4. The curves are then smoothed, generating two types of signal intensity curves. The software then analyzes these curves according to preset thresholds: first-type curves with an upward slope ≥ 0.5 signal units / hour and a peak value within 0.7-0.9 hours are marked as specific first-type enhanced signals; second-type curves with a time ≥ 0.5 hours to drop to half the peak intensity are marked as specific second-type enhanced signals. Throughout the process, the software displays the marking results in real time, and staff can view the signal distribution in suspected target tissue areas through the interface.

[0044] In the overall scheme of step S103 above, by utilizing the initial aqueous phase map and non-aqueous phase map to remove background and fat signals, interference from irrelevant tissue signals is effectively eliminated, allowing subsequent analysis to focus on the effective signals in the target area. By constructing a two-dimensional matrix, the dispersed signal data is structured according to the "space-time" dimension, facilitating systematic analysis of the signal change patterns at each location. Through curve fitting, discrete signal data is transformed into continuous signal intensity curves, clearly presenting the time response characteristics of the detection device at different locations. Furthermore, through channel fitting and weight optimization, the accuracy of the curves is improved. By marking specific signals with specific enhancement features of the curves, signals only related to the target tissue are accurately selected, avoiding the misidentification of normal tissue signals as target tissue signals. Overall, this step, through layer-by-layer processing and analysis, extracts the specific signals of the target tissue from complex signal data, providing a crucial basis for the subsequent generation of images of the target tissue and significantly improving the accuracy of target tissue signal recognition.

[0045] S104. Based on the specific first type of enhancement signal and the second type of enhancement signal, and combined with the time dimension information in the dynamic signal data, the MRI image is reconstructed using a compressed sensing algorithm to obtain a dynamic enhanced image of the target tissue. The specific evolution mode of the detection device in the time dimension is used to identify and suppress artifact interference in the dynamic enhanced image.

[0046] Optionally, such as Figure 2 As shown, step S104 may specifically include the following steps: S1041. The specific first type of enhancement signal and the specific second type of enhancement signal are combined into a joint enhancement signal, the spatial location of which is consistent with the dynamic signal data; S1042. Based on the time dimension information of the dynamic signal data, a time evolution constraint function is constructed, which describes the prior mode of the signal strength of the detection device changing with time. S1043. Using the joint enhancement signal as the initial input and the time evolution constraint function as the sparsity constraint, construct the objective optimization function, and update the signal estimation value of each spatial location at continuous time nodes by iteratively solving the objective optimization function. Specifically, step S1043 may include the following process: assigning the joint enhanced signal to the signal estimate as an initial value, constructing the target optimization function using the time evolution constraint function as the basis for sparsity evaluation; the target optimization function includes a data matching part between the signal estimate and the dynamic signal data, and a pattern conformance part of the signal estimate under the time evolution constraint function; updating the signal estimate through a cyclic calculation process, calculating the target optimization function value corresponding to the current signal estimate in each loop, and adjusting the signal estimate based on the target optimization function value; repeating the cyclic calculation process until the change in the target optimization function value is lower than a preset stopping condition, and outputting the updated signal estimate.

[0047] S1044. The updated signal estimate is compared with the dynamic signal data by residual. If the residual does not converge, the iteration is repeated. When the residual converges, the signal estimate is subjected to Fourier transform to reconstruct a dynamic enhanced image of the target tissue. S1045. During the reconstruction process, abnormal fluctuation points in the signal intensity curve that do not conform to the time evolution constraint function are detected, the spatial locations corresponding to the abnormal fluctuation points are marked as artifact interference regions, and the signal intensity of the artifact interference regions is set to zero in the dynamic enhancement image.

[0048] In the above steps, the joint enhancement signal is a comprehensive signal formed by merging the specific type I enhancement signal and the specific type II enhancement signal according to their spatial correspondence. Its spatial position perfectly matches the dynamic signal data, and it can centrally reflect the signal characteristics of the target tissue. The time evolution constraint function is a pre-constructed prior pattern describing the change of the signal intensity of the detection device over time, based on the physical characteristics of the detection device and the response law of the target tissue microenvironment, used to limit the reasonable range of signal variation. The target optimization function is a mathematical function used to iteratively optimize the signal estimate, which includes a data matching part and a pattern matching part. The data matching part measures the difference between the signal estimate and the dynamic signal data, and the pattern matching part measures the difference between the signal estimate and the time evolution constraint. The following parameters are considered: the fit of the bundle function; the signal estimate, which is a predicted value continuously updated during the iteration process to approximate the actual target tissue signal; the preset stopping condition, which is a pre-set criterion for determining whether the iteration terminates, usually when the change in the target optimization function value is lower than a certain minimum value; the residual comparison, which is the process of calculating the difference between the updated signal estimate and the dynamic signal data, and residual convergence, which means that the difference is less than a preset threshold, indicating that the signal estimate is close enough to the true value; abnormal fluctuation points, which are signal points in the signal intensity curve that deviate from the time evolution constraint function and have no reasonable physiological significance; and the artifact interference region, which is the spatial location corresponding to the abnormal fluctuation point. The signals in these regions do not originate from the target tissue, but are interference caused by factors such as motion and equipment noise.

[0049] In this embodiment of the application, firstly, the specific first type of enhancement signal and the specific second type of enhancement signal are merged into a joint enhancement signal through step S1041. The spatial location of the joint enhancement signal is consistent with the dynamic signal data. For example, when processing the abdominal target region of target region D, the specific first type of enhancement signal and the specific second type of enhancement signal marked by all spatial locations are extracted firstly, and then merged according to the "spatial location-time node" dimension: the two types of signals at the same spatial location and the same time node are superimposed according to a preset ratio, such as the first type of signal accounting for 0.5 and the second type of signal accounting for 0.5, to form the joint enhancement signal value at that location and time point; after the joint enhancement signal values ​​of all locations and time nodes are integrated, it is ensured that their spatial location distribution is completely consistent with the dynamic signal data, avoiding subsequent analysis errors due to position misalignment.

[0050] Secondly, based on the time dimension information of the dynamic signal data in step S1042, a time evolution constraint function is constructed. The time evolution constraint function describes the prior pattern of the signal strength of the detection device changing with time. For example, combining the historical response pattern of the detection device in the target tissue area in the dynamic signal data of the target area D, such as the rapid rise of the signal in the first 0.5-1 hours and the slow decline in the first 1-2 hours, a time evolution constraint function is constructed: the function takes time as the independent variable and signal strength as the dependent variable, and limits the rise rate of the signal in the first 0.5-1 hours to no less than a certain value and the decline rate in the first 1-2 hours to no more than a certain value, while excluding changes such as sudden changes and sudden drops that do not conform to the characteristics of the detection device. This function can be implemented in the form of polynomial fitting or piecewise function, providing a reasonable pattern reference for subsequent signal optimization.

[0051] Next, in step S1043, using the joint augmentation signal as the initial input and the time evolution constraint function as the sparsity constraint, a target optimization function is constructed. The target optimization function is solved iteratively to update the signal estimate at each spatial location at continuous time nodes. Specifically, all values ​​of the joint augmentation signal are first assigned to the signal estimate as initial values. Then, using the time evolution constraint function as the basis for sparsity evaluation, a target optimization function is constructed. The data matching part of this function is the mean square error between the signal estimate and the dynamic signal data, and the pattern matching part is the sum of squared deviations between the signal estimate and the time evolution constraint function. Both are weighted according to preset weights, such as 0.6 for the data matching part and 0.4 for the pattern matching part, and then added together to obtain the target optimization function. The target optimization function is calculated, and then a loop calculation process is initiated: In each loop, the target optimization function value corresponding to the current signal estimate is calculated first, and then the signal estimate is adjusted using the gradient descent algorithm. For example, for positions with large function values, the signal estimate is fine-tuned to reduce the function value. The loop calculation process is repeated until the change in the target optimization function value between two consecutive loops is lower than the preset stopping condition, such as a change of less than 0.001. The updated signal estimate is then output. For example, in target region D, at a certain spatial location, the initial signal estimate has a signal strength of 100 at 1.2 hours. After 3 iterations, it is adjusted to 92, and the target optimization function value decreases from 5.8 to 1.2, satisfying the stopping condition. Therefore, the signal estimate at this location and time point is determined to be 92.

[0052] Next, in step S1044, the updated signal estimate is compared with the dynamic signal data using residuals. If the residuals do not converge, the iteration is repeated. When the residuals converge, the signal estimate is subjected to a Fourier transform to reconstruct a dynamic enhanced image of the target tissue. For example, the residuals of the signal estimates at all spatial locations in the target region D and the dynamic signal data are calculated. The residual is the absolute value of the difference between the two. If the average residual at a certain location is 0.8, which is higher than the preset convergence threshold of 0.5, the process returns to step S1043 to iterate again. If the average residual at all locations drops to 0.3, satisfying the convergence condition, the signal estimates at all spatial locations are organized according to time nodes. A Fourier transform is performed on the signal estimate at each time node to convert the frequency domain signal into a spatial domain image signal. After contrast adjustment, edge enhancement, and other processing, a dynamic enhanced image is generated. This image only displays areas where the signal matches the characteristics of the target tissue. Normal tissue areas are low in brightness due to the lack of specific signals, clearly distinguishing the target tissue.

[0053] Finally, in step S1044, during the reconstruction process, abnormal fluctuation points in the signal intensity curve that do not conform to the time evolution constraint function are detected. The spatial locations corresponding to the abnormal fluctuation points are marked as artifact interference regions, and the signal intensity of the artifact interference regions is set to zero in the dynamic enhancement image. For example, in the dynamic enhancement image reconstruction process of the target region D, the signal intensity curve of each spatial location is monitored in real time: if the signal intensity of a certain spatial location suddenly rises from 80 to 150 in 1 hour, deviating from the "reaching peak and then slowly decreasing" pattern defined by the time evolution constraint function, the point is determined to be an abnormal fluctuation point; the spatial location corresponding to the point is marked as an artifact interference region; in the dynamic enhancement image, the signal intensity of the region is forcibly set to zero to eliminate the interference of artifacts on the image and ensure that only the signal region of the real target tissue is retained in the image.

[0054] In practical applications, firstly, staff use medical image processing software (brand A) to merge the specific first and second type enhancement signals of the liver region (target area E) into a combined enhancement signal based on spatial location, ensuring that the two types of signals are superimposed correspondingly for each voxel. Next, based on the dynamic signal data of target area E, the software constructs a time evolution constraint function at the initial detection times of 0.4 hours, 0.8 hours, 1.2 hours, and 1.6 hours, limiting the signal to a pattern of increasing from 0.4 to 0.8 hours and slowly decreasing from 0.8 to 1.6 hours. Then, using the combined enhancement signal as the initial value, the software constructs the target... The optimization function updates the signal estimate through iterative solutions. The function value is calculated after each iteration until the change in function value is less than 0.002, for a total of 5 iterations. Afterward, the software calculates the residual between the signal estimate and the dynamic signal data. The average residual is 0.28, satisfying the convergence threshold of 0.3. Fourier transform is then performed on the signal estimate to reconstruct the dynamic enhanced image. During reconstruction, the software detects a sudden drop in signal at the liver edge region after 1.0 hour, which does not conform to the constraint function. This region is marked as an artifact area and its signal is set to zero. The final generated dynamic enhanced image clearly shows two target tissue regions within the liver without obvious artifacts.

[0055] In the overall scheme of step S104 above, by merging two types of specific enhancement signals, the signal characteristics of the target tissue are concentrated, avoiding the limitations of a single signal type and providing a comprehensive signal foundation for accurate imaging; by constructing a time evolution constraint function, the signal variation range is limited by the actual response law of the detection device, eliminating unreasonable signal interference and ensuring the physiological rationality of the signal; through iterative optimization, the signal estimate is continuously adjusted to both fit the original dynamic signal data and conform to the evolution mode of the detection device, improving signal accuracy; through residual comparison and Fourier transform, the reliability of the reconstructed image is ensured, presenting only the target tissue; through artifact detection and elimination, the image quality is further purified, avoiding misdiagnosis caused by artifacts. Overall, this step, through a complete process of signal merging, constraint, optimization, reconstruction, and artifact removal, ultimately generates a high-precision, high-definition dynamic enhancement image of the target tissue.

[0056] The following is a complete embodiment for steps S101 to S104: like Figure 3As shown, firstly, using compressed sensing technology, the detection device precisely acquires dynamic signal data on T1-weighted and T2-weighted images at four time points: 0.2 hours, 0.4 hours, 0.6 hours, and 0.8 hours after the start of detection, focusing on capturing signal changes in small lesion areas. Next, the multi-echo water-fat separation sequence of the MRI machine is activated to dynamically scan the pancreas and surrounding areas of target region C, simultaneously acquiring multi-echo data. Based on the difference in chemical shift between water and fat, the water signal of the pancreatic parenchyma is precisely separated from the fat signal of the surrounding adipose tissue. Clear initial water and non-water phase images are then reconstructed using MRI images. Subsequently, time-resolved analysis was conducted by combining dynamic signal data, initial water map, and non-water phase map: the initial water map was used to remove background signals outside the pancreas, such as gastrointestinal gas areas, and the non-water phase map was used to filter out surrounding fat signals. Signal intensity curves for each spatial location were generated based on the processed data. It was found that the T1-weighted signal of small lesions within the pancreas increased rapidly from 0.2 to 0.4 hours, while the T2-weighted signal showed a delayed decrease from 0.4 to 0.8 hours, whereas normal pancreatic tissue did not exhibit this pattern. Based on this, specific T1-enhancing signals and specific T2-enhancing signals were separated. Finally, the temporal dimension information of the two types of specific enhancement signals was combined with the dynamic signal data, and the MRI image was reconstructed using a compressed sensing algorithm. This clearly presented an active pancreatic cancer lesion with a diameter of approximately 3 mm. At the same time, the temporal evolution mode of the detection device was used to identify and suppress artifact interference caused by abdominal peristalsis in the image, accurately distinguishing the lesion from the surrounding normal pancreatic tissue.

[0057] The real-time dynamic MRI tumor imaging method based on compressed sensing provided in this application addresses the problem of insufficient signal specificity in traditional MRI by using a detection device to target the microenvironment of the target tissue and combining compressed sensing technology to efficiently acquire specific dynamic signals. It eliminates water-lipid signal interference using a multi-echo water-lipid separation sequence to generate high-quality baseline images, laying the foundation for subsequent signal analysis. Time-resolved analysis accurately separates the specific enhanced signals of the target tissue, effectively distinguishing it from normal tissue. Finally, compressed sensing algorithms are used to reconstruct the dynamic enhanced image representing the target tissue and suppress artifacts, significantly improving the accuracy and clarity of target tissue imaging.

[0058] Figure 4 This is a schematic diagram illustrating a specific implementation of a real-time dynamic MRI tumor imaging system based on compressed sensing, as provided in this application. (Refer to...) Figure 4 The system may include: The acquisition module 41 uses compressed sensing technology to acquire dynamic signal data from the first type of weighted image and the second type of weighted image at different time points caused by structural changes in the detection device. The separation module 42 is used to dynamically scan the target area using a multi-echo water-fat separation sequence. During the dynamic scanning process, multi-echo data is acquired synchronously. The multi-echo data is separated according to the chemical shift difference between water and fat to obtain water signals and fat signals. An initial water map is generated by MRI image reconstruction based on the separated water signals, and a non-water phase map is generated by MRI image reconstruction based on the separated fat signals. The generation module 43 is used to perform time-resolved analysis on the multi-echo data based on the dynamic signal data, the initial water map and the non-aqueous phase map, generate the signal intensity curve of the detection device changing with time, and separate the specific first-type enhancement signal and the second-type enhancement signal according to the enhancement characteristics of the first-type weighted image signal and the second-type weighted image signal at different time points in the signal intensity curve. The reconstruction module 44 is used to reconstruct MRI images using a compressed sensing algorithm based on the specific first type of enhancement signal and the second type of enhancement signal, combined with the time dimension information in the dynamic signal data, to obtain a dynamic enhanced image of the target tissue, and to identify and suppress artifact interference in the dynamic enhanced image by utilizing the specific evolution mode of the detection device in the time dimension.

[0059] The real-time dynamic MRI tumor imaging system based on compressed sensing in this application embodiment is used to implement the aforementioned real-time dynamic MRI tumor imaging method based on compressed sensing. Therefore, the specific implementation of the real-time dynamic MRI tumor imaging system based on compressed sensing can be found in the embodiment section of the real-time dynamic MRI tumor imaging method based on compressed sensing mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0060] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described methods for real-time dynamic MRI tumor imaging based on compressed sensing.

[0061] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described real-time dynamic MRI tumor imaging methods based on compressed sensing.

[0062] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0063] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the embodiments of the real-time dynamic MRI tumor imaging method based on compressed sensing.

[0064] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0065] The above provides a detailed description of a real-time dynamic MRI tumor imaging method and system based on compressed sensing provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A real-time dynamic MRI tumor imaging method based on compressed sensing, characterized in that, include: Compressed sensing technology was used to collect dynamic signal data from the first and second type weighted images of the detection device at different time points caused by structural changes. A multi-echo water-fat separation sequence was used to dynamically scan the target area. Multi-echo data was acquired synchronously during the dynamic scanning process. The multi-echo data was separated according to the difference in chemical shift between water and fat to obtain water signals and fat signals. An initial water map was generated by MRI image reconstruction based on the separated water signals, and a non-water phase map was generated by MRI image reconstruction based on the separated fat signals. Based on the dynamic signal data, the initial water map, and the non-aqueous phase map, the multi-echo data is subjected to time-resolved analysis to generate a signal intensity curve of the detection device as a function of time. Based on the enhancement characteristics of the first type of weighted image signal and the second type of weighted image signal at different time points in the signal intensity curve, the specific first type of enhancement signal and the second type of enhancement signal are separated. Based on the specific first and second type enhancement signals, and combined with the time dimension information in the dynamic signal data, a compressed sensing algorithm is used to reconstruct MRI images to obtain dynamic enhanced images of the target tissue. The specific evolution pattern of the detection device in the time dimension is used to identify and suppress artifact interference in the dynamic enhanced images.

2. The method according to claim 1, characterized in that, Based on the specific first and second type enhancement signals, and combined with the temporal information in the dynamic signal data, a compressed sensing algorithm is used to reconstruct the MRI image, obtaining a dynamic enhanced image of the target tissue. Furthermore, utilizing the specific evolutionary pattern of the detection device in the temporal dimension, artifact interference in the dynamic enhanced image is identified and suppressed, including: The specific first type of enhancement signal and the specific second type of enhancement signal are combined into a joint enhancement signal, and the spatial location of the joint enhancement signal is consistent with the dynamic signal data. Based on the time dimension information of the dynamic signal data, a time evolution constraint function is constructed, which describes the prior pattern of the signal strength of the detection device changing with time. Using the joint enhanced signal as the initial input and the time evolution constraint function as the sparsity constraint, an objective optimization function is constructed. By iteratively solving the objective optimization function, the signal estimation value of each spatial location at continuous time nodes is updated. The updated signal estimate is compared with the dynamic signal data by residual. If the residual does not converge, the iteration is repeated. When the residual converges, the signal estimate is subjected to Fourier transform to reconstruct a dynamic enhanced image of the target tissue. During the reconstruction process, abnormal fluctuation points in the signal intensity curve that do not conform to the time evolution constraint function are detected. The spatial locations corresponding to the abnormal fluctuation points are marked as artifact interference regions, and the signal intensity of the artifact interference regions is set to zero in the dynamically enhanced image.

3. The method according to claim 2, characterized in that, Using the joint enhanced signal as the initial input and the time evolution constraint function as the sparsity constraint, a target optimization function is constructed. The signal estimate for each spatial location at consecutive time nodes is updated by iteratively solving the target optimization function, including: The joint enhanced signal is assigned to the signal estimate as the initial value, and the objective optimization function is constructed using the time evolution constraint function as the basis for sparsity evaluation. The objective optimization function includes a data matching part between the signal estimate and the dynamic signal data, and a pattern conformance part of the signal estimate under the time evolution constraint function; The signal estimate is updated through a cyclic calculation process. In each iteration, the target optimization function value corresponding to the current signal estimate is calculated, and the signal estimate is adjusted based on the target optimization function value. The calculation process is repeated until the change in the target optimization function value is lower than the preset stopping condition, and then the updated signal estimate is output.

4. The method according to claim 1, characterized in that, Based on the dynamic signal data, the initial water map, and the non-aqueous phase map, the multi-echo data is subjected to time-resolved analysis to generate a signal intensity curve of the detection device over time. Based on the enhancement characteristics of the first-type weighted image signal and the second-type weighted image signal at different time points in the signal intensity curve, specific first-type enhancement signals and second-type enhancement signals are separated, including: Using the initial water map as a spatial mask, the signal values ​​of the background region in the dynamic signal data are removed. The fat region is marked in the non-aqueous phase map, and the signal values ​​of the fat region in the dynamic signal data are filtered out. Based on the signal values ​​obtained after removing background and fat regions, a two-dimensional matrix of signal intensity at time nodes and spatial locations is constructed, wherein the rows of the two-dimensional matrix correspond to spatial locations and the columns correspond to time nodes. Curve fitting is performed on the signal intensity data representing the time change of each row in the two-dimensional matrix to generate the first type weighted image signal intensity curve and the second type weighted image signal intensity curve for each spatial location. Based on the rising slope and peak position of the first type of weighted image signal intensity curve, the region that meets the preset threshold condition is marked as a specific first type of enhanced signal; Based on the descent delay characteristics of the second type of weighted image signal intensity curve, regions that meet the preset time delay conditions are marked as specific second type enhanced signals.

5. The method according to claim 4, characterized in that, Curve fitting is performed on the signal intensity data representing the time variation of each row in the two-dimensional matrix to generate the first-type weighted image signal intensity curve and the second-type weighted image signal intensity curve for each spatial location, including: Extract the time-series signal data corresponding to each spatial location in the two-dimensional matrix, establish the correspondence between time nodes and signal strength values, and form a set of discrete data points; Based on the physical characteristics of the detection device, the signal data is divided into a first signal channel and a second signal channel, wherein the first signal channel corresponds to the fast response characteristic and the second signal channel corresponds to the delayed response characteristic. For the first signal channel, a polynomial function approximation method is used, and the fitting parameters are determined by the least squares method to obtain the continuous function expression of the first signal channel; For the second signal channel, an exponential decay function model is used for fitting, and the decay constant and initial amplitude parameters are solved by nonlinear regression to obtain the continuous function expression of the second signal channel; Based on the reliability of the signal strength, different weights are assigned to the two continuous function expressions, and the continuous function expressions processed by the weighting factors are smoothed to eliminate high-frequency noise, thereby generating the first type of weighted image signal strength curve and the second type of weighted image signal strength curve.

6. The method according to claim 1, characterized in that, The dynamic signal data of the first-type weighted image and the second-type weighted image caused by structural transformation at different time points are acquired using compressed sensing technology, including: At each time point, the first type of excitation pulse sequence is emitted using the radio frequency coil of the magnetic resonance device to excite the first type of signal response of the detection device with the frequency parameters corresponding to the first type of weighted image, and the first type of frequency domain data is acquired based on the random frequency sampling template. At the same time point, switch to the frequency parameters corresponding to the second type of weighted image, emit the second type of excitation pulse sequence to excite the second type of signal response of the detection device, and collect the second type of frequency domain data based on the random frequency sampling template; The first type of frequency domain data and the second type of frequency domain data at each time point are integrated in chronological order to form dynamic signal data containing spatial coordinates, time points, and signal type identifiers.

7. The method according to claim 1, characterized in that, A multi-echo water-fat separation sequence is used to dynamically scan the target region. Multi-echo data is acquired synchronously during the dynamic scan. Based on the chemical shift differences between water and fat, the multi-echo data is separated to obtain water and fat signals. An initial water map is generated based on the separated water signal through MRI image reconstruction, and a non-water phase map is generated based on the separated fat signal through MRI image reconstruction. This includes: The target area is continuously scanned using a multi-echo gradient echo sequence. The original frequency domain data is acquired synchronously at each echo time point, and the original frequency domain data at each echo time point is combined into a multi-channel multi-echo data block. Based on the preset chemical shift difference between water and fat, a phase difference evolution model is constructed that includes phase accumulation functions at different echo time points. The multi-echo data block is input into the phase difference evolution model, and the water signal and fat signal of each voxel are separated by solving a system of linear equations. The separated water signal is subjected to inverse Fourier transform to reconstruct a two-dimensional initial water map, and the separated fat signal is subjected to inverse Fourier transform to reconstruct a two-dimensional non-aqueous phase map.

8. A real-time dynamic MRI tumor imaging system based on compressed sensing, characterized in that, include: The acquisition module is used to acquire dynamic signal data from the first and second type weighted images of the detection device at different time points caused by structural changes, using compressed sensing technology. The separation module is used to dynamically scan the target area using a multi-echo water-fat separation sequence. During the dynamic scanning process, multi-echo data is acquired synchronously. The multi-echo data is separated according to the chemical shift difference between water and fat to obtain water signals and fat signals. An initial water map is generated by MRI image reconstruction based on the separated water signals, and a non-water phase map is generated by MRI image reconstruction based on the separated fat signals. The generation module is used to perform time-resolved analysis on the multi-echo data based on the dynamic signal data, the initial water map and the non-aqueous phase map, generate the signal intensity curve of the detection device changing with time, and separate the specific first-type enhancement signal and the second-type enhancement signal according to the enhancement characteristics of the first-type weighted image signal and the second-type weighted image signal at different time points in the signal intensity curve. The reconstruction module is used to reconstruct MRI images using a compressed sensing algorithm based on the specific first-type enhancement signal and the second-type enhancement signal, combined with the time dimension information in the dynamic signal data, to obtain a dynamic enhanced image of the target tissue, and to identify and suppress artifact interference in the dynamic enhanced image by utilizing the specific evolution pattern of the detection device in the time dimension.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the real-time dynamic MRI tumor imaging method based on compressed sensing as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the real-time dynamic MRI tumor imaging method based on compressed sensing as described in any one of claims 1 to 7.

Images