Original enhanced magnetic resonance nerve imaging system capable of accurately displaying complex forms of nerves

Abstract

The invention relates to the technical field of medical imaging, and discloses an original enhanced magnetic resonance nerve imaging system for accurately displaying a complex nerve form, which comprises the following modules: a hardware module for providing a physical scanning platform with high resolution and signal-to-noise ratio and scanning data; the sampling strategy module is used for analyzing the scanned data, designing a self-adaptive spatial sampling mode and outputting a sampling mask; the sequence design module is used for generating a high-level pulse sequence based on the output sampling mask; the data acquisition module is used for executing the advanced pulse sequence and acquiring under-sampling spatial data; the image reconstruction module is used for reconstructing a high-resolution image from the undersampled spatial data; and the post-processing and visualization module is used for analyzing the reconstructed image and generating a 3D rendered image in a neural complex form. The technical effect of greatly improving the signal-to-noise ratio is achieved by matching the 11.7 T high-field magnetic resonance imaging equipment with the 256-channel receiving coil array. Compared with a common 3T or 7T system in the prior art, the signal intensity is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging technology, and in particular to an original enhanced magnetic resonance neuroimaging system for accurately displaying complex neural morphologies. Background Art

[0002] Magnetic resonance imaging (MRI) technology is an important tool in neuroscience research and clinical diagnosis, and is widely used to non-invasively assess brain structure and function. In existing technologies, clinical MRI systems typically use a magnetic field strength of 1.5T or 3T, with limited image resolution, especially when displaying complex neural structures (such as cortical folds and tiny nerve fibers). Studies have shown that high-field MRI (such as 7T) can significantly improve the signal-to-noise ratio (SNR) and resolution, but it faces challenges such as uneven radiofrequency field, increased magnetic susceptibility artifacts, and tissue heating caused by excessive radiofrequency energy deposition. These problems directly limit the application of high-resolution imaging, especially in neural connectivity research and the diagnosis of neurological diseases (such as Alzheimer's disease).

[0003] In existing technologies, the number of receiving coil channels is generally 16 to 32, which limits the acceleration capability of parallel imaging and results in long scanning times, making it difficult to meet the needs of efficient imaging. Sampling strategies often adopt uniform or fixed undersampling modes, which fail to be optimized for complex neural areas and easily lead to image quality degradation or extended scanning times. Pulse sequence design is usually based on a fixed flip angle and fails to fully utilize quantum coherence to enhance signals or improve contrast, especially in high-field strength environments, resulting in insufficient signal accumulation.

[0004] Image reconstruction relies on compressed sensing technology, but reconstruction of undersampled data is prone to introduce artifacts. Although deep learning methods have made progress, they still require a large amount of high-quality data support, and their clinical applications are limited. Neural structure segmentation mostly relies on manual or semi-automatic methods, which are inefficient and inaccurate, especially when dealing with complex neural morphologies.

[0005] To this end, the present invention provides an original enhanced magnetic resonance neuroimaging system that accurately displays complex neural morphologies to address the deficiencies in the prior art. Summary of the Invention

[0006] In response to the shortcomings of the existing technology, the present invention provides an original enhanced magnetic resonance neuroimaging system that accurately displays complex neural morphologies, solving the problems of low resolution and insufficient efficiency in traditional magnetic resonance imaging systems.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an original enhanced magnetic resonance neuroimaging system for accurately displaying complex neural morphologies, comprising the following modules:

[0008] Hardware module: provides a physical scanning platform with high resolution and signal-to-noise ratio to scan data;

[0009] Sampling strategy module: analyzes and scans data, designs adaptive spatial sampling patterns, and outputs sampling masks;

[0010] Sequence design module: Generates advanced pulse sequences based on the output sampling mask;

[0011] Data acquisition module: executes advanced pulse sequences and collects undersampled spatial data;

[0012] Image reconstruction module: reconstructs high-resolution images from undersampled spatial data;

[0013] Post-processing and visualization module: Analyze the reconstructed image and generate 3D rendering images of the complex morphology of the nerve.

[0014] Preferably, the physical scanning platform includes: providing an 11.7T magnetic resonance imaging device with a magnetic field strength B0 = 11.7T and equipped with a 256-channel receiving coil array.

[0015] Preferably, the designing of the adaptive spatial sampling mode comprises the following steps:

[0016] S1, perform a low-resolution reconnaissance scan to acquire 1 mm resolution k-space data;

[0017] S2. Input the artificial intelligence model to analyze complex neural regions and generate variable density random k-space sampling trajectories.

[0018] Preferably, the advanced pulse sequence is mainly a radio frequency pulse sequence, and the flip angle changes with time as α(t)=60·e -0.01·t degrees, RF field strength B1=10μT, Rabi oscillation frequency ω=42.58·10 6 ·B1rad / s.

[0019] Preferably, the artificial intelligence model is a convolutional neural network that generates sampling masks based on low-resolution reconnaissance scan data to optimize the sampling density of complex neural areas.

[0020] Preferably, the duration of the radio frequency pulse sequence is 5 milliseconds, data is collected in parallel through 256-channel receiving coils, and the acceleration factor is 8.

[0021] Preferably, the convolutional neural network is trained on a data set containing 1,000 brain MRI images, using mean square error as the loss function.

[0022] Preferably, the image reconstruction module further includes using a U-Net model to accelerate the reconstruction process, wherein the U-Net is implemented based on PyTorch and trained on a simulated MRI dataset.

[0023] Preferably, the U-Net model generates T1 and T2 quantitative maps with a resolution of 0.2 mm from the undersampled spatial data as a supplement to the 3D rendered image.

[0024] Preferably, the 3D rendered image is segmented into cortexes through a graph neural network, and the neural network topology is analyzed based on the graph theory connectivity theorem.

[0025] The present invention provides an original enhanced magnetic resonance neuroimaging system that accurately displays complex neural morphologies. It has the following beneficial effects:

[0026] 1. The present invention uses high field strength to enhance signal strength, improving the signal-to-noise ratio by approximately 15.21 times compared to a common 3T system. The optimized RF pulse sequence enhances the coherent accumulation of the spin system, further improving the signal-to-noise ratio and resolving the issue of insufficient signal-to-noise ratio in high-resolution imaging with existing technologies. Traditional clinical MRI systems typically have a magnetic field strength of 1.5T or 3T, resulting in limited image resolution. While a high field strength of 7T can improve the signal-to-noise ratio and resolution, it faces numerous challenges that limit its application in high-resolution imaging.

[0027] 2. The present invention uses an AI-driven adaptive k-space sampling strategy and Monte Carlo method. The sampling strategy module is based on the low-resolution reconnaissance scan data provided by the hardware module, uses a convolutional neural network to analyze complex neural areas, outputs a sampling density mask, and uses the Monte Carlo method to generate a variable-density random k-space sampling trajectory to ensure that the center k-space sampling is dense and the edge k-space sampling is sparse, with a sampling rate of 50%. This overcomes the problem that the existing sampling strategies mostly use uniform or fixed undersampling modes, are not optimized for complex neural areas, and result in reduced image quality or extended scanning time.

[0028] 3. The present invention combines deep learning U-Net and graph neural network technology. The image reconstruction module uses the U-Net model to accelerate the reconstruction process. U-Net is implemented based on PyTorch and trained on a simulated MRI dataset. The post-processing and visualization module uses a graph neural network to segment the cortex and nerve fibers, and analyzes the neural network topology based on the graph theory connectivity theorem. This solves the problems in the existing technology that image reconstruction relies on compressed sensing technology and is prone to introducing artifacts, deep learning methods require a large amount of high-quality data support, and are limited in clinical application, and neural structure segmentation mostly relies on manual or semi-automatic methods, which are inefficient and have poor accuracy.

[0029] 4. The present invention combines the system with an integrated AI-driven adaptive k-space sampling strategy. The results of the post-processing and visualization modules can be fed back to the sampling strategy module to optimize subsequent scanning strategies. The hardware module can also be expanded to support a real-time motion correction system, which monitors the patient's head movement through optical tracking or additional sensors and adjusts the scanning parameters. This overcomes the problems of existing technologies such as the inability to adjust the scanning strategy in real time according to imaging results and the difficulty in effectively responding to patient head movement, significantly improving the ability of neural connection research and neurological disease analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0032] Please see the attached Figure 1 The present invention provides an original enhanced magnetic resonance neuroimaging system for accurately displaying complex neural morphologies, including the following modules:

[0033] Hardware module: provides a physical scanning platform with high resolution and signal-to-noise ratio to scan data;

[0034] In this embodiment, the hardware module uses an 11.7T magnetic resonance imaging (MRI) scanner as a core device, and its magnetic field strength is fixed at 11.7 Tesla (T) to provide a superior signal-to-noise ratio (SNR).

[0035] The high-field strength design effectively enhances the signal strength. Compared with the common 3T system, its signal-to-noise ratio is improved by about 15.21 times, laying the foundation for subsequent high-resolution imaging work.

[0036] Typically, the hardware module is equipped with a 256-channel receive coil array to capture multi-channel MRI signals. Leveraging parallel imaging technology, this configuration can achieve an acceleration factor of up to 8x, effectively reducing scan time while maintaining image quality.

[0037] The receive coil array is specifically optimized for neuroimaging and uses a head coil design to achieve full coverage of brain signals and ensure high signal sensitivity.

[0038] The hardware module is equipped with a high-performance gradient coil with a maximum gradient strength of 200mT / m and a switching rate of 200T / m / s. This gradient system can quickly switch gradient waveforms and can strongly support the complex k-space trajectories generated by the sequence design module, thereby ensuring the accuracy of spatial encoding.

[0039] The hardware module integrates an advanced control system that precisely manages the timing and synchronization of RF pulses and gradient waveforms.

[0040] The control system is capable of executing pulse sequences with nonlinear flip angle variations, such as the quantum coherence enhancement sequence designed by the sequence design module, where the flip angle varies with time according to the formula:

[0041] α(t)=α0e -λt

[0042] Where α(t) is the flip angle that changes with time, α0 is the initial flip angle, and λ = 0.01S -1 is the attenuation coefficient, and t is the time variable of the pulse sequence execution.

[0043] The field strength B1 of the RF pulse is set to 10 microtesla, and the oscillation frequency is calculated using the Rabi oscillation theorem:

[0044] ω=γB1

[0045] Where ω is the Rabi oscillation frequency, γ is the cyclotron ratio, and B1 is the RF field strength.

[0046] This design enhances the coherent accumulation of the spin system, improves the signal-to-noise ratio, and provides support for high-resolution imaging.

[0047] The hardware module supports parallel imaging techniques such as SENSE or GRAPPA. The 256-channel receiving coil array accelerates the data acquisition process through multi-channel data acquisition, combined with processing by the subsequent image reconstruction module, while maintaining image resolution at the 0.2 mm isotropic level and a field of view of 150 mm.

[0048] Generally speaking, to cope with the challenges of magnetic field inhomogeneity brought about by the high field strength of 11.7 T, the hardware module adopts active and passive field shimming technology.

[0049] Through high-order shim coils and ferromagnetic material adjustments, the magnetic field uniformity within the imaging volume is maintained, image distortion is reduced, and the accuracy of spatial encoding of complex neural morphologies is ensured.

[0050] The hardware module is equipped with a real-time specific absorption rate (SAR) monitoring system to ensure that RF energy deposition does not exceed the safety limit of 10W / kg. This function is particularly important in high-field MRI because its RF heating effect is more significant and directly guarantees patient safety.

[0051] For example, the hardware module can be expanded to support a real-time motion correction system that monitors patient head movement through optical tracking or additional sensors and adjusts scan parameters to compensate for motion artifacts. This feature, while not required, can further improve image quality in high-resolution imaging.

[0052] Specifically, the hardware module executes the pulse sequence generated by the sequence design module to generate raw k-space data. Its output data is directly transmitted to the data acquisition module for subsequent processing.

[0053] The hardware module's scanning capability supports a 50% undersampling rate, which, combined with compressed sensing technology, ensures that the image reconstruction module can restore images with a resolution of 0.2 mm.

[0054] The magnetic field strength of the hardware module can be adjusted to specific imaging needs, although 11.7 T is the preferred configuration to optimize the signal-to-noise ratio.

[0055] Similarly, the number of receiving coil channels can vary according to the application scenario. 256 channels is the optimal balance point for the current design, providing a compromise between high sensitivity and data processing complexity.

[0056] The hardware module works in conjunction with the data acquisition module to ensure the execution of pulse sequences and the acquisition of high-quality k-space data.

[0057] Sampling strategy module: analyzes and scans data, designs adaptive spatial sampling patterns, and outputs sampling masks;

[0058] The sampling strategy module performs analysis and optimization based on the low-resolution reconnaissance scan data provided by the hardware module to generate a variable-density random k-space sampling trajectory. The reconnaissance scan usually uses a standard gradient echo sequence to collect data with a resolution of 1 mm, which takes about 2 minutes.

[0059] The sampling strategy module uses an artificial intelligence model to identify complex neural areas. The model is a convolutional neural network trained on a public dataset containing 1,000 brain MRI images, and uses mean square error as the loss function.

[0060] A convolutional neural network analyzes reconnaissance scan data and outputs a sampling density mask that prioritizes sampling points in complex regions (such as cortical folds).

[0061] The sampling strategy module uses the Monte Carlo method to generate variable density random k-space sampling trajectories to ensure that the center k-space sampling is dense and the edge k-space sampling is sparse, and the sampling rate satisfies the formula:

[0062] M≥0.5·N

[0063] Where M is the number of sampling points, N = 15003 is the total number of image pixels, calculated based on a resolution of 0.2 mm and a field of view of 150 mm, 0.5 is the undersampling ratio, and the target is a 50% Nyquist sampling rate.

[0064] The sampling strategy module adjusts the sampling density according to the importance score of the output. The importance score is based on the curvature and texture complexity of the neural region, and the formula is:

[0065] D(r)=D0·e -β·(1-I(r))

[0066] Where D(r) represents the sampling density of position r, D0=1 is the baseline density, β=0.5 is the attenuation factor, and I(r) is the importance score of position I(r), ranging from 0 to 1, based on the output of the convolutional neural network.

[0067] The sampling strategy module supports dynamic adjustment of the sampling strategy, optimizes subsequent k-space trajectories based on the visualization results of the previous scan, and realizes adaptive imaging.

[0068] The output of the sampling strategy module is the k-space sampling mask, which is directly input into the sequence design module to guide the gradient waveform design and ensure that the subsequent pulse sequence can focus on complex neural structures.

[0069] The 50% sampling rate significantly shortens the scanning time, and combined with the compressed sensing theory (based on the Donahue-Kanders theorem), it ensures that the image reconstruction module can recover high-resolution images from undersampled data.

[0070] Sequence design module: Generates advanced pulse sequences based on the output sampling mask;

[0071] In this embodiment, the sequence design module receives the k-space sampling mask generated by the sampling strategy module and develops a 3D gradient echo sequence based on the mask to ensure that the imaging resolution reaches 0.2 mm and the field of view remains at 150 mm.

[0072] The mask defines a variable density sampling trajectory, with dense sampling in the center area and sparse sampling in the edge area, providing an optimized basis for subsequent data collection.

[0073] First, the sequence design module focuses on the design of RF pulses, using quantum coherence enhancement technology to improve the coherence accumulation of the spin system. The flip angle of the RF pulse changes nonlinearly with time, following the formula:

[0074] α(t)=α0e -λt

[0075] Where α(t) is the flip angle that changes with time, α0 is the initial flip angle, and λ = 0.01S -1 is the attenuation coefficient, and t is the time variable of the pulse sequence execution.

[0076] Secondly, the field strength of the RF pulse is set to 10 microtesla, and the oscillation frequency is calculated using the Rabi oscillation theorem, as follows:

[0077] ω=γB1

[0078] Where ω is the Rabi oscillation frequency, γ is the cyclotron ratio, and B1 is the RF field strength.

[0079] For example, the pulse duration is fixed at 5 milliseconds, optimized to ensure signal acquisition efficiency matches the gradient switching capabilities of the hardware module. The gradient waveform is adjusted based on the sampling mask to support the execution of variable density sampling trajectories.

[0080] Ultimately, the pulse sequence generated by the sequence design module is directly transmitted to the data acquisition module for execution on the 11.7 Tesla MRI machine. This pulse sequence is compatible with the hardware module's 256-channel receive coil array and utilizes parallel imaging technology to accelerate data acquisition by a factor of 8.

[0081] For example, the gradient waveform design takes into account the maximum gradient intensity of 200 mT / m and the switching rate of 200 T / m / s of the hardware module to ensure the accurate realization of the k-space trajectory. The waveform adjustment process corresponds one-to-one with the density distribution of the sampling mask to ensure the effectiveness of the sampling strategy.

[0082] The sequence design module supports multi-parameter adjustment, such as changing the attenuation coefficient or initial flip angle to adapt to different imaging requirements. This flexibility makes it possible to image specific neural areas.

[0083] In addition, the sequence design module must ensure that it matches the performance constraints of the hardware module, such as calibrating the RF pulse parameters through real-time feedback to compensate for magnetic field inhomogeneity under high field strength.

[0084] Data acquisition module: executes advanced pulse sequences and collects undersampled spatial data;

[0085] The data acquisition module runs the pulse sequence generated by the sequence design module on the 11.7T magnetic resonance imaging device to collect k-space data with a 50% undersampling rate. The device is equipped with a 256-channel receiving coil array and supports parallel imaging technology to improve data acquisition efficiency.

[0086] First, the pulse sequence includes radio frequency pulses and gradient waveforms. The radio frequency pulse flip angle varies with time as α(t) = 60·e -0.01·t The RF field strength B1 = 10 μ, and the gradient waveform is designed according to the variable density random trajectory of the sampling strategy module to ensure dense sampling of the central k-space and sparse sampling of the edge k-space.

[0087] Secondly, the data acquisition module uses 256 channels of receiving coils to collect signals in parallel, with an acceleration factor of 8, which significantly shortens the scanning time. The scanning time is controlled at 600 seconds, the field of view is 150 mm, the resolution target is 0.2 mm, the number of samples per dimension is 750, and the total number of pixels N = 7503

[0088] For example, the hardware module's gradient coil has a maximum intensity of 200 mT / m and a switching rate of 200 T / m / s, supporting rapid gradient waveform switching and variable-density sampling trajectories. The specific absorption rate (SAR) is monitored in real time during the acquisition process to ensure it does not exceed 10 W / kg, ensuring patient safety.

[0089] Next, the data acquisition module outputs undersampled k-space data with a data dimension of M = 0.5 N, where M is the number of sampling points, meeting the sparsity requirement of compressed sensing. The formula is:

[0090] M≥C·S·log(N)

[0091] Wherein, C=2 is the compressed sensing constant, S=0.1 is the signal sparsity, and log(N) is the natural logarithm based on N=750.

[0092] Finally, the acquired data is transmitted to the image reconstruction module through the hardware module's control system for subsequent processing. The data quality directly depends on the pulse parameters of the sequence design module and the performance of the hardware module.

[0093] For example, parallel imaging technology of a 256-channel receiving coil array is implemented using the SENSE or GRAPPA algorithm, and an acceleration factor of 8 ensures that the scan time is controlled within 10 minutes.

[0094] The data acquisition module supports real-time motion correction, monitors the patient's head movement through an optical tracking system, adjusts acquisition parameters, and reduces motion artifacts.

[0095] In general, the data acquisition module works closely with the pulse sequence of the sequence design module, utilizing the high field strength and multi-channel capabilities of the hardware module to efficiently acquire undersampled data, providing a foundation for high-resolution imaging.

[0096] In summary, the performance of the data acquisition module directly affects the imaging quality of the system, and its technical implementation ensures the accurate acquisition of complex neural morphologies.

[0097] Image reconstruction module: reconstructs high-resolution images from undersampled spatial data;

[0098] In the present invention, the image reconstruction module is responsible for reconstructing a high-resolution image from the undersampled k-space data provided by the data acquisition module, providing a basis for the post-processing and visualization modules.

[0099] This module is closely connected with the previous data acquisition module. It uses the 50% undersampling rate data it collects to ensure a resolution of 0.2 mm through compressed sensing and deep learning technology, meeting the requirements for precise display of complex neural morphologies.

[0100] In this embodiment, the image reconstruction module receives the undersampled k-space data output by the data acquisition module, and the data dimension is M=0.5·N, where N=750 3 Total number of fully sampled pixels, based on a resolution of 0.2 mm and a field of view of 150 mm.

[0101] First, the reconstruction process uses the iterative soft threshold algorithm (ISTA), the formula is:

[0102] x k+1 =S λ (x k +Φ T (y-Φx k ))

[0103] Among them, x k+1 is the image estimation of the k+1th iteration, represents the image estimation of the kth iteration, Φ is the measurement matrix, and maps the image space to the k space. T is the transpose of the measurement matrix, y is the undersampled k-space data, S λ is the soft threshold function, and the regularization parameter λ=0.01·||Φ T y||2, where ||Φ T y||2 is the two-norm with unit of None, and the value of λ is calculated by the formula to ensure the stability of reconstruction.

[0104] Secondly, the ISTA algorithm is based on compressed sensing theory and utilizes the sparsity of the signal in the wavelet domain. The sparse basis uses the Daubechies wavelet transform. The number of iterations is set to 100 to ensure convergence.

[0105] For example, Daubechies wavelet transform is used as a sparse basis to decompose the image into multi-scale representation, which meets the requirements of the Donahue-Kanders theorem. The formula is:

[0106] min||Ψx||1subjectto||y-Φx||2≤∈

[0107] Among them, Ψ is the wavelet transform matrix, ∈ is the error tolerance, and the empirical value is set to 0.01.

[0108] Next, the image reconstruction module combines deep learning technology and uses the U-Net model to accelerate the reconstruction process.

[0109] U-Net is implemented based on PyTorch and trained on a simulated MRI dataset. The input is undersampled k-space data and the output is an image with a resolution of 0.2 mm.

[0110] Finally, the reconstructed image is transferred to the post-processing and visualization module for neural structure analysis and 3D rendering. Reconstruction time is controlled within 5 minutes, ensuring the overall efficiency of the system.

[0111] For example, the loss function of the U-Net model uses mean square error, and the training data covers a variety of brain lesion scenarios, improving adaptability to complex neural morphologies.

[0112] In general, the image reconstruction module restores high-resolution images from undersampled data through the synergy of compressed sensing and deep learning, ensuring the imaging quality of the system.

[0113] In summary, the technical implementation of the image reconstruction module ensures the accurate reconstruction of the complex morphology of the nerve, and its performance directly affects the final visualization effect.

[0114] Post-processing and visualization module: Analyzes and reconstructs images to generate 3D rendering images of complex neural morphologies;

[0115] The post-processing and visualization module is mainly responsible for analyzing the high-resolution images output by the image reconstruction module and generating 3D rendered images of complex neural morphologies, providing intuitive display for research and clinical diagnosis.

[0116] This module is closely connected with the previous image reconstruction module. It uses the reconstructed image with a resolution of 0.2 mm to segment the neural structure through the graph neural network (GNN), ensuring that the post-processing results can be fed back to the sampling strategy module to optimize the subsequent scanning strategy.

[0117] In this embodiment, the post-processing and visualization module receives an image with a resolution of 0.2 mm output by the image reconstruction module, and the image dimensions are 750x750x750, which is calculated based on a field of view of 150 mm and a resolution of 0.2 mm.

[0118] First, the post-processing module uses a graph neural network (GNN) to segment the cortex and nerve fibers. Based on the graph connectivity theorem, the neural network topology is analyzed. The GNN is trained on a labeled cortical dataset to identify the connectivity of the neural structure and ensure segmentation accuracy.

[0119] Secondly, the segmentation process of GNN is defined based on the nodes and edges of the graph, and the formula is:

[0120]

[0121] Where C(v) is the connectivity score of node v, N(v) is the neighborhood of node v, w(u,v) is the edge weight based on spatial distance and signal strength, and f(u) is the feature vector of node u.

[0122] For example, GNN's training data covers a variety of brain pathology scenarios, improving its adaptability to complex neural morphologies.

[0123] Next, the visualization module imports the segmented image into ParaView or ITK-SNAP tools to generate a 3D rendered image in VTK format, which supports interactive operations.

[0124] Finally, 3D rendered images allow for magnification, rotation, and highlighting of complex neural areas, such as cortical folds or tiny nerve fibers. A resolution of 0.2 mm is maintained, ensuring clear details of neural structures.

[0125] For example, the output in VTK format facilitates further analysis by researchers and supports interactive visualization needs.

[0126] In general, the post-processing and visualization module presents complex neural morphology through GNN segmentation and 3D rendering, providing intuitive results for the system.

[0127] In summary, the present invention proposes an original enhanced magnetic resonance neuroimaging system for precise display of complex neural morphologies. It aims to achieve 0.2 mm resolution imaging of complex neural structures (such as the cortex and tiny nerve fibers) through an 11.7T high-field magnetic resonance imaging device and a 256-channel receiving coil array, with a scanning time of approximately 10 minutes.

[0128] The system integrates an AI-driven adaptive k-space sampling strategy, uses the Monte Carlo method to generate variable-density random sampling trajectories, and combines compressed sensing technology based on the Donahue-Kanders theorem to efficiently reconstruct images from 50% undersampled data. At the same time, it improves the signal-to-noise ratio through quantum coherence enhancement based on the Rabi oscillation theorem, and uses deep learning (U-Net) and graph neural networks (GNN) to optimize image reconstruction and neural structure segmentation, ultimately generating 3D rendered images in VTK format.

[0129] The system supports real-time adaptive adjustment, ensuring that the imaging process dynamically optimizes complex neural areas, significantly improving the ability to study neural connections and analyze neurological diseases (such as Alzheimer's disease).

[0130] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An original enhanced magnetic resonance neuroimaging system for accurate display of complex neural morphology, characterized by: Includes the following modules: Hardware module: provides a physical scanning platform with high resolution and signal-to-noise ratio to scan data; Sampling strategy module: analyzes and scans data, designs adaptive spatial sampling patterns, and outputs sampling masks; Sequence design module: Generates advanced pulse sequences based on the output sampling mask; Data acquisition module: executes advanced pulse sequences and collects undersampled spatial data; Image reconstruction module: reconstructs high-resolution images from undersampled spatial data; Post-processing and visualization module: Analyze the reconstructed image and generate 3D rendering images of the complex morphology of the nerve.

2. The original enhanced magnetic resonance neuroimaging system for accurate display of complex neural morphology according to claim 1 is characterized in that: The physical scanning platform includes: providing an 11.7T magnetic resonance imaging device with a magnetic field strength B0=11.7T and equipped with a 256-channel receiving coil array.

3. The original enhanced magnetic resonance neuroimaging system for accurate display of complex neural morphology according to claim 1 is characterized in that: The design of adaptive spatial sampling comprises the following steps: S1, perform a low-resolution reconnaissance scan to acquire 1 mm resolution k-space data; S2. Input the artificial intelligence model to analyze complex neural regions and generate variable density random k-space sampling trajectories.

4. The original enhanced magnetic resonance neuroimaging system for accurate display of complex neural morphology according to claim 1 is characterized in that: The advanced pulse sequence is mainly a radio frequency pulse sequence, and the flip angle changes with time as α(t)=60·e -0.01·t degrees, RF field strength B1=10μT, Rabi oscillation frequency ω=42.58·10 6 ·B1rad / s.

5. The original enhanced magnetic resonance neuroimaging system for accurately displaying complex neural morphology according to claim 3 is characterized in that: The artificial intelligence model is a convolutional neural network that generates sampling masks based on low-resolution reconnaissance scan data to optimize the sampling density of complex neural areas.

6. The original enhanced magnetic resonance neuroimaging system for accurate display of complex neural morphology according to claim 4, characterized in that: The duration of the radio frequency pulse sequence is 5 milliseconds, and data is collected in parallel through 256-channel receiving coils with an acceleration factor of 8.

7. The original enhanced magnetic resonance neuroimaging system for accurate display of complex neural morphology according to claim 5, characterized in that: The convolutional neural network was trained on a dataset of 1000 brain MRI images, using mean square error as the loss function.

8. The original enhanced magnetic resonance neuroimaging system for accurate display of complex neural morphology according to claim 1 is characterized in that: The image reconstruction module further includes using a U-Net model to accelerate the reconstruction process, wherein the U-Net is implemented based on PyTorch and trained on a simulated MRI dataset.

9. The original enhanced magnetic resonance neuroimaging system for accurately displaying complex neural morphology according to claim 8, characterized in that: The U-Net model generates T1 and T2 quantitative maps with a resolution of 0.2 mm from the undersampled spatial data to complement the 3D rendered images.

10. The original enhanced magnetic resonance neuroimaging system for accurate display of complex neural morphology according to claim 1, characterized in that: The 3D rendered image is segmented into cortexes using a graph neural network, and the neural network topology is analyzed based on the graph connectivity theorem.

Images