Multi-modal magnetic resonance imaging method and intelligent cloud computing platform
By acquiring and processing data in a magnetic resonance imager, using neural networks and biophysical models to generate high-quality quantitative and contrast images, the problem of long scanning time in traditional magnetic resonance imaging methods is solved, and imaging efficiency and image quality are improved.
Patent Information
- Application Number
- CN202411136642.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-12-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The scanning time of traditional magnetic resonance quantitative imaging methods is too long, resulting in a contradiction between spatial resolution and imaging speed, limiting its clinical application.
Using multimodal magnetic resonance imaging method, multimodal quantitative data acquisition based on biophysical parameters is carried out in a medical magnetic resonance image, and data is transmitted to the server in real time for neural network adaptive denoising and reconstruction of magnetic resonance biophysical model images. , generate multiple quantitative parameter images and synthesize multiple contrast-weighted images required for clinical use.
The total acquisition time of the magnetic resonance imager is shortened, image quality and accuracy is improved, patient discomfort is reduced, and more abundant multi-parameter quantitative images and contrast-weighted images are obtained in the same time.
Smart Images

Figure CN119090985A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information processing technology, and in particular to a multimodal magnetic resonance imaging method and an intelligent cloud computing platform. Background Art
[0002] Magnetic resonance imaging is a multi-parameter imaging technique without ionizing radiation and has a wide range of medical clinical applications. As a research direction of magnetic resonance imaging, quantitative magnetic resonance imaging mainly uses the changes in the intensity of the received signal after excitation with different radio frequency pulse sequences to calculate the images of quantitative parameters such as longitudinal relaxation time (T1), transverse relaxation time (T2) and proton density (PD) of different tissues in the body. This quantitative image can intuitively reflect the different physiological and physical properties of tissues and has been proven to be a sensitive biomarker for diagnosing diseases (such as epilepsy, multiple sclerosis, cancer, etc.). However, due to the long scanning time of traditional magnetic resonance quantitative imaging methods, these methods are rarely used in clinical applications. Spatial resolution and imaging speed are a pair of contradictory requirements in quantitative imaging. Although some progress has been made in recent studies, fast, stable and high-resolution quantitative imaging of the whole brain is still a challenge. This bottleneck limits the application of quantitative imaging in clinical practice. Summary of the invention
[0003] The purpose of the present invention is to provide a multimodal magnetic resonance imaging method and an intelligent cloud computing platform to effectively improve the efficiency and imaging quality of magnetic resonance quantitative imaging.
[0004] To achieve the above object, the present invention provides the following solution.
[0005] In one aspect, the present invention provides a multimodal magnetic resonance imaging method, comprising:
[0006] Perform multimodal quantitative magnetic resonance data acquisition based on biophysical parameters in medical magnetic resonance imaging devices;
[0007] The collected magnetic resonance data is transmitted in real time to a local or cloud-based server;
[0008] Performing adaptive intelligent denoising based on a neural network on the transmitted magnetic resonance data on the server to obtain denoised magnetic resonance data;
[0009] Perform image reconstruction based on the magnetic resonance biophysical model on the denoised magnetic resonance data to obtain multiple quantitative parameter images;
[0010] A variety of contrast-weighted images required clinically are synthesized based on a variety of quantitative parameter images.
[0011] Optionally, the multimodal quantitative magnetic resonance data acquisition based on biophysical parameters in a medical magnetic resonance imager specifically includes:
[0012] Different biophysical parameters are used in each excitation of the pulse sequence, and the acquisition method is three-dimensional variable density spiral trajectory acquisition; the pulse sequence includes inversion recovery pulse sequence, multiple inversion recovery pulse sequence, liquid attenuation inversion recovery pulse sequence, gradient echo pulse sequence, steady-state free precession pulse sequence, spin echo pulse sequence and planar echo imaging pulse sequence; different biophysical parameters include different repetition time, echo time and radio frequency pulse flip angle.
[0013] Optionally, the real-time transmission of the acquired magnetic resonance data to a local or cloud-based server specifically includes:
[0014] The collected magnetic resonance data is converted into a new format and then transmitted in real time to a local or cloud-based server via the TCP / IP protocol.
[0015] Optionally, performing adaptive intelligent denoising based on a neural network on the transmitted magnetic resonance data on the server to obtain denoised magnetic resonance data specifically includes:
[0016] Automatically identify the noise level of magnetic resonance data, including noise intensity and distribution, through pre-trained neural networks;
[0017] De-noising algorithms of different strengths are used for magnetic resonance data with different noise levels.
[0018] Optionally, the denoised magnetic resonance data is subjected to image reconstruction based on a magnetic resonance biophysical model to obtain multiple quantitative parameter images, specifically including:
[0019] generating a magnetic resonance signal dictionary based on the biophysical parameters and the magnetic resonance biophysical model;
[0020] The MRI signal dictionary is projected into the subspace, combined with the denoised MRI data, and iteratively reconstructed to obtain the subspace coefficient spectrum;
[0021] Back-project the obtained subspace coefficient spectrum back to the time domain;
[0022] The image back-projected back into the time domain is identified point by point with the magnetic resonance signal dictionary to obtain multiple quantitative parameter images.
[0023] Optionally, the multiple quantitative parameter images include longitudinal relaxation time T1 quantitative image, transverse relaxation time T2 quantitative image, proton density quantitative image, quantitative magnetic susceptibility image, myelin water fraction image, apparent diffusion coefficient image, anisotropy fraction image, main magnetic field quantitative image and radio frequency field quantitative image.
[0024] Optionally, synthesizing a plurality of contrast-weighted images required clinically according to a plurality of quantitative parameter images specifically includes:
[0025] Various contrast-weighted images of T1-weighted, T2-weighted and diffusion-weighted were synthesized based on various quantitative parameter images and magnetic resonance biophysical models.
[0026] On the other hand, the present invention also provides a multimodal magnetic resonance imaging intelligent cloud computing platform, including: a medical magnetic resonance imager and a server deployed locally or in the cloud, wherein the server is communicatively connected to the medical magnetic resonance imager to implement the multimodal magnetic resonance imaging method.
[0027] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0028] The multimodal magnetic resonance imaging method and intelligent cloud computing platform provided by the present invention can perform multimodal quantitative magnetic resonance data acquisition based on biophysical parameters; transmit the acquired magnetic resonance data to a local or cloud server in real time; perform fast and high-fidelity artificial intelligence denoising on the transmitted magnetic resonance data on the server; perform image reconstruction based on the magnetic resonance biophysical model on the denoised magnetic resonance data to obtain multiple quantitative parameter images; generate multiple contrast weighted images required for clinical use based on multiple quantitative images. Compared with the existing magnetic resonance imaging technology, the method proposed by the present invention can shorten the total acquisition time of magnetic resonance, improve the image quality, and obtain more accurate quantitative parameter imaging results and synthesize multi-contrast parameter imaging required for clinical use at the same acquisition time. The intelligent cloud computing platform proposed by the present invention can quickly obtain high-resolution magnetic resonance quantitative images and multiple contrast magnetic resonance weighted images required for clinical use, shorten the time for medical magnetic resonance imaging to scan patients, improve the efficiency of hospital magnetic resonance scanning, and reduce the discomfort of patients in medical magnetic resonance imaging; the real-time denoising algorithm proposed by the present invention can effectively remove noise and artifacts from medical images and improve the quality of magnetic resonance images. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0030] Figure 1 A schematic diagram of the overall architecture of the multimodal magnetic resonance imaging intelligent cloud computing platform provided by the present invention;
[0031] Figure 2 A flow chart of the multimodal magnetic resonance imaging method provided by the present invention;
[0032] Figure 3 This is a schematic diagram of the magnetic resonance quantitative image reconstruction process proposed by the present invention;
[0033] Figure 4 Schematic diagram of a whole-brain multi-parameter quantitative image and a synthesized whole-brain multi-contrast weighted image obtained by the method of the present invention;
[0034] Figure 5 A schematic diagram for comparing the multiple contrast-weighted images synthesized by the present invention and the multiple contrast-weighted images of traditional magnetic resonance imaging. DETAILED DESCRIPTION
[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0036] The present invention provides a multimodal magnetic resonance imaging method and an intelligent cloud computing platform to effectively improve the efficiency and quality of magnetic resonance quantitative imaging. In addition, the scalable system architecture of the present invention can build a new medical imaging experimental platform for users and provide cutting-edge medical imaging technology.
[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0038] Figure 1 The overall architecture diagram of the multi-modal magnetic resonance imaging intelligent cloud computing platform provided by the present invention is shown in FIG. Figure 1As shown, the multimodal magnetic resonance imaging intelligent cloud computing platform of the present invention includes a medical magnetic resonance imager and a server arranged locally or in the cloud. Among them, the medical magnetic resonance imager integrates a fast and high-resolution quantitative acquisition technology of magnetic resonance data, and a multimodal quantitative pulse sequence based on biophysical parameters is used in the medical magnetic resonance imager to quickly acquire magnetic resonance data. The medical magnetic resonance imager is connected to the server in communication, and the collected magnetic resonance data is transmitted to the server arranged locally or in the cloud in real time by using the real-time transmission technology of magnetic resonance data. The server integrates the intelligent denoising technology of magnetic resonance data, the quantitative magnetic resonance image reconstruction technology, and the multi-contrast weighted image synthesis technology of magnetic resonance. The magnetic resonance data denoising technology is used to perform fast and high-fidelity artificial intelligence denoising on the transmitted magnetic resonance data to obtain high-fidelity magnetic resonance data; further, the magnetic resonance quantitative image reconstruction technology is used to reconstruct the denoised magnetic resonance data based on the magnetic resonance biophysical model to obtain a variety of quantitative parameter images; further, the multi-contrast weighted image synthesis technology of magnetic resonance is used to synthesize a variety of contrast weighted images required for clinical use according to the obtained multiple quantitative images.
[0039] In addition, the multimodal magnetic resonance imaging intelligent cloud computing platform of the present invention adopts a scalable platform architecture. Users can not only use the data acquisition, transmission, denoising, image reconstruction and synthesis technologies of the platform to quickly obtain multi-parameter quantitative images and multi-contrast images required for clinical use, but also use the scalable characteristics and computing resources of the platform to develop and experiment with new technologies, and continuously update the medical imaging technology provided by the platform.
[0040] The present invention aims to solve the problems of long scanning time and low resolution in magnetic resonance quantitative imaging, and proposes a series of improved methods for magnetic resonance acquisition, data transmission, data denoising, image reconstruction and post-processing technology, and integrates them into an intelligent cloud computing platform. The platform proposed by the present invention can shorten the total acquisition time of the magnetic resonance imager, improve the acquisition efficiency and the quality of magnetic resonance images, and obtain more accurate and richer multi-parameter quantitative images and multi-contrast weighted images compared with traditional magnetic resonance imaging technology at the same acquisition time.
[0041] Based on the multimodal magnetic resonance imaging intelligent cloud computing platform provided by the present invention, a multimodal magnetic resonance imaging method can be implemented, such as Figure 2 As shown, the method includes:
[0042] Step 1: Perform multimodal quantitative magnetic resonance data acquisition based on biophysical parameters in a medical magnetic resonance imaging instrument.
[0043] A fast and high-resolution quantitative acquisition technology for magnetic resonance data is integrated in a real-time medical magnetic resonance imager. Different from the traditional magnetic resonance acquisition technology, the present invention develops a three-dimensional continuous radio frequency pulse excitation and fast acquisition technology in a medical magnetic resonance imager, so that high-resolution multi-parameter quantitative imaging can be obtained simultaneously. The fast and high-resolution quantitative acquisition technology for magnetic resonance data specifically refers to the use of different biophysical parameters such as repetition time, echo time and radio frequency pulse flip angle in each excitation of the pulse sequence, and the acquisition method is three-dimensional variable density spiral trajectory acquisition. The pulse sequence of this architecture includes but is not limited to diffusion-weighted pulse sequence, inversion recovery pulse sequence, multi-inversion recovery pulse sequence, balanced free precession steady-state pulse sequence, spin echo pulse sequence, liquid attenuated inversion recovery pulse sequence, gradient echo pulse sequence and planar echo imaging pulse sequence.
[0044] Step 2: The collected magnetic resonance data is transmitted in real time to a local or cloud-based server.
[0045] The real-time transmission technology of magnetic resonance data is integrated in the local or cloud-based server, which can transmit the magnetic resonance data collected by the medical magnetic resonance imaging device to the local or cloud-based server in real time and synchronously. The specific steps are as follows: The medical magnetic resonance imaging device can generate a large amount of image data during scanning. These data first need to be processed preliminarily on the device itself in order to be converted into a transmittable format, such as HDF5 or MRD format. Subsequently, the TCP / IP protocol can be used to transmit reliable, orderly and error-detected data to the cloud or locally deployed server to facilitate the next step of processing and analysis.
[0046] Step 3: On the server, the transmitted magnetic resonance data is subjected to adaptive intelligent denoising based on a neural network to obtain denoised magnetic resonance data.
[0047] The server also integrates intelligent denoising technology for magnetic resonance data. It uses a neural network-based intelligent denoising algorithm to achieve adaptive and rapid denoising according to the noise level of the transmitted magnetic resonance data, and obtains magnetic resonance data with high fidelity and high signal-to-noise ratio. The general steps are as follows: The algorithm first analyzes the transmitted magnetic resonance data and automatically identifies the noise intensity and distribution. This can be achieved through a pre-trained neural network to complete the learning of noise of different intensities and different noise distributions caused by the geometric factors of the receiving coil. Then, according to the identified noise level (noise intensity and distribution), the algorithm adjusts its denoising strategy. This means that for images with high noise intensity, the algorithm will apply stronger denoising processing, while for images with low noise intensity, lighter denoising processing is used to avoid losing too much image details. The use of efficient network architecture and algorithm optimization ensures the rapid execution of the denoising process, making it suitable for real-time or near-real-time medical image processing needs. The generated denoised magnetic resonance data will be saved in a medical image standard format, such as DICOM format or NIfTI format, for users to download.
[0048] Step 4: Perform image reconstruction based on the magnetic resonance biophysical model on the denoised magnetic resonance data to obtain multiple quantitative parameter images.
[0049] The server of the present invention also integrates a magnetic resonance quantitative image reconstruction technology, and uses a reconstruction algorithm based on a magnetic resonance biophysical model to reconstruct high-quality multi-parameter quantitative images. Figure 3 FIG. 2 shows a schematic diagram of the reconstruction algorithm process based on the magnetic resonance biophysical model proposed in the present invention. Figure 3 , the step 4 specifically comprises:
[0050] Step 4.1: Generate a magnetic resonance signal dictionary based on biophysical parameters and a magnetic resonance biophysical model.
[0051] The magnetic resonance biophysical model used in the present invention specifically refers to the signal model of the extended phase map. According to the biophysical parameters of the pulse sequence in step 1 (different repetition times, echo times and radio frequency pulse flip angles) and the signal model of the extended phase map, a magnetic resonance signal dictionary can be generated. Specifically, the extended phase map is a signal simulation model, which is realized by numerically solving the basic physical equation (Bloch equation) of magnetic resonance. Under the signal model of the extended phase map, according to the above-mentioned pulse sequence biophysical parameters and a set of given relaxation time ranges (such as longitudinal relaxation time T1 from 1-5000ms, transverse relaxation time T2 from 1-4000ms), a corresponding set of magnetic resonance signals can be numerically simulated, and this set of magnetic resonance signals covers all possible relaxation time ranges of human physiology, so this set of complete magnetic resonance signal sets is called a magnetic resonance signal dictionary.
[0052] The principle of generating a magnetic resonance signal dictionary by expanding the phase map is described in detail below.
[0053] The extended phase diagram and the Bloch equation are related by Fourier transform. In the extended phase diagram, the transverse state is defined as F + (k) and F - (k), the longitudinal state is Z(k), where k is the dephasing order. In general, the definition of the extended phase diagram F, the Z state matrix and the magnetization vector [M x ,M y ,M z ] T The following Discrete Fourier Transform (DFT) relationship exists:
[0054]
[0055] Where DFT stands for discrete Fourier transform and r represents the spatial coordinate. From equation (1), we can see that: (i) the spin state F + and F-conjugate; (ii) the magnetization vector of the Bloch equation is in the position space, and the F state matrix represents the Fourier transform of the position space. It can be seen that F is actually the harmonic components of the sum of a series of magnetization vectors in the position space, where the dephasing order k actually represents the order of different harmonics in the Fourier transform. In other words, the extended phase map algorithm describes the "state space" and the Bloch equation describes the "position space", which are a pair of Fourier transform pairs.
[0056] The effect of the sequence on signal changes is mainly determined by the RF pulse excitation, the phase dispersion caused by the unbalanced gradient and the relaxation effect during the acquisition process. The following introduces the influence of these factors on signal changes in the extended phase diagram.
[0057] (i) The effect of RF pulse excitation. Consider the flip angle of the nth RF pulse excitation to be α n , the phase is φ n , then define the operator T(φ n ,α n )for:
[0058]
[0059] From formula (2), we can see that the operator T(φ n ,α n ) is a complex matrix. Since T is independent of k, after the RF pulse excitation, the new state matrix can be expressed as:
[0060]
[0061] It can be seen from equations (2) and (3) that the RF pulse can change all elements of the state matrix.
[0062] (ii) The role of relaxation. Similar to the Bloch equation simulation, the transverse relaxation (T2 relaxation) and longitudinal relaxation (T1 relaxation) in the extended phase diagram are n The moments can be expressed as:
[0063]
[0064] and
[0065]
[0066] where TR n represents the nth repetition time; Z0 is the initial state of the longitudinal state Z(k). Combining equations (2) to (5), if only the RF pulse and relaxation effects are considered, for some gradient-completely balanced sequences, the state matrix can be expressed as:
[0067]
[0068] (iii) Unbalanced magnetic resonance gradient effect. The state matrix is a 3×k matrix and can be expressed as:
[0069]
[0070] For the gradient unbalanced sequence, the unbalanced gradient actually plays the role of dephasing in the plane, which is equivalent to the change of the dephasing order k of all the plane states F in the state matrix in equation (7). A unit positive gradient is represented by an increase of one dephasing order of F in the state matrix, and a unit negative gradient is represented by a decrease of one dephasing order of F. Since the effect of the gradient does not change the longitudinal magnetization vector, the state of the Z matrix does not change.
[0071] When the F matrix is in the 0 state, there is no dephasing effect and the echo will be generated. According to the role of the previous RF pulse, relaxation effect and magnetic resonance gradient in the extended phase map, the signal changes of the echo of the tissue with given relaxation parameters in multiple repetition times of the sequence can be simulated. Assuming that the pulse sequence has n repetition times (time of repetitions, TR), the signal matrix can be expressed as:
[0072]
[0073] The elements with a subscript of 0 in the F matrix represent the echo signals generated by n TRs. Therefore, it is only necessary to calculate the F0 matrix to simulate the signal change curve of the magnetic resonance. When a combination of multiple different relaxation times (T1, T2 values) is given, the magnetic resonance signal dictionary can be generated according to formula (8).
[0074] As mentioned above, the extended phase map and Bloch equation simulation are mathematically a pair of Fourier transforms, which respectively describe the changes of protons in the "state space" and "position space" due to RF pulse excitation, magnetic resonance gradient and relaxation, so the two are completely equivalent.
[0075] Step 4.2: Project the MRI signal dictionary into the subspace, combine it with the denoised MRI data, and iteratively reconstruct the subspace coefficient map.
[0076] Subspace reconstruction is an iterative reconstruction algorithm based on the spatial-temporal low-rank nature of magnetic resonance signals. The specific steps are: (1) Perform singular value decomposition on the magnetic resonance signal dictionary generated in step 4.1 to obtain right singular vectors; select the first few vectors with the highest energy among the truncated right singular vectors, which are called the temporal bases of the dictionary; at this time, the dictionary has been projected into the subspace through the temporal base; (2) The magnetic resonance data collected in step 1 are also projected into the subspace through the same temporal base in (1), so that the large amount of magnetic resonance data collected in step 1 can be represented by a linear combination of the temporal bases of the dictionary; the coefficients of each temporal base are called the coefficient map of the subspace, see for details. Figure 3 (3) Because the magnetic resonance data quickly acquired in step 1 are full of noise and artifacts, it is necessary to use iterative reconstruction to remove the noise and artifacts projected into the subspace coefficient map; the iterative reconstruction algorithm mainly uses common convex optimization algorithms such as the conjugate gradient method and the alternating direction multiplier algorithm (ADMM) to ensure that the reconstructed coefficient map and the actually acquired magnetic resonance signal meet the requirements in terms of data consistency.
[0077] Step 4.3: Back-project the obtained subspace coefficient spectrum back to the time domain.
[0078] After obtaining a high-quality subspace coefficient spectrum in step 4.2, the coefficient spectrum and the transpose of the time basis of step 4.2 can be used for high-dimensional matrix multiplication, and the coefficient spectrum can be back-projected back to the time domain. The purpose of this step is to back-project the magnetic resonance signal from the subspace back to the time domain, thereby restoring the high-quality magnetic resonance signal. As mentioned above, in step 1, a series of magnetic resonance signals were acquired using rapid acquisition technology. These signals are full of noise and image artifacts due to the short acquisition time. By utilizing the time-space low-rank characteristics of magnetic resonance, these magnetic resonance signals are projected into the subspace for iterative reconstruction, removing noise and artifacts, obtaining a high-quality coefficient spectrum, and then back-projecting it back to the time domain. In this way, the reconstructed signal can effectively remove the noise and image artifacts caused by rapid acquisition.
[0079] Step 4.4: The image back-projected back into the time domain is identified point by point with the magnetic resonance signal dictionary to obtain multiple quantitative parameter images.
[0080] As mentioned above, the magnetic resonance signal dictionary contains the range of all possible relaxation times T1 and T2 of the human body. If the signal curve of each pixel of the image projected back to the time domain is matched one by one with the magnetic resonance signal dictionary, the values of relaxation times T1 and T2 can be inferred. The present invention uses point multiplication to find the magnetic resonance signal dictionary entry that best matches the signal curve: the signal curve of each pixel is multiplied by all dictionary entries respectively, and the dictionary entry with the largest product is the most matching one. Find the most matching dictionary entry, and the corresponding T1 and T2 values can be obtained. By doing the same operation on each pixel, a quantitative spectrum of relaxation times T1 and T2 can be obtained. At the same time, a proton density spectrum can also be obtained. Since only T1 and T2 are considered in the extended phase map simulation, and the influence of PD is not considered, the signal itself is normalized. The collected signal has a weighted proton density, so the proton density of each pixel in the image can be obtained by the ratio of the square sum of the actual collected signal to the square sum of the corresponding elements of the magnetic resonance signal dictionary. After performing this operation on each pixel of the image, the proton density value of each pixel is divided by the maximum value of the PD of all pixels and normalized to the range of [0,1] to obtain a normalized proton density map. Through this method, T1, T2 and PD values can be identified at the same time.
[0081] Similarly, the extended phase map can be further used to expand the MRI signal dictionary, and other quantitative parameter maps can be obtained by adding a variety of different preparation pulses, such as quantitative susceptibility mapping (QSM), myelin-water fraction (MWF), apparent diffusion coefficient (ADC) image, fractional anisotropy (FA) image, main magnetic field (B0) quantitative image and radio frequency field (B1) quantitative image. The difference in obtaining these quantitative images is that different radio frequency pulses need to be applied in step 1, and according to the acquired magnetic resonance data, they can all be obtained using the quantitative image reconstruction technology described in step 3. Figure 4 Part (A) shows the whole brain multi-parameter quantitative image obtained by the present invention. Of course, the multiple quantitative parameter images obtained by the method of the present invention may include but are not limited to T1 quantitative image, T2 quantitative image, PD quantitative image, QSM image, MWF image, ADC image, FA image, B0 quantitative image and B1 quantitative image. All quantitative images generated in step 4 will be saved in a medical image standard format, such as DICOM format or NIfTI format, for users to download.
[0082] Step 5: Synthesize multiple contrast-weighted images required clinically based on multiple quantitative parameter images.
[0083] The server of the present invention also integrates a magnetic resonance multi-contrast weighted image synthesis technology, adopts an image synthesis algorithm based on magnetic resonance signal characteristics, utilizes the multiple quantitative parameter images obtained in step 4, the extended phase image signal model based on magnetic resonance physics, and the magnetic resonance signal dictionary based on human physiological parameters, to synthesize T1-weighted, T2-weighted, diffusion-weighted and other contrast-weighted magnetic resonance images, referred to as multi-contrast weighted images.
[0084] Multi-contrast weighted images refer to magnetic resonance images commonly used in clinical magnetic resonance imaging in hospitals and used for clinical diagnosis. These different contrast-weighted images include T1-weighted, T2-weighted, and diffusion-weighted. These commonly used weighted images in clinical practice are acquired using traditional magnetic resonance imaging sequences, such as inversion recovery pulse sequence, double inversion recovery pulse sequence, fluid-attenuated inversion recovery pulse sequence, gradient echo pulse sequence, steady-state free precession pulse sequence, fast spin echo pulse sequence, and planar echo imaging pulse sequence. These sequences can be used to obtain multi-contrast weighted images required for clinical diagnosis. However, these pulse sequences require a long scanning time, such as Figure 5As shown, it takes 22 minutes to obtain a low-resolution image. The rapid magnetic resonance acquisition + data transmission + image reconstruction technology proposed in the present invention is different from the traditional imaging methods of these pulse sequences. The present invention first obtains a multi-parameter quantitative image, and then synthesizes the multi-contrast weighted image required for clinical use based on the signal model of the extended phase image. In this way, it only takes 2.5 minutes to obtain the high-resolution image required for clinical use, which shortens the scanning time required for clinical use by 9 times, greatly improves imaging efficiency, reduces the scanning time of patients in the medical magnetic resonance imaging device, and improves comfort.
[0085] In addition, a magnetic resonance image with customized contrast can also be obtained according to user needs. Because the method of the present invention is based on the signal model of multi-parameter quantitative images and extended phase images to synthesize contrast-weighted images, the user only needs to input the required repetition time and echo time and the required contrast on the platform, and the platform can synthesize the magnetic resonance image required by the user through these input parameters and signal models.
[0086] Figure 4 Part (B) shows the whole brain multi-contrast weighted image synthesized by the method of the present invention. All multi-contrast weighted images generated in step 5 will be saved in a medical image standard format, such as DICOM format or NIfTI format, for users to download.
[0087] Compared with the traditional method, the platform and method proposed by the present invention have at least the following advantages:
[0088] (1) High-resolution quantitative magnetic resonance images can be obtained quickly, shortening the time for medical magnetic resonance imaging to scan patients, improving the efficiency of hospital magnetic resonance scanning, and reducing the discomfort of patients in medical magnetic resonance imaging. This is because in steps 1, 3, 4, and 5, the present invention proposes a rapid acquisition, denoising, reconstruction, and synthesis technology that is different from traditional magnetic resonance imaging, and it only takes 2.5 minutes to obtain a 1×1×1 mm whole brain image. 3 The high-resolution MRI quantitative images can be synthesized into multi-contrast parameter imaging required for clinical use. In contrast, the traditional MRI acquisition method takes 22 minutes to obtain a 1×1×5 mm 3 Resolution images, such as Figure 5 shown. Figure 5 The results show that the new technology proposed by the present invention improves the imaging speed by 8.8 times and the image resolution by 5 times compared with the traditional magnetic resonance imaging technology.
[0089] (2) Better magnetic resonance image quality can be obtained under the same scanning time; this is because in step 3, the present invention proposes a magnetic resonance data fast denoising algorithm based on artificial intelligence neural network, which can realize adaptive fast denoising according to the noise level of magnetic resonance data, thereby obtaining magnetic resonance data with higher fidelity and higher signal-to-noise ratio than traditional magnetic resonance imaging;
[0090] (3) The present invention is insensitive to patient motion artifacts. During the magnetic resonance imaging process, the patient's movement will cause motion artifacts in the magnetic resonance image, seriously affecting the image quality. In step 1 and step 4, the acquisition and reconstruction technology proposed by the present invention includes a motion correction algorithm for the magnetic resonance image, which effectively corrects the artifacts caused by the patient's movement. Specifically, in step 1 of the present invention, multiple motion navigation echoes (motion navigator) are added to the pulse sequence of data acquisition to track the patient's movement in real time. This motion navigation echo is excited by a set of small flip angle radio frequency pulses, and is collected by a low-resolution three-dimensional variable density spiral trajectory and inserted into the pulse sequence of step 1. The acquisition of a single motion navigation echo only takes 0.4 seconds. By inserting multiple motion navigation echoes in step 1, the patient's movement can be tracked in real time, and the patient's motion parameters between each two motion navigation echoes can be estimated. For example, the patient's rotation angle (rotation_x, rotation_y, rotation_z) and translation distance (translation_x, translation_y, translation_z) in three directions in space during the time between two motion navigation echoes. Based on these parameters, in step 4, the acquired magnetic resonance data are first calibrated in three dimensions by rotation and translation according to the estimated motion parameters, and then subsequent subspace reconstruction, time back projection and multi-parameter quantitative imaging are performed to correct artifacts caused by patient motion.
[0091] (4) The scalable system architecture of the present invention can build a customized cloud-based medical imaging experimental platform for users, providing the most cutting-edge medical imaging technology. Traditional magnetic resonance fingerprinting technology (Magnetic Resonance Fingerprinting) is also a new method for quickly and simultaneously obtaining multi-parameter quantitative imaging. However, since magnetic resonance fingerprinting technology only excites one two-dimensional layer at each excitation and only collects one two-dimensional spiral line, the signal collected by the magnetic resonance imager each time is highly undersampled, and the magnetic resonance quantitative image reconstructed by it is full of irrelevant undersampling artifacts. This noise-like undersampling artifact will affect the accuracy of dictionary matching. Therefore, magnetic resonance fingerprinting technology needs to ensure a sufficiently large number of excitations (generally the number of excitations needs to be greater than 1000) to ensure accurate multi-parameter quantitative results, which will increase the total acquisition time. Although in recent years there have been many technologies based on parallel imaging and artificial intelligence to reduce acquisition time, it currently still takes 12 minutes to obtain a 1×1×1mm whole brain image. 3 In contrast, the method proposed in this paper only takes 2.5 minutes to obtain a quantitative image of the whole brain with a resolution of 1×1×1 mm. 3 The high-resolution magnetic resonance quantitative image has an imaging speed 4.8 times higher than that of traditional magnetic resonance fingerprint imaging technology. In addition, traditional magnetic resonance fingerprint imaging technology is only a single acquisition technology and has not been integrated into a complete full-process computing platform, so it is currently difficult to achieve clinical application. The platform of the present invention provides an extensible user data transmission method and computing resources. Hospitals can use the provided platform to achieve standardization of magnetic resonance data acquisition, data transmission, and image reconstruction processes, thereby achieving standardized management of magnetic resonance scanning platforms. And because the intelligent cloud computing platform of the present invention can be deployed in multiple imaging centers, it can reduce the errors in scanning results of magnetic resonance images caused by different hospitals and different magnetic resonance imaging devices. The new magnetic resonance technology can be quickly deployed to clinical hospitals for pre-clinical trials through the unified interface and format of this platform, which effectively promotes the integration of industry, academia, and research, and has broad application prospects.
[0092] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A multimodal magnetic resonance imaging method, characterized in that: include: Perform multimodal quantitative magnetic resonance data acquisition based on biophysical parameters in medical magnetic resonance imaging devices; The collected magnetic resonance data is transmitted in real time to a local or cloud-based server; Performing adaptive intelligent denoising based on a neural network on the transmitted magnetic resonance data on the server to obtain denoised magnetic resonance data; Perform image reconstruction based on the magnetic resonance biophysical model on the denoised magnetic resonance data to obtain multiple quantitative parameter images; A variety of contrast-weighted images required clinically are synthesized based on a variety of quantitative parameter images.
2. The multimodal magnetic resonance imaging method according to claim 1, characterized in that: The multimodal quantitative magnetic resonance data acquisition based on biophysical parameters in the medical magnetic resonance imaging apparatus specifically includes: Different biophysical parameters are used in each excitation of the pulse sequence, and the acquisition method is three-dimensional variable density spiral trajectory acquisition; the pulse sequence includes inversion recovery pulse sequence, multiple inversion recovery pulse sequence, liquid attenuation inversion recovery pulse sequence, gradient echo pulse sequence, steady-state free precession pulse sequence, spin echo pulse sequence and planar echo imaging pulse sequence; different biophysical parameters include different repetition time, echo time and radio frequency pulse flip angle.
3. The multimodal magnetic resonance imaging method according to claim 1, characterized in that: The real-time transmission of the acquired magnetic resonance data to a local or cloud-based server specifically includes: The collected magnetic resonance data is converted into a new format and then transmitted in real time to a local or cloud-based server via the TCP / IP protocol.
4. The multimodal magnetic resonance imaging method according to claim 1, characterized in that: The step of performing adaptive intelligent denoising based on a neural network on the transmitted magnetic resonance data on the server to obtain denoised magnetic resonance data specifically includes: Automatically identify the noise level of magnetic resonance data, including noise intensity and distribution, through pre-trained neural networks; De-noising algorithms of different strengths are used for magnetic resonance data with different noise levels.
5. The multimodal magnetic resonance imaging method according to claim 1, characterized in that: The denoised magnetic resonance data is subjected to image reconstruction based on a magnetic resonance biophysical model to obtain multiple quantitative parameter images, specifically including: generating a magnetic resonance signal dictionary based on the biophysical parameters and the magnetic resonance biophysical model; The MRI signal dictionary is projected into the subspace, combined with the denoised MRI data, and iteratively reconstructed to obtain the subspace coefficient spectrum; Back-project the obtained subspace coefficient spectrum back to the time domain; The image back-projected back into the time domain is identified point by point with the magnetic resonance signal dictionary to obtain multiple quantitative parameter images.
6. The multimodal magnetic resonance imaging method according to claim 5, characterized in that: The multiple quantitative parameter images include longitudinal relaxation time T1 quantitative image, transverse relaxation time T2 quantitative image, proton density quantitative image, quantitative magnetic susceptibility image, myelin water fraction image, apparent diffusion coefficient image, anisotropy fraction image, main magnetic field quantitative image and radio frequency field quantitative image.
7. The multimodal magnetic resonance imaging method according to claim 1, characterized in that: The method of synthesizing a plurality of contrast-weighted images required clinically according to a plurality of quantitative parameter images specifically includes: Various contrast-weighted images of T1-weighted, T2-weighted and diffusion-weighted were synthesized based on various quantitative parameter images and magnetic resonance biophysical models.
8. A multimodal magnetic resonance imaging intelligent cloud computing platform, characterized in that: include: A medical magnetic resonance imager and a server located locally or in the cloud, wherein the server is communicatively connected to the medical magnetic resonance imager to implement the multimodal magnetic resonance imaging method according to any one of claims 1 to 7.
Citation Information
Cited By
Quantitative magnetic resonance imaging mode conversion method based on artificial intelligence
CN120107396A
Modal Conversion Method for Artificial Intelligence-Based Quantitative Magnetic Resonance Imaging
CN120107396B
