A method and system for improving low-field magnetic resonance imaging contrast

Abstract

The present invention provides a method and system for improving the contrast of low-field magnetic resonance imaging, which belongs to the field of image processing technology and includes: step S1, collecting multi-contrast low-field magnetic resonance data, the low-field magnetic resonance data including T1-weighted images and T2-weighted images. Step S2, predicting T1 maps, T2 maps and proton density maps based on T1-weighted images and T2-weighted images; step S3, performing contrast enhancement processing on the predicted T1 maps and T2 maps, and the contrast of the proton density map remains unchanged; step S4, synthesizing the proton density map with the contrast-enhanced T1 map and T2 map to obtain a corresponding contrast image. Beneficial effect: The present invention can improve the contrast of different tissues in magnetic resonance images under low field strength, thereby improving the visual perception and diagnostic difficulty of the image, and improving the diagnostic accuracy.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and system for improving the contrast of low-field magnetic resonance imaging. Background Art

[0002] Magnetic resonance imaging (MRI) is a new diagnostic technique based on the principle that atomic nuclei with magnetic moments can undergo energy level transitions under the influence of a magnetic field. MRI utilizes an external high-frequency magnetic field, causing matter in the body to radiate energy into the surrounding environment, generating signals. The imaging process is similar to image reconstruction and CT, but MRI relies neither on external radiation, absorption, and reflection, nor on gamma radiation from radioactive substances within the body. Instead, it uses the interaction between the external magnetic field and the object to generate images. High-energy magnetic fields are harmless to the human body.

[0003] Low-field MRI generally refers to magnetic resonance imaging with a field strength lower than a certain value (such as 0.5T). Compared with high-field MRI, low-field MRI has low examination costs and is easy to popularize. However, under low field strength, the contrast between different tissues (such as gray matter and white matter) will decrease to a certain extent, making it difficult to distinguish, which brings certain difficulties to disease diagnosis and data analysis.

[0004] In the existing technology, high-field MRI images are generated based on low-field MRI images. However, this method requires the acquisition of paired low-field MRI and high-field MRI images, which is costly. Alternatively, low-field MRI images are simulated based on high-field MRI images. However, there are inevitable inconsistencies between the simulated images and the real images, which can easily introduce deviations. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a method and system for improving the contrast of low-field magnetic resonance imaging.

[0006] The technical problem solved by the present invention can be achieved by adopting the following technical solutions:

[0007] A method for improving low-field magnetic resonance imaging contrast, comprising:

[0008] Step S1, collecting multi-contrast low-field magnetic resonance data, wherein the low-field magnetic resonance data includes a T1-weighted image and a T2-weighted image;

[0009] Step S2, predicting and obtaining a T1 map, a T2 map, and a proton density map based on the T1-weighted image and the T2-weighted image;

[0010] Step S3, performing contrast enhancement processing on the predicted T1 map and the T2 map, while keeping the contrast of the proton density map unchanged;

[0011] Step S4: synthesize the proton density map with the contrast-enhanced T1 map and the T2 map to obtain a corresponding contrast image, and display the image.

[0012] Preferably, the step S2 specifically includes:

[0013] The low-field magnetic resonance data is input into a pre-trained neural network model, and the neural network model performs prediction based on the T1-weighted image and the T2-weighted image to obtain the T1 map, the T2 map and the proton density map.

[0014] Preferably, it also includes: providing a neural network framework, training the neural network framework to obtain the neural network model in step S2, wherein the input of the neural network framework is the synthesized contrast image, and the output of the neural network framework is the T1 map, the T2 map and the proton density map.

[0015] Preferably, in step S2, the neural network model inversely obtains the T1 map, the T2 map and the proton density map based on the T1 weighted image and the T2 weighted image.

[0016] Preferably, the neural network framework includes at least any one of a convolutional neural network, a deep self-attention transformation network or a fully connected network.

[0017] Preferably, in step S3, the method of performing contrast enhancement processing includes at least any one of linear change, gamma transformation, histogram equalization algorithm, or limited contrast adaptive histogram equalization algorithm.

[0018] Preferably, in step S4, the proton density map is synthesized with the contrast-enhanced T1 map and the T2 map using the Bloch equation.

[0019] Preferably, in step S4, the weights of the T1 image and the T2 image in the synthesized contrast image are adjusted by adjusting the echo time and / or repetition time.

[0020] The present invention further provides a system for improving the contrast of low-field magnetic resonance imaging, which is used to implement the above-mentioned method for improving the contrast of low-field magnetic resonance imaging, comprising:

[0021] An acquisition unit, configured to acquire multi-contrast low-field magnetic resonance data, wherein the low-field magnetic resonance data includes a T1-weighted image and a T2-weighted image;

[0022] a processing unit, connected to the acquisition unit, configured to predict a T1 map, a T2 map, and a proton density map based on the T1-weighted image and the T2-weighted image;

[0023] a contrast enhancement processing unit, connected to the processing unit, configured to perform contrast enhancement processing on the predicted T1 map and the T2 map, while maintaining the contrast of the proton density map unchanged;

[0024] A synthesis unit is connected to the contrast enhancement processing unit and is used to synthesize the proton density map with the contrast-enhanced T1 map and T2 map to obtain a corresponding contrast image and display it.

[0025] Preferably, the synthesis unit further comprises: an adjustment module for adjusting the echo time and / or repetition time.

[0026] The advantages or beneficial effects of the technical solution of the present invention are:

[0027] The present invention can enhance the contrast of different tissues in magnetic resonance images under low field strength, thereby improving the visual perception of the image and the difficulty of diagnosis, and increasing the accuracy of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a method for improving the contrast of low-field magnetic resonance imaging in a preferred embodiment of the present invention;

[0029] Figure 2 FIG. 4 is a structural block diagram of a system for enhancing the contrast of low-field magnetic resonance imaging in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 any creative efforts shall fall within the scope of protection of the present invention.

[0031] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0032] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0033] Compared to high-field MRI, the relative differences between T1 and T2 of various anatomical structures (such as gray matter and white matter in the brain) decrease at low field strengths. This is the core reason for the reduced contrast of low-field MRI. In addition, in the medical diagnosis process, T1 and T2 are mainly relied upon to determine whether a certain area is normal tissue or abnormal tissue, and not much attention is paid to proton density. Therefore, the embodiments of the present invention need to increase the difference between T1 and T2. Proton density will never change with field strength and will not be changed by any contrast enhancement algorithm.

[0034] In the prior art, if a contrast enhancement algorithm is designed to directly enhance the contrast of T1-weighted images or T2-weighted images, there may be a region of tissue with good contrast due to large changes in proton density, causing the algorithm to mistakenly believe that the contrast of this region does not need to be enhanced. However, in reality, the T1 and T2 contrast reflected in the T1-weighted image or T2-weighted image is insufficient, which poses certain difficulties for disease diagnosis and data analysis.

[0035] In a preferred embodiment of the present invention, based on the above problems existing in the prior art, a method for improving the contrast of low-field magnetic resonance imaging is provided, which belongs to the field of image processing technology. Figure 1 Shown, including:

[0036] Step S1, collecting multi-contrast low-field magnetic resonance data, the low-field magnetic resonance data including T1-weighted images and T2-weighted images;

[0037] Specifically, first, T1-weighted images and T2-weighted images are acquired, wherein the T1-weighted images can be acquired using a 2D spin echo sequence, or an MPRAGE sequence, or a 2D FLASH sequence, or a 3D FLASH sequence; and the T2-weighted images can be acquired using a 2D spin echo sequence or a 3D spin echo sequence.

[0038] Furthermore, the low-field MRI data may also include one or more image data from sequences such as FLAIR and DWI. FLAIR can be acquired using a 2D or 3D inversion recovery sequence, and DWI can be acquired using a 2D or 3D EPI sequence.

[0039] Step S2, predicting and obtaining a T1 map, a T2 map, and a proton density map based on the T1-weighted image and the T2-weighted image;

[0040] Specifically, the signals of the T1-weighted image and the T2-weighted image are functions of the proton density map, and the specific function is determined by the acquisition sequence. For example, if the T1-weighted image and the T2-weighted image in step S1 are acquired using a 2D spin echo sequence, the formula for the spin echo sequence is:

[0041] S=K·[H]·(1-e -TR / TI )·e -TE / T2 ;

[0042] Where K represents the scaling factor;

[0043] H represents the spin proton density;

[0044] TR indicates repetition time;

[0045] TE represents echo time;

[0046] T1 represents the relaxation time of the longitudinal magnetization vector;

[0047] T2 represents the transverse magnetization vector relaxation time;

[0048] S represents the total signal intensity of the spinecho sequence.

[0049] According to the above formula, the proton density map can be theoretically obtained by reverse engineering from T1-weighted images and T2-weighted images. In the embodiment of the present invention, other existing acquisition sequences can also be used to acquire T1-weighted images and T2-weighted images, and the proton density map can be obtained by reverse engineering from T1-weighted images and T2-weighted images.

[0050] Step S3, performing contrast enhancement processing on the predicted T1 image and T2 image, while keeping the contrast of the proton density image unchanged;

[0051] As a preferred embodiment, in step S3, the method of performing contrast enhancement processing includes at least any one of linear change, gamma transformation, histogram equalization algorithm, or contrast limited adaptive histogram equalization algorithm (CLAHE).

[0052] It should be noted that T1 maps, T2 maps, and proton density maps (PD maps) are quantitative estimates of physical parameters, while T1-weighted images and T2-weighted images are non-quantitative measurements. The pixel brightness of their images mainly reflects the distribution of T1 and T2, respectively, and is also mixed with other factors such as proton density.

[0053] Step S4: synthesize the proton density map with the contrast-enhanced T1 map and T2 map to obtain a corresponding contrast image, and display it.

[0054] As a preferred embodiment, in step S4, the proton density map and the contrast-enhanced T1 map and T2 map are synthesized using the Bloch equation.

[0055] As a preferred embodiment, in step S4, the weights of the T1 image and the T2 image in the synthesized contrast image are adjusted by adjusting the echo time and / or the repetition time.

[0056] Specifically, a new weighted image is synthesized and displayed to medical staff to assist them in diagnosis. Preferably, the existing Bloch equation can be used to synthesize the proton density map and the T1 map and T2 map. Further, the synthesized new weighted image can match the image of the low-field magnetic resonance data collected in step S1, that is, the weights of the longitudinal magnetization vector relaxation time T1 and the transverse magnetization vector relaxation time T2 are consistent with those in step S1, which is equivalent to using the synthesized contrast image to replace the low-field magnetic resonance data collected in step S1 to show it to medical staff. Compared with the image collected in step S1, the contrast of the synthesized new weighted image is improved.

[0057] Furthermore, different repetition times TR and echo times TE can be adjusted to adjust the weights of T1 and T2, and other weighted images can be synthesized to assist doctors in diagnosis and improve the accuracy of diagnosis.

[0058] As a preferred embodiment, step S2 specifically includes:

[0059] The low-field magnetic resonance data are input into a pre-trained neural network model, which makes predictions based on T1-weighted images and T2-weighted images to obtain T1 maps, T2 maps, and proton density maps.

[0060] Specifically, since the process of inverting the proton density map from T1-weighted images and T2-weighted images is relatively difficult and generally requires the constraint of prior information, the embodiment of the present invention is based on deep learning technology and predicts the corresponding T1 maps (T1 maps), T2 maps (T2 maps) and proton density maps (PDmaps) through a neural network model, thereby simplifying the calculation process.

[0061] As a preferred embodiment, it also includes: providing a neural network framework, training the neural network framework to obtain the neural network model in step S2, wherein the input of the neural network framework is a synthesized contrast image, and the output of the neural network framework is a T1 map, a T2 map and a proton density map.

[0062] As a preferred embodiment, in step S2, the neural network model inversely obtains the T1 map, T2 map and proton density map based on the T1 weighted image and the T2 weighted image.

[0063] Furthermore, the training process of the neural network model is as follows:

[0064] First, training data is synthesized to create a large number of different T1 maps, T2 maps, and proton density maps in the anatomical model. The anatomical model can be simulated using an existing database, such as the simulated brain database (Brain Web).

[0065] For each image, different anatomical structures, such as brain regions that can be divided into white matter, gray matter, and cerebrospinal fluid, are assigned reasonable values that are consistent with the current field strength and are randomly distributed within a certain range.

[0066] Random stretching, compression, distortion and other deformation processing are performed on T1 images, T2 images and proton density images to simulate anatomical differences between different people;

[0067] synthesizing the T1 map, the T2 map, and the proton density map into a multi-contrast image in the low-field magnetic resonance data in step S1 by using the Bloch equation;

[0068] The synthesized multi-contrast image is used as the input of the neural network framework (connected along the channel direction), and the T1 map, T2 map and proton density map are used as the output of the neural network framework (connected along the channel direction) to train the neural network model.

[0069] As a preferred embodiment, the neural network framework includes at least any one of a convolutional neural network, a deep self-attention transformation network or a fully connected network.

[0070] Specifically, in this embodiment, the neural network framework can use a convolutional neural network, such as the classic ResNet50 architecture, or a transformer network, such as swin-transformer, or a structure based on a fully connected network, such as Automap.

[0071] Furthermore, the neural network model can use the adam or rmsprop optimizer, and use the L1 norm or L2 norm as the loss function of the neural network model, or the weighted sum of the L1 norm and the L2 norm can also be used as the loss function.

[0072] The present invention also provides a system for improving the contrast of low-field magnetic resonance imaging, which is used to implement the method for improving the contrast of low-field magnetic resonance imaging as described above. Figure 2 Shown, including:

[0073] Acquisition unit 1, used for acquiring multi-contrast low-field magnetic resonance data, the low-field magnetic resonance data including T1-weighted images and T2-weighted images;

[0074] The processing unit 2 is connected to the acquisition unit 1 and is used to predict a T1 map, a T2 map and a proton density map based on the T1-weighted image and the T2-weighted image;

[0075] a contrast enhancement processing unit 3, connected to the processing unit 2, for performing contrast enhancement processing on the predicted T1 map and T2 map, while keeping the contrast of the proton density map unchanged;

[0076] The synthesis unit 4 is connected to the contrast enhancement processing unit 3 and is used to synthesize the proton density map with the contrast-enhanced T1 map and T2 map to obtain a corresponding contrast image and display it.

[0077] As a preferred embodiment, the synthesis unit 4 further includes: an adjustment module 41 for adjusting the echo time and / or repetition time.

[0078] The technical solution of the present invention has the following advantages or beneficial effects: the present invention can enhance the contrast of different tissues in magnetic resonance images under low field strength, thereby improving the visual perception and diagnostic difficulty of the image and increasing the diagnostic accuracy.

[0079] The above description is only a preferred embodiment of the present invention and does not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. A method for improving the contrast of low-field magnetic resonance imaging, characterized in that: include: Step S1, collecting multi-contrast low-field magnetic resonance data, wherein the low-field magnetic resonance data includes a T1-weighted image and a T2-weighted image; Step S2, predicting and obtaining a T1 map, a T2 map, and a proton density map based on the T1-weighted image and the T2-weighted image; Step S3, performing contrast enhancement processing on the predicted T1 map and the T2 map, while keeping the contrast of the proton density map unchanged, so as to increase the difference between the T1 map and the T2 map and retain the original proton density map; Step S4: synthesize the proton density map with the contrast-enhanced T1 map and the T2 map to obtain a corresponding contrast image, and display the image.

2. The method for improving low-field magnetic resonance imaging contrast according to claim 1, characterized in that: The step S2 specifically includes: The low-field magnetic resonance data is input into a pre-trained neural network model, and the neural network model performs prediction based on the T1-weighted image and the T2-weighted image to obtain the T1 map, the T2 map and the proton density map.

3. The method for improving low-field magnetic resonance imaging contrast according to claim 2, characterized in that: Also includes: A neural network framework is provided and trained to obtain the neural network model in step S2, wherein the input of the neural network framework is the synthesized contrast image, and the output of the neural network framework is the T1 map, the T2 map, and the proton density map.

4. The method for improving low-field magnetic resonance imaging contrast according to claim 3, characterized in that: In step S2, the neural network model inversely obtains the T1 map, the T2 map, and the proton density map based on the T1 weighted image and the T2 weighted image.

5. The method for improving low-field magnetic resonance imaging contrast according to claim 3, characterized in that: The neural network framework includes at least any one of a convolutional neural network, a deep self-attention transformation network or a fully connected network.

6. The method for improving low-field magnetic resonance imaging contrast according to claim 1, characterized in that: In step S3, the contrast enhancement processing method includes at least one of linear change, gamma transformation, histogram equalization algorithm, or limited contrast adaptive histogram equalization algorithm.

7. The method for improving low-field magnetic resonance imaging contrast according to claim 1, characterized in that: In step S4, the proton density map is synthesized with the contrast-enhanced T1 map and the T2 map using the Bloch equation.

8. The method for improving low-field magnetic resonance imaging contrast according to claim 1, characterized in that: In step S4, the weights of the T1 map and the T2 map in the synthesized contrast image are adjusted by adjusting the echo time and / or repetition time.

9. A system for improving low-field magnetic resonance imaging contrast, for implementing the method for improving low-field magnetic resonance imaging contrast according to any one of claims 1 to 8, characterized in that: include: An acquisition unit, configured to acquire multi-contrast low-field magnetic resonance data, wherein the low-field magnetic resonance data includes a T1-weighted image and a T2-weighted image; a processing unit, connected to the acquisition unit, configured to predict a T1 map, a T2 map, and a proton density map based on the T1-weighted image and the T2-weighted image; a contrast enhancement processing unit, connected to the processing unit, configured to perform contrast enhancement processing on the predicted T1 map and the T2 map, while maintaining the contrast of the proton density map unchanged; A synthesis unit is connected to the contrast enhancement processing unit and is used to synthesize the proton density map with the contrast-enhanced T1 map and T2 map to obtain a corresponding contrast image and display it.

10. The system for enhancing low-field magnetic resonance imaging contrast according to claim 9, characterized in that: The synthesis unit further includes: an adjustment module for adjusting the echo time and / or repetition time.

Images