R2* image synthesis method based on generative adversarial network

Abstract

The invention provides an R2 * image synthesis method based on a generative adversarial network. The method comprises the following steps: S1, data preprocessing: calculating and correcting an R2 * value from an MEGRE sequence; s2, ROI segmentation: after matching the R2 * graph with an AALv3 template, extracting an average R2 * value of a region of interest from the composite graph and the real graph; s3, a GAN model: inputting a T1 weighted image and a T2 weighted image by a generator, and generating a corresponding R2 * image; the discriminator distinguishes the synthesized R2 * image generated by the generator from a real R2 * image obtained from a real MEGRE sequence; s4, performing quantitative evaluation on the generated image: evaluating the similarity between the synthetic image and the real R2 * graph by adopting a normalized mean square error, a peak signal-to-noise ratio and a structural similarity index; and S5, statistical analysis: analyzing and comparing the R2 * value of the region of interest in the composite graph and the real R2 * graph. According to the invention, powerful auxiliary diagnosis is provided for neurodegenerative diseases similar to Parkinson's disease from the perspective of medical image synthesis and processing.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image synthesis, and in particular to an R2* image synthesis method based on a generative adversarial network. Background Art

[0002] Parkinson's disease (PD) is a chronic neurodegenerative disease characterized by the progressive loss of dopaminergic neurons in the substantia nigra pars compacta (SNpc). Both postmortem histological analysis and in vivo magnetic resonance imaging studies have demonstrated that PD patients have elevated iron levels in the nigrostriatal dopaminergic system, especially in the substantia nigra.

[0003] Since this iron deposition leads to local magnetic field inhomogeneities, it affects spin-spin interactions and ultimately accelerates the transverse relaxation of magnetic resonance imaging (MRI) signals. Therefore, MRI can indirectly estimate iron content by observing changes in the T2* relaxation time or its reciprocal, the transverse relaxation rate constant (R2*=1 / T2*).

[0004] Traditional methods of acquiring R2* images, including acquisition and image reconstruction using a multi-echo gradient recall (MEGRE) sequence, are time-consuming and computationally intensive. In addition, the MEGRE sequence is sensitive to both susceptibility and motion artifacts. Macroscopic B0 magnetic field inhomogeneities can adversely affect the quantitative measurement of relaxation parameters in the MEGRE sequence. During the acquisition of the MEGRE signal, the effects of macroscopic magnetic field inhomogeneities on unconscious body movements and subtle anatomical fluctuations may also introduce motion artifacts into these quantitative maps. For this reason, despite their important clinical value, R2* images cannot be incorporated into routine standard clinical protocols. Summary of the invention

[0005] The technical problem to be solved by the present invention is how to quickly and accurately generate an R2* image from a conventional nuclear magnetic resonance sequence.

[0006] In order to solve the above technical problems, the present invention provides an R2* image synthesis method based on a generative adversarial network, the method comprising the following steps:

[0007] S1. Data preprocessing: Calculate and correct the R2* value from the MEGRE sequence using a single exponential decay model;

[0008] S2, ROI segmentation: registering the first echo image generated by the MEGRE sequence to the AALv3 template; matching the R2* image with the AALv3 template to obtain the region of interest;

[0009] S3, GAN model: design and generate a GAN model, the GAN model includes a generator and a discriminator, the generator inputs a T1-weighted image and a T2-weighted image to generate a corresponding R2* image; the discriminator distinguishes between the synthetic R2* image generated by the generator and the real R2* image obtained by the real MEGRE sequence;

[0010] S4. Quantitative evaluation of generated images: Normalized mean square error, peak signal-to-noise ratio and structural similarity index are used to evaluate the similarity between the synthesized images and the real R2* images;

[0011] S5. Statistical analysis: extract the average R2* value of the region of interest from the synthetic image and the real R2* image; use the Kolmogorov-Smirnov test to evaluate the normality of the data; and calculate the Pearson correlation coefficient and Bland-Altman plot to compare the R2* values ​​of the region of interest in the synthetic image and the real R2* image.

[0012] Furthermore, the R2* value is calculated from the MEGRE sequence using a single exponential decay model, which satisfies the relationship:

[0013] S(t)=S0.exp(-R2 * ·TE+i·2π·Δf·TE)·F(TE)

[0014] where Δf is the frequency shift and depends on the tissue structure as well as the macroscopic magnetic field generated primarily by the tissue / air interface; F(TE) is the voxel spread function; TE is the echo time; and i is the phase portion of the signal.

[0015] Furthermore, the S1 also includes using a FLIRT tool to register the T1-weighted image and the T2-weighted image to a model image of the first echo corresponding to the MEGRE sequence, so that different sequences have spatial consistency.

[0016] Furthermore, the S1 also includes obtaining data, wherein:

[0017] List the database data as external data to prove the generalization of the GAN model;

[0018] The collected data are listed as internal data, and the internal data are divided into a training set and a test set; wherein 80% of the internal data are used for the training set, and 20% of the internal data are used for the test set.

[0019] Furthermore, the regions of interest in S2 include: substantia nigra reticulata, substantia nigra compacta, putamen, globus pallidus and caudate nucleus.

[0020] Furthermore, in S3, the R2* image is synthesized using a pix2pix image-to-image conversion framework, and the network architecture of the generator and the discriminator is the same as the network architecture of the pix2pix image-to-image conversion framework.

[0021] Furthermore, the loss function of the generator is a combination of adversarial loss, pixel-level loss and perceptual loss:

[0022] The adversarial loss satisfies the following relationship:

[0023] L cGAN (G,D)=E c,x [logD(c,x)]+E c [log(1-D(c,G(c)))]

[0024] Wherein, c represents the original image; x represents the R2* image generated by the MEGRE sequence; G represents the generator; D represents the discriminator; G(c) represents the synthetic data;

[0025] The pixel-level loss is a combination of an L1-weighted reconstruction loss and a multi-scale structural similarity index to represent the difference between the generated image and the desired target image; wherein the multi-scale structural similarity index is defined as:

[0026] L MS-SSIM (x,G(c))=1-MS-SSIM(x,G(c))

[0027] Among them, MS-SSIM(x,G(c)) is an indicator to measure structural similarity at multiple scales, which can incorporate image details at different resolutions;

[0028] Then the pixel-level loss is defined as:

[0029] L con (x,G(c))=ω·L MS-SSIM (x,G(c))+(1-ω)·L L1 (x,G(c))

[0030] Where, ω = 0.16;

[0031] The perceptual loss is increased using a pre-trained VGG-16 model, and the perceptual loss is defined as:

[0032]

[0033] Wherein, N represents the number of convolutional layers of the VGG-16 model used to calculate the perceptual loss; represents the feature representation of the i-th convolutional layer; M iIndicates the number of channels of the i-th feature layer.

[0034] Furthermore, the loss function of the generator satisfies the following relationship:

[0035]

[0036] Wherein, G represents the generator; D represents the discriminator; L cGAN represents the adversarial loss; L per represents the perceived loss; L con Indicates pixel-level loss.

[0037] Furthermore, the training details of the GAN model in S3 include:

[0038] The weights are updated using the minibatch stochastic gradient descent method and the Adam optimizer; wherein the batchsize value of the minibatch stochastic gradient descent method is 16; the learning rate of the Adam optimizer is 0.0002, and the momentum parameters b1=0.5 and b2=0.999;

[0039] Data augmentation techniques are used to address the data imbalance between the PD patient group and the healthy control group; wherein the data augmentation techniques include random cropping and horizontal flipping.

[0040] Furthermore, in S4:

[0041] The normalized mean square error is used to evaluate the difference between the generated image and the original image at the pixel level;

[0042] The peak signal-to-noise ratio is defined as the ratio between the maximum possible power of the signal and the noise level in the image;

[0043] The structural similarity index measures the similarity of two images from three aspects: brightness, contrast and structure;

[0044] in:

[0045] The higher the peak signal-to-noise ratio and the structural similarity index are, the better the reconstruction effect is; the lower the value of the normalized mean square error is, the higher the reconstruction fidelity is.

[0046] Furthermore, the S5 further includes:

[0047] The two-sample t-test was used to compare the difference in R2* values ​​between the PD patient group and the healthy control group;

[0048] The R2* values ​​of the synthetic and real R2* images of SNpc were used to perform area under the receiver operating characteristic curve analysis to distinguish the PD patient group from the healthy control group;

[0049] In the external data, AUC analysis was performed using the synthesized R2 plot to distinguish the PD patient group from the healthy control group.

[0050] Furthermore, the generator input also includes an arterial spin labeling image and / or a diffusion weighted image to generate a more accurate R2* image.

[0051] Compared with the prior art, the present invention has the following beneficial effects:

[0052] The present invention is based on an artificial intelligence model of a generative adversarial network, and generates a synthetic transverse relaxation rate constant image, i.e., a synthetic R2* image, from T1-weighted images and T2-weighted images, and the generated map has a good correlation with the real R2* map. It provides a powerful auxiliary diagnosis for neurodegenerative lesions similar to Parkinson's disease from the perspective of medical image synthesis and processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of synthesizing an R2* image from a T1-weighted image and a T2-weighted image disclosed in the present invention;

[0054] Figure 2 This is an image of an embodiment disclosed in the present invention. DETAILED DESCRIPTION

[0055] In order to make the technical scheme and technical effect of the present invention clearer, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments.

[0056] Glossary:

[0057] T1-weighted images (T1W) and T2-weighted images (T2W) are two basic sequences in magnetic resonance imaging (MRI).

[0058] in:

[0059] T1-weighted images are mainly used to show normal brain tissue structure. On T1-weighted images, long T1 signals are usually darker changes, while fat or hemorrhage will show high signals at certain times. This sequence is particularly effective for observing anatomical structures, because short T1 substances such as fat and protein appear as high signals (i.e. brighter) on the image, while water and other long T1 substances appear as low signals (i.e. darker).

[0060] T2-weighted images are mainly used to show some lesions, especially those that show high signals on T2-weighted images, such as cerebral infarction, demyelination of white matter, multiple sclerosis, etc. On T2-weighted images, the cerebrospinal fluid in the ventricles and sulci shows a relatively high signal, that is, a relatively whitish signal. This sequence is very sensitive to detecting tissues or lesions with high water content.

[0061] The present invention aims to provide an R2* image synthesis method based on a generative adversarial network, which can generate a synthetic transverse relaxation rate constant image, namely a synthetic R2* image, from a T1-weighted image and a T2-weighted image, thereby providing a new idea for the auxiliary diagnosis of neurodegenerative diseases such as Parkinson's disease (PD).

[0062] The method provided by the present invention mainly comprises the following steps:

[0063] 1. Data Collection

[0064] 1.1. Research sample collection

[0065] PD patients and age-matched healthy subjects who underwent MRI examinations in the hospital were screened, and the inclusion criteria of the PD patient group and the healthy control group were restricted.

[0066] 1.2 MRI imaging parameters

[0067] In-house MRI studies were performed on a 3T scanner (Discovery 750w, GE Medical Systems, USA) using a standard 24-channel head coil.

[0068] T1-weighted images (T1W) were acquired by 3D T1 BRAVO with the following parameters: matrix size = 256 × 256; field of view (FOV) = 25.6 cm; repetition time (TR) = 8.5 ms; inversion time (TI) = 450 ms; slice thickness = 1.0 mm; echo time (TE) = 3.5 ms; flip angle = 12°.

[0069] The parameters of T2-weighted images (T2W) were: matrix size = 512×512; field of view = 24 cm; repetition time = 5869 ms; slice thickness = 5 mm; echo time = 116.2 ms; flip angle = 142°.

[0070] The MEGRE parameters are: matrix size = 256×256; field of view = 25.6 cm; repetition time = 31 ms; slice thickness = 1.0 mm; number of echoes = 12; first echo time (TE_1) = 1.9 ms; echo interval (ΔTE) = 2.3 ms; flip angle = 12°.

[0071] 1.3 Data Classification

[0072] The database data is listed as external data to prove the generalization of the GAN model; the external data comes from the Parkinson's Progression Markers Initiative (PPMI) database.

[0073] The collected data are listed as internal data, and the internal data are divided into a training set and a test set; 80% of the internal data are used for the training set, and 20% of the internal data are used for the test set.

[0074] All MRI images were center-cropped and resized to 256×256.

[0075] 2. Data Preprocessing

[0076] 2.1. Calculate the R2* value from the MEGRE sequence using the single exponential decay model, which satisfies the relationship:

[0077] S(t)=S0.exp(-R2 * ·TE+i·2π·Δf·TE)·F(TE)

[0078] where Δf is the frequency shift and depends on the tissue structure as well as the macroscopic magnetic field generated primarily by the tissue / air interface; F(TE) is the voxel spread function; TE is the echo time; and i is the phase portion of the signal.

[0079] 2.2. Use the FLIRT tool in the FMRIB Software Library (FSL) to align the T1-weighted image and the T2-weighted image to the model of the first echo corresponding to the MEGRE sequence to make different sequences have spatial consistency.

[0080] 2.3. Use data augmentation technology to solve the data imbalance between the PD patient group and the healthy control group; data augmentation technology includes random cropping and horizontal flipping, which are used to increase the diversity of data when training neural networks.

[0081] Random cropping: This technique increases the diversity of the training dataset by randomly cropping sub-images of different sizes and positions from the original image. This technique can effectively expand the size of the training dataset, prevent the model from overfitting, and improve the generalization ability of the model.

[0082] Horizontal flip: In horizontal flip, the image is flipped along its width, that is, from left to right. This technique can simulate various situations in which images may appear in the real world, thereby helping the model learn more general features.

[0083] 3. ROI Segmentation

[0084] The first echo image generated by the MEGRE sequence was registered to the AALv3 template using an affine registration method with 12 degrees of freedom and the FMRIB Linear Image Registration Tool (FLIRT). The linear transformation was then optimized using a nonlinear warp implemented by the FMRIB Nonlinear Image Registration Tool (FNIRT), which included bias field correction.

[0085] After matching the R2* map with the AALv3 template, the regions of interest were obtained, including the substantia nigra pars reticulata (SNpr), substantia nigra pars compacta (SNpc), putamen (PUT), globus pallidus (GP), and caudate nucleus (CAU).

[0086] IV. GAN Model

[0087] Figure 1 The process of synthesizing R2* images from T1-weighted images and T2-weighted images is shown. The R2* image is synthesized using the pix2pix image-to-image conversion framework. The GAN model includes a generator and a discriminator.

[0088] The generator inputs T1-weighted images and T2-weighted images and generates corresponding R2* images. The task of the generator is to create images that are so realistic that the discriminator cannot distinguish them from real images.

[0089] The discriminator distinguishes between synthetic R2* images generated by the generator and real R2* images obtained from real MEGRE sequences. The task of the discriminator is to evaluate the authenticity of the input image.

[0090] The network architecture of the generator and discriminator is the same as that of the pix2pix image-to-image translation framework. The difference is that the loss function of the generator is a combination of adversarial loss, pixel-level loss, and perceptual loss. Specifically:

[0091] The adversarial loss satisfies the following relationship:

[0092] L cGAN (G,D)=E c,x [logD(c,x)]+E c [log(1-D(c,G(c)))]

[0093] Among them, c represents the original image; x represents the R2* image generated by the MEGRE sequence; G represents the generator; D represents the discriminator; G(c) represents the synthetic data.

[0094] The pixel-level loss is a combination of the L1-weighted reconstruction loss and a multi-scale structural similarity index to represent the difference between the generated image and the desired target image; the multi-scale structural similarity index is defined as:

[0095] L MS-SSIM (x,G(c))=1-MS-SSIM(x,G(c))

[0096] Among them, MS-SSIM(x,G(c)) is an indicator to measure structural similarity at multiple scales, which can incorporate image details at different resolutions;

[0097] Then the pixel-level loss is defined as:

[0098] L con (x,G(c))=ω·L MS-SSIM (x,G(c))+(1-ω)·L L1 (x,G(c))

[0099] Among them, ω=0.16.

[0100] The pre-trained VGG-16 model is used to increase the perceptual loss, which is defined as:

[0101]

[0102] Where N represents the number of convolutional layers of the VGG-16 model used to calculate the perceptual loss; represents the feature representation of the i-th convolutional layer; M i Indicates the number of channels of the i-th feature layer.

[0103] In summary, the loss function of the generator satisfies the following relationship:

[0104]

[0105] Among them, G represents the generator; D represents the discriminator; L cGAN represents adversarial loss; L per Indicates perceived loss; L con Indicates pixel-level loss.

[0106] 5. Training Details

[0107] The minibatch stochastic gradient descent method and Adam optimizer are used to update the weights. The batchsize value of the minibatch stochastic gradient descent method is 16. The learning rate of the Adam optimizer is 0.0002, and the momentum parameters b1=0.5 and b2=0.999.

[0108] Data augmentation includes random crops and horizontal (left-to-right) flipping.

[0109] Furthermore, the generative adversarial network model (GAN model) was trained for 70 epochs. The Ubuntu 18.04 (Canonical) operating system and the PyTorch deep learning framework were used, and CUDA 11.3 (NVIDIA Corporation) was used for graphics processing unit acceleration. The model building work was carried out on a workstation equipped with two NVIDIA Pascal architecture GeForce GTX 2080Ti graphics processing units.

[0110] 6. Quantitative Evaluation of Generated Images

[0111] Normalized mean square error (NMSE), peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) are used to evaluate the similarity between the synthetic image and the real R2* map. Specifically:

[0112] The normalized mean square error is used to evaluate the degree of pixel-level difference between the generated image and the original image.

[0113] The peak signal-to-noise ratio is defined as the ratio between the maximum possible power of the signal and the noise level in the image.

[0114] The structural similarity index measures the similarity between two images in terms of brightness, contrast, and structure.

[0115] Among them: Reference Figure 2 For the example images shown, higher peak signal-to-noise ratio and structural similarity index indicate better reconstruction results; lower normalized mean square error values ​​indicate higher reconstruction fidelity.

[0116] 7. Statistical Analysis

[0117] The average R2* value of the region of interest was extracted from the synthetic images and the real R2* images, and the Kolmogorov-Smirnov test was used to assess the normality of the data.

[0118] Pearson correlation coefficients and Bland-Altman plots were calculated to compare the R2* values ​​of the regions of interest in the synthetic images with the true R2* images.

[0119] Two-sample t-test was used to compare the differences in R2* values ​​between the PD patient group and the healthy control group.

[0120] The area under the receiver operating characteristic curve analysis was performed using the R2* values ​​of the synthetic and real R2* images of SNpc to distinguish the PD patient group from the healthy control group.

[0121] In the external data, AUC analysis was performed using the synthesized R2* plot to distinguish the PD patient group from the healthy control group.

[0122] The Spearman correlation test was used to test the correlation between ΔR2* and the percentage change in the Unified Parkinson's Disease Rating Scale Part III (UPDRSIII) deterioration (ΔUPDRS, %). The 95% confidence interval (CI) was calculated using the bootstrap method (bootstrap sample size = 1000). Wherein, ΔR2* is defined as the difference between R2*(t1)–R2*(t0); t0 is the baseline MRI scan; t1 is the longitudinal scan three years later; P < 0.05 was considered statistically significant.

[0123] It should be noted that, in addition to T1-weighted images and T2-weighted images, the present invention can also use other modal medical images (such as proton density-weighted images, diffusion-weighted images, etc.) as input to generate a more accurate R2* image. This multimodal image fusion technology can expand the scope of application of the model and improve the quality and accuracy of the generated images.

[0124] In addition, the method provided by the present invention is also applicable to the R2* image generation network model of other neurodegenerative diseases, such as Alzheimer's disease.

[0125] Although 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 the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A R2* image synthesis method based on a generative adversarial network, characterized in that: The method comprises the following steps: S1. Data preprocessing: Calculate and correct the R2* value from the MEGRE sequence using a single exponential decay model; S2, ROI segmentation: registering the first echo image generated by the MEGRE sequence to the AALv3 template; matching the R2* image with the AALv3 template to obtain the region of interest; S3, GAN model: design and generate a GAN model, the GAN model includes a generator and a discriminator, the generator inputs a T1-weighted image and a T2-weighted image to generate a corresponding R2* image; the discriminator distinguishes between the synthetic R2* image generated by the generator and the real R2* image obtained by the real MEGRE sequence; S4. Quantitative evaluation of generated images: Normalized mean square error, peak signal-to-noise ratio and structural similarity index are used to evaluate the similarity between the synthesized images and the real R2* images; S5. Statistical analysis: extract the average R2* value of the region of interest from the synthetic image and the real R2* image; use the Kolmogorov-Smirnov test to evaluate the normality of the data; and calculate the Pearson correlation coefficient and Bland-Altman plot to compare the R2* values ​​of the region of interest in the synthetic image and the real R2* image.

2. The R2* image synthesis method based on generative adversarial network according to claim 1, characterized in that: The R2* value is calculated from the MEGRE sequence using a single exponential decay model, which satisfies the relationship: S(t)=S0.exp(-R2 * ·TE+i·2π·Δf·TE)·F(TE) where Δf is the frequency shift and depends on the tissue structure as well as the macroscopic magnetic field generated primarily by the tissue / air interface; F(TE) is the voxel spread function; TE is the echo time; and i is the phase portion of the signal.

3. The R2* image synthesis method based on generative adversarial network according to claim 1, characterized in that: The S1 also includes using the FLIRT tool to register the T1-weighted image and the T2-weighted image to the model of the first echo corresponding to the MEGRE sequence, so that different sequences have spatial consistency.

4. The R2* image synthesis method based on generative adversarial network according to claim 1, characterized in that: The S1 also includes obtaining data, wherein: List the database data as external data to prove the generalization of the GAN model; The collected data are listed as internal data, and the internal data are divided into a training set and a test set; wherein 80% of the internal data are used for the training set, and 20% of the internal data are used for the test set.

5. The R2* image synthesis method based on generative adversarial network according to claim 1, characterized in that: The regions of interest in S2 include: substantia nigra pars reticulata, substantia nigra pars compacta, putamen, globus pallidus and caudate nucleus.

6. The R2* image synthesis method based on generative adversarial network according to claim 1, characterized in that: In S3, the R2* image is synthesized using a pix2pix image-to-image conversion framework, and the network architecture of the generator and the discriminator is the same as the network architecture of the pix2pix image-to-image conversion framework.

7. The R2* image synthesis method based on generative adversarial network according to claim 1 or 6, characterized in that: The loss function of the generator is a combination of adversarial loss, pixel-level loss, and perceptual loss: The adversarial loss satisfies the following relationship: L cGAN (G,D)=E c,x [logD(c,x)]+E c [log(1-D(c,G(c)))] Wherein, c represents the original image; x represents the R2* image generated by the MEGRE sequence; G represents the generator; D represents the discriminator; G(c) represents the synthetic data; The pixel-level loss is a combination of an L1-weighted reconstruction loss and a multi-scale structural similarity index to represent the difference between the generated image and the desired target image; wherein the multi-scale structural similarity index is defined as: L MS-SSIM (x,G(c))=1-MS-SSIM(x,G(c)) Among them, MS-SSIM(x, G(c)) is an indicator to measure structural similarity at multiple scales, which can incorporate image details at different resolutions; Then the pixel-level loss is defined as: L con (x,G(c))=ω·L MS-SSIM (x,G(c))+(1-ω)·L L1 (x,G(c)) Where, ω = 0.16; The perceptual loss is increased using a pre-trained VGG-16 model, and the perceptual loss is defined as: Wherein, N represents the number of convolutional layers of the VGG-16 model used to calculate the perceptual loss; represents the feature representation of the i-th convolutional layer; M i Indicates the number of channels of the i-th feature layer.

8. The R2* image synthesis method based on generative adversarial network according to claim 7, characterized in that: The loss function of the generator satisfies the following relationship: Wherein, G represents the generator; D represents the discriminator; L cGAN represents the adversarial loss; L per represents the perceived loss; L con Represents pixel-level loss.

9. The R2* image synthesis method based on generative adversarial network according to claim 4, characterized in that: The training details of the GAN model in S3 include: The weights are updated using the minibatch stochastic gradient descent method and the Adam optimizer; wherein the batchsize value of the minibatch stochastic gradient descent method is 16; the learning rate of the Adam optimizer is 0.0002, and the momentum parameters b1=0.5 and b2=0.999; Data augmentation techniques are used to address the data imbalance between the PD patient group and the healthy control group; wherein the data augmentation techniques include random cropping and horizontal flipping.

10. The R2* image synthesis method based on generative adversarial network according to claim 1, characterized in that: In S4: The normalized mean square error is used to evaluate the difference between the generated image and the original image at the pixel level; The peak signal-to-noise ratio is defined as the ratio between the maximum possible power of the signal and the noise level in the image; The structural similarity index measures the similarity of two images from three aspects: brightness, contrast and structure; in: The higher the peak signal-to-noise ratio and the structural similarity index are, the better the reconstruction effect is; the lower the value of the normalized mean square error is, the higher the reconstruction fidelity is.

11. The R2* image synthesis method based on generative adversarial network according to claim 9, characterized in that: The S5 further includes: The two-sample t-test was used to compare the difference in R2* values ​​between the PD patient group and the healthy control group; The R2* values ​​of the synthetic and real R2* images of SNpc were used to perform area under the receiver operating characteristic curve analysis to distinguish the PD patient group from the healthy control group; In the external data, AUC analysis was performed using the synthesized R2* plot to distinguish the PD patient group from the healthy control group.

12. The R2* image synthesis method based on generative adversarial network according to claim 1, characterized in that: The generator input also includes arterial spin labeling images and / or diffusion weighted images to generate more accurate R2* images.

Images