Image generation method and device, computer device and storage medium

Abstract

The application relates to an image generation method and device, computer equipment, a storage medium and a computer program product. The method comprises the following steps: obtaining a diffusion weighted image of a first signal-to-noise ratio corresponding to a to-be-tested tissue, wherein the diffusion weighted image of the first signal-to-noise ratio is obtained by scanning the to-be-tested tissue; inputting the diffusion weighted image of the first signal-to-noise ratio into an image enhancement model to obtain a diffusion weighted image of a second signal-to-noise ratio corresponding to the to-be-tested tissue; and the image enhancement model is used for enhancing the signal-to-noise ratio of the diffusion weighted image. The signal-to-noise ratio of the diffusion weighted image in a mute mode can be improved by using the method.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to an image generation method and device, a computer device, a storage medium, and a computer program product. BACKGROUND

[0002] Magnetic resonance diffusion weighted imaging can detect the diffusion movement of water molecules in living tissues. By measuring the signal intensity change of the tissue before and after the diffusion sensitive gradient field is applied, the diffusion state (freedom and direction) of water molecules in the tissue is detected, and the detection result can indirectly reflect the microstructure characteristics and changes of the tissue. In the magnetic resonance diffusion weighted imaging technology in the quiet mode, due to the long echo chain of the quiet scanning sequence, the collected signal is small, and the noise is large, so that the signal-to-noise ratio of the reconstructed diffusion weighted image is low.

[0003] In the conventional technology, in order to improve the signal-to-noise ratio of the diffusion weighted image in the quiet mode, the diffusion weighted image is usually obtained by multiple acquisition and averaging.

[0004] However, using the conventional technology, the signal-to-noise ratio is limited, and the improvement effect is not ideal. SUMMARY

[0005] Therefore, it is necessary to provide an image generation method and device, a computer device, a computer readable storage medium, and a computer program product capable of improving the signal-to-noise ratio of the diffusion weighted image in the quiet mode.

[0006] In a first aspect, the present application provides an image generation method. The method comprises:

[0007] obtaining a diffusion weighted image of a first signal-to-noise ratio corresponding to a to-be-detected tissue;

[0008] inputting the diffusion weighted image of the first signal-to-noise ratio into an image enhancement model to obtain a diffusion weighted image of a second signal-to-noise ratio corresponding to the to-be-detected tissue;

[0009] The image enhancement model is used to enhance the signal-to-noise ratio of the diffusion weighted image, and the second signal-to-noise ratio is greater than the first signal-to-noise ratio.

[0010] In one embodiment, the diffusion weighted image of the first signal-to-noise ratio includes a first diffusion weighted image of a first diffusion sensitive coefficient and a second diffusion weighted image of a second diffusion sensitive coefficient, both having the same signal-to-noise ratio; and the diffusion weighted image of the second signal-to-noise ratio includes a third diffusion weighted image of the first diffusion sensitive coefficient and a fourth diffusion weighted image of the second diffusion sensitive coefficient, both having the same signal-to-noise ratio.

[0011] In one embodiment, the training method of the image enhancement model comprises:

[0012] obtain a third signal-to-noise ratio diffusion weighted image sample and a fourth signal-to-noise ratio diffusion weighted image sample, wherein the third signal-to-noise ratio diffusion weighted image sample is reconstructed based on third signal-to-noise ratio K-space data, and the fourth signal-to-noise ratio diffusion weighted image sample is reconstructed based on partial diffusion weighted imaging data in the third signal-to-noise ratio K-space data;

[0013] train an initial image enhancement model by taking the fourth signal-to-noise ratio diffusion weighted image sample as input data, wherein a loss function of the initial image enhancement model is constructed according to a similarity between a predicted diffusion weighted image output by the initial image enhancement model and the third signal-to-noise ratio diffusion weighted image sample.

[0014] In one of the embodiments, the third signal-to-noise ratio diffusion weighted image sample includes a first diffusion weighted image sample of a first diffusion sensitivity coefficient and a second diffusion weighted image sample of a second diffusion sensitivity coefficient with the same signal-to-noise ratio, and the fourth signal-to-noise ratio diffusion weighted image sample includes a third diffusion weighted image sample of the first diffusion sensitivity coefficient and a fourth diffusion weighted image sample of the second diffusion sensitivity coefficient with the same signal-to-noise ratio.

[0015] In one of the embodiments, the diffusion weighted imaging data includes diffusion weighted imaging data acquired under different channel numbers, and for each channel number, corresponding K-space data is generated.

[0016] In one of the embodiments, the diffusion weighted imaging data includes diffusion weighted imaging data samples acquired under different acquisition time lengths, and for each acquisition time length, corresponding K-space data is generated.

[0017] In a second aspect, the present application further provides an image generation device. The device includes:

[0018] an image acquisition module configured to acquire a first signal-to-noise ratio diffusion weighted image corresponding to a to-be-tested tissue;

[0019] an image generation module configured to input the first signal-to-noise ratio diffusion weighted image into an image enhancement model to obtain a second signal-to-noise ratio diffusion weighted image corresponding to the to-be-tested tissue;

[0020] wherein the image enhancement model is configured to enhance the signal-to-noise ratio of the diffusion weighted image.

[0021] In a third aspect, the present application further provides an image generation method. The method includes:

[0022] acquiring a first group of K-space data corresponding to a to-be-tested tissue, wherein the first group of K-space data can be used to reconstruct a first signal-to-noise ratio diffusion weighted image;

[0023] input the first set of K-space data into an image enhancement model to obtain a second set of K-space data corresponding to the to-be-tested tissue, wherein the second set of K-space data can be used to reconstruct a diffusion weighted image with a second signal-to-noise ratio.

[0024] In a fourth aspect, the present application provides a computer device. The computer device comprises a memory and a processor. The memory stores a computer program. The processor implements the following steps when executing the computer program:

[0025] obtaining a diffusion weighted image with a first signal-to-noise ratio corresponding to the to-be-tested tissue;

[0026] inputting the diffusion weighted image with the first signal-to-noise ratio into an image enhancement model to obtain a diffusion weighted image with a second signal-to-noise ratio corresponding to the to-be-tested tissue;

[0027] The image enhancement model is used to enhance the signal-to-noise ratio of the diffusion weighted image.

[0028] In a fifth aspect, the present application provides a computer readable storage medium. The computer readable storage medium stores a computer program. The computer program is executed by a processor to implement the following steps:

[0029] obtaining a diffusion weighted image with a first signal-to-noise ratio corresponding to the to-be-tested tissue;

[0030] inputting the diffusion weighted image with the first signal-to-noise ratio into an image enhancement model to obtain a diffusion weighted image with a second signal-to-noise ratio corresponding to the to-be-tested tissue;

[0031] The image enhancement model is used to enhance the signal-to-noise ratio of the diffusion weighted image.

[0032] In a sixth aspect, the present application provides a computer program product. The computer program product comprises a computer program. The computer program is executed by a processor to implement the following steps:

[0033] obtaining a diffusion weighted image with a first signal-to-noise ratio corresponding to the to-be-tested tissue;

[0034] inputting the diffusion weighted image with the first signal-to-noise ratio into an image enhancement model to obtain a diffusion weighted image with a second signal-to-noise ratio corresponding to the to-be-tested tissue;

[0035] The image enhancement model is used to enhance the signal-to-noise ratio of the diffusion weighted image.

[0036] The aforementioned image generation method, apparatus, computer equipment, storage medium, and computer program product input a low signal-to-noise ratio (SNR) diffusion-weighted image of the tissue under test, acquired in silent mode, into an image enhancement model for SNR enhancement, thereby obtaining a high SNR diffusion-weighted image. This image enhancement model can effectively learn the mapping relationship between the low SNR diffusion-weighted image and the high SNR diffusion-weighted image, thus significantly improving the image SNR. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating an image generation method in one embodiment;

[0038] Figure 2 (a) shows the K-space filling method for gold standard data; (b) shows the K-space filling method for the actual input data of the image enhancement model after gold standard downsampling.

[0039] Figure 3 Here is a diagram of the AIFI network structure in one embodiment;

[0040] Figure 4 This is a schematic diagram illustrating the operation of an image enhancement model in one embodiment;

[0041] Figure 5 This is a structural block diagram of an image generation device in one embodiment;

[0042] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0044] In one embodiment, such as Figure 1 As shown, an image generation method is provided. This embodiment illustrates the method applied to a server. It is understood that this method can also be applied to a terminal, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0045] Step S102: Obtain the diffusion-weighted image of the tissue under test with the first signal-to-noise ratio.

[0046] The first diffusion-weighted image with the first signal-to-noise ratio is obtained by scanning the to-be-detected tissue by using a silent scan sequence.

[0047] Specifically, in the silent acquisition mode formed by applying the silent sequence, the signal-to-noise ratio of the acquired diffusion-weighted image is low, and therefore the image quality is not high. For example, in the K-space filling mode of PROPELLER or ARMS, the silent effect is achieved by modifying the diffusion gradient and the acquisition gradient, reducing the climbing rate and maximum amplitude of the radio frequency pulse, thereby reducing the noise generated when the magnetic field is switched, and the distortion of the magnetic resonance image can be avoided. However, the magnetic resonance signal is weakened at the same time, resulting in a decrease in the signal-to-noise ratio of the image. Therefore, the server obtains the first diffusion-weighted image with the first signal-to-noise ratio corresponding to the to-be-detected tissue, and the first signal-to-noise ratio corresponds to a low signal-to-noise ratio.

[0048] In step S104, the first diffusion-weighted image with the first signal-to-noise ratio is input into the image enhancement model to obtain a second diffusion-weighted image with a second signal-to-noise ratio corresponding to the to-be-detected tissue.

[0049] The image enhancement model is used to enhance the signal-to-noise ratio of the diffusion-weighted image. The second signal-to-noise ratio is greater than the first signal-to-noise ratio.

[0050] Specifically, in order to obtain a diffusion-weighted image with a high signal-to-noise ratio, the server inputs a diffusion-weighted image with a low signal-to-noise ratio into the trained image enhancement model, and outputs a diffusion-weighted image with a high signal-to-noise ratio corresponding to the to-be-detected tissue. It can be understood that the second signal-to-noise ratio corresponds to a high signal-to-noise ratio.

[0051] In the above image generation method, the diffusion-weighted image with a low signal-to-noise ratio corresponding to the to-be-detected tissue acquired in the silent mode is input into the image enhancement model for signal-to-noise ratio enhancement, and a diffusion-weighted image with a high signal-to-noise ratio can be obtained. The image enhancement model can well learn the mapping relationship between the diffusion-weighted image with a low signal-to-noise ratio and the diffusion-weighted image with a high signal-to-noise ratio, which may be, for example, pixel contrast, pixel gray value, etc. This is conducive to greatly improving the image signal-to-noise ratio.

[0052] In one embodiment, the first diffusion-weighted image with the first signal-to-noise ratio includes a first diffusion-weighted image with a first diffusion sensitivity coefficient and a second diffusion-weighted image with a second diffusion sensitivity coefficient. The second diffusion-weighted image with the second signal-to-noise ratio includes a third diffusion-weighted image with the first diffusion sensitivity coefficient and a fourth diffusion-weighted image with the second diffusion sensitivity coefficient. It can be understood that the signal-to-noise ratio corresponding to the first diffusion-weighted image and the second diffusion-weighted image is the first signal-to-noise ratio. The signal-to-noise ratio corresponding to the third diffusion-weighted image and the fourth diffusion-weighted image is the second signal-to-noise ratio.

[0053] Based on this, in one embodiment, a possible implementation of the above-mentioned step S104 "inputting the first signal-to-noise ratio diffusion weighted image into the image enhancement model to obtain the second signal-to-noise ratio diffusion weighted image corresponding to the tissue to be measured" is involved. On the basis of the above-mentioned embodiment, step S104 can be implemented by the following steps:

[0054] Step S1042, inputting the first diffusion weighted image of the first diffusion sensitivity coefficient into the first image enhancement sub-model in the image enhancement model to obtain the third diffusion weighted image of the first diffusion sensitivity coefficient corresponding to the tissue to be measured;

[0055] Step S1044, inputting the second diffusion weighted image of the second diffusion sensitivity coefficient into the second image enhancement sub-model in the image enhancement model to obtain the fourth diffusion weighted image of the second diffusion sensitivity coefficient corresponding to the tissue to be measured.

[0056] Specifically, the server selects the respective image enhancement sub-models corresponding to different diffusion sensitivity coefficients, i.e. b values, to perform signal-to-noise ratio enhancement on the corresponding diffusion weighted images, so that each image enhancement sub-model outputs a respective high signal-to-noise ratio diffusion weighted image, i.e. the third diffusion weighted image and the fourth diffusion weighted image.

[0057] Further, in one embodiment, the method further comprises the following steps:

[0058] According to the signal intensity of the tissue to be measured in the third diffusion weighted image, the signal intensity of the tissue to be measured in the fourth diffusion weighted image, the first diffusion sensitivity coefficient and the second diffusion sensitivity coefficient, a second signal-to-noise ratio apparent diffusion coefficient image corresponding to the tissue to be measured is obtained.

[0059] Specifically, the server calculates a high signal-to-noise ratio apparent diffusion coefficient (ADC) image corresponding to the tissue to be measured according to the formula ADC = ln(SI 低 / SI 高 ) / (b 高 -b 低 ), wherein ln represents the natural logarithm, SI 低 represents the signal intensity of the tissue to be measured in the diffusion weighted image under the low diffusion sensitivity coefficient, SI 高 represents the signal intensity of the tissue to be measured in the diffusion weighted image under the high diffusion sensitivity coefficient, b 高 represents the high diffusion sensitivity coefficient, and b 低 represents the low diffusion sensitivity coefficient.

[0060] In this embodiment, different diffusion-weighted images with different diffusion sensitivity coefficients are obtained by using different image enhancement sub-models, and then high signal-to-noise ratio apparent diffusion coefficient images are generated, which is beneficial to improve the quality of the apparent diffusion coefficient images.

[0061] In one embodiment, a training process related to the image enhancement model is involved. On the basis of the above-mentioned embodiment, the training process can be specifically implemented by the following steps:

[0062] In step S112, K-space data with a third signal-to-noise ratio is obtained.

[0063] In step S114, part of the diffusion-weighted imaging data is extracted from the K-space data with the third signal-to-noise ratio to construct K-space data with a fourth signal-to-noise ratio.

[0064] In step S116, image reconstruction is performed according to the K-space data with the fourth signal-to-noise ratio to obtain a diffusion-weighted image sample with the fourth signal-to-noise ratio.

[0065] In step S118, image reconstruction is performed according to the K-space data with the third signal-to-noise ratio to obtain a diffusion-weighted image sample with the third signal-to-noise ratio.

[0066] In step S120, the diffusion-weighted image sample with the fourth signal-to-noise ratio is taken as input data to train the initial image enhancement model.

[0067] The K-space data is obtained by filling the diffusion-weighted imaging data collected after multiple excitation pulses. The fourth signal-to-noise ratio is less than the third signal-to-noise ratio. It can be understood that the fourth signal-to-noise ratio corresponds to a low signal-to-noise ratio, and the third signal-to-noise ratio corresponds to a high signal-to-noise ratio. The loss function of the initial image enhancement model is constructed according to the similarity between the predicted diffusion-weighted image output by the initial image enhancement model and the diffusion-weighted image sample with the third signal-to-noise ratio.

[0068] Optionally, the K-space data with the third signal-to-noise ratio is collected by using a K-space filling mode of propeller filling technology (PROPELLER). The PROPELLER DWI technology is based on the PROPELLER acquisition technology to realize 2D diffusion application. A diffusion module is added before each shot, and the data of multiple shots is used to fill the K-space center area to form a blade or blade of PROPELLER. Each blade fills the entire k-space radially. The non-deformation high-resolution diffusion application is realized. The K-space data with the third signal-to-noise ratio can be obtained by non-silence acquisition or multiple average acquisition (i.e. increasing the number of scanning averages) in a silent manner, and used as a gold standard. In this way, high signal-to-noise ratio data can be obtained.

[0069] Specifically, in order to enable the pixels in the low signal-to-noise ratio diffusion weighted image sample (i.e., the fourth signal-to-noise ratio diffusion weighted image sample) and the high signal-to-noise ratio diffusion weighted image sample (i.e., the third signal-to-noise ratio diffusion weighted image sample) of the model input to correspond, the server randomly or uniformly extracts part of the diffusion weighted imaging data from the third signal-to-noise ratio K-space data after obtaining the third signal-to-noise ratio K-space data, to construct the fourth signal-to-noise ratio K-space data. And by adjusting the extraction ratio, such as extracting according to 1:4, the network input image is made to be as close as possible to the image obtained under a low number of scans, that is, a low signal-to-noise ratio diffusion weighted image sample obtained under a short scan time, as shown in Figure 2

[0070] Then, the server takes the fourth signal-to-noise ratio diffusion weighted image sample as input data, and trains the initial image enhancement model based on the loss function constructed based on the similarity between the predicted diffusion weighted image and the third signal-to-noise ratio diffusion weighted image sample, to enable the initial image enhancement model to learn the mapping relationship between the low signal-to-noise ratio diffusion weighted image sample and the high signal-to-noise ratio diffusion weighted image sample, so as to identify the noise pattern brought by the PROPELLER filled K-space, and then remove the noise.

[0071] In this embodiment, first, the gold standard data of the image enhancement model (artificial intelligence network) is generated, and then the K-space of the gold standard data is uniformly extracted, noise is added, and the input data of the AI network is generated. After the AI network is trained, the low signal-to-noise ratio diffusion weighted image can be processed to obtain a high signal-to-noise ratio diffusion weighted image. The AI network is suitable for processing whole body magnetic resonance data of each part under the quiet scan mode, and optimizing the signal-to-noise ratio of the magnetic resonance image.

[0072] In the training process of the image enhancement model, the loss function driving the model parameter update needs to consider the similarity problem of the predicted diffusion weighted image and the third signal-to-noise ratio diffusion weighted image sample in the image domain and the frequency domain, to ensure that the network output result is as close as possible to the gold standard. Based on this idea, in an embodiment, in the AIFI, the AI module takes the low signal-to-noise ratio image reconstructed from the undersampled K-space data (for example, the low signal-to-noise ratio image reconstructed after the undersampled part is filled with zeros) as the network input, takes the corresponding high signal-to-noise ratio and / or high resolution data (for example, the parallel reconstruction, compressed sensing reconstruction of the undersampled K-space data; or the full-sampled K-space data reconstruction) as the learning standard, trains the neural network, to suppress the noise in the image, and realizes the high frequency interpolation of the k-space.

[0073] Specifically, the AI module selects the RIDNet network for training, and the network structure is as shown in Figure 3 ​As shown, it can be divided into three parts of feature extraction, feature learning and image reconstruction. Among them, the feature extraction part is composed of a 3x3 convolution layer; the feature learning part is composed of four Enhancement Attention Modules (EAM) modules. The structure of the EAM module is as shown in the dashed box Figure 3 As shown in the dashed box, it is composed of four parts:

[0074] The first part contains two branches composed of dilated convolution layers, with dilated coefficients of 1, 2 and 3, 4 respectively; the second part is a residual module composed of two 3x3 convolution layers; the third part is a strengthened version of the residual module, composed of two 3x3 convolution layers and one 1x1 convolution layer; the fourth part is composed of a global average pooling and two 1x1 convolution layers. The reconstruction part is composed of a 3x3 convolution layer. The entire RIDNet network uses multiple long residual connections, short residual connections and local connections, which aims to speed up the training of the network, prevent gradient disappearance and improve the accuracy of the model.

[0075] Further, in order to be able to constrain the model in the image domain and the frequency domain at the same time, both to alleviate the problem of over-smoothing of the network output image caused by the L1 loss function, and to suppress the abnormal noise and texture that the perceptual loss function may produce, the loss function calculation network of the AI network in the AIFI calculates the pixel-by-pixel difference between the network prediction image and the gold standard image in the image domain and the frequency domain, to ensure that the error between the final output image of the network and the gold standard image is as small as possible, so that the network output result is as close as possible to the gold standard. The loss function used in training includes three parts, as shown in formula (1-1), the L1 norm loss function of the network output image and the gold standard image in the image domain is calculated, as shown in formula (1-2); the perceptual loss function of the network output image and the gold standard image in the image domain is calculated, as shown in formula (1-3); the L1 norm loss function of the network output image and the gold standard image in the frequency domain is calculated, as shown in formula (1-4).

[0076]

[0077] L1 = ||I output -I GT ||1 (1-2)

[0078]

[0079]

[0080] wherein: I output is the network output result, I GT is the corresponding gold standard image, represents the output feature map of the nth layer of the pre-trained network, an amplitude map of the network output image in the frequency domain, an amplitude map of the gold standard in the frequency domain.

[0081] In addition, in the training process, the image enhancement model can also extract high-order features of the image through an existing network (VGG16), construct a loss function, increase constraints, and enhance the similarity between the network output image and the gold standard image in a larger field of view, so that the output image is more realistic.

[0082] In one embodiment, the third diffusion-weighted image sample with the same signal-to-noise ratio includes a first diffusion-weighted image sample with a first diffusion sensitivity coefficient and a second diffusion-weighted image sample with a second diffusion sensitivity coefficient; and the fourth diffusion-weighted image sample with the same signal-to-noise ratio includes a third diffusion-weighted image sample with the first diffusion sensitivity coefficient and a fourth diffusion-weighted image sample with the second diffusion sensitivity coefficient. Based on this, in one embodiment, a possible implementation of the above step S120 “training the initial image enhancement model by taking the fourth diffusion-weighted image sample with the same signal-to-noise ratio as input data” is involved. Based on the above embodiment, step S120 can be implemented by the following steps:

[0083] Step S1202, training a first initial image enhancement sub-model in the initial image enhancement model by taking the third diffusion-weighted image sample with the first diffusion sensitivity coefficient as input data, wherein the loss function of the first initial image enhancement sub-model is constructed according to the similarity between the first predicted diffusion-weighted image output by the first initial image enhancement sub-model and the first diffusion-weighted image sample with the first diffusion sensitivity coefficient;

[0084] Step S1204, training a second initial image enhancement sub-model in the initial image enhancement model by taking the fourth diffusion-weighted image sample with the second diffusion sensitivity coefficient as input data, wherein the loss function of the second initial image enhancement sub-model is constructed according to the similarity between the second predicted diffusion-weighted image output by the second initial image enhancement sub-model and the second diffusion-weighted image sample with the second diffusion sensitivity coefficient.

[0085] Specifically, as shown in Figure 4 For the diffusion-weighted image sample, two diffusion sensitivity coefficients, i.e., the first diffusion sensitivity coefficient and the second diffusion sensitivity coefficient, are set, for example, b0 for the low b value and b1000 for the high b value. Since the diffusion-weighted images under the two b values are different, two initial image enhancement sub-models can be used for training to obtain the diffusion-weighted image with a high signal-to-noise ratio under each b value.

[0086] In this embodiment, different initial image enhancement sub-models are trained respectively, and the noise patterns caused by propeller filling K-space under different diffusion sensitivity coefficients can be recognized, and the corresponding noise can be removed.

[0087] In one embodiment, the diffusion weighted imaging data includes diffusion weighted imaging data acquired under different channel numbers.

[0088] Among them, for each channel number, the corresponding K-space data is generated.

[0089] Specifically, in order to increase the diversity of training samples, the server selects clinically commonly used channel number coils (such as 24 channels, 32 channels, 48 channels, etc.), and acquires diffusion weighted imaging data under different channel numbers as training data.

[0090] In this embodiment, the diffusion weighted imaging data acquired under different channel numbers is selected as training data, which is beneficial to improve the generalization ability of the model, so as to be applicable to various application scenarios.

[0091] In one embodiment, the diffusion weighted imaging data includes diffusion weighted imaging data samples acquired under different acquisition time lengths.

[0092] Among them, for each acquisition time length (imaging time length), the corresponding K-space data is generated.

[0093] Specifically, considering that in an emergency, medical workers may appropriately shorten the magnetic resonance scan time to adapt to the needs of patients, therefore, the time margin problem needs to be considered in the model training process, and the quality of low signal-to-noise ratio diffusion weighted images in a certain time period is improved. For example, the server acquires gold standard data of different time lengths such as 14 minutes and 16 minutes, and obtains model training data of “3.5 minutes” and “4 minutes” after decimation by 4 times (i.e. extracting by 1 / 4 ratio). These training data correspond to real 3.5 minutes and 4 minutes.

[0094] In this embodiment, the diffusion weighted imaging data samples acquired under different acquisition time lengths are selected as training data, which is beneficial to improve the generalization ability of the model, so as to be applicable to various application scenarios.

[0095] Based on the same inventive concept, the embodiments of the present application also provide an image generation method, which comprises the following steps:

[0096] Step S132, acquiring a first group of K-space data corresponding to the to-be-measured tissue, wherein the first group of K-space data can reconstruct a diffusion weighted image with a first signal-to-noise ratio;

[0097] In step S134, the first set of K-space data is input into the image enhancement model to obtain a second set of K-space data corresponding to the to-be-measured tissue, wherein the second set of K-space data can be used to reconstruct a diffusion-weighted image with a second signal-to-noise ratio.

[0098] Specifically, since the diffusion-weighted image can be obtained by reconstructing the K-space data, on the basis of the implementation principle of the above-mentioned embodiments, the initial first set of K-space data can also be directly input into the image enhancement model for enhancement to obtain the second set of K-space data. Subsequently, the second set of K-space data is reconstructed to obtain the diffusion-weighted image with a high signal-to-noise ratio. It can be understood that the present embodiment has the same inventive concept as the above-mentioned embodiments, and only the data at different stages is enhanced. Since there is a close corresponding relationship between the data at different stages, the ultimate purpose is the same, and the image signal-to-noise ratio can be improved.

[0099] It should be understood that, although each step in the flowchart involved in each of the above-mentioned embodiments is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each of the above-mentioned embodiments can include multiple steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0100] Based on the same inventive concept, the present application also provides an image generation device for implementing the above-mentioned image generation method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above-mentioned method, and therefore the specific limitations in one or more image generation device embodiments provided below can refer to the limitations of the image generation method described above, which will not be described here.

[0101] In one embodiment, as shown in Figure 5 An image generation device is provided, comprising:

[0102] The image acquisition module 202 is configured to acquire a diffusion-weighted image with a first signal-to-noise ratio corresponding to the to-be-measured tissue, wherein the diffusion-weighted image with the first signal-to-noise ratio is obtained by scanning the to-be-measured tissue.

[0103] The image generation module 204 is configured to input the diffusion-weighted image with the first signal-to-noise ratio into an image enhancement model to obtain a diffusion-weighted image with a second signal-to-noise ratio corresponding to the to-be-measured tissue.

[0104] The image enhancement model is used for enhancing the signal-to-noise ratio of the diffusion weighted image.

[0105] In one embodiment, the diffusion weighted image with the first signal-to-noise ratio includes a first diffusion weighted image with the same signal-to-noise ratio of a first diffusion sensitivity coefficient and a second diffusion weighted image with the same signal-to-noise ratio of a second diffusion sensitivity coefficient; the diffusion weighted image with the second signal-to-noise ratio includes a third diffusion weighted image with the same signal-to-noise ratio of the first diffusion sensitivity coefficient and a fourth diffusion weighted image with the same signal-to-noise ratio of the second diffusion sensitivity coefficient; the image generation module 204 is specifically configured to input the first diffusion weighted image with the first diffusion sensitivity coefficient into a first image enhancement sub-model in the image enhancement model to obtain the third diffusion weighted image with the first diffusion sensitivity coefficient corresponding to the to-be-detected tissue; input the second diffusion weighted image with the second diffusion sensitivity coefficient into a second image enhancement sub-model in the image enhancement model to obtain the fourth diffusion weighted image with the second diffusion sensitivity coefficient corresponding to the to-be-detected tissue; the device further includes an image determination module configured to obtain an apparent diffusion coefficient image with the second signal-to-noise ratio corresponding to the to-be-detected tissue according to the signal intensity of the to-be-detected tissue in the third diffusion weighted image, the signal intensity of the to-be-detected tissue in the fourth diffusion weighted image, the first diffusion sensitivity coefficient and the second diffusion sensitivity coefficient.

[0106] In one embodiment, the device further includes a model training module configured to obtain K-space data with a third signal-to-noise ratio, wherein the K-space data is obtained by filling diffusion weighted imaging data collected after multiple excitation pulses; extract part of the diffusion weighted imaging data from the K-space data with the third signal-to-noise ratio to construct K-space data with a fourth signal-to-noise ratio, wherein the fourth signal-to-noise ratio is less than the third signal-to-noise ratio; perform image reconstruction according to the K-space data with the fourth signal-to-noise ratio to obtain a diffusion weighted image sample with the fourth signal-to-noise ratio, and perform image reconstruction according to the K-space data with the third signal-to-noise ratio to obtain a diffusion weighted image sample with the third signal-to-noise ratio; train an initial image enhancement model by taking the diffusion weighted image sample with the fourth signal-to-noise ratio as input data, wherein a loss function of the initial image enhancement model is constructed according to a similarity between a predicted diffusion weighted image output by the initial image enhancement model and the diffusion weighted image sample with the third signal-to-noise ratio.

[0107] In one embodiment, the third diffusion-weighted image sample with the same signal-to-noise ratio comprises a first diffusion-weighted image sample with the same signal-to-noise ratio of a first diffusion sensitivity coefficient and a second diffusion-weighted image sample with the same signal-to-noise ratio of a second diffusion sensitivity coefficient; the fourth diffusion-weighted image sample with the same signal-to-noise ratio comprises a third diffusion-weighted image sample with the same signal-to-noise ratio of the first diffusion sensitivity coefficient and a fourth diffusion-weighted image sample with the same signal-to-noise ratio of the second diffusion sensitivity coefficient; the model training module is specifically configured to take the third diffusion-weighted image sample with the first diffusion sensitivity coefficient as input data to train a first initial image enhancement sub-model in the initial image enhancement model, wherein a loss function of the first initial image enhancement sub-model is constructed according to a similarity between a first predicted diffusion-weighted image output by the first initial image enhancement sub-model and the first diffusion-weighted image sample with the first diffusion sensitivity coefficient; and take the fourth diffusion-weighted image sample with the second diffusion sensitivity coefficient as input data to train a second initial image enhancement sub-model in the initial image enhancement model, wherein a loss function of the second initial image enhancement sub-model is constructed according to a similarity between a second predicted diffusion-weighted image output by the second initial image enhancement sub-model and the second diffusion-weighted image sample with the second diffusion sensitivity coefficient.

[0108] The above-mentioned modules in the image generation apparatus can be realized by software, hardware, and combinations thereof, in whole or in part. The above-mentioned modules can be embedded in or independent of a processor in a computer device in hardware form, or stored in a memory in a computer device in software form, so as to be called and executed by a processor to perform operations corresponding to the above-mentioned modules.

[0109] In one embodiment, a computer device, which can be a server, has an internal structure diagram as shown in Figure 6 The computer device includes a processor, a memory, and a network interface connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement an image generation method.

[0110] Those skilled in the art can understand that Figure 6 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. Specifically, the computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0111] In one embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0112] obtaining a first signal-to-noise ratio diffusion weighted image corresponding to the to-be-tested tissue, wherein the first signal-to-noise ratio diffusion weighted image is obtained by scanning the to-be-tested tissue;

[0113] inputting the first signal-to-noise ratio diffusion weighted image into the image enhancement model to obtain a second signal-to-noise ratio diffusion weighted image corresponding to the to-be-tested tissue;

[0114] wherein the image enhancement model is used to enhance the signal-to-noise ratio of the diffusion weighted image.

[0115] In one embodiment, the processor further implements the following steps when executing the computer program:

[0116] inputting the first diffusion weighted image of the first diffusion sensitivity coefficient into a first image enhancement sub-model in the image enhancement model to obtain a third diffusion weighted image of the first diffusion sensitivity coefficient corresponding to the to-be-tested tissue; inputting the second diffusion weighted image of the second diffusion sensitivity coefficient into a second image enhancement sub-model in the image enhancement model to obtain a fourth diffusion weighted image of the second diffusion sensitivity coefficient corresponding to the to-be-tested tissue; and obtaining an apparent diffusion coefficient image of a second signal-to-noise ratio corresponding to the to-be-tested tissue according to the signal intensity of the to-be-tested tissue in the third diffusion weighted image, the signal intensity of the to-be-tested tissue in the fourth diffusion weighted image, the first diffusion sensitivity coefficient, and the second diffusion sensitivity coefficient.

[0117] In one embodiment, the processor further implements the following steps when executing the computer program:

[0118] obtaining K-space data of a third signal-to-noise ratio, wherein the K-space data is obtained by filling diffusion weighted imaging data collected after multiple excitation pulses; extracting part of the diffusion weighted imaging data from the K-space data of the third signal-to-noise ratio to construct K-space data of a fourth signal-to-noise ratio, wherein the fourth signal-to-noise ratio is less than the third signal-to-noise ratio; performing image reconstruction according to the K-space data of the fourth signal-to-noise ratio to obtain a diffusion weighted image sample of the fourth signal-to-noise ratio, and performing image reconstruction according to the K-space data of the third signal-to-noise ratio to obtain a diffusion weighted image sample of the third signal-to-noise ratio; and training an initial image enhancement model by taking the diffusion weighted image sample of the fourth signal-to-noise ratio as input data, wherein a loss function of the initial image enhancement model is constructed according to a similarity between a predicted diffusion weighted image output by the initial image enhancement model and the diffusion weighted image sample of the third signal-to-noise ratio.

[0119] In one embodiment, the processor further implements the following steps when executing the computer program:

[0120] The third diffusion weighted image sample of the first diffusion sensitivity coefficient is input as input data to train a first initial image enhancement sub-model in the initial image enhancement model, wherein a loss function of the first initial image enhancement sub-model is constructed according to similarity between a first predicted diffusion weighted image output by the first initial image enhancement sub-model and the first diffusion weighted image sample of the first diffusion sensitivity coefficient; the fourth diffusion weighted image sample of the second diffusion sensitivity coefficient is input as input data to train a second initial image enhancement sub-model in the initial image enhancement model, wherein a loss function of the second initial image enhancement sub-model is constructed according to similarity between a second predicted diffusion weighted image output by the second initial image enhancement sub-model and the second diffusion weighted image sample of the second diffusion sensitivity coefficient.

[0121] In one embodiment, a computer readable storage medium is provided, and the computer readable storage medium has stored thereon a computer program, and the computer program is executed by a processor to implement the following steps:

[0122] The first signal-to-noise ratio diffusion weighted image corresponding to the to-be-tested tissue is obtained, wherein the first signal-to-noise ratio diffusion weighted image is obtained by scanning the to-be-tested tissue;

[0123] The first signal-to-noise ratio diffusion weighted image is input into the image enhancement model to obtain a second signal-to-noise ratio diffusion weighted image corresponding to the to-be-tested tissue;

[0124] The image enhancement model is used to enhance the signal-to-noise ratio of the diffusion weighted image.

[0125] In one embodiment, the computer program is executed by the processor to further implement the following steps:

[0126] The first diffusion weighted image of the first diffusion sensitivity coefficient is input into a first image enhancement sub-model in the image enhancement model to obtain a third diffusion weighted image of the first diffusion sensitivity coefficient corresponding to the to-be-tested tissue; the second diffusion weighted image of the second diffusion sensitivity coefficient is input into a second image enhancement sub-model in the image enhancement model to obtain a fourth diffusion weighted image of the second diffusion sensitivity coefficient corresponding to the to-be-tested tissue; and the second signal-to-noise ratio apparent diffusion coefficient image corresponding to the to-be-tested tissue is obtained according to signal intensity of the to-be-tested tissue in the third diffusion weighted image, signal intensity of the to-be-tested tissue in the fourth diffusion weighted image, the first diffusion sensitivity coefficient and the second diffusion sensitivity coefficient.

[0127] In one embodiment, the computer program is executed by the processor to further implement the following steps:

[0128] The process involves acquiring K-space data for a third signal-to-noise ratio (SNR), which is obtained by filling in diffusion-weighted imaging data acquired after multiple excitation pulses. A portion of the diffusion-weighted imaging data for the third SNR is extracted from the K-space data to construct K-space data for a fourth SNR, where the fourth SNR is less than the third SNR. Image reconstruction is performed based on the K-space data for the fourth SNR to obtain diffusion-weighted image samples for the fourth SNR. Similarly, image reconstruction is performed based on the K-space data for the third SNR to obtain diffusion-weighted image samples for the third SNR. The diffusion-weighted image samples for the fourth SNR are used as input data to train an initial image enhancement model. The loss function of the initial image enhancement model is constructed based on the similarity between the predicted diffusion-weighted image output by the initial image enhancement model and the diffusion-weighted image samples for the third SNR.

[0129] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0130] Using the third diffusion-weighted image sample of the first diffusion sensitivity coefficient as input data, the first initial image enhancement sub-model in the initial image enhancement model is trained. The loss function of the first initial image enhancement sub-model is constructed based on the similarity between the first predicted diffusion-weighted image output by the first initial image enhancement sub-model and the first diffusion-weighted image sample of the first diffusion sensitivity coefficient. Using the fourth diffusion-weighted image sample of the second diffusion sensitivity coefficient as input data, the second initial image enhancement sub-model in the initial image enhancement model is trained. The loss function of the second initial image enhancement sub-model is constructed based on the similarity between the second predicted diffusion-weighted image output by the second initial image enhancement sub-model and the second diffusion-weighted image sample of the second diffusion sensitivity coefficient.

[0131] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0132] Acquire a diffusion-weighted image of the tissue under test with a first signal-to-noise ratio, wherein the diffusion-weighted image of the tissue under test is obtained by scanning the tissue under test;

[0133] The diffusion-weighted image with the first signal-to-noise ratio is input into the image enhancement model to obtain the diffusion-weighted image with the second signal-to-noise ratio corresponding to the tissue under test.

[0134] Among them, the image enhancement model is used to enhance the signal-to-noise ratio of the diffusion-weighted image.

[0135] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0136] inputting the first diffusion weighted image of the first diffusion sensitivity coefficient into a first image enhancement sub-model in the image enhancement model to obtain a third diffusion weighted image of the first diffusion sensitivity coefficient corresponding to the to-be-detected tissue; inputting the second diffusion weighted image of the second diffusion sensitivity coefficient into a second image enhancement sub-model in the image enhancement model to obtain a fourth diffusion weighted image of the second diffusion sensitivity coefficient corresponding to the to-be-detected tissue; and obtaining an apparent diffusion coefficient image of a second signal-to-noise ratio corresponding to the to-be-detected tissue according to a signal intensity of the to-be-detected tissue in the third diffusion weighted image, a signal intensity of the to-be-detected tissue in the fourth diffusion weighted image, the first diffusion sensitivity coefficient, and the second diffusion sensitivity coefficient.

[0137] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0138] obtaining K-space data of a third signal-to-noise ratio, wherein the K-space data is obtained by filling diffusion weighted imaging data collected after multiple excitation pulses; extracting part of the diffusion weighted imaging data from the K-space data of the third signal-to-noise ratio to construct K-space data of a fourth signal-to-noise ratio, wherein the fourth signal-to-noise ratio is less than the third signal-to-noise ratio; performing image reconstruction according to the K-space data of the fourth signal-to-noise ratio to obtain a diffusion weighted image sample of the fourth signal-to-noise ratio, and performing image reconstruction according to the K-space data of the third signal-to-noise ratio to obtain a diffusion weighted image sample of the third signal-to-noise ratio; and taking the diffusion weighted image sample of the fourth signal-to-noise ratio as input data to train an initial image enhancement model, wherein a loss function of the initial image enhancement model is constructed according to a similarity between a predicted diffusion weighted image output by the initial image enhancement model and the diffusion weighted image sample of the third signal-to-noise ratio.

[0139] In one embodiment, the computer program, when executed by the processor, further implements the following steps:

[0140] taking the third diffusion weighted image sample of the first diffusion sensitivity coefficient as input data to train a first initial image enhancement sub-model in the initial image enhancement model, wherein a loss function of the first initial image enhancement sub-model is constructed according to a similarity between a first predicted diffusion weighted image output by the first initial image enhancement sub-model and the first diffusion weighted image sample of the first diffusion sensitivity coefficient; and taking the fourth diffusion weighted image sample of the second diffusion sensitivity coefficient as input data to train a second initial image enhancement sub-model in the initial image enhancement model, wherein a loss function of the second initial image enhancement sub-model is constructed according to a similarity between a second predicted diffusion weighted image output by the second initial image enhancement sub-model and the second diffusion weighted image sample of the second diffusion sensitivity coefficient.

[0141] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0142] It can be understood by those skilled in the art that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing related hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments. Any reference to memory, database or other medium used in the embodiments provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0143] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0144] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. An image generation method, characterized in that, The method includes: Obtain the diffusion-weighted image of the tissue under test with the first signal-to-noise ratio; The diffusion-weighted image with the first signal-to-noise ratio is input into the image enhancement model to obtain the diffusion-weighted image with the second signal-to-noise ratio corresponding to the tissue under test. The image enhancement model is used to enhance the signal-to-noise ratio of the diffusion-weighted image.

2. The method according to claim 1, characterized in that, The diffusion-weighted image of the first signal-to-noise ratio includes a first diffusion-weighted image with the same signal-to-noise ratio and a first diffusion sensitivity coefficient, and a second diffusion-weighted image with the same second diffusion sensitivity coefficient; the diffusion-weighted image of the second signal-to-noise ratio includes a third diffusion-weighted image with the same signal-to-noise ratio and a first diffusion sensitivity coefficient, and a fourth diffusion-weighted image with the same second diffusion sensitivity coefficient.

3. The method according to claim 1, characterized in that, The training method for the image enhancement model includes: Acquire diffusion-weighted image samples with a third signal-to-noise ratio and diffusion-weighted image samples with a fourth signal-to-noise ratio, wherein the diffusion-weighted image samples with the third signal-to-noise ratio are reconstructed based on K-space data of the third signal-to-noise ratio, and the diffusion-weighted image samples with the fourth signal-to-noise ratio are reconstructed based on partial diffusion-weighted imaging data in the K-space data of the third signal-to-noise ratio. The diffusion-weighted image samples with the fourth signal-to-noise ratio are used as input data to train the initial image enhancement model. The loss function of the initial image enhancement model is constructed based on the similarity between the predicted diffusion-weighted image output by the initial image enhancement model and the diffusion-weighted image samples with the third signal-to-noise ratio.

4. The method according to claim 3, characterized in that, The diffusion-weighted image samples with the third signal-to-noise ratio include first diffusion-weighted image samples with the same first diffusion sensitivity coefficient and second diffusion-weighted image samples with the same second diffusion sensitivity coefficient; the diffusion-weighted image samples with the fourth signal-to-noise ratio include third diffusion-weighted image samples with the same first diffusion sensitivity coefficient and fourth diffusion-weighted image samples with the same second diffusion sensitivity coefficient.

5. The method according to claim 3, characterized in that, The diffusion-weighted imaging data includes diffusion-weighted imaging data acquired at different numbers of channels, wherein K-space data is generated for each number of channels.

6. The method according to claim 3, characterized in that, The diffusion-weighted imaging data includes diffusion-weighted imaging data samples acquired at different acquisition durations, wherein K-space data is generated for each acquisition duration.

7. An image generation apparatus, characterized in that, The device includes: The image acquisition module is used to acquire the diffusion-weighted image of the tissue under test with the first signal-to-noise ratio. The image generation module is used to input the diffusion-weighted image with the first signal-to-noise ratio into the image enhancement model to obtain the diffusion-weighted image with the second signal-to-noise ratio corresponding to the tissue under test. The image enhancement model is used to enhance the signal-to-noise ratio of the diffusion-weighted image.

8. An image generation method, characterized in that, The method includes: Acquire the first set of K-space data corresponding to the tissue to be tested, wherein the first set of K-space data can reconstruct a diffusion-weighted image with a first signal-to-noise ratio; The first set of K-space data is input into the image enhancement model to obtain the second set of K-space data corresponding to the tissue under test. The second set of K-space data can reconstruct a diffusion-weighted image with a second signal-to-noise ratio.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images