Image denoising method and device, electronic equipment and medium

Abstract

The application provides a denoising method and device of an image, electronic equipment and medium. The method acquires myelin water image data of human brain tissue based on a preset multi-echo gradient echo sequence, and calculates amplitude image data acquired at each echo time point; determines a first noise level metric value of the multi-echo gradient echo sequence based on a noise value in the amplitude image data acquired at the last echo time point and an amplitude of the amplitude image data corresponding to the first echo time point; and obtains a corresponding second noise level metric value based on an error value between the first noise level metric value and the second noise level metric value. The amplitude image data at each echo time point is subjected to a preset denoising convolutional neural network to obtain a noise value of the amplitude image data at each echo time point, thereby denoising the amplitude image data acquired at the corresponding echo time point, obtaining noise-free amplitude image data at each echo time point, and improving the denoising effect.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and more specifically, to an image noise reduction method, apparatus, electronic device, and medium. Background Technology

[0002] Myelin water imaging is a magnetic resonance imaging technique that is of great significance for diagnosing lesions occurring in the myelin sheath and quantifying myelin sheath integrity. The specific procedure for acquiring T2* signals using multi-echo gradient echo sequences (mGRE) involves immediately employing a series of alternating polarity readout gradients after a small-angle excitation pulse and a phase-encoded gradient. The small-angle excitation pulse is used to minimize the first echo time (TE1). To further shorten TE1 and the echo interval, signals (echoes) are acquired during the readout gradient plateau, as well as at the rising and falling edges. This method offers several advantages, including low specific absorption rates, short echo intervals, 3D excitation capability, and minimum-phase radio frequency excitation. Notably, using minimum-phase radio frequency excitation can reduce the echo time of the first echo, thereby improving the accuracy of sampling true information about myelin water.

[0003] The multi-echo gradient echo sequence used to acquire myelin water image data conforms to a Rician distribution in the images acquired at each echo time. The T2* signal is a transverse magnetization vector signal composed of the amplitude values ​​corresponding to the voxels of the image at each echo time.

[0004] Currently, the main methods for denoising mGRE-acquired images are anisotropic diffusion filters and nonlocal mean filters. Before using these filters, the denoising parameters (noise variance) need to be determined. The denoising parameters are typically the noise value of the first echo image in the mGRE sequence. Subsequently, the anisotropic diffusion filter or nonlocal mean filter is applied to denoise each echo image of the mGRE.

[0005] However, for images obtained from mGRE sequences, the noise parameters applied by the anisotropic diffusion filter and the nonlocal mean filter only consider the noise value of the first echo image. Since the noise values ​​of images acquired at different echo time points in an mGRE sequence are not the same, selecting only the noise reduction parameters based on the first echo image is ineffective for image denoising, leading to inaccurate calculation of myelin water content. Summary of the Invention

[0006] The purpose of this application is to provide an image denoising method, apparatus, electronic device, and medium to solve the above-mentioned problems existing in the prior art, improve the image denoising effect, and thereby improve the accuracy of myelin water content calculation.

[0007] Firstly, an image noise reduction method is provided, which may include:

[0008] Based on a preset multi-echo gradient echo sequence, myelin water imaging data of voxels in human brain tissue are collected to obtain amplitude image data at each echo time point; the preset multi-echo gradient echo sequence includes multiple echo time points.

[0009] Based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point, the first noise level metric of the multi-echo gradient echo sequence is determined.

[0010] Obtain the error values ​​between the first noise level metric and each of the stored second noise level metric values; the second noise level metric values ​​are determined based on the initial recorded amplitude image data containing noise values ​​after adding new noise values;

[0011] If the target error value is less than a preset error threshold, then the target second noise level metric corresponding to the target error value is obtained; the target error value is one of the obtained error values.

[0012] The amplitude image data collected at each echo time point is used as input data and input into a preset denoising convolutional neural network corresponding to the target second noise level metric value to obtain the noise value of the amplitude image data at each echo time point output by the denoising convolutional neural network; the denoising convolutional neural network is obtained by training a convolutional neural network based on residual learning based on the noise distribution corresponding to different second noise level metric values.

[0013] The noise values ​​corresponding to each echo time point are used to denoise the amplitude image data collected at the corresponding echo time points, so as to obtain noise-free amplitude image data at each echo time point.

[0014] In an optional implementation, before determining the first noise level metric of the multi-echo gradient echo sequence based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point, the method further includes:

[0015] The variance of the background data in the amplitude image data at the last echo time point is determined as the noise value in the amplitude image data at the last echo time point;

[0016] The average value of the amplitude in the amplitude image data at the first echo time point is determined as the amplitude of the amplitude image data at the first echo time point.

[0017] In an optional implementation, a first noise level metric for the multi-echo gradient echo sequence is determined based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point, including:

[0018] The first noise level metric of the multi-echo gradient echo sequence is determined by dividing the noise value in the amplitude image data acquired at the last echo time point by the amplitude value of the amplitude image data at the first echo time point.

[0019] In an optional implementation, the training process of the denoising convolutional neural network includes:

[0020] Based on a preset multi-echo gradient echo sequence, myelin water image data of human brain tissue is acquired, and the amplitude image data of myelin water image acquired at each echo time point is calculated.

[0021] For the first echo time point and the last echo time point, different preset noise values ​​are added to the initial recorded amplitude image data of the corresponding echo time point, and different new recorded amplitude image data corresponding to the echo time point and the second noise level metric value corresponding to the different new recorded amplitude image data are obtained, and the second noise level metric value is stored.

[0022] Based on the different newly recorded amplitude image data and the corresponding initial recorded amplitude image data, the noise distribution corresponding to different second noise level metrics is determined;

[0023] Using the noise distributions corresponding to the different second noise level metrics, a convolutional neural network based on residual learning is trained to obtain a denoising convolutional neural network corresponding to the different second noise level metrics.

[0024] In an optional implementation, for the first echo time point and the last echo time point, preset different noise values ​​are added to the initial recorded amplitude image data of the corresponding echo time points, and different new recorded amplitude image data corresponding to the echo time points and a second noise level metric value corresponding to the different new recorded image data are obtained, including:

[0025] The variance of the background data in the initial recorded amplitude image data corresponding to the first echo time point and the last echo time point is increased by the preset different noise values ​​to obtain new recorded amplitude image data corresponding to the first echo time point and the last echo time point; wherein, the new recorded amplitude image data corresponding to the first echo time point and the last echo time point contains new noise values;

[0026] Based on the new noise value in the newly recorded amplitude image data at the last echo time point and the amplitude of the newly recorded amplitude image data at the first echo time point, a second noise level metric for the multi-echo gradient echo sequence is determined.

[0027] In an optional implementation, the amplitude image data at each echo time point all follow a Rice distribution;

[0028] The variance of the background data in the initial recorded amplitude image data samples corresponding to the first echo time point and the last echo time point is increased by the preset different noise values ​​to obtain new recorded amplitude image data samples corresponding to the first echo time point and the last echo time point, including:

[0029] The variance of the Rice distribution followed by the initial recorded amplitude image data samples corresponding to the first echo time point and the last echo time point is increased by the preset different noise values ​​to obtain new recorded amplitude image data samples corresponding to the first echo time point and the last echo time point.

[0030] In an optional implementation, the noise value corresponding to each echo time point is used to denoise the amplitude image data acquired at the corresponding echo time point to obtain noise-free amplitude image data for each echo time point, including:

[0031] Subtract the noise value corresponding to the echo time point from the amplitude image data of each echo time point to obtain the noise-free amplitude image data of each echo time point.

[0032] Secondly, an image noise reduction apparatus is provided, which may include:

[0033] The acquisition unit is used to acquire myelin water image data of voxels in human brain tissue based on a preset multi-echo gradient echo sequence, and obtain amplitude image data acquired at each echo time point; the preset multi-echo gradient echo sequence includes multiple echo time points.

[0034] The determining unit is used to determine the first noise level metric of the multi-echo gradient echo sequence based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point.

[0035] The acquisition unit is used to acquire the error values ​​between the first noise level measurement value and each of the stored second noise level measurement values; the second noise level measurement value is determined based on the initial recorded amplitude image data containing noise values ​​after adding new noise values;

[0036] Furthermore, if the target error value is less than a preset error threshold, then a target second noise level metric value corresponding to the target error value is obtained; the target error value is one of the obtained error values.

[0037] Furthermore, the amplitude image data acquired at each echo time point is used as input data and input into a preset denoising convolutional neural network corresponding to the target second noise level metric, to obtain the noise value of the amplitude image data at each echo time point output by the denoising convolutional neural network; the denoising convolutional neural network is obtained by training a convolutional neural network based on residual learning based on the noise distribution corresponding to different second noise level metric values.

[0038] The noise reduction unit is used to reduce the noise of the amplitude image data collected at each echo time point by using the noise value corresponding to each echo time point, so as to obtain noise-free amplitude image data at each echo time point.

[0039] Thirdly, an electronic device is provided, which includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0040] Memory, used to store computer programs;

[0041] When a processor executes a program stored in memory, it implements any of the steps described in the first aspect above.

[0042] Fourthly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements the steps of any of the methods described in the first aspect above.

[0043] The image denoising method provided in this application takes into account the noise impact of amplitude image data acquired at different echo time points of the mGRE sequence. Instead of using the noise in the amplitude image data acquired at the first echo time point as a reference, it calculates the noise level of the entire mGRE sequence and identifies the noise in the amplitude image data corresponding to each echo time point through a trained denoising convolutional neural network. This maximizes the consideration of the oscillation amplitude of the entire T2* decay curve, thereby improving the accuracy of denoising and the accuracy of myelin water content calculation. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A schematic flowchart illustrating an image noise reduction method provided in an embodiment of this application;

[0046] Figure 2 A schematic diagram of a T2* attenuation curve provided in an embodiment of this application;

[0047] Figure 3 A distribution map of myelin water content provided in an embodiment of this application;

[0048] Figure 4 A schematic diagram of the structure of an image noise reduction device provided in an embodiment of this application;

[0049] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0050] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0051] For ease of understanding, the terms used in the embodiments of this application are explained below:

[0052] A voxel is short for volume pixel. It is the smallest unit of digital data in three-dimensional space, used in fields such as 3D imaging, scientific data, and medical imaging. Some true 3D displays use voxels to describe resolution, such as displays that can display 512×512×512 voxels. Voxels do not contain spatial location data (i.e., their coordinates), but their position can be inferred from their position relative to other voxels, that is, their position in the data structure that makes up a single volumetric image.

[0053] The content of myelin sheath is generally estimated by estimating the water content of the myelin sheath in the tissue, because the water content of the myelin sheath has unique magnetic resonance properties that differ from those of surrounding tissues. However, myelin sheath water imaging data often contains significant noise, resulting in a low signal-to-noise ratio, and direct estimation of myelin sheath content typically leads to large estimation errors.

[0054] The noise level in images (or "amplitude image data") acquired by magnetic resonance imaging (MRI) equipment based on multi-echo gradient echo sequences (mGRE) is determined by the number of receiving coils. Noise acquired by a single-channel coil is typically considered Rician noise (more precisely, in low-signal regions, the noise approximates a Rician distribution; in high-signal regions, the noise approximates a Gaussian distribution, but overall it follows a Rician distribution), meaning the noise in the image is considered to follow a Rician distribution. However, noise acquired by multi-channel coils typically follows a non-central chi-square distribution. Due to the instability of the non-central chi-square distribution, images acquired by multi-channel coils are generally assumed to still follow a Rician distribution unless otherwise specified.

[0055] Noisy images are typically abstracted into the following form:

[0056] y = x + e (1)

[0057] Where x and y represent noise-free image data and noisy image data, respectively, and e represents the noise value. The noise reduction process can be described as finding a specific transformation T(·) such that the noisy image is transformed into a noise-free image, i.e., T(y) = x, or using the residual idea, such that T(y) = e, thus x = T(y) - y.

[0058] The distribution characteristics of Rician noise are as follows:

[0059]

[0060] Both N1 and N2 follow a normal distribution, i.e., N1~N(0,σ1) 2 ), N2~N(0,σ2) 2 Therefore, when using conventional filters for noise reduction, it is necessary to perform noise variance detection on the image to be denoised.

[0061] The business data processing method provided in this application embodiment can be applied on a server or on a terminal. The server can be an application server or a cloud server; to ensure the accuracy of detection, the terminal can be a user equipment (UE) such as a mobile phone, smartphone, laptop, digital broadcast receiver, personal digital assistant (PDA), tablet computer (PAD), handheld device, vehicle-mounted device, wearable device, computing device or other processing device connected to a wireless modem, mobile station (MS), mobile terminal, etc.

[0062] Existing anisotropic diffusion filters and nonlocal mean filters only select denoising parameters based on the first echo image during the denoising process, without applying the inherent correlation properties between the corresponding images in the mGRE sequence. This results in an uneven attenuation curve of the transverse magnetization vector (T2*) composed of the amplitudes corresponding to voxels in the image at each echo time point after denoising, leading to deviations in the calculation of myelin water content. This application takes into account the noise impact of amplitude image data acquired at different echo time points of the mGRE sequence. Instead of using the noise in the amplitude image data acquired at the first echo time point as a reference, it calculates the noise level of the entire mGRE sequence and uses a trained denoising convolutional neural network to identify the noise in the amplitude image data corresponding to each echo time point. This maximizes the consideration of the oscillation amplitude of the entire T2* attenuation curve, thereby improving the accuracy of denoising. The T2* attenuation curve represents the inherent relationship between the amplitude image data corresponding to each echo time point in the entire sequence.

[0063] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0064] Figure 1 This is a schematic flowchart illustrating an image noise reduction method provided in an embodiment of this application. Figure 1 As shown, the method may include:

[0065] Step S110: Based on the preset multi-echo gradient echo sequence, collect myelin water image data of voxels in human brain tissue to obtain amplitude image data collected at each echo time point.

[0066] The preset multi-echo gradient echo sequence includes multiple echo time points. At each echo time point, myelin water image data in human brain tissue is collected to obtain amplitude image data collected at each echo time point.

[0067] Since the acquired amplitude image data contains noise values, and the images acquired in the multi-echo gradient echo sequence at each echo time conform to a Rician distribution, the amplitude image data follows a Rician distribution, i.e., it satisfies: Where x and y represent the noise-free image and the noisy image, respectively, and the noise values ​​follow normal distributions N1 and N2.

[0068] Step S120: Based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point, determine the first noise level metric of the multi-echo gradient echo sequence.

[0069] In practice, the variance of the background data in the amplitude image data at the last echo time point can be determined as the noise value in the amplitude image data at the last echo time point; the average value of each amplitude in the amplitude image data at the first echo time point can be determined as the amplitude of the amplitude image data at the first echo time point.

[0070] It should be noted that the average value of the noise in the amplitude image data corresponding to each echo time point can also be obtained as the noise value of the amplitude image data of the last echo time point; or, the amplitude corresponding to the median of each amplitude in the amplitude image data of the first echo time point can be determined as the amplitude of the amplitude image data of the first echo time point. The above determination method is only one method, and the specific choice can be made according to the actual business situation. This application does not limit it here.

[0071] Then, the noise value in the amplitude image data acquired at the last echo time point is divided by the amplitude value of the amplitude image data at the first echo time point to determine the first noise level metric of the multi-echo gradient echo sequence, which is the current noise level metric.

[0072] It should be noted that other calculations can also be performed on the noise value corresponding to the last echo time point and the amplitude corresponding to the first echo time point. For example, the noise value corresponding to the last echo time point can be divided by the sum of the noise value and the amplitude corresponding to the first echo time point to determine the current noise level metric of the multi-echo gradient echo sequence. This application does not limit this.

[0073] Understandably, the current noise level metric here is analogous to the signal-to-noise ratio of an image: the ratio of signal to noise indicates the quality of the image. Therefore, the current noise level metric is defined as the ratio of the noise value in the amplitude image data acquired at the last echo time point to the amplitude of the amplitude image data at the first echo time point to characterize the noise level of the entire multi-echo gradient echo sequence.

[0074] Step S130: Based on the first noise level metric value and each of the stored second noise level metric values, obtain the target second noise level metric value corresponding to the target error value.

[0075] In specific implementation, the error values ​​between the first noise level metric and each stored second noise level metric are obtained; wherein, the second noise level metric is determined based on the initial recorded amplitude image data containing noise values ​​after adding new noise values. The process of obtaining the second noise level metric includes:

[0076] Based on a preset multi-echo gradient echo sequence, myelin water image data of human brain tissue is collected, and the amplitude image data of myelin water image collected at each echo time point is calculated.

[0077] In the initial recorded amplitude image data containing noise values, the variance of the background data in the initial recorded amplitude image data corresponding to the first and last echo time points is increased by a preset different noise value to obtain new recorded amplitude image data corresponding to the first and last echo time points. Specifically, since the amplitude image data at each echo time point all follow a Rice distribution, the variance of the Rice distribution followed by the initial recorded amplitude image data samples corresponding to the first and last echo time points is increased by a preset different noise value to obtain new recorded amplitude image data samples corresponding to the first and last echo time points. That is, combined with formula (2), it can be seen that only the value of at least one variance in the formula needs to be changed to make the new recorded amplitude image data corresponding to the first and last echo time points contain new noise values.

[0078] Subsequently, based on the new noise value in the newly recorded amplitude image data sample at the last echo time point and the amplitude of the newly recorded amplitude image data at the first echo time point, the second noise level metric of the multi-echo gradient echo sequence is determined and stored.

[0079] Furthermore, it detects whether any of the acquired error values ​​are less than a preset error threshold.

[0080] If the target error value is less than the preset error threshold, then the target second noise level metric value corresponding to the target error value is obtained.

[0081] Step S140: Take the amplitude image data collected at each echo time point as input data, input the preset residual learning convolutional neural network corresponding to the target second noise level metric, and obtain the noise value of the amplitude image data at each echo time point output by the noise reduction convolutional neural network.

[0082] The convolutional neural network for residual learning is trained based on the noise distribution corresponding to different second noise level metrics.

[0083] Before performing this step, a denoising convolutional neural network needs to be trained. The training process for the denoising convolutional neural network includes:

[0084] Obtain the second noise level metric value corresponding to the initial recorded amplitude image data collected in step S130 above, and the new recorded amplitude image data after adding different preset noise values;

[0085] Based on different newly recorded image data and corresponding initial recorded amplitude image data samples, the noise distribution corresponding to different second noise level measures is determined; for example, the newly recorded amplitude image data is the data sample of image SS, and its corresponding initial recorded amplitude image data sample is the data sample of image S. By subtracting the corresponding data sample of image S from the data sample of image SS using formula (2), the noise distribution corresponding to the second noise level measures can be obtained, which does not include image content data.

[0086] By using the noise distributions corresponding to the different second noise level metrics obtained above, and iteratively training the convolutional neural network based on residual learning, we can obtain the denoising convolutional neural networks corresponding to the different second noise level metrics.

[0087] The trained denoising convolutional neural network can identify the overall distribution of noise in the amplitude image data corresponding to each echo time point for specific noise, that is, noise that meets each second noise level metric, rather than the numerical relationship between neighborhoods in the amplitude image.

[0088] It is understood that if no error value is less than the preset error threshold in step S130, the noise distribution corresponding to the first noise level metric can be obtained to train a denoising convolutional neural network that satisfies the first noise level metric, or the denoising process can be terminated directly. This application does not limit this.

[0089] Step S150: Use the noise value corresponding to each echo time point to denoise the amplitude image data collected at the corresponding echo time point to obtain noise-free amplitude image data at each echo time point.

[0090] Subtract the corresponding noise value from the amplitude image data at each echo time point to obtain noise-free amplitude image data at each echo time point.

[0091] Based on the above embodiments of this application, since the noise influence of amplitude image data acquired at different echo time points of the mGRE sequence is considered, that is, the intrinsic relationship between amplitude image data acquired at different echo time points is considered, the T2* attenuation curve is smoothed, that is, the noise reduction effect of the image is improved, thereby improving the accuracy of myelin water value calculation.

[0092] like Figure 2 As shown, taking the T2* decay curve corresponding to a voxel in the mGRE data as an example, after passing through the denoising convolutional neural network with a specific noise level according to this application and the corresponding denoising processing, the original T2* decay curve, i.e., the original decay curve with different oscillation amplitudes in the figure, is smoothed to obtain the smoothed curve after denoising, i.e., the decay curve after denoising. The original MWF of the myelin water fraction corresponding to the selected voxel is 0.27287, and the denoised MWF is 0.17078. Figure 2 In the text, TE represents the echo time, and Signal represents the signal quantity, i.e., the amplitude of the voxel.

[0093] Due to the signal characteristics S(r,TE) of the mGRE sequence and its characteristic S(t) fitted by the complex three-pool model:

[0094]

[0095]

[0096] Where r represents the spatial position of each voxel, A / M represents the amplitude value, TE represents the echo time, ω0 represents the proton precession frequency, my, ax, and ex represent myelin water, axon water, and extracellular water, respectively, and f represents the frequency and This represents the initial phase value. The myelin water fraction can be expressed as:

[0097] MWF=A my / (A my +A ax +A ex (5)

[0098] It is evident that there is a correlation between the amplitude image data acquired at each echo time. By using formula (4) to fit the result of formula (3), the final myelin water content is obtained, and the accuracy of noise reduction will directly affect the estimation of myelin water content.

[0099] like Figure 3As shown, taking the myelin water content values ​​corresponding to 30 voxels as an example, after applying the noise reduction convolutional neural network with a specific noise level as described in this application and the corresponding noise reduction processing, the discretization range (unreasonable range) of the original myelin water content values ​​(solid circles in the figure) is adjusted to a reasonable range (hollow circles in the figure). To increase the accuracy of the statistics, a two-tailed independent samples t-test can be used here to test the difference between the two groups of data. Figure 3 The MWF values ​​of 30 randomly selected voxels are displayed, showing data before and after denoising (solid circles in the figure). MWF: Myelin water fraction; Original mean: Mean of original myelin water fraction; Denoised mean: Mean of denoising myelin water fraction; SD: Standard deviation; p: Difference value from independent samples t-test.

[0100] This method takes into account the noise impact of amplitude image data acquired at different echo time points of the mGRE sequence. Instead of using the noise in the amplitude image data acquired at the first echo time point as a reference, it calculates the noise level of the entire mGRE sequence and uses a trained denoising convolutional neural network to identify the noise in the amplitude image data corresponding to each echo time point. This approach maximizes the consideration of the oscillation amplitude of the entire T2* decay curve, thereby improving the accuracy of denoising and the accuracy of myelin water content calculation.

[0101] Corresponding to the above method, embodiments of this application also provide an image noise reduction device, such as... Figure 4 As shown, the noise reduction device for the image includes: an acquisition unit 410, a determination unit 420, an acquisition unit 430, and a noise reduction unit 440;

[0102] The acquisition unit 410 is used to acquire myelin water image data of voxels in human brain tissue based on a preset multi-echo gradient echo sequence, and obtain amplitude image data acquired at each echo time point; the preset multi-echo gradient echo sequence includes multiple echo time points.

[0103] The determining unit 420 is used to determine the first noise level metric of the multi-echo gradient echo sequence based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point.

[0104] The acquisition unit 430 is used to acquire the error values ​​between the first noise level measurement value and each of the stored second noise level measurement values; the second noise level measurement value is determined based on the initial recorded amplitude image data containing noise values ​​after adding new noise values;

[0105] Furthermore, if the target error value is less than a preset error threshold, then a target second noise level metric value corresponding to the target error value is obtained; the target error value is one of the obtained error values.

[0106] Furthermore, the amplitude image data collected at each echo time point is used as input data and input into a preset denoising convolutional neural network corresponding to the target second noise level metric value to obtain the noise value of the amplitude image data at each echo time point output by the denoising convolutional neural network; the denoising convolutional neural network is obtained by training a convolutional neural network based on residual learning based on the noise distribution corresponding to different second noise level metric values.

[0107] The noise reduction unit 440 is used to reduce the noise of the amplitude image data collected at the corresponding echo time point using the noise value corresponding to each echo time point, so as to obtain noise-free amplitude image data at each echo time point.

[0108] In an optional implementation, unit 420 is also used for:

[0109] The variance of the background data in the amplitude image data at the last echo time point is determined as the noise value in the amplitude image data at the last echo time point;

[0110] Furthermore, the average value of each amplitude in the amplitude image data at the first echo time point is determined as the amplitude of the amplitude image data at the first echo time point.

[0111] In an optional implementation, the determining unit 420 is specifically used to determine the first noise level metric of the multi-echo gradient echo sequence by dividing the noise value in the amplitude image data acquired at the last echo time point by the amplitude of the amplitude image data at the first echo time point.

[0112] In an optional implementation, the apparatus further includes a training unit 450; the training unit 450 is configured to:

[0113] Based on a preset multi-echo gradient echo sequence, myelin water image data of human brain tissue is collected, and the amplitude image data of myelin water image collected at each echo time point is calculated.

[0114] For the first echo time point and the last echo time point, different preset noise values ​​are added to the initial recorded amplitude image data of the corresponding echo time point, and different new recorded amplitude image data corresponding to the echo time point and the second noise level metric value corresponding to the different new recorded amplitude image data are obtained, and the second noise level metric value is stored.

[0115] Based on the different newly recorded amplitude image data and the corresponding initial recorded amplitude image data, the noise distribution corresponding to different second noise level metrics is determined;

[0116] Using the noise distributions corresponding to the different second noise level metrics, a convolutional neural network based on residual learning is trained to obtain a denoising convolutional neural network corresponding to the different second noise level metrics.

[0117] In an optional implementation, unit 420 is further specifically used for:

[0118] The variance of the background data in the initial recorded amplitude image data corresponding to the first echo time point and the last echo time point is increased by the preset different noise values ​​to obtain new recorded amplitude image data corresponding to the first echo time point and the last echo time point; wherein, the new recorded amplitude image data corresponding to the first echo time point and the last echo time point contains new noise values;

[0119] Based on the new noise value in the newly recorded amplitude image data at the last echo time point and the amplitude of the newly recorded amplitude image data at the first echo time point, a second noise level metric for the multi-echo gradient echo sequence is determined.

[0120] In an optional implementation, the amplitude image data at each echo time point all follow a Rice distribution;

[0121] The acquisition unit 430 is further configured to add the preset different noise values ​​to the variance of the Rice distribution followed by the initial recorded amplitude image data corresponding to the first echo time point and the last echo time point, so as to obtain new recorded amplitude image data corresponding to the first echo time point and the last echo time point.

[0122] In an optional implementation, the noise reduction unit 440 is specifically used to subtract the noise value corresponding to the echo time point from the amplitude image data at each echo time point to obtain noise-free amplitude image data at each echo time point.

[0123] The functions of each functional unit in the image noise reduction device provided in the above embodiments of this application can be implemented through the above method steps. Therefore, the specific working process and beneficial effects of each unit in the image noise reduction device provided in the embodiments of this application will not be repeated here.

[0124] This application also provides an electronic device, such as... Figure 5As shown, it includes a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other through the communication bus 540.

[0125] Memory 530 is used to store computer programs;

[0126] When the processor 510 executes the program stored in the memory 530, it performs the following steps:

[0127] Based on a preset multi-echo gradient echo sequence, myelin water imaging data of voxels in human brain tissue are collected to obtain amplitude image data at each echo time point; the preset multi-echo gradient echo sequence includes multiple echo time points.

[0128] Based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point, the first noise level metric of the multi-echo gradient echo sequence is determined.

[0129] Obtain the error values ​​between the first noise level metric and each of the stored second noise level metric values; the second noise level metric values ​​are determined based on the initial recorded amplitude image data containing noise values ​​after adding new noise values;

[0130] If the target error value is less than a preset error threshold, then the target second noise level metric corresponding to the target error value is obtained; the target error value is one of the obtained error values.

[0131] The amplitude image data collected at each echo time point is used as input data and input into a preset denoising convolutional neural network corresponding to the target second noise level metric value to obtain the noise value of the amplitude image data at each echo time point output by the denoising convolutional neural network; the denoising convolutional neural network is obtained by training a convolutional neural network based on residual learning based on the noise distribution corresponding to different second noise level metric values.

[0132] The noise values ​​corresponding to each echo time point are used to denoise the amplitude image data collected at the corresponding echo time points, so as to obtain noise-free amplitude image data at each echo time point.

[0133] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0134] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0135] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0136] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0137] The implementation methods and beneficial effects of the various components of the electronic device in the above embodiments for solving the problem can be found in [reference needed]. Figure 1 The steps in the illustrated embodiments are used to implement the electronic device. Therefore, the specific working process and beneficial effects of the electronic device provided in this application will not be repeated here.

[0138] In another embodiment provided in this application, a computer-readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform the image noise reduction method described in any of the above embodiments.

[0139] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform the image noise reduction method described in any of the above embodiments.

[0140] Those skilled in the art will understand that the embodiments in this application can be provided as methods, systems, or computer program products. Therefore, the embodiments in this application can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, the embodiments in this application can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0141] This application describes embodiments of methods, apparatus (systems), and computer program products according to embodiments of this application with reference to flowchart illustrations and / or block diagrams. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0142] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0143] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0144] Although preferred embodiments have been described in this application, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of this application.

[0145] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims in this application and their equivalents, then this application also intends to include these modifications and variations.

Claims

1. A method for denoising images, characterized in that, The method includes: Based on a preset multi-echo gradient echo sequence, myelin water imaging data of voxels in human brain tissue are collected to obtain amplitude image data at each echo time point; the preset multi-echo gradient echo sequence includes multiple echo time points. Based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point, the first noise level metric of the multi-echo gradient echo sequence is determined. Obtain the error values ​​between the first noise level metric and each of the stored second noise level metric values; the second noise level metric values ​​are determined based on the initial recorded amplitude image data containing noise values ​​after adding new noise values; If the target error value is less than a preset error threshold, then the target second noise level metric corresponding to the target error value is obtained; the target error value is one of the obtained error values. The amplitude image data collected at each echo time point is used as input data and input into a preset denoising convolutional neural network corresponding to the target second noise level metric value to obtain the noise value of the amplitude image data at each echo time point output by the denoising convolutional neural network; the denoising convolutional neural network is obtained by training a convolutional neural network based on residual learning based on the noise distribution corresponding to different second noise level metric values. The noise values ​​corresponding to each echo time point are used to denoise the amplitude image data collected at the corresponding echo time points to obtain noise-free amplitude image data at each echo time point. The process of obtaining the second noise level metric includes: Based on a preset multi-echo gradient echo sequence, myelin water image data of human brain tissue is collected, and the amplitude image data of myelin water image collected at each echo time point is calculated. For the first echo time point and the last echo time point, different preset noise values ​​are added to the initial recorded amplitude image data of the corresponding echo time point to obtain the new recorded amplitude image data corresponding to the first echo time point and the last echo time point. Based on the new noise value in the newly recorded amplitude image data sample at the last echo time point and the amplitude of the newly recorded amplitude image data at the first echo time point, a second noise level metric of the multi-echo gradient echo sequence is determined, and the corresponding second noise level metric is stored.

2. The method as described in claim 1, characterized in that, Before determining the first noise level metric of the multi-echo gradient echo sequence based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point, the method further includes: The variance of the background data in the amplitude image data at the last echo time point is determined as the noise value in the amplitude image data at the last echo time point; The average value of the amplitude in the amplitude image data at the first echo time point is determined as the amplitude of the amplitude image data at the first echo time point.

3. The method as described in claim 1 or 2, characterized in that, Based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point, a first noise level metric value for the multi-echo gradient echo sequence is determined, including: The first noise level metric of the multi-echo gradient echo sequence is determined by dividing the noise value in the amplitude image data acquired at the last echo time point by the amplitude value of the amplitude image data at the first echo time point.

4. The method as described in claim 1, characterized in that, The training process of the noise-reducing convolutional neural network includes: Based on the different newly recorded amplitude image data and the corresponding initial recorded amplitude image data, the noise distribution corresponding to different second noise level metrics is determined; Using the noise distributions corresponding to the different second noise level metrics, a convolutional neural network based on residual learning is trained to obtain a denoising convolutional neural network corresponding to the different second noise level metrics.

5. The method as described in claim 4, characterized in that, For the first echo time point and the last echo time point, preset different noise values ​​are added to the initial recorded amplitude image data of the corresponding echo time points. Different newly recorded amplitude image data corresponding to the echo time points and a second noise level metric corresponding to the different newly recorded amplitude image data samples are obtained, including: The variance of the background data in the initial recorded amplitude image data corresponding to the first echo time point and the last echo time point is increased by the preset different noise values ​​to obtain new recorded amplitude image data corresponding to the first echo time point and the last echo time point; wherein, the new recorded amplitude image data corresponding to the first echo time point and the last echo time point contains new noise values; Based on the new noise value in the newly recorded amplitude image data at the last echo time point and the amplitude of the newly recorded amplitude image data at the first echo time point, a second noise level metric for the multi-echo gradient echo sequence is determined.

6. The method as described in claim 5, characterized in that, The amplitude image data at each echo time point all follow a Rice distribution; The variance of the background data in the initial recorded amplitude image data corresponding to the first echo time point and the last echo time point is increased by the preset different noise values ​​to obtain new recorded amplitude image data corresponding to the first echo time point and the last echo time point, including: The variance of the Rice distribution followed by the initial recorded amplitude image data corresponding to the first echo time point and the last echo time point is increased by the preset different noise values ​​to obtain new recorded amplitude image data corresponding to the first echo time point and the last echo time point.

7. The method as described in claim 1, characterized in that, Using the noise values ​​corresponding to each echo time point, the amplitude image data acquired at the corresponding echo time point is denoised to obtain noise-free amplitude image data for each echo time point, including: Subtract the noise value corresponding to the echo time point from the amplitude image data of each echo time point to obtain the noise-free amplitude image data of each echo time point.

8. An image noise reduction device, characterized in that, The device includes: The acquisition unit is used to acquire myelin water image data of voxels in human brain tissue based on a preset multi-echo gradient echo sequence, and obtain amplitude image data acquired at each echo time point; the preset multi-echo gradient echo sequence includes multiple echo time points. The determining unit is used to determine the first noise level metric of the multi-echo gradient echo sequence based on the noise value in the amplitude image data acquired at the last echo time point and the amplitude of the amplitude image data corresponding to the first echo time point. The acquisition unit is used to acquire the error values ​​between the first noise level measurement value and each of the stored second noise level measurement values; the second noise level measurement value is determined based on the initial recorded amplitude image data containing noise values ​​after adding new noise values; Furthermore, if the target error value is less than a preset error threshold, then a target second noise level metric value corresponding to the target error value is obtained; the target error value is one of the obtained error values. Furthermore, the amplitude image data acquired at each echo time point is used as input data and input into a preset denoising convolutional neural network corresponding to the target second noise level metric, to obtain the noise value of the amplitude image data at each echo time point output by the denoising convolutional neural network; the denoising convolutional neural network is obtained by training a convolutional neural network based on residual learning based on the noise distribution corresponding to different second noise level metric values. The noise reduction unit is used to reduce the noise of the amplitude image data collected at the corresponding echo time points using the noise values ​​corresponding to each echo time point, so as to obtain noise-free amplitude image data at each echo time point. Specifically, the acquisition unit is used to acquire the second noise level metric value through the following process: Based on a preset multi-echo gradient echo sequence, myelin water image data of human brain tissue is collected, and the amplitude image data of myelin water image collected at each echo time point is calculated. For the first echo time point and the last echo time point, different preset noise values ​​are added to the initial recorded amplitude image data of the corresponding echo time point to obtain the new recorded amplitude image data corresponding to the first echo time point and the last echo time point. Based on the new noise value in the newly recorded amplitude image data sample at the last echo time point and the amplitude of the newly recorded amplitude image data at the first echo time point, a second noise level metric of the multi-echo gradient echo sequence is determined, and the corresponding second noise level metric is stored.

9. An electronic device, characterized in that, The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method described in any one of claims 1-7.

Images