Method for generating magnetic resonance proton density fat fraction and R2 star quantitative graph

Abstract

The invention relates to the field of image processing and medicine, and discloses a method for generating a magnetic resonance proton density fat fraction and R2 star quantitative graph, which comprises the following steps of: acquiring mDixon sequence image data, DWI sequence image data and corresponding qDixon sequence graph data, and integrating the image data to form a data set; preprocessing the acquired image data; constructing a deep learning model for generating a magnetic resonance proton density fat fraction and R2 star quantitative graph from the mDixon sequence image and the DWI sequence image; training the established deep learning model by using the preprocessed image data to obtain a trained deep learning model; and acquiring mDixon sequence image data and DWI sequence image data of a to-be-tested person, inputting the mDixon sequence image data and the DWI sequence image data into the trained deep learning model, and generating a proton density fat fraction and R2 star quantitative graph. The problem that PDFF deviation is large in an iron overload patient is solved.

Description

Technical Field

[0001] The invention relates to the fields of image processing and medicine, and in particular to a method for generating a magnetic resonance proton density fat fraction and R2 star quantitative image. Background Art

[0002] Quantitative Dixon imaging (qDixon) is an important technique used to accurately measure the proton density fat fraction (PDFF) and R2 star (R2 star) of the liver, two parameters that are critical for the diagnosis of hepatic steatosis (i.e., hepatic fat deposition). By utilizing multiple echo signals, qDixon is able to accurately separate fat and water components, thereby providing a non-invasive quantitative assessment of liver fat content. Due to its high accuracy and non-invasive nature, qDixon is used in the clinical diagnosis of liver diseases, especially in the evaluation of diseases such as fatty liver, non-alcoholic fatty liver disease (NAFLD), and cirrhosis.

[0003] However, a major drawback of the qDixon technique is that its scanning time is too long, which results in a long examination time for patients. This not only increases the examination cost and affects the patient experience, but also increases the chance of voluntary or involuntary movement during the examination due to decreased patient tolerance, affecting image quality. In addition, the qDixon method is usually only implemented in high-end MRI equipment, which also limits the application and popularization of this method in routine clinical practice.

[0004] In contrast, mDixon is a more simplified technology. Compared with qDixon, mDixon requires fewer echo signals, which significantly shortens the acquisition time and reduces costs, making the technology more applicable in routine clinical examinations. The advantage of the mDixon method is that it not only reduces the burden of equipment, but also improves patient comfort. Especially in cases where rapid examinations are required, mDixon can provide an effective alternative.

[0005] However, since mDixon uses two-point echo signals and lacks correction for T2 attenuation, it may produce deviations when there are tissue changes such as iron deposition. Iron deposition, especially in patients with iron overload, is one of the important factors affecting the accuracy of mDixon. Iron deposition can lead to a decrease in the T2 value of the liver, thereby affecting the accuracy of water-lipid separation, and thus leading to inaccurate estimation of PDFF values. Therefore, the application effect of mDixon in patients with iron overload is far less ideal than in the normal population, and PDFF estimation may produce significant deviations. Therefore, there is an urgent need for a PDFF and R2 star quantification method that can improve efficiency while maintaining high accuracy. Summary of the invention

[0006] The main purpose of the present invention is to solve the technical problem that the application effect of mDixon in patients with iron overload is far less than that in normal people in the prior art. A method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map comprises the following steps: S1, obtaining mDixon sequence image data and DWI sequence image data and corresponding qDixon sequence image data, and integrating the image data to form a data set; S2, preprocessing the collected image data, wherein the data preprocessing includes image standardization, image denoising and data amplification; S3, construct a deep learning model to generate magnetic resonance proton density fat fraction and R2 star quantitative map from mDixon sequence images and DWI sequence images; S4, using the image data pre-processed by S2, training the deep learning model established by S3 to obtain a trained deep learning model; S5. Obtain the mDixon sequence image data and DWI sequence image data of the subject, input the trained deep learning model, and generate the proton density fat fraction and R2 star quantitative map.

[0007] As a preferred technical solution, in step S1, the mDixon sequence image is used to provide separation information of fat and water components. Four images are obtained through the mDixon sequence image, namely, an in-phase image (IP), an anti-phase image (OP), a water image, and a fat image; the qDixon sequence image data obtains accurate PDFF and R2 star quantitative images as the benchmark truth value during model training and evaluation.

[0008] As a preferred technical solution, in step S2, the specific steps of image standardization, image denoising and data enhancement are: all images are uniformly adjusted to the same resolution; then, the Z-Score standardization method is applied to normalize the pixel values ​​of the images; the images are automatically segmented to remove background noise, and only the target area is retained for subsequent analysis; random flipping, rotation, cropping, brightness adjustment, contrast change and random noise addition are performed.

[0009] As a preferred technical solution, in step S3, the deep learning model for generating magnetic resonance proton density fat fraction and R2 star quantitative map uses U-Net convolutional neural network as the basic architecture. The U-Net network structure includes multiple convolution layers, pooling layers, upsampling layers and jump connection layers, and the network adopts multi-branch input and multi-channel output. As a preferred technical solution, the U-Net model input includes multiple branches, which respectively receive IP and OP images and DWI sequence images from the mDixon sequence images. The input image is first extracted through a series of convolution layers and pooling layers of each branch, and the features of each branch are fused in the middle layer of the network; in the upsampling stage, the multi-scale features of each branch are fused, and the final feature map will be sent to two different convolution layers, and each layer generates PDFF and R2 star quantitative map.

[0010] As a preferred technical solution, in step S4, the training of a deep learning model for generating magnetic resonance proton density fat fraction and R2 star quantitative map includes the following steps: during the training process, the model is iteratively trained using labeled training data; the labels include PDFF and R2 star quantitative map from qDixon technology, and the model gradually optimizes the network weights by minimizing the loss function between the predicted value and the true label; and the cross-validation method is used to verify the generalization ability of the model.

[0011] A second aspect of the present invention provides a system for generating a magnetic resonance proton density fat fraction and R2 star quantitative map, comprising: An acquisition module is used to acquire mDixon sequence image data and DWI sequence image data and corresponding qDixon sequence image data, and integrate the image data to form a data set; A preprocessing module, used to preprocess the collected image data, wherein the data preprocessing includes image standardization, image denoising and data amplification; A deep learning model building module is used to build a deep learning model for generating magnetic resonance proton density fat fraction and R2 star quantitative images from mDixon sequence images and DWI sequence images; A training module is used to train the deep learning model established by S3 using the image data preprocessed by S2 to obtain a trained deep learning model; The generation module is used to obtain the mDixon sequence image data and DWI sequence image data of the subject, input the trained deep learning model, and generate the proton density fat fraction and R2 star quantitative map.

[0012] A third aspect of the present invention provides an electronic device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via lines; the at least one processor calls the instructions in the memory so that the electronic device executes the method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map as described above.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map as described above.

[0014] The present invention has the following beneficial effects: The present invention generates PDFF and R2 star quantitative images by using conventional MRI sequences (mDixon, DWI), which reduces the scanning time by at least 30% compared to the traditional qDixon technology and avoids the discomfort caused by long-term exposure of patients to MRI scanning.

[0015] By combining multimodal data including diffusion weighted imaging DWI that provides iron deposition information with deep learning models, the estimation accuracy of PDFF and R2 star parameters can be significantly improved. Compared with mDixon technology and single modality input data, the present invention solves the problem of large PDFF deviation in patients with iron overload. Experiments show that the PDFF value predicted by the present invention is highly consistent with the qDixon method.

[0016] The present invention is not only applicable to the estimation of liver fat fraction, but can also be extended to quantitative imaging of other organs or tissues, and has broad clinical application prospects. By combining with other MRI sequences and quantitative analysis methods, the present invention can provide strong support for the early diagnosis of various diseases and promote the development of precision medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is an overall flow chart of the method for generating magnetic resonance proton density fat fraction and R2 star quantitative map based on deep learning of the present invention; Figure 2 is a network architecture diagram of the present invention; Figure 3 A comparison diagram of the PDFF and R2 star quantitative graphs generated by the model of the present invention and the quantitative graphs obtained by the qDixon technology; Figure 4 are the real PDFF data distribution diagram and the real R2 star data distribution diagram, A is the real PDFF data distribution diagram in the training set and the test set, and B is the real R2 star data distribution diagram in the training set and the test set; Figure 5is the regression analysis graph of MBCP-Net PDFF (No DWI) and MBCP-Net PDFF (With DWI) in the group without moderate iron deposition, in which qDixon PDFF was used as the reference standard; Figure 6 This is the same regression analysis graph in the moderate to severe iron deposition group; Figure 7 Bland-Altman analysis of MBCP-Net PDFF (No DWI) and qDixon PDFF in two iron deposition groups; Figure 8 Bland-Altman analysis of MBCP-Net PDFF (With DWI) and qDixon PDFF in two iron deposition groups; Fig. 9 The distribution diagrams of qDixon PDFF minus MBCP-Net PDFF (With DWI) and MBCP-Net PDFF (No DWI) in the moderate to severe iron deposition group are shown. DETAILED DESCRIPTION

[0018] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] Proton Density Fat Fraction (PDFF) PDFF represents the percentage of fat proton signal in unit volume of tissue to total proton signal (fat + water), and is a dimensionless quantitative index (range: 0%~100%). Through multi-echo gradient echo (multi-echo GRE) or chemical shift encoding (IDEAL / Dixon) sequence, signals with different echo times (TE) or different chemical shifts (difference in frequency between fat and water) are collected. The law of signal change with TE in the fat-water mixing model is used, combined with complex algorithms (such as multi-peak fat spectrum model, R2 star correction), to separate the signal contribution of fat and water and calculate PDFF.

[0020] R2 Star R2* is the transverse relaxation rate, measured in seconds⁻¹ (s⁻¹), which reflects the rate at which the transverse magnetization signal in the tissue decays and is directly related to magnetic field inhomogeneities (such as iron deposition, hemorrhage, calcification, etc.).

[0021] qDixon sequence image qDixon (Quantitative Dixon) is an improved magnetic resonance imaging technology based on the physical model of the traditional Dixon method. It quantitatively generates proton density fat fraction (PDFF) and R2 (transverse relaxation rate) maps through a multi-echo gradient echo (multi-echo GRE) sequence and complex post-processing algorithms.

[0022] mDixon sequence image mDixon is an optimized version of the traditional Dixon technique that achieves more robust fat-water separation and generates multiple contrast images through improved scanning sequences and post-processing algorithms.

[0023] DWI sequence images DWI (Diffusion-Weighted Imaging) is a magnetic resonance imaging technology based on the diffusion movement of water molecules. It detects the restrictions of tissue microstructure (such as cell density and membrane integrity) on the diffusion of water molecules by applying diffusion-sensitive gradient pulses.

[0024] A method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map based on deep learning includes the following steps: S1: Use magnetic resonance imaging equipment to collect the mDixon sequence and diffusion weighted imaging DWI sequence of the subject and the corresponding qDixon sequence image data, and integrate the data to form a data set; S2: performing data preprocessing on the acquired MRI images, wherein the data preprocessing includes image standardization, image denoising and data amplification; S3: Construct a deep learning model to generate magnetic resonance proton density fat fraction and R2 star quantitative images from mDixon sequence images and DWI sequence images; S4: Using the magnetic resonance imaging data preprocessed in S2, the network model established in S3 is trained to obtain the trained model weights.

[0025] S5: Use the mDixon sequence and diffusion weighted imaging DWI sequence image data collected by various magnetic resonance equipment, and input the magnetic resonance proton density fat fraction and R2 star quantitative map to generate a deep learning model, and use the weights obtained in S4 to generate the corresponding PDFF and R2 star quantitative map.

[0026] Furthermore, in step S1, the mDixon sequence is used to provide separation information of fat and water components. Four images will be obtained through the mDixon sequence, namely, in-phase image (IP), anti-phase image (OP), water and fat images; while the DWI sequence is used to provide signal attenuation and tissue change information caused by iron deposition to improve the sensitivity to iron deposition; the qDixon sequence obtains accurate PDFF and R2 star quantitative images as the benchmark truth value for model training and evaluation.

[0027] Furthermore, in step S2, the specific steps of image standardization, image denoising and data enhancement are as follows: All images are uniformly adjusted to the same resolution to ensure that different images have consistent spatial scales. Then, the Z-Score normalization method is applied to normalize the pixel values ​​of the images to eliminate deviations and inconsistencies caused by different devices or scanning conditions, ensuring the uniformity of input data and the stability of model training; Automatically segment the image, remove background noise, and retain only the target area for subsequent analysis to ensure the accuracy and focus of the model processing; Random flipping, rotation, cropping, brightness adjustment, contrast change, random noise addition, etc. These enhancement methods can simulate the changes that may occur under different scanning conditions, effectively prevent overfitting, and improve the performance and adaptability of the model in practical applications.

[0028] Furthermore, in step S3, the deep learning model for generating magnetic resonance proton density fat fraction and R2 star quantitative image uses U-Net convolutional neural network as the basic architecture, and the U-Net network structure includes multiple convolution layers, pooling layers, upsampling layers and jump connection layers. The network adopts a multi-branch input and multi-channel output method to simultaneously process mDixon sequence and DWI sequence image data. The network effectively captures the detailed information in the image through the feature extraction module, and uses the jump connection layer to fuse features of different scales, thereby maintaining the spatial information of the image and enhancing the feature learning ability.

[0029] The U-Net model input includes multiple branches, which receive IP and OP images and DWI images from the mDixon sequence respectively. The input image is first extracted through a series of convolutional layers and pooling layers of each branch, and the features of each branch are fused in the middle layer of the network. In the upsampling stage, the multi-scale features of each branch are fused to make full use of multimodal information. The final feature map will be sent to two different convolutional layers, and each layer will generate PDFF and R2 star quantitative maps.

[0030] Further, in step S4, the deep learning model training for generating magnetic resonance proton density fat fraction and R2 star quantitative image includes the following steps: During the training process, the model is iteratively trained using labeled training data. The labels include PDFF and R2 star quantitative graphs from the qDixon technique. The model gradually optimizes the network weights by minimizing the loss function between the predicted value and the true label. The cross-validation method is used to verify the generalization ability of the model to ensure that the trained weights can achieve consistent and accurate results on different data sets.

[0031] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 The first embodiment of the method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map in the embodiment of the present invention includes: like Figure 1 As shown, this embodiment provides a method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map based on deep learning, comprising the following steps: S1: Use magnetic resonance imaging equipment to collect the mDixon sequence and diffusion weighted imaging DWI sequence and the corresponding qDixon sequence abdominal image data of the subject, and integrate the data to form a data set.

[0032] Specifically, in this embodiment, mDixon sequences, diffusion weighted imaging DWI (b=800) sequences and corresponding qDixon sequence abdominal image data of 301 subjects collected by United Imaging MRI equipment are used as a training set, and mDixon sequences, diffusion weighted imaging DWI (b=800) sequences and corresponding qDixon sequence abdominal image data of 135 subjects collected by Siemens, Philips and some United Imaging equipment are used as a test set.

[0033] S2: performing data preprocessing on the acquired MRI images, wherein the data preprocessing includes image standardization, image denoising and data enhancement.

[0034] Specifically, in step S2, all images are uniformly adjusted to the same resolution to ensure that different images have consistent spatial scales. Then, the Z-Score normalization method is applied to normalize the pixel values ​​of the images. The nnUNet model is used to accurately segment the abdominal images. By automatically segmenting the abdominal images, the background noise outside the abdomen is effectively removed, and only the target abdominal area is retained. The data is randomly flipped, rotated, cropped, brightness adjusted, contrast changed, and random noise added to enhance the data.

[0035] S3: Construct a deep learning model for generating magnetic resonance proton density fat fraction and R2 star quantitative images.

[0036] Specifically, in step S3, the network structure is as follows: Figure 2As shown in the figure. The deep learning model for constructing magnetic resonance proton density fat fraction and R2 star quantitative map uses 2D U-Net convolutional neural network as the basic architecture. The U-Net network structure includes multiple convolutional layers, pooling layers, upsampling layers and jump connection layers. The network adopts a multi-branch input and multi-channel output method to simultaneously process mDixon sequence and DWI sequence image data. The network effectively captures the detailed information in the image through the feature extraction module and uses the jump connection layer to fuse features of different scales, thereby maintaining the spatial information of the image and enhancing the feature learning ability.

[0037] S4: In the network model established in S3, the magnetic resonance imaging data preprocessed by S2 is trained to obtain trained weights.

[0038] Specifically, in step S4, during the training process, the model is iteratively trained using labeled training data. The labels include PDFF and R2 star quantitative maps from the qDixon technology. The model gradually optimizes the network weights by minimizing the loss function between the predicted value and the true label. The loss function is the L1 norm loss. The 5-fold cross-validation method is used to verify the generalization ability of the model to ensure that the trained weights can achieve consistent and accurate results on different data sets.

[0039] S5: Use the mDixon sequence and DWI (b=800) sequence abdominal imaging data of the test set, and input the magnetic resonance proton density fat fraction and R2 star quantitative map to generate a deep learning model. Use the weights obtained in S4 to generate the corresponding PDFF and R2 star quantitative map.

[0040] In the specific implementation: We obtained ethics committee approval from Renji Hospital and retrospectively collected 436 abdominal MR imaging cases, all of which acquired T1w Dixon positive and negative water-fat phase images, DWI, PDFF, and R2 star sequences, and these case data came from MR devices from three manufacturers: Siemens, Philips, and UIH.

[0041] The four phase images of positive and negative water and fat and DWI are the routine sequences of liver MR imaging scans.

[0042] The PDFF value reflects the fat content, while the R2 star value reflects the degree of iron deposition. Fatty changes and iron deposition are the manifestations of common metabolic diseases of the liver on images. Due to the characteristics of the MR sequence technology principle, the two are usually inseparable in the form of a "twin combination", but to obtain PDFF and R2 star values, special sequences need to be used for acquisition.

[0043] Adopting Multi-branch Conformal Prediction Network (MBCP-Net) The preprocessed in-phase (IP), anti-phase (OP) and diffusion-weighted imaging (DWI) data are input into the multi-branch conformal prediction network (MBCP-Net), which can simultaneously generate quantitative proton density fat fraction (PDFF) and transverse relaxivity (R2 star) maps, and output the corresponding upper and lower bounds of the predicted values ​​and uncertainty maps.

[0044] The network is built on an improved 2D U-Net framework and has the following core features: Multi-branch feature extraction: IP, OP, and DWI trimodal data are processed through independent branches respectively. Each branch contains cascaded convolutional layers and pooling layers to complete primary feature extraction.

[0045] Cross-modal feature fusion: Deep fusion of three-branch features is achieved in the middle layer of the network, and multi-modal information is fully integrated through a cross-connection architecture.

[0046] Multi-scale feature enhancement: The skip connection mechanism is used in the upsampling stage to dynamically fuse the multi-scale features of different branches and effectively utilize the complementary information of multimodal data.

[0047] Introduce conformal prediction mechanism at the output stage: Three independent convolutional layers are configured for PDFF and R2 star maps to generate predicted values, upper bounds and lower bounds respectively. The predicted values ​​are strictly constrained within the upper and lower bounds. The interval span is visualized through the uncertainty map. The conformal quantile regression algorithm is used to dynamically adjust the confidence interval range to ensure statistical reliability.

[0048] The network training adopts a dual-loss collaborative optimization strategy: (1) Quantile regression loss: accurately constrains the confidence interval boundaries of the predicted values; (2) L1 loss: used to generate accurate pixel values.

[0049] in, represents the quantile to be optimized, represents the quantile estimator. The upper and lower bounds can be estimated by the and conditional quantiles. m is the weight coefficient of the PDFF part, and n is the weight coefficient of the R2 star part.

[0050] experiment For the entire dataset, data from 301 patients from UIH were used for training, validation, and calibration phases, while the remaining data were used as a test set, including 54 UIH subjects, 61 Siemens Healthcare (SIE) subjects, and 20 Philips Healthcare (PHI) subjects.

[0051] In the training phase (N=990 slices) and the validation phase (N=265 slices), 5 slices were randomly selected in each case as input for MBCP-Net. The model used the Adam optimizer with an initial learning rate of 0.001 and was trained for 500 epochs with a batch size of 2. To make the learning process more stable, a cosine learning rate scheduler was used for dynamic adjustment. In the loss function, the weights m=0.7 and n=0.3 balance the error parts of PDFF and R2 stars. In order to estimate the 90% uncertainty interval for each pixel, the model was trained to predict the 95% and 5% quantiles with α=10%. During the validation process, the peak signal-to-noise ratio (PSNR) metric was used to monitor the model performance and the best performing epoch results were saved to achieve the best quantitative prediction.

[0052] During the calibration phase, the model was fine-tuned using a calibration set of 250 slices to ensure that the predicted uncertainty intervals accurately covered pixels outside the expected quantile range. This included adjusting the model predictions so that pixels outside the prediction bounds were appropriately included in the uncertainty intervals.

[0053] In addition, in order to evaluate the role of DWI, we removed the DWI input branch in MBCP-Net for comparative experiments while keeping other conditions unchanged. At this time, the network only takes IP and OP images as input, so that we can evaluate the contribution of multimodal information in DWI images to model performance.

[0054] Automated trusted parameter extraction In order to evaluate the quality of the generated quantitative maps, we developed an automatic parameter extraction method that replaces the traditional manual region extraction work and uses the uncertainty quantification function of MBCP-Net to provide statistical confidence for the extracted parameters. This method abandons the tedious operation of manually annotating the region of interest in the traditional method, and through the uncertainty quantification results provided by MBCP-Net, it can provide a reliability assessment for the extracted parameters.

[0055] Specifically, we first segmented the liver region using TotalSegmentator to obtain the liver region mask. Next, we normalized the uncertainty map output by the model to map it to the range of [0,1]. By subtracting the normalized uncertainty value from 1, we obtained the credibility map, thereby converting uncertainty into a credibility metric. For each case, we selected the region with a credibility level of more than 90% to form the credibility mask. Finally, the parameter extraction region (i.e., the credible liver region) was defined as the intersection of the liver mask and the credibility mask, ensuring that we only considered highly reliable regions in the liver. In this region, we extracted the average values ​​of the PDFF and R2 star maps generated by the model and compared them with the standard reference values ​​of the qDixon method. Through this automated processing flow, we can not only obtain evaluation results with high statistical reliability, but also the repeatability of the entire process is well guaranteed. More importantly, no manual operation is required, which greatly improves research efficiency.

[0056] Statistical analysis To evaluate the performance of the proposed method, we mainly analyze two groups: (1) PDFF estimates generated by the MBCP-Net model using IP, OP, and DWI images as input (MBCP-Net PDFF With DWI for short), and (2) PDFF estimates generated by the MBCP-Net model using only IP and OP images as input (MBCP-Net PDFF No DWI for short), with qDixonPDFF as the reference standard.

[0057] When performing a quantitative comparison between MBCP-Net PDFF With DWI and MBCP-Net PDFF No DWI, we selected the intersection of two reliable liver regions to ensure that the two methods performed parameter extraction in the same anatomical region. In addition, according to the reference R2 star value, patients were divided into two categories: no or mild iron deposition (≤62.5) and moderate to severe iron deposition (>62.5). Subsequently, we performed a quantitative comparison between MBCP-Net PDFF With DWI and MBCP-Net PDFF No DWI in different iron deposition groups.

[0058] For the two sets of PDFF estimates, we used paired t-test and Wilcoxon signed-rank test to analyze their bias. In addition, we calculated the intraclass correlation coefficient (ICC) to evaluate the consistency between the PDFF estimates and the qDixon PDFF. At the same time, linear regression analysis and Bland-Altman analysis were also used to evaluate the consistency between the PDFF estimates and the qDixon PDFF.

[0059] All statistical analyses were performed at a significance level of p < 0.05. The results are presented as mean ± standard deviation. Statistical calculations were performed using Python's SciPy and statsmodels libraries to ensure the reproducibility of the results.

[0060] result In the mild iron deposition group, both MBCP-Net PDFF (With DWI) and MBCP-Net PDFF (No DWI) showed high consistency and reliability with qDixon PDFF, with the DWI group showing slightly better results. Specifically, the intraclass correlation coefficients (ICC) of PDFF were 0.976 (With DWI) and 0.949 (No DWI), respectively. Linear regression analysis further confirmed that With DWI performed better in maintaining the slope (1.026) and intercept (-0.200), while the slope of the No DWI group was 1.106 and the intercept was -1.201. Bland-Altman analysis also showed that the agreements limits of the With DWI group (-3.709 to 3.686) were tighter than those of the No DWI group (-5.188 to 4.792).

[0061] Table 1 is a comparison between MBCP-Net PDFF With DWI and MBCP-Net PDFF No DWI (using qDixonPDFF as the standard). For the moderate to severe iron deposition group, MBCP-Net PDFF (With DWI) showed a significant advantage. In this group, the ICC of With DWI reached 0.960, which was better than 0.949 in the No DWI group, reflecting a higher consistency. Linear regression analysis revealed that the slope of the With DWI group (0.987, intercept 1.080) was closer to the ideal value, indicating that the systematic bias was significantly reduced. Bland-Altman analysis further highlighted the advantage of With DWI in the moderate to severe iron deposition group, with a bias of 0.940% and tighter agreement limits (-7.242% to 9.121%), while the bias of the No DWI group was 2.224% and the agreement limits were (-6.605% to 11.052%). Paired t-test and Wilcoxon signed-rank test showed that in the moderate to severe iron deposition group, there were significant differences between the No DWI and With DWI groups (P<0.0001).

[0062] Table 1

[0063]

[0064] The consistency of qDixon PDFF with two methods (PDFF With DWI and PDFF No DWI) in different iron deposition test sets is shown in Figure 3-9 : Figure 3 A comparison diagram of the PDFF and R2 star quantitative graphs generated by the model of the present invention and the quantitative graphs obtained by the qDixon technology; Figure 4 are the real PDFF data distribution diagram and the real R2 star data distribution diagram, A is the real PDFF data distribution diagram in the training set and the test set, and B is the real R2 star data distribution diagram in the training set and the test set; Figure 5 is the regression analysis of MBCP-Net PDFF (No DWI) and MBCP-Net PDFF (With DWI) in the group without moderate iron deposition, in which qDixon PDFF was used as the reference standard; Figure 6 The same regression analysis was performed in the moderate to severe iron deposition group; Figure 7 Bland-Altman analysis of MBCP-Net PDFF (No DWI) and qDixon PDFF in the two iron deposition groups; Figure 8 Bland-Altman analysis of MBCP-Net PDFF (With DWI) and qDixon PDFF in two iron deposition groups; Fig. 9 The distribution of qDixonPDFF minus MBCP-Net PDFF (With DWI) and MBCP-Net PDFF (No DWI) in the moderate and severe iron deposition group is shown. Paired t-test and Wilcoxon signed-rank test showed that there was a significant difference between MBCP-Net PDFF (With DWI) and PDFF (No DWI).

[0065] The above describes the method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map in an embodiment of the present invention. The following describes the device for generating a magnetic resonance proton density fat fraction and R2 star quantitative map in an embodiment of the present invention: An acquisition module is used to acquire mDixon sequence image data and DWI sequence image data and corresponding qDixon sequence image data, and integrate the image data to form a data set; A preprocessing module, used to preprocess the collected image data, wherein the data preprocessing includes image standardization, image denoising and data amplification; A deep learning model building module is used to build a deep learning model for generating magnetic resonance proton density fat fraction and R2 star quantitative images from mDixon sequence images and DWI sequence images; A training module is used to train the deep learning model established by S3 using the image data preprocessed by S2 to obtain a trained deep learning model; The generation module is used to obtain the mDixon sequence image data and DWI sequence image data of the subject, input the trained deep learning model, and generate the proton density fat fraction and R2 star quantitative map.

[0066] An embodiment of the present invention also provides an electronic device, which may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) (for example, one or more processors) and memories, and one or more storage media for storing applications or data (for example, one or more mass storage devices). Among them, the memory and the storage medium may be short-term storage or permanent storage. The program stored in the storage medium may include one or more modules, each module may include a series of instruction operations in the electronic device. Furthermore, the processor may be configured to communicate with the storage medium and execute a series of instruction operations in the storage medium on the electronic device.

[0067] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input and output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that the electronic device structure in this embodiment does not constitute a limitation on the electronic device, and may include more or fewer components, or combine certain components, or arrange components differently.

[0068] An embodiment of the present invention provides a structure of an electronic device, which may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) (for example, one or more processors) and memories, and one or more storage media for storing applications or data (for example, one or more mass storage devices). Among them, the memory and the storage medium can be short-term storage or permanent storage. The program stored in the storage medium may include one or more modules, each module may include a series of instruction operations in the electronic device. Furthermore, the processor can be configured to communicate with the storage medium and execute a series of instruction operations in the storage medium on the electronic device.

[0069] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input and output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will appreciate that the electronic device structure does not constitute a limitation on the electronic device, and may include more or less components than the aforementioned, or combine certain components, or arrange the components differently.

[0070] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium, and when the instructions are executed on a computer, the computer executes the steps of the aforementioned method.

[0071] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0072] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0073] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map, characterized in that: The following steps are involved: S1, obtaining mDixon sequence image data, DWI sequence image data and qDixon sequence image data, and integrating the image data to form a data set; S2, preprocessing the collected image data, wherein the data preprocessing includes image standardization, image denoising and data amplification; S3, construct a deep learning model to generate magnetic resonance proton density fat fraction and R2 star quantitative map from mDixon sequence images and DWI sequence images; S4, using the image data pre-processed by S2, training the deep learning model established by S3 to obtain a trained deep learning model; S5. Obtain the mDixon sequence image data and DWI sequence image data of the subject, input the trained deep learning model, and generate the proton density fat fraction and R2 star quantitative map.

2. The method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map according to claim 1, characterized in that: In step S1, the mDixon sequence image is used to provide separation information of fat and water components. Four images are obtained through the mDixon sequence image, namely, in-phase image (IP), anti-phase image (OP), water and fat images; the qDixon sequence image data obtains accurate PDFF and R2 star quantitative images as the benchmark truth value for model training and evaluation.

3. The method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map according to claim 1, characterized in that: In step S2, the specific steps of image standardization, image denoising and data enhancement are as follows: Resize all images to the same resolution; Next, the Z-Score normalization method is applied to normalize the pixel values ​​of the image; Automatically segment the image, remove background noise, and retain only the target area for subsequent analysis; Random flipping, rotation, cropping, brightness adjustment, contrast changes, and random noise addition.

4. The method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map according to claim 1, characterized in that: In step S3, the deep learning model for generating magnetic resonance proton density fat fraction and R2 star quantitative image uses U-Net convolutional neural network as the basic architecture. The U-Net network structure includes multiple convolutional layers, pooling layers, upsampling layers and jump connection layers. The network adopts a multi-branch input and multi-channel output method.

5. The method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map according to claim 1, characterized in that: The U-Net model input includes multiple branches, which respectively receive IP and OP images from the mDixon sequence images and DWI sequence images. The input image is firstly subjected to feature extraction through a series of convolutional layers and pooling layers of each branch, and the features of each branch are fused in the middle layer of the network; in the upsampling stage, the multi-scale features of each branch are fused, and the final feature map is sent to two different convolutional layers, and each layer generates PDFF and R2 star quantitative maps.

6. The method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map according to claim 1, characterized in that: In step S4, the deep learning model training for generating magnetic resonance proton density fat fraction and R2 star quantitative image includes the following steps: During the training process, the model is iteratively trained using labeled training data; the labels include PDFF and R2 star quantitative maps from the qDixon technique, and the model gradually optimizes the network weights by minimizing the loss function between the predicted value and the true label; The cross-validation method was used to verify the generalization ability of the model.

7. A system for generating a magnetic resonance proton density fat fraction and R2 star quantitative map, characterized in that: The system comprises: An acquisition module is used to acquire mDixon sequence image data and DWI sequence image data and corresponding qDixon sequence image data, and integrate the image data to form a data set; A preprocessing module, used to preprocess the collected image data, wherein the data preprocessing includes image standardization, image denoising and data amplification; A deep learning model building module is used to build a deep learning model for generating magnetic resonance proton density fat fraction and R2 star quantitative images from mDixon sequence images and DWI sequence images; A training module is used to train the deep learning model established by S3 using the image data preprocessed by S2 to obtain a trained deep learning model; The generation module is used to obtain the mDixon sequence image data and DWI sequence image data of the subject, input the trained deep learning model, and generate the proton density fat fraction and R2 star quantitative map.

8. An electronic device, comprising a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute each step of the method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the various steps of the method for generating a magnetic resonance proton density fat fraction and R2 star quantitative map as described in any one of claims 1 to 6 are implemented.

Images