Method and system for MRI imaging in liver cancer prognosis based on IVIM-DWI

Abstract

The invention provides a method and a system for MRI (Magnetic Resonance Imaging) based on IVIM-DWI in liver cancer prognosis, and relates to the technical field of medical image analysis. The method comprises the steps of obtaining image data of a patient; an initial IVIM parameter of the IVIM-DWI imaging data is extracted by using an adaptive threshold segmentation fitting method; the initial IVIM parameters are optimized through an improved Levenberg-Marquardt algorithm, and the final IVIM parameters are obtained; performing region segmentation on the image data to obtain a segmented image of the target region; performing feature extraction on the segmented image of the target region by using a three-dimensional convolutional neural network to obtain a target region feature set; fusing the target region feature set to obtain a first fused feature set; performing feature extraction on the final IVIM parameters to obtain parameter feature vectors; splicing the parameter feature vector with the first fusion feature set to obtain a second fusion feature set; and inputting the second fusion feature set into a prognosis discrimination model to obtain a prognosis result. The problem of low prognosis evaluation accuracy is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image analysis, and in particular to a method and system for liver cancer prognosis based on IVIM-DWI MRI imaging. Background Art

[0002] Hepatocellular carcinoma (HCC) is one of the most common malignant tumors in the world. Its early diagnosis and prognostic assessment are of great significance to improving the survival rate of patients. Traditional prognostic assessment methods mainly rely on clinical indicators and imaging features, but due to the heterogeneity of tumors, the accuracy and individualization of these methods are limited, resulting in low accuracy of prognostic assessment. Summary of the invention

[0003] In order to overcome the deficiencies of the prior art, the object of the present invention is to provide a method and system for liver cancer prognosis based on IVIM-DWI MRI imaging.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for prognosis of liver cancer based on IVIM-DWI MRI imaging, comprising:

[0006] Acquiring imaging data of the patient, wherein the imaging data includes: IVIM-DWI imaging data, T1-weighted imaging data, and T2-weighted imaging data;

[0007] Extracting initial IVIM parameters of the IVIM-DWI imaging data using an adaptive threshold segmented fitting method;

[0008] The initial IVIM parameters are optimized using an improved Levenberg-Marquardt algorithm to obtain final IVIM parameters;

[0009] Performing region segmentation on the image data to obtain a segmented image of the target region;

[0010] Using a three-dimensional convolutional neural network to extract features from the segmented image of the target area to obtain a target area feature set;

[0011] Fusing the target area feature set to obtain a first fused feature set;

[0012] Performing feature extraction on the final IVIM parameters to obtain a parameter feature vector;

[0013] Concatenating the parameter feature vector with the first fused feature set to obtain a second fused feature set;

[0014] The second fused feature set is input into the prognosis discrimination model to obtain a prognosis result.

[0015] Preferably, the step of extracting the initial IVIM parameters of the IVIM-DWI imaging data using an adaptive threshold segmented fitting method comprises:

[0016] Performing overall fitting on the IVIM-DWI imaging data to obtain overall IVIM parameters;

[0017] Determining a parameter fitting curve according to the overall IVIM parameters;

[0018] Determine the signal change rate of each parameter according to the parameter fitting curve;

[0019] Determining a threshold value according to the rate of change of the signal;

[0020] According to the threshold, high-value area parameter extraction and low-value area parameter extraction, low-value parameters and high-value parameters are realized;

[0021] The initial IVIM parameters are obtained according to the low value parameters and the high value parameters.

[0022] Preferably, the expression of the overall IVIM parameter is:

[0023] S(b)=S0[fexp(-bD * )+(1-f)exp(-bD)];

[0024] Where S(b) is the signal intensity when b is b, S0 is the signal intensity when b is 0, f is the perfusion fraction, D is the pure diffusion coefficient, and D * is the pseudo-diffusion coefficient and b is the gradient factor.

[0025] Preferably, the expression of the parameter fitting curve is:

[0026]

[0027] Among them, S fit (b) is the value of the fitting curve corresponding to the b value, exp is the exponential function, D total 、f total and They are the pure diffusion coefficient, perfusion fraction and pseudo-diffusion coefficient in the overall parameters, respectively.

[0028] Preferably, the expression of the final IVIM parameter is:

[0029]

[0030] Among them, θ finalis the final IVIM parameter, θ is the parameter vector, r is the residual vector, S obs,i is the observed signal under the i-th b value.

[0031] Preferably, performing region segmentation on the image data to obtain a segmented image of the target region includes:

[0032] Preprocessing the image data to obtain preprocessed data;

[0033] Inputting the preprocessed data into an improved three-dimensional U-Net model to obtain a segmented image of the target area;

[0034] Among them, the improved three-dimensional U-Net model is based on the traditional U-Net model, including a symmetrical encoder and decoder structure; using three-dimensional depth-separable convolution instead of standard convolution; adding residual connections between the layers of the encoder and decoder; and adding an attention module at the jump connection of U-Net.

[0035] Preferably, the expression of the prognostic discriminant model is:

[0036]

[0037] in, For prognostic results, is the prognostic discriminant model function, F fusion is the second fusion feature set, and Θ is the parameter set of the model.

[0038] A system for liver cancer prognosis based on IVIM-DWI MRI imaging, comprising:

[0039] A data acquisition module, used to acquire the patient's imaging data, wherein the imaging data includes: IVIM-DWI imaging data, T1-weighted imaging data and T2-weighted imaging data;

[0040] A first extraction module is used to extract initial IVIM parameters of the IVIM-DWI imaging data using an adaptive threshold segmented fitting method;

[0041] A second extraction module is used to optimize the initial IVIM parameters using an improved Levenberg-Marquardt algorithm to obtain final IVIM parameters;

[0042] A segmentation module, used to perform regional segmentation on the image data to obtain a segmented image of the target area;

[0043] A third extraction module is used to extract features from the segmented image of the target area using a three-dimensional convolutional neural network to obtain a target area feature set;

[0044] A first fusion module, used for fusing the target area feature set to obtain a first fused feature set;

[0045] A fourth extraction module, used for performing feature extraction on the final IVIM parameters to obtain a parameter feature vector;

[0046] A second fusion module, used for concatenating the parameter feature vector with the first fusion feature set to obtain a second fusion feature set;

[0047] The prognosis module is used to input the second fusion feature set into the prognosis discrimination model to obtain a prognosis result.

[0048] The present invention discloses the following technical effects:

[0049] The present invention provides a method and system for liver cancer prognosis based on IVIM-DWI MRI imaging, the method comprising: acquiring patient imaging data, wherein the imaging data comprises: IVIM-DWI imaging data, T1-weighted imaging data and T2-weighted imaging data; extracting initial IVIM parameters of the IVIM-DWI imaging data by using an adaptive threshold segmented fitting method; optimizing the initial IVIM parameters by using an improved Levenberg-Marquardt algorithm to obtain final IVIM parameters; performing regional segmentation on the imaging data to obtain a segmented image of a target region; performing feature extraction on the segmented image of the target region by using a three-dimensional convolutional neural network to obtain a target region feature set; fusing the target region feature set to obtain a first fused feature set; performing feature extraction on the final IVIM parameters to obtain a parameter feature vector; splicing the parameter feature vector with the first fused feature set to obtain a second fused feature set; inputting the second fused feature set into a prognosis discrimination model to obtain a prognosis result. The present invention combines the functional and morphological information of liver cancer by integrating the quantitative parameters of IVIM-DWI with the deep features of multimodal MRI images, optimizes the accuracy of IVIM parameter extraction by using an improved algorithm, and extracts high-dimensional features of the target area through a three-dimensional convolutional neural network, and finally constructs an accurate prognostic discrimination model. The most important beneficial effect is that this method significantly improves the accuracy and reliability of liver cancer prognosis assessment, provides a scientific basis for the formulation of individualized treatment plans, and has important clinical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0051] Figure 1 A flow chart of a method for using IVIM-DWI-based MRI imaging to predict liver cancer prognosis is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] like Figure 1 As shown, the present invention provides a method for liver cancer prognosis based on IVIM-DWI MRI imaging, comprising:

[0055] Step 100: Acquire imaging data of a patient, wherein the imaging data includes: IVIM-DWI imaging data, T1-weighted imaging data, and T2-weighted imaging data;

[0056] Step 200: extracting initial IVIM parameters of the IVIM-DWI imaging data using an adaptive threshold segmented fitting method;

[0057] Step 300: Optimizing the initial IVIM parameters using an improved Levenberg-Marquardt algorithm to obtain final IVIM parameters;

[0058] Step 400: performing region segmentation on the image data to obtain a segmented image of the target region;

[0059] Step 500: extracting features from the segmented image of the target area using a three-dimensional convolutional neural network to obtain a target area feature set;

[0060] Step 600: fusing the target region feature set to obtain a first fused feature set;

[0061] Step 700: extracting features of the final IVIM parameters to obtain parameter feature vectors;

[0062] Step 800: concatenate the parameter feature vector with the first fused feature set to obtain a second fused feature set;

[0063] Step 900: Input the second fusion feature set into a prognosis discrimination model to obtain a prognosis result.

[0064] Furthermore, the method of extracting the initial IVIM parameters of the IVIM-DWI imaging data using an adaptive threshold segmented fitting method includes:

[0065] Performing overall fitting on the IVIM-DWI imaging data to obtain overall IVIM parameters;

[0066] Determining a parameter fitting curve according to the overall IVIM parameters;

[0067] Determine the signal change rate of each parameter according to the parameter fitting curve;

[0068] Determining a threshold value according to the rate of change of the signal;

[0069] According to the threshold, high-value area parameter extraction and low-value area parameter extraction, low-value parameters and high-value parameters are realized;

[0070] The initial IVIM parameters are obtained according to the low value parameters and the high value parameters.

[0071] Furthermore, the expression of the overall IVIM parameter is:

[0072] S(b)=S0[fexp(-bD * )+(1-f)exp(-bD)];

[0073] Where S(b) is the signal intensity when b is b, S0 is the signal intensity when b is 0, f is the perfusion fraction, D is the pure diffusion coefficient, and D * is the pseudo-diffusion coefficient and b is the gradient factor.

[0074] Furthermore, the expression of the parameter fitting curve is:

[0075]

[0076] Among them, S fit (b) is the value of the fitting curve corresponding to the b value, exp is the exponential function, D total 、f total and They are the pure diffusion coefficient, perfusion fraction and pseudo-diffusion coefficient in the overall parameters, respectively.

[0077] Specifically, by analyzing the curve of signal change rate versus b value, the b value with an inflection point or a significant change in change rate is found as a threshold. The threshold is used to distinguish between perfusion-dominated areas and pure diffusion-dominated areas.

[0078] In the low b-value region, the signal is significantly affected by the perfusion effect, and a biexponential model is used for fitting. In the high b-value region, the perfusion effect can be ignored and is simplified to a single exponential model.

[0079] Furthermore, the expression of the final IVIM parameter is:

[0080]

[0081] Among them, θ final is the final IVIM parameter, θ is the parameter vector, r is the residual vector, S obs,i is the observed signal under the i-th b value.

[0082] Specifically, the Levenberg-Marquardt algorithm is an optimization algorithm for nonlinear least squares fitting, which combines the advantages of the gradient descent method and the Gauss-Newton method. However, when dealing with the IVIM model, the standard LM algorithm may converge slowly or fall into a local minimum due to the high nonlinearity and correlation between parameters.

[0083] This expression indicates that the final IVIM parameters are obtained by minimizing the square difference between the observed signal and the model predicted signal using the improved LM algorithm.

[0084] Furthermore, the image data is segmented into regions to obtain a segmented image of the target region, including:

[0085] Preprocessing the image data to obtain preprocessed data;

[0086] Image resampling and normalization: Resample the image data to a uniform voxel size to ensure spatial consistency. For example, resample all image data to a voxel size of 1×1×11×1×1 mm.

[0087] Standardization: Perform intensity standardization on the image data, such as zero mean unit variance standardization, to eliminate the effects of different scanners and imaging parameters.

[0088] Use non-local mean filtering (Non-local Means Filtering) or wavelet denoising to reduce image noise and improve segmentation accuracy.

[0089] Inputting the preprocessed data into an improved three-dimensional U-Net model to obtain a segmented image of the target area;

[0090] Among them, the improved three-dimensional U-Net model is based on the traditional U-Net model, including a symmetrical encoder and decoder structure; using three-dimensional depth-separable convolution instead of standard convolution; adding residual connections between the layers of the encoder and decoder; and adding an attention module at the jump connection of U-Net.

[0091] Specifically, the 3D segmentation network based on deep learning:

[0092] Encoder-decoder structure: retains the symmetric structure of the classic U-Net to facilitate feature fusion.

[0093] Depthwise separable convolution: Use three-dimensional depthwise separable convolution to reduce the number of parameters and computation, and speed up training and inference.

[0094] Residual connection: Introduce residual connection to alleviate the gradient vanishing problem of deep networks.

[0095] Self-Attention Module: Add a self-attention module in the middle layer of the encoder and decoder to emphasize the feature representation of key areas.

[0096] Channel and Spatial Attention (CBAM): Combining channel attention and spatial attention to enhance the model’s ability to learn important features.

[0097] Joint segmentation and classification

[0098] Segmentation task: locate the precise position of the target area.

[0099] Classification task: Determine the pathological characteristics of the target area, such as benign or malignant.

[0100] Shared features: Utilize the shared encoder to promote information sharing between different tasks and improve the overall performance of the model.

[0101] Furthermore, the three-dimensional convolutional neural network is used to extract features from the segmented image of the target area to obtain a target area feature set, including:

[0102] Normalizing the segmented image of the target area to obtain a standardized sub-image of the target area;

[0103] Inputting the target area sub-image into the constructed 3D-CNN model;

[0104] The 3D-CNN model includes: multiple layers of three-dimensional convolutional layers, pooling layers and fully connected layers, and each convolutional layer is followed by batch normalization and activation functions.

[0105] Furthermore, the expression of the prognostic discriminant model is:

[0106]

[0107] in, For prognostic results, is the prognostic discriminant model function, F fusion is the second fusion feature set, and Θ is the parameter set of the model.

[0108] Specifically, the first fusion feature set (Fimage):

[0109] Source: Feature extraction of segmented images of target areas using a three-dimensional convolutional neural network (3D-CNN).

[0110] Composition: Fusion of deep features extracted from multimodal images (T1-weighted, T2-weighted, IVIM-DWI).

[0111] Fimage=[fT1,fT2,fIVIM];

[0112] Parameter feature vector (FIVIM):

[0113] Source: Feature extraction of final IVIM parameters.

[0114] Composition: Includes key parameters in the IVIM model, reflecting the microscopic diffusion and perfusion properties of the tissue.

[0115] Construct the second fusion feature set:

[0116] The first fusion feature set is concatenated with the parameter feature vector to form the final feature set for prognostic discrimination.

[0117] The specific model expression can be a deep neural network, including the following parts:

[0118] Input Layer:

[0119] Receive F fusion As input, hidden layer: multiple layers of nonlinear transformations are used to extract high-order features and patterns.

[0120] Layer 1: h1 = σ(W1F fusion +b1)

[0121] Layer 2: h2 = σ(W2h1 + b2)

[0122] Parameter explanation: W i is the weight matrix of the i-th layer. i is the bias vector of the i-th layer. σ(·) is the activation function, such as ReLU, Sigmoid, etc.

[0123] Output layer:

[0124] The output of the last layer is transformed linearly or nonlinearly to obtain the final prognosis prediction result.

[0125] expression:

[0126]

[0127] Specifically, W out represents the weight matrix of the output layer, which is used to map the feature representation of the hidden layer to the final prediction result. h1h2 represents the output of the hidden layer, which is the intermediate result after the input feature undergoes layer-by-layer nonlinear transformation. out is an additive term.

[0128] This embodiment also provides a system for liver cancer prognosis based on IVIM-DWI MRI imaging, including:

[0129] A data acquisition module, used to acquire the patient's imaging data, wherein the imaging data includes: IVIM-DWI imaging data, T1-weighted imaging data and T2-weighted imaging data;

[0130] A first extraction module is used to extract initial IVIM parameters of the IVIM-DWI imaging data using an adaptive threshold segmented fitting method;

[0131] A second extraction module is used to optimize the initial IVIM parameters using an improved Levenberg-Marquardt algorithm to obtain final IVIM parameters;

[0132] A segmentation module, used to perform regional segmentation on the image data to obtain a segmented image of the target area;

[0133] A third extraction module is used to extract features from the segmented image of the target area using a three-dimensional convolutional neural network to obtain a target area feature set;

[0134] A first fusion module, used for fusing the target area feature set to obtain a first fused feature set;

[0135] A fourth extraction module, used for performing feature extraction on the final IVIM parameters to obtain a parameter feature vector;

[0136] A second fusion module, used for concatenating the parameter feature vector with the first fusion feature set to obtain a second fusion feature set;

[0137] The prognosis module is used to input the second fusion feature set into the prognosis discrimination model to obtain a prognosis result.

[0138] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0139] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for the prognosis of liver cancer based on IVIM-DWI MRI imaging, characterized in that: include: Acquiring imaging data of the patient, wherein the imaging data includes: IVIM-DWI imaging data, T1-weighted imaging data, and T2-weighted imaging data; Extracting initial IVIM parameters of the IVIM-DWI imaging data using an adaptive threshold segmented fitting method; The initial IVIM parameters are optimized using an improved Levenberg-Marquardt algorithm to obtain final IVIM parameters; Performing region segmentation on the image data to obtain a segmented image of the target region; Using a three-dimensional convolutional neural network to extract features from the segmented image of the target area to obtain a target area feature set; Fusing the target area feature set to obtain a first fused feature set; Performing feature extraction on the final IVIM parameters to obtain a parameter feature vector; Concatenating the parameter feature vector with the first fused feature set to obtain a second fused feature set; The second fused feature set is input into the prognosis discrimination model to obtain a prognosis result.

2. The method for liver cancer prognosis based on IVIM-DWI MRI imaging according to claim 1, characterized in that: The method of extracting initial IVIM parameters of the IVIM-DWI imaging data by using an adaptive threshold segmented fitting method includes: Performing overall fitting on the IVIM-DWI imaging data to obtain overall IVIM parameters; Determining a parameter fitting curve according to the overall IVIM parameters; Determine the signal change rate of each parameter according to the parameter fitting curve; Determining a threshold value according to the rate of change of the signal; According to the threshold, high-value area parameter extraction and low-value area parameter extraction, low-value parameters and high-value parameters are realized; The initial IVIM parameters are obtained according to the low value parameters and the high value parameters.

3. The method for liver cancer prognosis based on IVIM-DWI MRI imaging according to claim 2, characterized in that: The expression of the overall IVIM parameter is: S(b)=S0[fexp(-bD * )+(1-f)exp(-bD)]; Where S(b) is the signal intensity when b is b, S0 is the signal intensity when b is 0, f is the perfusion fraction, D is the pure diffusion coefficient, and D * is the pseudo-diffusion coefficient and b is the gradient factor.

4. The method for liver cancer prognosis based on IVIM-DWI MRI imaging according to claim 3, characterized in that: The expression of the parameter fitting curve is: Among them, S fit (b) is the value of the fitting curve corresponding to the b value, exp is the exponential function, D total 、f total and They are the pure diffusion coefficient, perfusion fraction and pseudo-diffusion coefficient in the overall parameters, respectively.

5. The method for liver cancer prognosis based on IVIM-DWI MRI imaging according to claim 4, characterized in that: The expression of the final IVIM parameter is: Among them, θ final is the final IVIM parameter, θ is the parameter vector, r is the residual vector, S obs,i is the observed signal under the i-th b value.

6. The method for liver cancer prognosis based on IVIM-DWI MRI imaging according to claim 4, characterized in that: Performing region segmentation on the image data to obtain a segmented image of the target region includes: Preprocessing the image data to obtain preprocessed data; Inputting the preprocessed data into an improved three-dimensional U-Net model to obtain a segmented image of the target area; Among them, the improved three-dimensional U-Net model is based on the traditional U-Net model, including a symmetrical encoder and decoder structure; using three-dimensional depth-separable convolution instead of standard convolution; adding residual connections between the layers of the encoder and decoder; and adding an attention module at the jump connection of U-Net.

7. The method for liver cancer prognosis based on IVIM-DWI MRI imaging according to claim 1, characterized in that: The method of extracting features from the segmented image of the target area using a three-dimensional convolutional neural network to obtain a target area feature set includes: Normalizing the segmented image of the target area to obtain a standardized sub-image of the target area; Inputting the target area sub-image into the constructed 3D-CNN model; The 3D-CNN model includes: multiple layers of three-dimensional convolutional layers, pooling layers and fully connected layers, and each convolutional layer is followed by batch normalization and activation functions.

8. The method for liver cancer prognosis based on IVIM-DWI MRI imaging according to claim 1, characterized in that: The expression of the prognostic discriminant model is: in, For prognostic results, is the prognostic discriminant model function, F fusion is the second fusion feature set, and Θ is the parameter set of the model.

9. A system for liver cancer prognosis based on IVIM-DWI MRI imaging, characterized in that: include: A data acquisition module, used to acquire the patient's imaging data, wherein the imaging data includes: IVIM-DWI imaging data, T1-weighted imaging data and T2-weighted imaging data; A first extraction module is used to extract initial IVIM parameters of the IVIM-DWI imaging data using an adaptive threshold segmented fitting method; A second extraction module is used to optimize the initial IVIM parameters using an improved Levenberg-Marquardt algorithm to obtain final IVIM parameters; A segmentation module, used to perform regional segmentation on the image data to obtain a segmented image of the target area; A third extraction module is used to extract features from the segmented image of the target area using a three-dimensional convolutional neural network to obtain a target area feature set; A first fusion module, used for fusing the target area feature set to obtain a first fused feature set; A fourth extraction module, used for performing feature extraction on the final IVIM parameters to obtain a parameter feature vector; A second fusion module, used for concatenating the parameter feature vector with the first fusion feature set to obtain a second fusion feature set; The prognosis module is used to input the second fusion feature set into the prognosis discrimination model to obtain a prognosis result.

Images