A method for constructing a magnetic resonance diffusion multi-component spectroscopy analysis model for tumor assessment

Abstract

The present invention relates to a method for constructing a magnetic resonance diffusion multi-component spectrum analysis model for tumor assessment, specifically comprising: 1) acquiring DWI images with multiple b values ​​and multiple diffusion directions; 2) denoising the DWI image signals; 3) delineating the tumor region of interest (ROI) and the corresponding healthy tissue ROI, and obtaining the signal values ​​in the ROI with different b values; 4) fitting the multi-component spectrum analysis model for the signal values ​​of the voxels in the delineated ROI, and constructing models of different component spectra; 5) using BIC to analyze and compare the different component spectrum models. The present invention takes into account that water molecules in a biological body are constrained by the surrounding medium, and their diffusion motion is restricted. The diffusion signal is modeled as a mixture reflecting different diffusion components, and multiple fixed diffusion coefficients ( D i ) values ​​and the corresponding different D i Compared with the traditional single exponential attenuation DWI method, the model constructed in the present invention has a better fitting goodness of fit to the DWI signal and is more conducive to distinguishing benign and malignant tumors.

Description

Technical Field

[0001] The invention belongs to the field of image processing and medical technology, and relates to tumor assessment using a magnetic resonance diffusion weighted imaging multi-component spectrum analysis method, and specifically is a method for constructing a magnetic resonance diffusion multi-component spectrum analysis model applied to tumor assessment. Background Art

[0002] Diffusion-weighted imaging (DWI) is a functional magnetic resonance imaging technology that can detect the diffusion motion of water molecules in human tissues. It is also the only in vivo imaging technology that can reflect the motion state of water molecules in tissues. DWI can indirectly detect changes in the microstructure inside tumor tissues by monitoring changes in the free diffusion motion state of water molecules in the intracellular and extracellular environments, thereby reflecting the cell structure and microenvironment. Generally, malignant tumor cells are denser in structure than normal tissues, and have more cancerous interstitial components, which leads to a decrease in the diffusion motion space of water molecules and a slowing of the diffusion motion, which appears as a high signal on the DWI image. The single exponential attenuation DWI model is currently commonly used in clinical applications. It is assumed that human tissue is a single uniform substance and the diffusion motion of molecules follows a Gaussian distribution, so that the DWI signal attenuation satisfies:

[0003] S b =S0*exp(-b·ADC) (1)

[0004] Among them, is the DWI signal value when the diffusion sensitivity coefficient value is b, is the DWI signal value when the diffusion sensitivity coefficient value is 0 (no diffusion sensitivity gradient), and ADC is the apparent diffusion coefficient (ADC), which reflects the average diffusion coefficient within the voxel. By acquiring two DWI images with different diffusion sensitivity coefficient values ​​and fitting the DWI image signal values ​​of the two b values ​​using formula (1), the ADC value can be obtained as a quantitative functional parameter for distinguishing benign and malignant tumors. Conventional DWI shows that the characteristic of cancerous lesions is a decrease in ADC. However, the expected change in ADC accompanying cancerous lesions may be confused with edema or necrosis and may not be detected from ADC alone.

[0005] In addition, traditional DWI based on ADC images also has some significant limitations, including: the model is based on the assumption that human tissue is a single uniform structure. In fact, due to the complex structure of human tissue, the single index model can only obtain the average diffusion coefficient within the voxel, and cannot distinguish between extracellular and intracellular water molecules, and cannot provide more information reflecting changes in composition and structure. In addition, studies have found that 1) ADC generally collects low b values ​​(b < 1000s / mm 2) is obtained by fitting the DWI signal, but as the b value increases, the signal intensity change of the DWI image does not follow the law of single exponential decay; 2) The ADC value is sensitive to the b value, which also limits the consistency of the ADC value. Summary of the invention

[0006] In order to overcome the limitations of the single exponential model DWI, the present invention adopts a DWI multi-component spectrum analysis method, a linear mixed model. It is characterized in that the DWI signal is decomposed into the signal from restricted water molecules, blocked water molecules and free water molecules. The present invention includes the following steps:

[0007] A method for constructing a magnetic resonance diffusion multi-component spectral analysis model for tumor assessment, the method assuming that the tissue in a single voxel is not uniform and contains multiple diffusion components, and using a multi-component spectral analysis method, specifically comprising the following steps:

[0008] Step 1: Use at least three gradient directions and multiple b values ​​to collect diffusion-weighted imaging (DWI) data; the b value is the diffusion weighting coefficient, and the number of b values ​​is at least the number of spectra in the multi-component spectrum analysis. In order to obtain non-Gaussian diffusion signals, the maximum b value is at least 1000 s / mm 2 ;

[0009] Step 2: Remove Rician noise from DWI data;

[0010] Step 3: At b = 1000s / mm 2 The tumor area and the healthy tissue corresponding to the tumor area are delineated on the DWI image, and the delineated area is copied to the DWI image with different b values ​​to obtain the signal value of the same delineated area at different b values;

[0011] Step 4: Model fitting is performed by minimizing the difference between the signal values ​​of all delineated areas at different b values ​​and the DWI signal values ​​of the magnetic resonance diffusion multi-component spectrum analysis model to be constructed when the diffusion weighting coefficient is b. The values ​​of the diffusion spectrum components are calculated using the non-negative least squares method to construct a magnetic resonance diffusion multi-component spectrum analysis model for all voxels in the region of interest.

[0012] Step 5: Evaluate the goodness of fit of each model and calculate the Bayesian information criterion (BIC) for each model. By comparison, the model with the lowest BIC value is the best fitting model, which is the magnetic resonance diffusion multi-component spectroscopy analysis model finally adopted.

[0013] Step 2 is to remove Rician noise from the DWI data, specifically:

[0014]

[0015] Where S raw is the original DWI signal, δ n is the standard deviation of the real or imaginary part of the noise, S corr is the DWI signal after Rician noise removal; S noise is the background noise intensity, so in the calculation, use:

[0016]

[0017] As the DWI signal data after Rician noise removal.

[0018] A fixed diffuse spectral component value D is used for all regions of interest. i , the volume fractions of tumor tissue and normal tissue corresponding to different diffusion coefficient values ​​in voxels are different. The magnetic resonance diffusion multi-component spectrum analysis model described in step 4 is expressed as:

[0019]

[0020] And meet:

[0021] Among them, S b is the DWI signal value when the diffusion weighting coefficient is b, S0 is the DWI signal value when the diffusion weighting coefficient is 0, n is the number of multi-component spectra, C i is the volume ratio of the i-th diffusion spectrum component in the voxel, D i is the value of the i-th diffusion spectrum component, and b is the diffusion weighting coefficient.

[0022] To evaluate the goodness of fit of each model, the Bayesian information criterion (BIC) was calculated for each multicomponent spectral analysis model. The BIC described in step 5 is expressed as:

[0023]

[0024] Where N is the sample size, n is the number of tissue multi-component spectra in the model, and r is the residual sum of squares of the model fitting; the BIC of each multi-component spectrum analysis model corresponds to ΔBIC:

[0025] ΔBIC=BIC i -BIC min

[0026] BIC i is the BIC value of any multi-component spectral analysis model, BIC minThe minimum BIC observed for all multi-component spectral analysis models; the optimal model is selected by comparing the size of the ΔBIC value, and the optimal model selects the model with the smallest ΔBIC value.

[0027] Different from the traditional single-index DWI method that assumes uniform tissue structure and single diffusion coefficient within the voxel, the present invention takes into account that water molecules in the body are constrained by the surrounding medium and their diffusion motion is subject to certain restrictions. The diffusion signal is modeled as a mixture reflecting different diffusion components. By adopting a fixed, global optimal D i The multi-component spectral analysis method takes into account the diffusion properties of each voxel of the tissue, regardless of whether it is cancerous or not, and obtains a fixed D i Values ​​and each voxel corresponds to a different D i The volume fraction of the single exponential model is thus overcome, and the ADC value is sensitive to the b value, which leads to the inconsistency of the ADC value; the signal intensity change of the high b value DWI image does not follow the law of single exponential decay; and the signal is decomposed into restricted water molecules, obstructed water molecules and free water molecules, so the new multi-component spectrum analysis method is more conducive to distinguishing benign and malignant tumors. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of the present invention;

[0029] Figure 2 This is a comparison diagram of the fitting residuals of the optimal four-component spectrum analysis model and the single exponential decay DWI model through BIC analysis in the embodiment of the present invention applied to the breast. DETAILED DESCRIPTION

[0030] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0031] See also Figure 1 The present invention comprises the following specific steps:

[0032] (1) Use multiple gradient directions and multiple b-values ​​to acquire diffusion-weighted imaging (DWI) data;

[0033] (2) Remove Rician noise from DWI data;

[0034]

[0035] Where S raw is the original DWI signal, S noise is the background noise intensity, S corr It is the DWI signal after Rician noise removal;

[0036] (3) At b = 1000s / mm 2The tumor area and the healthy tissue corresponding to the tumor area are delineated on the DWI image of the same b value, and the delineated area is copied to other DWI images with different b values ​​to obtain the signal values ​​of the same delineated area at different b values;

[0037] (4) construct a multi-component spectral analysis model for all voxels within the ROI, using a multi-component spectral analysis method with a fixed, globally optimal diffusion coefficient value;

[0038]

[0039] And meet:

[0040] Among them, S b is the DWI signal value when the diffusion weighting coefficient is b, S0 is the DWI signal value when the diffusion weighting coefficient is 0, n is the number of multi-component spectra, C i is the volume ratio of the i-th diffusion spectrum component in the voxel, D i is the value of the i-th diffusion spectrum component, and b is the diffusion weighting coefficient.

[0041] (5) Evaluate the goodness of fit of each model and calculate the Bayesian Information Criterion (BIC) for each model. The model with the lowest value is the best fitting model.

[0042]

[0043] Where N is the sample size, n is the number of tissue multi-component spectra in the model, and r is the residual sum of squares of the model fitting. In addition, the relative BIC (ΔBIC) of each multi-component spectra model is:

[0044] ΔBIC=BIC i -BIC min

[0045] BIC i is the BIC value of any multi-component spectral analysis model, BIC min is the minimum BIC observed for all multicomponent spectral analysis models studied.

[0046] Example

[0047] 1.1 Multi-b-value diffusion data acquisition

[0048] In this embodiment, the data used are derived from 37 breast DWI data collected by a 3T magnetic resonance device (Siemens 3T MAGNETOM Prisma). The acquisition parameters are as follows: DWI images with 7 b values ​​are collected in 6 diffusion gradient directions, and the b values ​​are b=0, 250 (6), 500 (6), 1000 (6), 1500 (6), 2000 (6), 3000 (6) s / mm2 , Repetition time (TR) / echo time (TE) = 6000ms / 93ms, FOV = 166*380mm 2 , acquisition matrix size: 84×192, layer thickness: 5 mm (the parameters can be adjusted according to the specific scanning area).

[0049] 1.2 Data Denoising

[0050] Since the noise in MRI will cause deviation in the acquired DWI signal, the signal intensity in the diffusion data was first calibrated to eliminate noise interference. The specific process is as follows:

[0051] 1) Estimate the average background signal intensity of each voxel;

[0052] 2) The observed signal intensity of each voxel minus the average background signal intensity, and the corrected signal intensity S is obtained using the following formula: cor r;

[0053]

[0054] Where S raw is the original DWI signal intensity, S noise is the average background noise intensity, S corr is the DWI signal intensity after removing Rician noise;

[0055] 3) Assuming that the diffusion of tissue in the region of interest is isotropic, the DWI image signal intensities in the six diffusion gradient directions acquired at each b value are averaged to obtain the DWI signal intensities at different b values.

[0056] 1.3 Establishment of diffusion model for multi-component spectrum analysis

[0057] The model of this embodiment is defined by the following formula:

[0058]

[0059] Where b is the diffusion weighting coefficient, S corr (b) represents the DWI signal when the diffusion weighting coefficient value is b after noise correction, n is the number of multi-component spectra, D i is the value of the i-th diffuse spectrum component, C i is the volume ratio of the i-th diffusion spectrum component in the voxel. i The values ​​are sorted from low to high, so the first diffusion spectrum has the lowest value, indicating restricted diffusion. To determine the optimal value of n and D for the region of interest iTo obtain the value, it is necessary to perform global fitting on the DWI signals of all regions of interest (including cancer tissues and all normal control tissues). corr (b) Divide by S corr (0) Signal normalization is performed. The value range of n is usually set to 2 to 5 (the number of diffusion spectra is less than the number of b values ​​used for image acquisition). For each value of n, this embodiment performs model fitting by minimizing the difference between the observed signal value and the signal value obtained by model fitting, and uses non-negative least squares method to calculate the global D i value, and the D i The corresponding C i value.

[0060] 1.4 Model testing and verification

[0061] To evaluate the degree of agreement between various multi-component spectral analysis models and DWI data and determine the optimal decision model, the Bayesian Information Criterion (BIC) was calculated for each model as follows:

[0062]

[0063] Where N is the sample size, n is the number of multicomponent spectra, and r is the residual sum of squares of the model fit. In addition, the relative BIC (ΔBIC) of each multicompartment diffusion model was calculated according to the following formula:

[0064] ΔBIC=BIC i -BIC min

[0065] BIC i is the BIC value of a specific model, BIC min The minimum BIC observed for any model. Paired t-test (α = 0.05) was used to compare whether the fitting residual of the multi-component spectrum analysis model was smaller than the fitting residual of the conventional single attenuation exponential DWI model. Two-sample t-test (α = 0.05) was used to compare the volume fractions of the multi-component spectrum between normal tissue and tumor.

[0066] In this embodiment, the BIC value is the smallest when the number of multi-component spectra is 4, and the ΔBIC values ​​are 14.91, 2.80, 0, and 2.98 respectively when the number of multi-component spectra is 2, 3, 4, and 5. For this embodiment, the multi-component spectrum analysis model with n=4 is the optimal multi-component spectrum analysis model.

[0067] The above embodiment is only an application of the invention and does not limit the scope of application of the invention. Through the analysis of the embodiment, compared with the traditional single exponential attenuation DWI model, the new multi-component spectrum analysis model has a better fit to the DWI signal and is more conducive to distinguishing benign and malignant tumors. The present invention is not limited to a specific body part and can be widely used in tumor imaging in other parts.

Claims

1. A method for constructing a magnetic resonance diffusion multi-component spectroscopy analysis model for tumor assessment, characterized in that: The method specifically comprises the following steps: Step 1: Use at least three gradient directions and multiple b values ​​to collect diffusion-weighted imaging (DWI) data; the b value is the diffusion weighting coefficient, and the number of b values ​​is at least the number of spectra in the multi-component spectrum analysis. In order to obtain non-Gaussian diffusion signals, the maximum b value is at least 1000 s / mm 2 ; Step 2: Remove Rician noise from DWI data; Step 3: At b = 1000s / mm 2 The tumor area and the healthy tissue corresponding to the tumor area are delineated on the DWI image, and the delineated area is copied to the DWI image with different b values ​​to obtain the signal value of the same delineated area at different b values; Step 4: The model is fitted by minimizing the difference between the signal values ​​of all delineated areas at different b values ​​and the DWI signal value of the magnetic resonance diffusion multi-component spectrum analysis model to be constructed when the diffusion weighting coefficient is b, and the diffusion spectrum component value D is calculated using the non-negative least squares method. i , construct a multi-component spectral analysis model of magnetic resonance diffusion for all voxels in the region of interest; Step 5: Evaluate the goodness of fit of each model and calculate the Bayesian information criterion (BIC) for each model. By comparison, the model with the lowest BIC value is the best fitting model, which is the magnetic resonance diffusion multi-component spectroscopy analysis model finally adopted.

2. The method for constructing a magnetic resonance diffusion multi-component spectroscopy analysis model according to claim 1, characterized in that: Step 2 is to remove Rician noise from the DWI data, specifically: Where S raw is the original DWI signal, δ n is the standard deviation of the real or imaginary part of the noise, S corr is the DWI signal after Rician noise removal; S noise is the background noise intensity, so in the calculation, use: As the DWI signal data after Rician noise removal.

3. The method for constructing a magnetic resonance diffusion multi-component spectroscopy analysis model according to claim 1, characterized in that: The magnetic resonance diffusion multi-component spectrum analysis model described in step 4 is expressed as: And meet: Among them, S b is the DWI signal value when the diffusion weighting coefficient is b, S0 is the DWI signal value when the diffusion weighting coefficient is 0, n is the number of multi-component spectra, C i is the volume ratio of the i-th diffusion spectrum component in the voxel, D i is the value of the i-th diffusion spectrum component, and b is the diffusion weighting coefficient.

4. The method for constructing a magnetic resonance diffusion multi-component spectroscopy analysis model according to claim 1, characterized in that: The BIC described in step 5 is expressed as: Where N is the sample size, n is the number of tissue multi-component spectra in the model, and r is the residual sum of squares of the model fitting; the BIC of each multi-component spectrum analysis model corresponds to ΔBIC: ΔBIC=BIC i -BIC min BIC i is the BIC value of any multi-component spectral analysis model, BIC min The minimum BIC observed for all multi-component spectral analysis models; the optimal model is selected by comparing the size of the ΔBIC value, and the optimal model selects the model with the smallest ΔBIC value.

Images