Method and device for quantitative calculation of tumor shrinkage deformation after liver ablation

Abstract

The quantitative calculation method and device for the degree of tumor shrinkage after liver ablation can solve the problem of quantitative calculation of shrinkage deformation of tumors of different properties under different ablation conditions, laying the foundation for accurate postoperative evaluation of tumor ablation. The method includes: (1) using MRI images with clear tumor residual shadows after ablation, selecting multiple phases of images, segmenting the preoperative liver, tumor area, postoperative tumor residual shadow and ablation area respectively, and calculating the volumes of the preoperative tumor area, postoperative tumor residual shadow and ablation area; (2) extracting radiomic features for the preoperative liver and tumor segmentation area of ​​each phase; (3) performing correlation tests on all radiomic features, removing relevant radiomic features, adding patient clinical information, and using the LASSO algorithm for feature dimensionality reduction screening; (4) using ablation energy parameters combined with radiomic features to establish a regression model for tumor volume shrinkage.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a method for quantitatively calculating the degree of tumor shrinkage after liver ablation, and a device for quantitatively calculating the degree of tumor shrinkage after liver ablation. Background Art

[0002] In recent years, image-guided percutaneous thermal ablation technology has become one of the most promising minimally invasive treatment methods for solid tumors such as liver, kidney, and breast. Microwave ablation is guided by ultrasound, CT, and other images to insert the ablation needle into the tumor. By releasing electromagnetic waves, the polar molecules in the local area vibrate and rub to generate high temperatures, ultimately achieving the purpose of inactivating the tumor. However, unlike the clear visualization of the tumor and treatment area during open surgery, minimally invasive ablation is performed under the guidance of two-dimensional images during surgery. Whether the treatment area completely covers the tumor and achieves a sufficient safety margin needs to be evaluated by comparing preoperative and postoperative images. However, due to the dehydration, shrinkage, and deformation of the tumor during the patient's ablation treatment, it is impossible to accurately evaluate the efficacy of the treatment only by comparing the two-dimensional images before and after surgery or by rigid alignment.

[0003] Non-rigid registration methods have lower errors in postoperative evaluation than rigid registration. However, the non-rigid registration methods currently used for post-ablation evaluation at home and abroad are limited by the universality of the tissue deformation energy functional construction. In particular, for the uneven local tissue deformation characteristics after liver tumor ablation, there is no method to quantitatively calculate tumor shrinkage deformation, nor is there a method to correct the non-rigid registration evaluation error caused by tumor shrinkage. Summary of the Invention

[0004] In order to overcome the defects of the existing technology, the technical problem to be solved by the present invention is to provide a method for quantitatively calculating the shrinkage deformation of tumors after liver ablation, which can solve the problem of quantitative calculation of the shrinkage deformation of tumors of different properties under different ablation conditions, and lay the foundation for accurate postoperative evaluation of tumor ablation.

[0005] The technical solution of the present invention is: a method for quantitatively calculating tumor shrinkage deformation after liver ablation, comprising the following steps:

[0006] (1) Using MRI images with clear tumor residual shadows after ablation, multiple phase images were selected to segment the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation area, and the volumes of the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation area were calculated.

[0007] (2) extracting radiomic features based on the preoperative tumor segmentation region for each stage;

[0008] (3) All imaging omics features were tested for correlation, and the omics features with correlation were removed. The patient's clinical information was added, and the LASSO algorithm was used for feature dimensionality reduction screening;

[0009] (4) A regression model of tumor volume shrinkage was established using ablation energy parameters combined with imaging genomics features.

[0010] The present invention utilizes MRI images with clear tumor residual shadows after ablation, selects multiple phase images such as T2 and arterial phase, and separately segments the preoperative tumor area, postoperative tumor residual shadow and ablation area, and calculates the volumes of the three; extracts imaging genomics features for the preoperative tumor segmentation area of ​​each phase; performs correlation test on all imaging genomics features, removes genomics features with correlation, adds patient clinical information, and uses lasso algorithm to perform feature dimensionality reduction screening; finally, uses ablation energy parameters combined with genomics features to establish a regression model for tumor volume shrinkage; therefore, it can solve the problem of quantitative calculation of shrinkage deformation of tumors of different properties under different ablation conditions, laying the foundation for accurate postoperative evaluation of tumor ablation.

[0011] Also provided is a device for quantitatively calculating the degree of tumor shrinkage after liver ablation, comprising:

[0012] An image segmentation module is configured to use MRI images with clear tumor residual shadows after ablation to select multiple phases of images, segment the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation zone, and calculate the volumes of the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation zone;

[0013] a feature extraction module configured to extract radiomic features for the preoperative tumor segmentation region of each phase;

[0014] The feature screening module is configured to perform correlation tests on all imaging omics features, remove relevant omics features, add patient clinical information, and use the LASSO algorithm to perform feature dimensionality reduction screening;

[0015] The model building module is configured to use ablation energy parameters in combination with imaging omics features to establish a regression model for tumor volume shrinkage. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 4 is a flow chart of the method for quantitatively calculating the degree of tumor shrinkage after liver ablation according to the present invention.

[0017] Figure 2 2 is a schematic diagram of the operation of the method for quantitatively calculating the degree of tumor shrinkage after liver ablation according to the present invention.

[0018] Figure 3 This is an effect diagram of processing the liver area in images before and after ablation using a point cloud registration algorithm in the existing technology.

[0019] Figure 4 According to the present invention Figure 3 Correction effect diagram. DETAILED DESCRIPTION

[0020] like Figure 1 As shown, this quantitative calculation method for tumor shrinkage deformation after liver ablation is applied to the correction of three-dimensional registration, which includes the following steps:

[0021] (1) Using MRI images with clear tumor residual shadows after ablation, multiple phase images were selected to segment the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation area, and the volumes of the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation area were calculated.

[0022] (2) extracting radiomic features based on the preoperative tumor segmentation region for each stage;

[0023] (3) All imaging omics features were tested for correlation, and the relevant omics features were removed. The patient's clinical information was added, and the LASSO (least absolute shrinkage and selection operator) algorithm was used for feature dimensionality reduction screening.

[0024] (4) A regression model of tumor volume shrinkage was established using ablation energy parameters combined with imaging genomics features.

[0025] The present invention utilizes MRI images with clear tumor residual shadows after ablation, selects multiple phase images such as T2 and arterial phase, and separately segments the preoperative liver and tumor area, postoperative tumor residual shadow and ablation area, and calculates the volumes of the three; extracts imaging genomics features for the preoperative liver and tumor segmentation areas of each phase; performs a correlation test on all imaging genomics features, removes genomics features with correlation, adds patient clinical information, and uses the lasso algorithm for feature dimensionality reduction screening; finally, uses the ablation energy parameter combined with genomics features to establish a regression model for tumor volume shrinkage; therefore, it can solve the problem of quantitative calculation of shrinkage deformation of tumors of different properties under different ablation conditions, laying the foundation for accurate postoperative evaluation of tumor ablation.

[0026] More than 200 patients' enhanced MRI images were collected before and after surgery, with clear tumor residual images in the postoperative images as the inclusion criteria. Preferably, in step (1), three-dimensional tumor segmentation was performed on the preoperative T2, DWI, arterial phase, and delayed phase to calculate the tumor volume V before ; Perform three-dimensional segmentation of the tumor residual image on the delayed enhanced MRI image after surgery and calculate the tumor residual image volume V after and the ablation zone volume V ablation (This can be achieved through image segmentation software such as ITK-snap; MITK); calculate the tumor shrinkage ratio (V before -V after ) / V before .

[0027] Preferably, in step (2), each phase of MRI images is standardized; and PyRadiomics is used to extract 873 three-dimensional imaging features of the tumor in each phase of T2, DWI, arterial phase, and delayed phase in each patient's preoperative MRI, including 13 shape features, 180 first-order features, and 680 texture features, i.e., a total of 3492 imaging features.

[0028] Preferably, in step (3), correlation analysis is used for screening, and 617 image features remain.

[0029] Preferably, in step (3), LASSO is a generalized linear model expressed as Y=∑ω i X i +b, where i=1,2,3,...n,ω i Each feature X i The regression coefficient is obtained by making the loss function formula (1)

[0030]

[0031] Minimize the unimportant feature coefficients and return them to 0 to achieve the purpose of feature screening. The lambda is calculated through a ten-fold cross-validation cycle optimization process. After LASSO feature screening, 23 omics features related to tumor shrinkage were finally included. They include:

[0032] exponential_firstorder_Minimum 1.04

[0033] gradient_glcm_Idn -0.002

[0034] gradient_glcm_Imc1 0.32

[0035] gradient_glcm_InverseVariance 0.51

[0036] gradient_glcm_MCC -0.55

[0037] gradient_glszm_SmallAreaLowGrayLevelEmphasis 2.41

[0038] Lbp-2D_glszm_SmallAreaEmphasis 0.69

[0039] logarithm_glcm_MaximumProbability -0.58

[0040] Wavelet-LLH_glcm_ClusterShade -1.0

[0041] Wavelet-LHL_firstorder_90Percentile 0.13

[0042] Wavelet-LHL_glszm_LowGrayLevelZoneEmphasis 1.33

[0043] Wavelet-LHH_firstorder_InterquartileRange 0.46

[0044] Wavelet-LHH_glszm_GrayLevelNonUniformity 0.86

[0045] Wavelet-LHH_glszm_LargeAreaEmphasis -0.87

[0046] Wavelet-LHH_glszm_SizeZoneNonUniformityNormalized 0.73

[0047] Wavelet-HLL_glcm_ClusterShade 0.89

[0048] Wavelet-HLL_glszm_LargeAreaLowGrayLevelEmphasis -0.59

[0049] Wavelet-HHL_glszm_GrayLevelNonUniformity 1.82

[0050] Wavelet-HHL_glszm_LowGrayLevelZoneEmphasis 0.36

[0051] Wavelet-HHL_glszm_ZoneEntropy 0.19

[0052] Wavelet-HHH_firstorder_Mean 0.13

[0053] Wavelet-HHH_glszm_SmallAreaLowGrayLevelEmphasis 0.67

[0054] Wavelet-LLL_glcm_MaximumProbability -1.56

[0055] The radiomics score Rad score of each case was calculated based on the screened texture features.

[0056] Preferably, in step (3), the ten-fold cross validation is to divide all cases into 10 parts, randomly select one of them as a validation set, and the other 9 as training sets, repeat the cross validation 10 times, and average the results of the 10 times to obtain the final prediction result.

[0057] According to the team's previous research and analysis, the degree of tissue shrinkage is also affected by the microwave energy absorbed by the tissue. Therefore, we used the imaging genomics score and each patient's ablation energy parameters (the microwave energy absorbed per unit volume of ablation area was calculated based on the ablation time, power, and ablation volume) and clinical information (age, gender, etc.) as factors affecting tumor shrinkage for regression analysis, and established a linear regression model to predict the degree of tumor volume shrinkage Y.

[0058] Preferably, in step (4), the imaging genomics score is used as a factor affecting tumor shrinkage, along with the ablation energy parameter and clinical information of each patient for regression analysis, and a linear regression model is established to predict the degree of tumor volume shrinkage Y; the ablation energy parameter is the microwave energy absorbed by the ablation zone per unit volume calculated based on the ablation time, power, and ablation volume, and the clinical information includes age, gender, and the ratio of the actual liver volume to the standard liver volume.

[0059] Preferably, the linear regression model in step (4) is formula (2)

[0060] Y=0.4*X1+1.02*X2+36.6 (2)

[0061] Where Y represents the degree of tumor shrinkage, X1 represents the energy absorbed per unit volume, and X2 represents the radiomics score.

[0062] Preferably, in step (4), in the postoperative registration evaluation, the point cloud elastic registration based on differential homeomorphism is used as the evaluation method (in addition, the shrinkage method is also applicable to other traditional elastic registration methods, such as the free-form deformation model, the kernel function-based registration model, etc.), and according to the degree of tumor volume shrinkage, the tumor boundary line of the initial non-rigid registration is uniformly shrunk centripetally along the normal vector in the three-dimensional space to perform registration tumor volume shrinkage correction, and the minimum distance between the tumor boundary and the ablation zone boundary in space is used as the measure of the ablation efficacy.

[0063] The following is to verify the technical effect of the present invention relative to the prior art.

[0064] Figure 3 This is the effect of processing the liver area of ​​the pre- and post-ablation images using the point cloud registration algorithm in the existing technology. The liver area before and after ablation is registered using the point cloud registration algorithm to obtain the transformation matrix. The registration result of the pre-operative liver tumor area corresponding to the post-operative CT / MRI image area is obtained according to the liver surface transformation matrix. The registration result is as follows: Figure 3 The initial point cloud registration results are shown in Figure 1. The right side shows the 3D result (the innermost layer is the residual image of the inactivated tumor, the middle layer is the preoperative tumor registration transformation result, and the outermost layer is the postoperative ablation area). It can be seen that the initial registration results are different from the actual tumor morphology.

[0065] Figure 4 According to the present invention Figure 3 The present invention is used to predict the morphological changes of liver tumors after ablation surgery, and to correct Figure 3 The obtained tumor morphology shows that the corrected tumor deformation essentially coincides with the actual morphology after inactivation. Therefore, compared to existing technologies, this invention can solve the problem of quantitatively calculating the shrinkage deformation of tumors of different properties under different ablation conditions, laying the foundation for accurate postoperative assessment of tumor ablation.

[0066] Those skilled in the art will appreciate that all or part of the steps in the above-described method can be implemented by instructing the relevant hardware through a program. The program can be stored in a computer-readable storage medium. When executed, the program includes the steps of the above-described method. The storage medium can be ROM / RAM, a magnetic disk, an optical disk, a memory card, etc. Therefore, corresponding to the method of the present invention, the present invention also includes a device for quantitatively calculating the degree of tumor shrinkage after liver ablation. The device is generally represented in the form of functional modules corresponding to the steps of the method. The device includes:

[0067] Also provided is a device for quantitatively calculating the degree of tumor shrinkage after liver ablation, comprising:

[0068] An image segmentation module is configured to use MRI images with clear tumor residual shadows after ablation to select multiple phases of images, segment the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation zone, and calculate the volumes of the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation zone;

[0069] a feature extraction module configured to extract radiomic features for the preoperative tumor segmentation region of each phase;

[0070] The feature screening module is configured to perform correlation tests on all imaging omics features, remove relevant omics features, add patient clinical information, and use the LASSO algorithm to perform feature dimensionality reduction screening;

[0071] The model building module is configured to use ablation energy parameters in combination with imaging omics features to establish a regression model for tumor volume shrinkage.

[0072] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of protection of the technical solution of the present invention.

Claims

1. A method for quantitatively calculating the degree of tumor shrinkage after liver ablation, characterized by: It includes the following steps: (1) Using MRI images with clear tumor residual shadows after ablation, select multiple phase images to segment the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation area, and calculate the volumes of the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation area; (2) Extract radiomic features based on the preoperative tumor segmentation region for each stage; (3) All imaging omics features were tested for correlation, and the relevant omics features were removed. The patient's clinical information was added, and the LASSO algorithm was used for feature dimensionality reduction screening. (4) Establishing a regression model for tumor volume shrinkage using ablation energy parameters combined with imaging omics features; In step (4), the radiomics score of each case is calculated based on the screened texture features, and the radiomics score is used as a factor affecting tumor shrinkage for regression analysis with the ablation energy parameter and clinical information of each patient, and a linear regression model is established to predict the degree of tumor volume shrinkage Y; the ablation energy parameter is the microwave energy absorbed by the ablation zone per unit volume calculated based on the ablation time, power and ablation volume, and the clinical information includes age, gender, and the ratio of the actual liver volume to the standard liver volume; The linear regression model in step (4) is formula (2) Y=0.4*X1+1.02*X2+36.6 (2) Where Y represents the degree of tumor shrinkage, X1 represents the energy absorbed per unit volume, and X2 represents the radiomics score.

2. The method for quantitatively calculating the degree of tumor shrinkage after liver ablation according to claim 1, characterized in that: In step (1), three-dimensional segmentation of the tumor is performed on the preoperative T2, DWI, arterial phase, and delayed phase, and the tumor volume V is calculated. before ; Perform three-dimensional segmentation of the tumor residual image on the delayed enhanced MRI image after surgery and calculate the tumor residual image volume V after and the ablation zone volume V ablation ; Calculate the tumor shrinkage ratio (V before -V after ) / V before .

3. The method for quantitatively calculating the degree of tumor shrinkage after liver ablation according to claim 2, characterized in that: In step (2), each phase of MRI images is standardized; PyRadiomics is used to extract 873 three-dimensional imaging features of tumors in each phase of T2, DWI, arterial phase, and delayed phase in each patient's preoperative MRI, including 13 shape features, 180 first-order features, and 680 texture features.

4. The method for quantitatively calculating the degree of tumor shrinkage after liver ablation according to claim 3, characterized in that: In step (3), correlation analysis was used for screening, and 617 image features remained.

5. The method for quantitatively calculating the degree of tumor shrinkage after liver ablation according to claim 4, characterized in that: In the step (4), in the postoperative registration evaluation, the point cloud elastic registration based on differential homeomorphism is used as the evaluation method. According to the degree of tumor volume shrinkage, the tumor boundary line of the initial non-rigid registration is uniformly shrunk centripetally along the normal vector in the three-dimensional space to perform registration tumor volume shrinkage correction. The minimum distance between the tumor boundary and the ablation zone boundary in space is used as the measure of the ablation efficacy.

6. A device for quantitatively calculating the degree of tumor shrinkage after liver ablation, characterized by: It includes: An image segmentation module is configured to use MRI images with clear tumor residual shadows after ablation to select multiple phases of images, segment the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation zone, and calculate the volumes of the preoperative liver, tumor area, postoperative tumor residual shadow, and ablation zone; a feature extraction module configured to extract radiomic features for the preoperative tumor segmentation region of each phase; The feature screening module is configured to perform correlation tests on all imaging omics features, remove relevant omics features, add patient clinical information, and use the LASSO algorithm to perform feature dimensionality reduction screening; a model building module configured to establish a regression model of tumor volume shrinkage using ablation energy parameters combined with radiomics features; In the model building module, a radiomics score is calculated for each case based on the screened texture features. The radiomics score is then used in a regression analysis with each patient's ablation energy parameter and clinical information as factors influencing tumor shrinkage, and a linear regression model is established to predict the degree of tumor volume shrinkage Y. The ablation energy parameter is the microwave energy absorbed by the ablation zone per unit volume calculated based on the ablation time, power, and ablation volume. Clinical information includes age, gender, and the ratio of actual liver volume to standard liver volume. The linear regression model in the model building module is formula (2) Y=0.4*X1+1.02*X2+36.6 (2) Where Y represents the degree of tumor shrinkage, X1 represents the energy absorbed per unit volume, and X2 represents the radiomics score.

Images