Construction of efficacy prediction model for patients with wild-type raf and braf gene colorectal liver metastasis based on radiomics features

Abstract

The present application belongs to the technical field of intelligent medical treatment, and particularly relates to a RAS, BRAF gene wild type colorectal cancer liver metastasis patient curative effect prediction model based on image group characteristics and application thereof. The present application successfully constructs a model for realizing curative effect prediction of RAS, BRAF gene wild type colorectal cancer liver metastasis patients receiving bevacizumab combined with chemotherapy treatment in the late first line based on CT image group characteristics before treatment: the enhanced CT image group data before initial treatment of patients with advanced colorectal cancer liver metastasis at the time of initial diagnosis is collected, 7 image group characteristics are obtained by using 1000 times Lasso-Logistic analysis, and an image group prediction model is constructed by using a multi-factor logistic regression method. The model constructed by the present application has important clinical application and popularization value from the clinical actual problem.

Description

Technical Field

[0001] This invention belongs to the field of intelligent medical technology, specifically relating to a predictive model for the efficacy of first-line bevacizumab combined with chemotherapy in patients with advanced colorectal cancer liver metastases based on radiomics features and its application. Background Technology

[0002] Colorectal cancer (CRC) is the third most common and deadliest cancer worldwide. According to the World Health Organization (WHO), 1.8 million new cases of CRC were diagnosed globally in 2018, with nearly 861,000 deaths. In recent years, the mortality rate of CRC has gradually declined in many countries due to improved early detection and more effective treatments. However, overall survival (OS) remains poor for patients with advanced CRC. The main sites of metastasis for advanced CRC are the liver, lungs, and bones, with liver metastasis being the leading cause of death in CRC patients, accounting for over 50% of CRC deaths. Even 5%–25% of patients have synchronous liver metastases at initial diagnosis, and another 50% develop liver metastases during disease progression.

[0003] Bevacizumab is a humanized anti-VEGF (vascular endothelial growth factor) monoclonal antibody. Its mechanism of action involves binding to VEGF and preventing its interaction with VEGF receptors (Flt-1 and KDR) on the surface of endothelial cells. In in vitro angiogenesis models and nude mouse (athymic) colon cancer xenograft models, administration of bevacizumab reduced microvascular growth and inhibited the progression of metastatic disease. Bevacizumab combined with chemotherapy is a first-line treatment option for patients with metastatic colorectal cancer, providing longer median progression-free survival (PFS) and overall survival (OS).

[0004] In patients with advanced colorectal cancer receiving first-line bevacizumab combined with chemotherapy, there are currently no highly effective and accurate predictive indicators of treatment efficacy. This issue has been a major focus for oncologists, oncologists, and pharmaceutical researchers. Therefore, accurately predicting patient efficacy before treatment can greatly help clinicians weigh the benefits and risks of their treatment plans, preventing patients with unsatisfactory treatment outcomes from incurring high treatment costs and adverse reactions. This provides a reference for clinical decision-making, enabling personalized and precise treatment. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a predictive model for the treatment efficacy of RAS and BRAF wild-type colorectal cancer with liver metastases based on radiomics features, and its application. This model can be used to predict the efficacy of first-line bevacizumab combined with chemotherapy in patients with advanced RAS and BRAF wild-type colorectal cancer with liver metastases, assisting physicians in making more accurate disease assessments and treatment decisions.

[0006] To achieve the above objectives, the present invention provides the following technical solution.

[0007] This invention provides a method for constructing a predictive model of the efficacy of first-line bevacizumab combined with chemotherapy in patients with RAS and BRAF wild-type colorectal cancer liver metastases. The method is characterized by being based on radiomics features, and the specific steps are as follows:

[0008] S1. Collect enhanced CT image data of patients with advanced colorectal cancer and liver metastases who were wild-type RAS and BRAF genes at initial diagnosis before treatment;

[0009] S2. Analyze the enhanced CT imaging data collected in S1, segment the region of interest, and extract image features;

[0010] S3. Using 1000 Lasso-Logistic analyses, the combination of 7 image features and parameters was stable more than 850 times;

[0011] S4. Using the multifactor logistic regression method, a corresponding efficacy prediction model based on the seven radiomics features obtained in S3 was constructed.

[0012] Furthermore, the region of interest segmentation described in S2 employs a three-dimensional semi-automatic segmentation method to segment the region of interest in the portal venous phase CT image.

[0013] Preferably, the specific steps of the three-dimensional semi-automatic segmentation method are as follows:

[0014] (1) Import the DICOM format PVP image into the 3D-Slicer software, and then use the soft tissue window (window width: 350HU, window level: 40HU) to outline the selected ROIs. Use the FAST-MARCHING semi-automatic fast segmentation algorithm to segment all PVP images of each patient layer by layer to analyze the ROIs as a whole.

[0015] (2) Manually adjust and refine the outlined area layer by layer again, outlining along the visible and clear boundaries of the lesion, and erasing adjacent normal tissue structures, such as bile ducts and blood vessels.

[0016] (3) The hollow module was used to delineate a 1cm thick shell-like area around the liver metastasis lesion;

[0017] (4) The completed ROIs segmentation results were reviewed and further refined by two experienced senior radiologists. Finally, the ROIs were exported in NRRD and MRML formats for storage and further analysis.

[0018] Furthermore, the extraction of image features described in S2 includes:

[0019] (1) First-order features describe the distribution of voxel intensity in ROIs;

[0020] (2) Shape-based features describe the stereoscopic visual characteristics of ROIs, including size and shape at both two-dimensional and three-dimensional levels;

[0021] (3) Texture features extracted based on five texture matrices: a) gray-level co-occurrence matrix, b) gray-level region size matrix, c) gray-level run length matrix, d) neighborhood gray-level difference matrix, and e) gray-level correlation matrix;

[0022] (4) Wavelet features: Wavelet filtering is added to the original image to reduce noise and extract detailed high-dimensional image group features.

[0023] Furthermore, the seven image group features obtained through screening in S3 include:

[0024] ball_wavelet.HHL_gldm_SmallDependenceHighGrayLevelEmphasis;

[0025] ball_wavelet.HHL_glszm_LowGrayLevelZoneEmphasis;

[0026] ball_wavelet.LHH_firstorder_Median;

[0027] shell_wavelet.LHH_firstorder_Maximum;

[0028] shell_wavelet.LHH_glszm_GrayLevelNonUniformityNormalized;

[0029] shell_wavelet.LHL_glcm_DifferenceVariance, shell_wavelet.LLH_firstorder_Kurtosis. Where "ball_" represents the radiomic characteristics of non-fused liver metastases with the largest cross-sectional area and clear boundaries, and "shell_" represents the radiomic characteristics of the shell-like region 1 cm thick surrounding the metastatic lesion.

[0030] Furthermore, the prediction model described in S4 is a radiomics prediction model. The formula for calculating the radiomics score is: 107.190 + 126.263 * ball_wavelet.HHL_gldm_SmallDependenceHighGrayLevelEmphasis + 19.815 * ball_wavelet.HHL_glszm_LowGrayLevelZoneEmphasis - 339.866 * ball_wavelet.LHH_firstorder_Median - 152.667 * shell_wavelet.LHH_firstorder_Maximum + 1. The formula is calculated as 115 * shell_wavelet.LHH_glszm_GrayLevelNonUniformityNormalized + 13.488 * shell_wavelet.LHL_glcm_DifferenceVariance + 1.085 * wavelet.LLH_firstorder_Kurtosis. The cutoff value is set to 50%. A score greater than 50% predicts disease progression in the patient after 4 cycles of treatment, while a score less than 50% predicts effective disease control after 4 cycles of treatment.

[0031] The present invention also provides a storage medium storing a plurality of instructions, characterized in that the instructions are adapted to be loaded and executed by a processor to implement the steps of the above-described method for predicting the efficacy of RAS and BRAF gene wild-type colorectal cancer liver metastasis based on radiomics feature models.

[0032] The present invention also provides a terminal, characterized in that it includes: a processor and a storage medium communicatively connected to the processor, the storage medium being adapted to store multiple instructions; the processor being adapted to call the instructions in the storage medium to execute the steps of implementing the above-described method for predicting the efficacy of RAS and BRAF gene wild-type colorectal cancer liver metastasis based on radiomics feature models.

[0033] This invention also provides a predictive model for the efficacy of first-line bevacizumab combined with chemotherapy in patients with RAS and BRAF wild-type colorectal cancer liver metastases, characterized in that it is constructed through the steps of the above-described method for predicting the efficacy of RAS and BRAF wild-type colorectal cancer liver metastases based on radiomics feature models.

[0034] The present invention also provides a device for predicting the efficacy of first-line bevacizumab combined with chemotherapy in patients with RAS and BRAF wild-type colorectal cancer liver metastases, characterized in that it includes the above-mentioned efficacy prediction model for RAS and BRAF wild-type colorectal cancer liver metastases.

[0035] The beneficial effects of the present invention compared with the prior art.

[0036] As described above, based on pre-treatment radiomics characteristics, this invention constructs a predictive model for the efficacy of first-line bevacizumab combined with chemotherapy in patients with RAS and BRAF wild-type colorectal cancer liver metastases. This invention has the following beneficial effects.

[0037] First, the method of this invention can be used to more accurately predict the treatment effect, i.e. whether patients with RAS and BRAF wild-type advanced colorectal cancer liver metastases will benefit from the treatment regimen by utilizing the radiomics features of enhanced CT before receiving first-line bevacizumab combined with chemotherapy. This helps doctors make more accurate judgments about the disease and treatment decisions.

[0038] Secondly, most known predictors of treatment efficacy rely on dynamic changes before and after treatment, and these indicators are often obtained through invasive examinations. In this invention, we utilize pre-treatment CT images, and enhanced CT is one of the most common examinations for cancer patients. The method of this invention can predict treatment efficacy economically and non-invasively.

[0039] The method of this invention, as an auxiliary diagnostic and treatment method for clinicians, can help clinicians make precise and individualized treatment decisions. Attached Figure Description

[0040] Figure 1. Using 1000 Lasso-Logistic analyses, the combination of 7 image features and parameters was repeated more than 850 times. They are (1) parameters, (2) wavelet.LHH_firstorder_Maximum

[0041] , (3) ball_wavelet.LHH_firstorder_Median, (4) shell_wavelet.LLH_firstorder_Kurtosis, (5) ball_wavelet.HHL_glszm_LowGrayLevelZoneEmphasis, (6) shell_wavelet.L HL_glcm_DifferenceVariance, (7) shell_wavelet.LHH_glszm_GrayLevelNonUniformityNormalized, (8) ball_wavelet.HHL_gldm_SmallDependenceHighGrayLevelEmphasis.

[0042] Figure 2. ROC curves of the radiomics model in the training set (black solid line) and validation set (grey dashed line). Detailed Implementation

[0043] The present invention will now be described in detail with reference to specific embodiments. The following embodiments will help to understand the present invention, but these embodiments are only for illustrative purposes, and the present invention is not limited thereto. The operating methods in the embodiments are all conventional operating methods in this technical field.

[0044] Example 1.

[0045] I. Research Methods and Results

[0046] 1.1 Case collection.

[0047] In accordance with the international ISBER criteria, patients with advanced colorectal cancer liver metastases who received first-line bevacizumab combined with chemotherapy between January 2014 and October 2019 were retrospectively collected. Inclusion criteria: (1) Patients with pathologically confirmed advanced colorectal cancer liver metastases; (2) Age 18–80 years; (3) ECOG score 0–1; (4) RAS and BRAF gene mutation status of surgical or biopsy specimens were tested and the gene status was wild-type; (5) Patients who received first-line bevacizumab combined with standardized chemotherapy after diagnosis and had known efficacy evaluation of 4 cycles of treatment; (6) Enhanced CT images of the lungs and abdomen before the start of first-line treatment were available, with a CT slice thickness ≤2 mm; (7) The time interval between the enhanced CT examination of the lungs and abdomen and the acquisition of surgical or biopsy specimens was no more than 30 days (range 4–30 days).

[0048] Exclusion criteria: (1) Patients with other tumors; (2) CT images with artifacts caused by implanted metal or movement; (3) The boundaries of liver metastases in CT images are too blurred to accurately delineate the edges.

[0049] Data Collection: Patients' names, genders, ages, and other clinicopathological information were collected from hospital medical records. Based on RECIST criteria, lesion size before and after treatment was recorded and compiled, and treatment efficacy was assessed. Informed consent was obtained from all included patients. The collection and testing of tumor patient specimens and patient follow-up in this study were approved by the ethics committee of the hospital where the patients were treated. Informed consent was obtained from all included patients.

[0050] Based on the inclusion and exclusion criteria, a total of 42 patients were ultimately included (7 patients with PD and 35 patients without PD). All patients were randomly divided into validation and training sets in a 7:3 ratio.

[0051] 1.2 CT image acquisition.

[0052] CT image acquisition is the first step in radiomics. First, a large number of images (DICOM format) are acquired from CT images. Then, these images are preprocessed, including image reconstruction, noise reduction, and grayscale normalization, to ensure the standardization and consistency of feature data acquisition and reconstruction parameters, including radiation dose, scanning scheme, reconstruction algorithm, and slice thickness.

[0053] The patient underwent a contrast-enhanced CT scan (CECT) of the lungs and abdomen prior to the initial systemic treatment. Specific CECT acquisition parameters and conditions were as follows: Following standard operating procedures, various 64-slice spiral CT scanners (GE, Phillips, Siemens, and Toshiba) were used; tube voltage was 120 kVp (range 100–140 kVp); tube current was 333 mA (range 100–752 mA); CT slice thickness was 2 mm; and standard reconstruction methods were applied. The contrast agent iohexol dosage was calculated at 1.2–1.5 mL / kg body weight, administered intravenously at a rate of 2.5 mL / s, followed by an infusion of 20–30 mL of normal saline. All patients underwent CT image scanning in a supine position with breath-holding during inspiration. Portal venous phase (PVP) images were scanned at approximately 60–70 seconds. During image screening, image quality was visually assessed and checked layer by layer. Finally, DICOM cleaner software was used to desensitize all CT images, removing private information such as patient name, gender, age, examination date, hospital registration number, CT examination number, and hospital name. Different patients were labeled with random numbers. All CT images were stored in DICOM format.

[0054] 1.3 CT image segmentation.

[0055] The CT image segmentation process primarily utilizes 3D-Slicer software (www.slicer.org), employing a three-dimensional semi-automatic segmentation method to segment regions of interest (ROIs) in portal venous phase (PVP) CT images. ROIs generally refer to regions delineated and segmented from the original CT image during image processing using methods such as rectangles, circles, ellipses, or irregular polygons for subsequent analysis. In this study, we selected the non-fused liver metastatic lesions with the largest cross-sectional area and well-defined boundaries, along with the surrounding 1 cm thick shell-like region, as the ROIs.

[0056] First, DICOM format PVP images were imported into 3D-Slicer software. Then, a soft tissue window (window width: 350 HU, window level: 40 HU) was used to outline the selected ROIs. The FAST-MARCHING semi-automatic fast segmentation algorithm was used to segment all PVP images for each patient layer-by-layer for overall ROI analysis. Next, we manually adjusted and refined the outlined areas layer by layer, outlining along the clearly visible boundaries of the lesions and erasing adjacent normal tissue structures, such as bile ducts and blood vessels. Based on this, we used the hollow module to outline a 1 cm thick shell-like region surrounding the liver metastases. The final ROI segmentation results were reviewed and further refined by two experienced senior radiologists. Finally, the ROIs were exported to NRRD and MRML formats for storage and further analysis.

[0057] 1.4 Image group feature extraction and screening of CT images.

[0058] Radiomic features of Regions of Interest (ROIs) for each patient were extracted using Pyradiomics methods. These features, extracted from the delineated ROIs, allow for the quantitative assessment of tumor intensity, shape, and texture. These radiomic features can be categorized into three types: 1) first-order features, shape-based features, and texture features. First-order features describe the distribution of voxel intensity within the ROIs; 2) shape-based features describe the stereoscopic visual characteristics of the ROIs, including size and shape at both two-dimensional and three-dimensional levels. These features are not the same as the gray intensity distribution in ROIs; 3) Texture features extracted based on 5 texture matrices: (1) Gray Level Co-occurrence Matrix (GLCM), (2) Gray Level Size Zone Matrix (GLSZM), (3) Gray Level Run Length Matrix (GLLM), (4) Neighborhood Gray Tone Difference Matrix (NGTDM) and (5) Gray Level Dependence Matrix (GLDM). In addition, we added wavelet filtering to the original image, which can extract detailed high-dimensional image group features while reducing noise.

[0059] After extracting the radiographic features of ROIs for each patient, we performed dimensionality reduction on all radiographic features. First, we removed radiographic features with identical values ​​across all patients, lacking discriminative power. Second, to verify the robustness and reproducibility of the radiographic features, we randomly selected 30 ROIs and calculated the intraclass correlation coefficient (ICC) and concordance correlation coefficient (CCC) for each feature. To calculate the ICC, two radiologists performed semi-automatic, layer-by-layer delineation of the ROIs for the same patient (using the same method as before). These two radiologists were unaware of the patient's treatment efficacy beyond the clinical diagnosis of colorectal cancer liver metastasis. To calculate the CCC, after delineating and segmenting the ROIs of the randomly selected 30 patients, the same delineation and segmentation steps were repeated by the same physician two weeks later. Finally, we excluded radiographic features with ICC or CCC values ​​below 0.75 from subsequent analyses.

[0060] 1.5 Constructing a prediction model based on image group features.

[0061] In this study, we constructed a RAS gene status prediction model based on radiomics features. All patients were randomly divided into training and validation sets in a 7:3 ratio. Using the Lasso-Logistic algorithm, 1000 analyses were performed, and the combination of the seven radiomics features and parameters consistently occurred more than 850 times. Seven imaging features were obtained through screening. Three of these features belonged to spherical regions of interest (ROIs) resembling liver metastases, and the other four belonged to shell-like regions (ROIs) 1 cm thick surrounding liver metastases. These seven features are: ball_wavelet.HHL_gldm_SmallDependenceHighGrayLevelEmphasis, ball_wavelet.HHL_glszm_LowGrayLevelZoneEmphasis, ball_wavelet.LHH_firstorder_Median, shell_wavelet.LHH_firstorder_Maximum, shell_wavelet.LHH_glszm_GrayLevelNonUniformityNormalized, shell_wavelet.LHL_glcm_DifferenceVariance, and shell_wavelet.LLH_firstorder_Kurtosis. Figure 1 ).

[0062] The constructed radiomics prediction model calculation formula is: 107.190 + 126.263 * ball_wavelet.HHL_gldm_SmallDependenceHighGrayLevelEmphasis + 19.815 * ball_wavelet.HHL_glszm_LowGrayLevelZoneEmphasis - 339.866 * ball_wavelet.LHH_firstorder_Median - 152.667 * shell_wavelet.LHH_firstorder_Maximum + 1.115 * shell_wavelet.LHH_glszm_GrayLevelNonUniformityNormalized + 13.488 * shell_wavelet.LHL_glcm_DifferenceVariance + 1.085 * wavelet.LLH_firstorder_Kurtosis.

[0063] The cutoff value is set to 50%. A score greater than 50% calculated by the above formula predicts that the patient will experience disease progression after 4 cycles of treatment, while a score less than 50% predicts that the patient's disease will still be effectively controlled after 4 cycles of treatment.

[0064] 1.6 Comparison of Predictive Performance of Radiomics Therapeutic Effect Prediction Models

[0065] The area under the curve (AUC) of the receiver operating characteristic (ROC) was used to evaluate the predictive ability of a model based on radiomics features for the efficacy of first-line bevacizumab combined with chemotherapy in patients with RAS and BRAF wild-type colorectal cancer liver metastases and to determine the cutoff value. A radiomics efficacy prediction model was constructed based on these seven radiomics features. The AUC of this radiomics model was 0.95 on the training set and 0.69 on the validation set. Figure 2 ).

Claims

1. A method for constructing a predictive model for the efficacy of first-line bevacizumab combined with chemotherapy in patients with RAS and BRAF wild-type colorectal cancer liver metastases, characterized in that, The method is based on radiomics features, and the specific steps are as follows: S1. Collect enhanced CT image data of patients with advanced colorectal cancer and liver metastases who were wild-type RAS and BRAF genes at initial diagnosis before treatment; S2. The enhanced CT imaging data collected in S1 were analyzed. A three-dimensional semi-automatic segmentation method was used to segment the portal venous phase CT images into two types of regions of interest: (1) non-fused liver metastatic lesions with the largest cross-sectional area and clear boundaries, and (2) a shell-like region 1 cm thick around the lesion. The imaging features were then extracted. S3. Using 1000 Lasso-Logistic analyses, the following 7 image group features were selected: ball_wavelet.HHL_gldm_SmallDependenceHighGrayLevelEmphasis; ball_wavelet.HHL_glszm_LowGrayLevelZoneEmphasis; ball_wavelet.LHH_firstorder_Median; shell_wavelet.LHH_firstorder_Maximum; shell_wavelet.LHH_glszm_GrayLevelNonUniformityNormalized; shell_wavelet.LHL_glcm_DifferenceVariance; shell_wavelet.LLH_firstorder_Kurtosis; Where "ball_" represents the radiomics features of a non-fused liver metastasis with the largest cross-sectional area and clear boundaries, and "shell_" represents the radiomics features of a shell-like region 1 cm thick around the lesion. S4. Using the multifactor logistic regression method, weight coefficients are assigned to the seven image group features to construct a radiomics scoring prediction model. The calculation formula of the model is as follows: 107.190+126.263*ball_wavelet.HHL_gldm_SmallDependenceHighGrayLevelEmphasis+19.815*ball_wavelet.HHL_glszm_LowGrayLevelZoneEmphasis-339.866*ball_wavelet.LHH_firstorder_Median-152.667*shell_wave let.LHH_firstorder_Maximum+1.115*shell_wavelet.LHH_glszm_GrayLevelNonUniformityNormalized+13.488*shell_wavelet.LHL_glcm_DifferenceVariance+1.085*wavelet.LLH_firstorder_Kurtosis, set the cutoff value to 50%.

2. The construction method according to claim 1, characterized in that, The specific steps of the 3D semi-automatic segmentation method described in S2 are as follows: (1) Import the DICOM format portal venous phase image into 3D-Slicer software, use the soft tissue window to outline the selected region of interest, wherein the window width of the soft tissue window is 350HU and the window level is 40HU, and use the FAST-MARCHING semi-automatic fast segmentation algorithm to segment the entire portal venous phase image of each patient layer by layer to analyze the region of interest as a whole. (2) Manually adjust and refine the outlined area layer by layer, outline along the visible and clear boundary of the lesion, and erase the adjacent normal tissue structure; (3) Use the hollow module to outline a 1cm thick shell-like area around the liver metastasis lesion; (4) The segmentation results of the completed region of interest were checked and improved by two senior radiologists. Finally, the region of interest was exported in NRRD and MRML formats for storage and further analysis.

3. The construction method according to claim 1, characterized in that, The image features extracted in S2 include: (1) first-order features; (2) shape-based features; (3) texture features extracted based on gray-level co-occurrence matrix, gray-level region size matrix, gray-level run length matrix, neighborhood gray-level difference matrix and gray-level correlation matrix; and (4) wavelet features.

4. A storage medium storing a plurality of instructions, characterized in that, The instructions are adapted to be loaded and executed by a processor to implement the steps of constructing a predictive model for the efficacy of first-line bevacizumab combined with chemotherapy in patients with RAS and BRAF wild-type colorectal cancer liver metastases based on radiomics features, as described in any one of claims 1-3.

5. A terminal, characterized in that, include: A processor and a storage medium communicatively connected to the processor, the storage medium being adapted to store multiple instructions; the processor being adapted to invoke the instructions in the storage medium to execute the steps of implementing the efficacy prediction model for constructing a treatment for patients with RAS and BRAF gene wild-type colorectal cancer liver metastases based on radiomics features according to any one of claims 1-3, which is used for first-line bevacizumab combined with chemotherapy in advanced patients.

Images