Thrombus composition visual navigation method based on CT image

Abstract

The invention belongs to the field of medical image processing, and provides a thrombus composition visual navigation method based on a CT image, which comprises the following steps: step 1, evaluating the CT image, and sketching thrombus according to the CT image of a stroke patient; 2, feature selection and model construction, wherein thrombus is divided into different areas; 3, performing histological evaluation, and associating different regions with tissue components; 4, outputting to a user: extracting the characteristics of each region to predict the prognosis of the patient; and fusing the CTA image and the DSA image, performing thrombus extraction navigation, and automatically and visually presenting different regions of thrombus components. The invention provides a visualization method for helping doctors to evaluate thrombus components of ischemic stroke patients.

Description

Technical Field

[0001] This application belongs to the field of medical image processing. Background Art

[0002] Stroke is the leading cause of disability and mortality in adults worldwide, with ischemic stroke accounting for 62.4% of all strokes in 2019. Robotic thrombectomy (MT) offers an opportunity to restore blood flow and salvage brain tissue, but may fail to achieve complete recanalization in approximately 20% of cases. Different thrombi exhibit varying sensitivities to thrombolytics and mechanical disruption, leading to varying outcomes after MT. The consensus is that, compared with red blood cell (RBC)-rich thrombi, fibrin- and platelet (FP)-rich thrombi are resistant to thrombolysis, difficult to retrieve, and less responsive to revascularization during MT. For FP-rich thrombi, advanced devices are needed to facilitate retrieval of stent retrievers. During MT, guidewires should be placed as far as possible across RBC-rich regions to minimize fragment-induced thrombus escape. Therefore, the ability to noninvasively and accurately quantify thrombus composition could provide valuable prognostic information. Currently, no histopathologically based image segmentation and quantification methods are available in clinical practice.

[0003] When admitted to the hospital, special signs of thrombosis are observed through imaging examinations to reflect different thrombosis components. Doctors can only rely on observing the images, and there is no visual method to assist doctors in their evaluation. Summary of the Invention

[0004] In order to solve the above problems, the present application proposes a thrombus composition visualization navigation method based on CT images.

[0005] Technical solution of this application: The thrombus composition visualization navigation method based on CT images includes the following steps: Step 1. CT image assessment: outline the thrombus based on the CT image of the stroke patient; Step 2. Feature selection and model building to divide the thrombus into different regions; Step 3. Histological evaluation to correlate different regions with tissue components; Step 4. Output to the user: Extract regional features to predict patient prognosis; fuse CTA images with DSA images for thrombus removal navigation, and automatically visualize different regions of thrombus composition.

[0006] Beneficial effects of this application: The present invention provides a visualization method for helping doctors evaluate thrombus composition in ischemic stroke patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 Schematic diagram of the method process of an embodiment of the present invention. DETAILED DESCRIPTION

[0008] The technical solution provided by this application will be further described below in conjunction with specific embodiments and accompanying drawings. The advantages and features of this application will become more apparent with reference to the following description.

[0009] Example 1 This embodiment describes a thrombus composition visualization navigation method based on dual-energy CT images, which includes the following steps: Step 1. Dual-energy CT image evaluation, outlining the thrombus based on the dual-energy CT images of stroke patients.

[0010] Step 1.1 Five parametric images were reconstructed based on dual-energy CT, including hybrid images, iodine concentration (IC) images, virtual single-energy (VM) images (40 keV and 190 keV), and virtual non-contrast (VNC) images.

[0011] Step 1.2: All images were normalized, image intensities were scaled to 0-600, and resampled to the same resolution (1 × 1 × 1 mm).

[0012] Step 1.3 discretizes and normalizes the voxel intensity values using a filter that reduces image noise.

[0013] Step 1.4 Use ITK-SNAP software to manually delineate the three-dimensional thrombus area.

[0014] Step 2. Feature selection and model building to divide the thrombus into different regions.

[0015] Step 2.1 Extract features: PyRadiomics software was used to extract 107 radiomic features from the entire thrombus and each subregion and normalized using Z scores.

[0016] Step 2.2 Feature screening: Feature selection was performed using the t-test (features with p < 0.05 were selected), Pearson correlation analysis (only one feature with correlation > 0.90 was retained), and LASSO (by introducing L1 regularization, the coefficients of unimportant features were reduced to zero, and non-zero features were retained). The results were then subjected to 10-fold cross-validation (to ensure stable results).

[0017] Step 2.3 Model construction: The non-zero features (obtained in step 2.2) were selected and the Radscore was calculated based on logistic regression, and then the whole thrombus and habitat models were constructed. Joint Model (DECT_Combined (combined model) = Habitat_c*2.8478+Habitat_d*(-1.09809)+Habitats*4.77103+Radiomics*4.34301-4.78027).

[0018] Examples of sub-regions are as follows: DECT-habitat_c (subregion c) screens 18 features (the result of step 2.2) and calculates the Radscore formula as follows Radscore = 0.35238095238095213 + +0.001487 * 40keV_habitat_c_original_firstorder_Minimum (Screen out different features) +0.031579 * 40keV_habitat_c_original_firstorder_RobustMeanAbsoluteDeviation -0.001231 * 40keV_habitat_c_original_firstorder_Skewness +0.079467 * 40keV_habitat_c_original_firstorder_TotalEnergy +0.074329 * 40keV_habitat_c_original_glrlm_ShortRunLowGrayLevelEmphasis -0.002762 * 40keV_habitat_c_original_glszm_SmallAreaEmphasis -0.029959 * IC_habitat_c_original_firstorder_RootMeanSquared -0.008022 * IC_habitat_c_original_glcm_Imc2 +0.041922 * IC_habitat_c_original_glcm_MaximumProbability -0.087525 * IC_habitat_c_original_glrlm_LongRunHighGrayLevelEmphasis +0.021877 * IC_habitat_c_original_glrlm_ShortRunLowGrayLevelEmphasis +0.022124 * Mix_habitat_c_original_firstorder_Kurtosis +0.073967 * Mix_habitat_c_original_firstorder_TotalEnergy +0.004918* Mix_habitat_c_original_glcm_InverseVariance +0.028064 * Mix_habitat_c_original_glcm_MCC -0.000804 * Mix_habitat_c_original_gldm_LargeDependenceLowGrayLevelEmphasis +0.038355 * VNC_habitat_c_original_glcm_Idn -0.005288 *VNC_habitat_c_original_ngtdm_Contrast. The results of the prognostic feature screening are as follows: Dual-energy CT DECT-habitat_c (sub-area c) screens 18 features and calculates the radscore formula as follows Radscore = 0.35238095238095213 + +0.001487 * 40keV_habitat_c_original_firstorder_Minimum +0.031579 * 40keV_habitat_c_original_firstorder_RobustMeanAbsoluteDeviation -0.001231 * 40keV_habitat_c_original_firstorder_Skewness +0.079467 * 40keV_habitat_c_original_firstorder_TotalEnergy +0.074329 * 40keV_habitat_c_original_glrlm_ShortRunLowGrayLevelEmphasis -0.002762 * 40keV_habitat_c_original_glszm_SmallAreaEmphasis -0.029959 * IC_habitat_c_original_firstorder_RootMeanSquared -0.008022 * IC_habitat_c_original_glcm_Imc2 +0.041922 * IC_habitat_c_original_glcm_MaximumProbability-0.087525 * IC_habitat_c_original_glrlm_LongRunHighGrayLevelEmphasis +0.021877 * IC_habitat_c_original_glrlm_ShortRunLowGrayLevelEmphasis +0.022124* Mix_habitat_c_original_firstorder_Kurtosis +0.073967 * Mix_habitat_c_original_firstorder_TotalEnergy +0.004918 * Mix_habitat_c_original_glcm_InverseVariance +0.028064 * Mix_habitat_c_original_glcm_MCC -0.000804 * Mix_habitat_c_original_gldm_LargeDependenceLowGrayLevelEmphasis +0.038355 * VNC_habitat_c_original_glcm_Idn -0.005288 * VNC_habitat_c_original_ngtdm_Contrast. DECT-habitat_d (sub-area d) screens 4 features to calculate the Radscore formula as follows Radscore = 0.3523809523809524 + +0.036381 * 40keV_habitat_d_original_gldm_DependenceVariance +0.004236 * IC_habitat_d_original_glcm_ClusterShade -0.033471 * Mix_habitat_d_original_gldm_LargeDependenceHighGrayLevelEmphasis +0.052526 * VNC_habitat_d_original_gldm_LargeDependenceHighGrayLevelEmphasis DECT-radiomics (entire thrombus) screens 17 features and calculates the radscore formula as follows Radscore = 0.3523809523809524 + +0.095157 * 40keV_original_firstorder_RobustMeanAbsoluteDeviation -0.004094 * 40keV_original_glcm_InverseVariance +0.081543 * 40keV_original_glrlm_ShortRunLowGrayLevelEmphasis-0.013625 * 40keV_original_glszm_SmallAreaEmphasis -0.057676 * IC_original_glcm_SumAverage +0.003294 * IC_original_glrlm_GrayLevelNonUniformityNormalized +0.005728 * IC_original_glrlm_LowGrayLevelRunEmphasis -0.030762 * IC_original_glszm_SmallAreaEmphasis -0.003816 * Mix_original_gldm_SmallDependenceLowGrayLevelEmphasis -0.026314 * VNC_original_firstorder_Minimum -0.018390* VNC_original_glcm_MaximumProbability -0.043460 * VNC_original_gldm_SmallDependenceLowGrayLevelEmphasis -0.025927 * VNC_original_glrlm_ShortRunLowGrayLevelEmphasis -0.027315 * VNC_original_glszm_GrayLevelNonUniformityNormalized -0.009792 * VNC_original_glszm_LowGrayLevelZoneEmphasis +0.025086 * VNC_original_shape_Elongation +0.004321 * VNC_original_shape_Flatness DECT-habitats (habitat_c + habitat_d) screens 16 features to calculate the Radscore formula as follows Radscore = 0.35238095238095246 + +0.117752 * 40keV_habitat_c_original_firstorder_TotalEnergy +0.000573 * 40keV_habitat_c_original_ngtdm_Coarseness -0.004304 * 190keV_habitat_c_original_ngtdm_Busyness -0.012334 *IC_habitat_c_original_glcm_Imc2 -0.063545 * IC_habitat_c_original_glcm_SumAverage -0.036393 * IC_habitat_c_original_glrlm_LongRunHighGrayLevelEmphasis +0.009408 * IC_habitat_c_original_glszm_GrayLevelNonUniformityNormalized +0.009381 * Mix_habitat_c_original_firstorder_TotalEnergy +0.001150 * Mix_habitat_c_original_glcm_MCC -0.003359 * Mix_habitat_c_original_gldm_LargeDependenceLowGrayLevelEmphasis +0.092919 *40keV_habitat_d_original_gldm_DependenceVariance +0.005351 * IC_habitat_d_original_glcm_ClusterShade -0.001776 * IC_habitat_d_original_glcm_InverseVariance -0.039243 * Mix_habitat_d_original_gldm_LargeDependenceHighGrayLevelEmphasis +0.069095 * VNC_habitat_d_original_gldm_LargeDependenceHighGrayLevelEmphasis -0.009490 * VNC_habitat_d_original_gldm_LargeDependenceLowGrayLevelEmphasis. Step 3. Histological evaluation to correlate the different regions with tissue components.

[0019] Step 3.1 further explored the correlation with subregional distribution using histopathological analysis.

[0020] Step 3.2 Use image review software to photograph the hematoxylin and eosin-stained specimens of the removed thrombus sections.

[0021] Step 3.3: Manually analyze the area and proportion of RBCs and FPs using ImageJ software.

[0022] Step 4. Output to the user: Extract regional features to predict patient prognosis; fuse CTA images with DSA images for thrombus removal navigation, and automatically visualize different regions of thrombus composition.

[0023] The results are displayed on the display interface for the doctor's reference, including: Step 4.1 Mark different areas of thrombus on the image and display it to the user; Step 4.2: Mark each region of the thrombus as an RBC-rich region or a FP-rich region to predict the patient's prognosis; Step 4.3 uses existing medical image segmentation technology to segment the size and characteristics of each component of the thrombus and display them to the user.

[0024] Example 2 This embodiment describes a thrombus composition visualization navigation method based on conventional CT images, including the following steps: Step 1. Routine CT image assessment: thrombus delineation based on conventional CT images of stroke patients.

[0025] Step 1.1 Five parametric images were reconstructed based on conventional CT, including hybrid images, iodine concentration (IC) images, virtual single-energy (VM) images (40 keV and 190 keV), and virtual non-contrast (VNC) images.

[0026] Step 1.2: All images were normalized, image intensities were scaled to 0-600, and resampled to the same resolution (1 × 1 × 1 mm).

[0027] Step 1.3 discretizes and normalizes the voxel intensity values using a filter that reduces image noise.

[0028] Step 1.4 Use ITK-SNAP software to manually delineate the three-dimensional thrombus area.

[0029] Step 2. Feature selection and model building to divide the thrombus into different regions.

[0030] Step 2.1 Extract features: PyRadiomics software was used to extract 107 radiomic features from the entire thrombus and each subregion and normalized using Z scores.

[0031] Step 2.2 Feature screening: Feature selection was performed using the t-test (features with p < 0.05 were selected), Pearson correlation analysis (only one feature with correlation > 0.90 was retained), and LASSO (by introducing L1 regularization, the coefficients of unimportant features were reduced to zero, and non-zero features were retained). The results were then subjected to 10-fold cross-validation (to ensure stable results).

[0032] Step 2.3 Model construction: The non-zero features (obtained in step 2.2) were selected and the Radscore was calculated based on logistic regression, and then the whole thrombus and habitat models were constructed. Joint model, the joint model Type: CT_Combined=Habitat_a*(0.75346)+Habitat_b*(1.03235)+Habitats*7.26958+Radiomics*7.1291-5.26 Examples of sub-regions are as follows: CT-habitat_a (subregion a) screens 16 features and calculates the Radscore formula as follows Radscore = 0.3828125 + +0.043061 * CTA_habitat_a_firstorder_RootMeanSquared +0.058264 * CTA_habitat_a_firstorder_Skewness +0.003599 *CTA_habitat_a_glcm_InverseVariance +0.014099 * CTA_habitat_a_glrlm_GrayLevelVariance -0.017174 * CTA_habitat_a_glszm_LargeAreaHighGrayLevelEmphasis +0.071547 * CTA_habitat_a_ngtdm_Strength -0.014092 * NCCT_habitat_a_glcm_Correlation +0.029708 * NCCT_habitat_a_gldm_SmallDependenceLowGrayLevelEmphasis +0.000782 * NCCT_habitat_a_glszm_SizeZoneNonUniformityNormalized +0.020013 * NCCT_habitat_a_glszm_SmallAreaEmphasis +0.076785 * NCCT_habitat_a_glszm_ZonePercentage +0.009154 *NCCT_habitat_a_ngtdm_Contrast +0.019402 * NCCT_habitat_a_shape_Elongation +0.018980 * NCCT_habitat_a_shape_Flatness -0.043715 * NCCT_habitat_a_shape_Maximum2DDiameterColumn +0.170869 * NCCT_habitat_a_shape_VoxelVolume The formula for calculating Radscore by screening 16 features of CT-habitat_b (sub-region b) is as follows Radscore= 0.3828125 + +0.052601 * CTA_habitat_b_firstorder_Kurtosis -0.018983 * CTA_habitat_b_glcm_ClusterShade +0.001749 * CTA_habitat_b_glrlm_ShortRunHighGrayLevelEmphasis -0.057697 * CTA_habitat_b_glszm_LowGrayLevelZoneEmphasis +0.003328 * CTA_habitat_b_glszm_SizeZoneNonUniformityNormalized -0.000921 * CTA_habitat_b_glszm_SmallAreaLowGrayLevelEmphasis -0.012801 * NCCT_habitat_b_glcm_Idm +0.029571 *NCCT_habitat_b_glcm_Imc1 +0.025800 * NCCT_habitat_b_gldm_SmallDependenceLowGrayLevelEmphasis +0.040498 * NCCT_habitat_b_glszm_LargeAreaLowGrayLevelEmphasis +0.011794 * NCCT_habitat_b_glszm_SizeZoneNonUniformityNormalized +0.186759 * NCCT_habitat_b_glszm_SmallAreaEmphasis +0.030926 * NCCT_habitat_b_glszm_SmallAreaHighGrayLevelEmphasis -0.011968 * NCCT_habitat_b_glszm_ZoneVariance+0.047109 * NCCT_habitat_b_shape_Elongation +0.006321 * NCCT_habitat_b_shape_Flatness CT-radiomics (entire thrombus) screens 29 features and calculates the Radscore formula as follows Radscore = 0.3828125 + +0.000489 * CTA_original_firstorder_Skewness +0.020449 * CTA_original_glcm_DifferenceVariance -0.007742 * CTA_original_glcm_InverseVariance +0.005325 * CTA_original_glcm_MCC +0.061884 * CTA_original_gldm_DependenceVariance -0.005516 * CTA_original_glrlm_ShortRunLowGrayLevelEmphasis -0.108543 * CTA_original_glszm_GrayLevelNonUniformityNormalized +0.003116 * CTA_original_glszm_ZonePercentage -0.066804 *CTA_original_ngtdm_Busyness +0.033773 * NCCT_original_firstorder_10Percentile+0.004707 * NCCT_original_glcm_ClusterShade -0.007591 * NCCT_original_glcm_Idn +0.014942 * NCCT_original_glcm_InverseVariance +0.076770 * NCCT_original_glcm_MCC -0.092878 * NCCT_original_gldm_DependenceVariance +0.091351 * NCCT_original_gldm_SmallDependenceLowGrayLevelEmphasis +0.002434 * NCCT_original_glrlm_GrayLevelVariance +0.089165 * NCCT_original_glrlm_LowGrayLevelRunEmphasis +0.244005 * NCCT_original_glszm_SmallAreaEmphasis +0.038753 * NCCT_original_glszm_ZonePercentage +0.014454 * NCCT_original_ngtdm_Busyness +0.010654 * NCCT_original_ngtdm_Coarseness +0.085922 * NCCT_original_shape_Flatness -0.012143 * NCCT_original_shape_MajorAxisLength +0.023526 * NCCT_original_shape_Maximum2DDiameterRow -0.000935 * NCCT_original_shape_Maximum3DDiameter -0.048120 * NCCT_original_shape_MinorAxisLength +0.032352 * NCCT_original_shape_Sphericity +0.170552 * NCCT_original_shape_VoxelVolume. CT-habitats(habitat_a+habitat_b) includes 31 buildings Radscore= 0.38281250000000006 + +0.040767 * CTA_habitat_a_firstorder_Skewness +0.013049 * CTA_habitat_a_glcm_MCC -0.003579 * CTA_habitat_a_glrlm_GrayLevelNonUniformityNormalized -0.031086 * CTA_habitat_a_glszm_LargeAreaHighGrayLevelEmphasis -0.008793 * CTA_habitat_a_glszm_LowGrayLevelZoneEmphasis -0.000919 * CTA_habitat_a_glszm_SmallAreaLowGrayLevelEmphasis +0.077897 * CTA_habitat_a_ngtdm_Strength +0.002525 * NCCT_habitat_a_glcm_Imc1 +0.033532 * NCCT_habitat_a_gldm_SmallDependenceLowGrayLevelEmphasis +0.018574 * NCCT_habitat_a_glszm_SizeZoneNonUniformityNormalized +0.047894 * NCCT_habitat_a_glszm_ZonePercentage +0.019448 * NCCT_habitat_a_ngtdm_Contrast +0.021389 * NCCT_habitat_a_shape_Elongation +0.008654 * NCCT_habitat_a_shape_Flatness -0.031182 * NCCT_habitat_a_shape_Maximum2DDiameterColumn +0.118277 * NCCT_habitat_a_shape_VoxelVolume +0.020560 * CTA_habitat_b_firstorder_Kurtosis -0.031155 * CTA_habitat_b_glcm_ClusterShade +0.018330 * CTA_habitat_b_glrlm_ShortRunHighGrayLevelEmphasis -0.043590 * CTA_habitat_b_glszm_LowGrayLevelZoneEmphasis -0.008599 * CTA_habitat_b_glszm_SmallAreaLowGrayLevelEmphasis +0.005978 * CTA_habitat_b_ngtdm_Contrast +0.027956 * NCCT_habitat_b_firstorder_Skewness -0.002354 * NCCT_habitat_b_glcm_Idm +0.020820 * NCCT_habitat_b_glcm_Imc1 +0.054483 * NCCT_habitat_b_glszm_LargeAreaLowGrayLevelEmphasis +0.017017 * NCCT_habitat_b_glszm_SizeZoneNonUniformityNormalized +0.160695 * NCCT_habitat_b_glszm_SmallAreaEmphasis -0.019681 * NCCT_habitat_b_glszm_ZoneVariance +0.019863 *NCCT_habitat_b_shape_Elongation +0.017589 * NCCT_habitat_b_shape_Flatness. Step 3. Histological evaluation to correlate different regions with tissue components; Step 3.1 further explored the correlation with subregional distribution using histopathological analysis.

[0033] Step 3.2 Use image review software to photograph the hematoxylin and eosin-stained specimens of the removed thrombus sections.

[0034] Step 3.3: Manually analyze the area and proportion of RBCs and FPs using ImageJ software.

[0035] Step 4. Output to the user: Extract regional features to predict patient prognosis; fuse CTA images with DSA images for thrombus removal navigation, and automatically visualize different regions of thrombus composition.

[0036] The results are displayed on the display interface for the doctor's reference, including: Step 4.1 Mark different areas of thrombus on the image and display it to the user; Step 4.2: Mark each region of the thrombus as an RBC-rich region or a FP-rich region to predict the patient's prognosis; Step 4.3 uses existing medical image segmentation technology to segment the size and characteristics of each component of the thrombus and display them to the user.

[0037] Example 3 Through internal and external validation, it was found that the CT-habitat_b and DECT-habitat_c sub-region models had better predictive effects than the entire thrombosis (radiomics) model, and the combined model was found to be the most effective.

[0038] The joint model is calculated as follows: CT_Combined=Habitat_a*(0.75346)+Habitat_b*(1.03235)+Habitats*7.26958+Radiomics*7.1291-5.26 DECT_Combined=Habitat_c*2.8478+Habitat_d*(-1.09809)+Habitats*4.77103+Radiomics*4.34301-4.78027 The above description is only a description of the preferred embodiments of the present application and does not limit the scope of the present application. Any changes or modifications made by any person skilled in the art based on the above disclosed technical content should be regarded as equivalent valid embodiments and fall within the scope of protection of the technical solution of the present application.

Claims

1. A thrombus composition visualization navigation method based on CT images, characterized in that: The steps include: Step 1. CT image assessment: outline the thrombus based on the CT image of the stroke patient; Step 2. Feature selection and model building to divide the thrombus into different regions; Step 3. Histological evaluation to correlate different regions with tissue components; Step 4. Output to the user: Extract regional features to predict patient prognosis; fuse CTA images with DSA images for thrombus removal navigation, and automatically visualize different regions of thrombus composition.

2. The thrombus composition visualization navigation method based on CT images according to claim 1, characterized in that: The step 1 comprises: Step 1.1 Reconstruct five parametric images based on CT, including hybrid images, iodine concentration (IC) images, virtual single-energy (VM) images, and virtual non-contrast (VNC) images; Step 1.2: Normalize all images, scale the image intensities to 0-600, and resample to the same resolution. Step 1.3: discretize and normalize the voxel intensity values using a filter that reduces image noise. Step 1.4 Use ITK-SNAP software to manually delineate the three-dimensional thrombus area.

3. The thrombus composition visualization navigation method based on CT images according to claim 1, characterized in that: The step 2 includes: Step 2.1 Extract features: 107 radiomic features were extracted from the entire thrombus and each subregion using PyRadiomics software and normalized using Z scores; Step 2.2 Feature screening: The t-test, Pearson correlation analysis, and LASSO were used for feature selection, and the results were subjected to 10-fold cross validation; Step 2.3 Model construction: Based on step 2.2, the Radscore of the selected non-zero features was calculated based on logistic regression, and then the entire thrombus and habitat models were constructed. A joint model of the entire thrombus and each habitat model was constructed through logistic regression.

4. The thrombus composition visualization navigation method based on CT images according to claim 1, characterized in that: The step 3 comprises: Step 3.1: Further explore the correlation with subregional distribution using histopathological analysis; Step 3.2: Use image review software to photograph the hematoxylin and eosin-stained specimens of the removed thrombus sections. Step 3.3: Manually analyze the area and proportion of RBCs and FPs using ImageJ software.

5. The thrombus composition visualization navigation method based on CT images according to claim 1, characterized in that: In step 4, the results are displayed on the display interface for the doctor's reference, and the displayed content includes: Step 4.1 Mark different areas of thrombus on the image and display it to the user; Step 4.2: Mark each region of the thrombus as either an RBC-rich region or a FP-rich region to predict the patient's prognosis; Step 4.3: Using existing medical image segmentation technology, the sizes and features of the various components of the thrombus are segmented and displayed to the user.

Images