A hepatocellular carcinoma blood vessel pattern analysis system and application thereof

Abstract

The application relates to the technical field of digital pathological image analysis, and particularly discloses a hepatocellular carcinoma blood vessel pattern analysis system and application thereof. The hepatocellular carcinoma blood vessel pattern analysis system comprises an image acquisition module, a tumor region detection module, a blood vessel segmentation module, a blood vessel subtype classification module and a blood vessel pattern analysis module. The blood vessel subtype classification module is used for identifying blood vessel subtypes in a blood vessel binary mask image. The blood vessel pattern analysis module is used for identifying a blood vessel pattern, and the hepatocellular carcinoma is classified and prognosis evaluation is performed according to the blood vessel pattern. The system constructed by the application can realize automatic identification of blood vessel structures and subtypes in hepatocellular carcinoma tissues, and can stably and accurately complete hepatocellular carcinoma classification and determination based on the hepatocellular carcinoma blood vessel pattern, thereby solving the technical bottlenecks that the hepatocellular carcinoma blood vessel evaluation in the prior art depends on artificial interpretation, is highly subjective, has poor consistency, and is difficult to realize quantitative analysis of blood vessel morphological characteristics, and lacks precise hepatocellular carcinoma classification and prognosis evaluation based on the blood vessel pattern.

Description

Technical Field

[0001] This application relates to the field of digital pathological image analysis technology, and in particular to a hepatocellular carcinoma vascular pattern analysis system and its application. Background Technology

[0002] Hepatocellular carcinoma (HCC) is one of the most common malignant tumors worldwide, and its invasion and metastasis are closely related to tumor angiogenesis. Tumor angiogenesis plays a crucial role in tumor growth, microenvironment remodeling, and immune escape, and has become an important entry point for research on the pathological mechanisms of this disease. Among existing assessment methods, microvessel density is widely used as a quantitative indicator to measure the degree of tumor angiogenesis. However, this indicator only reflects the number of blood vessels per unit area and cannot characterize the spatial structural features and morphological heterogeneity of blood vessels. Therefore, it has significant limitations in revealing the structure-function relationship between angiogenesis and tumor progression.

[0003] In recent years, with the deepening of histopathological research, tumor-encircling vascular structures (VETCs) have been identified and attracted widespread attention in hepatocellular carcinoma tissues. These structures exist as clusters of tumor cells surrounded by blood vessels and play an important role in the non-invasive metastasis of tumors. Currently, the identification and quantitative assessment of these vascular structures still mainly rely on manual observation and judgment by pathologists. However, manual assessment is not only time-consuming and labor-intensive, but also easily affected by differences in observer experience, exhibiting strong subjectivity and making it difficult to achieve large-scale, standardized quantitative analysis in clinical practice. Furthermore, in analyses such as tumor prognosis prediction, existing technologies mainly use a single vascular type as an analytical indicator, lacking combined analysis and comprehensive judgment of different vascular types.

[0004] With the continuous maturation of digital pathology scanning technology and the widespread application of deep learning in medical image analysis, automated analysis based on full-view pathology slide images has become possible. Therefore, constructing an analysis system capable of automatically identifying different vascular structures in hepatocellular carcinoma regions, quantitatively describing their morphological characteristics, and assessing tumor prognosis based on the abundance of these different vascular structures is of great significance for advancing research on tumor angiogenesis mechanisms and clinical pathological prognostic assessment, and is a technical problem urgently needing to be solved in this field. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a hepatocellular carcinoma vascular pattern analysis system and its application, which is used to solve the technical bottleneck of the prior art that the assessment of hepatocellular carcinoma vascular patterns is based on a single vascular morphology and relies on manual interpretation, which is highly subjective, inconsistent, and difficult to achieve quantitative, comprehensive analysis and refined prognostic assessment of multiple different vascular morphological features.

[0006] To achieve the above technical objectives, this application provides a hepatocellular carcinoma vascular pattern analysis system, including an image acquisition module, a tumor region detection module, a vascular segmentation module, a vascular subtype classification module, and a vascular pattern analysis module; The image acquisition module is used to identify tissue sections and prepare full-view pathological section images; The tumor region detection module is used to identify tumor regions in full-view pathological slide images and output tumor region mask images; The blood vessel segmentation module is used to segment the blood vessel structure within the tumor region mask image and generate a binary mask image of blood vessels. The vessel subtype classification module is used to identify vessel subtypes in binary mask images of vessels; The vascular pattern analysis module is used to calculate the microvessel density of vascular subtypes. It uses a robust normalization method to standardize the microvessel density, classifies the abundance level of vascular subtypes and identifies vascular patterns based on preset thresholds, and performs hepatocellular carcinoma classification and prognostic assessment based on vascular patterns.

[0007] Furthermore, the vascular subtypes include linear vessels, branched vessels, dilated vessels, and tumor-encircled vessels; among them, linear vessels are slender capillary-like structures without obvious branches; branched vessels are vascular structures with at least one branch point; dilated vessels are vascular structures with a maximum lumen diameter exceeding 20 μm; and tumor-encircled vessels are network vascular structures formed by surrounding tumor cell clusters.

[0008] Furthermore, the vascular pattern analysis module has a preset threshold of 0.5, and the abundance level of vascular subtypes is divided according to the preset threshold; among them, a microvessel density value greater than or equal to 0.5 obtained by standardization is considered high abundance, and a microvessel density value less than 0.5 obtained by standardization is considered low abundance.

[0009] Furthermore, based on vascular patterns, hepatocellular carcinoma (HCC) is classified into HCC type 1, HCC type 2, HCC type 3, and HCC type 4. When the vascular pattern shows low abundance of all four vascular subtypes, HCC is classified as HCC type 1. When the vascular pattern shows high abundance of linear and / or branched vessels, and low abundance of dilated vessels and tumor-encased vessels, HCC is classified as HCC type 2. When the vascular pattern shows high abundance of dilated vessels and low abundance of tumor-encased vessels, HCC is classified as HCC type 3. When the vascular pattern shows high abundance of tumor-encased vessels, HCC is classified as HCC type 4.

[0010] Furthermore, the vascular pattern analysis system of the present invention includes a prognostic prediction module; the prognostic prediction module is used to classify patients with type 1 or type 2 liver cancer into a low-risk group and patients with type 3 or type 4 liver cancer into a high-risk group; the low-risk group corresponds to a longer recurrence-free survival time, and the high-risk group corresponds to a shorter recurrence-free survival time.

[0011] Furthermore, the prognostic prediction module classifies type 2 hepatocellular carcinoma into the lowest risk subgroup, type 1 hepatocellular carcinoma into the low-to-medium risk subgroup, type 3 hepatocellular carcinoma into the medium-to-high risk subgroup, and type 4 hepatocellular carcinoma into the highest risk subgroup.

[0012] Furthermore, the tumor region detection module is signal-connected to the image acquisition module; the tumor region detection module has a built-in tumor region detection model based on the ResNet-50 backbone network. The tumor region detection model is used to receive full-view pathological slide images, magnify and pixel-block the full-view pathological slide images, identify tumor regions, and output a tumor region mask map.

[0013] Furthermore, the blood vessel segmentation module is signal-connected to the tumor region detection module; the blood vessel segmentation module has a built-in blood vessel segmentation model based on the U-Net architecture combined with the ResNet-50 encoder. The blood vessel segmentation model is used to receive the tumor region mask image, enlarge the image, perform pixel block processing, segment the blood vessel structure within the tumor region, and generate a binary mask image of blood vessels.

[0014] Furthermore, the vascular pattern analysis module is signal-connected to the vascular subtype classification module; the vascular pattern analysis module has a built-in statistical analysis unit adapted to R language, Python, and GraphPad Prism software.

[0015] Furthermore, the hepatocellular carcinoma vascular pattern analysis system provided by the present invention is a computer system; even further, the system is a computer system based on artificial intelligence.

[0016] This application provides a hepatocellular carcinoma prognostic assessment device, including a hepatocellular carcinoma vascular pattern analysis system.

[0017] In summary, this application provides a hepatocellular carcinoma vascular pattern analysis system, including an image acquisition module, a tumor region detection module, a blood vessel segmentation module, a blood vessel subtype classification module, and a blood vessel pattern analysis module. The image acquisition module is used to identify tissue sections and prepare full-view pathological slide images. The tumor region detection module is used to identify tumor regions in the full-view pathological slide images and output tumor region mask images. The blood vessel segmentation module is used to segment vascular structures within the tumor region mask images and generate binary vascular mask images. The blood vessel subtype classification module is used to identify vascular subtypes in the binary vascular mask images. The blood vessel pattern analysis module is used to identify vascular patterns and perform hepatocellular carcinoma subtyping and prognostic assessment based on vascular patterns. The hepatocellular carcinoma vascular pattern analysis system constructed by this invention can achieve automated identification of different vascular structures and their subtypes in hepatocellular carcinoma tissues, and can stably and accurately complete hepatocellular carcinoma subtyping and recurrence risk determination based on hepatocellular carcinoma vascular patterns, providing standardized and repeatable technical support for clinical prognostic assessment.

[0018] Compared to existing technologies, the hepatocellular carcinoma vascular pattern analysis system can operate automatically, accurately quantify different vascular features, construct tumor vascular patterns, and classify liver cancer and determine recurrence risk based on vascular patterns, providing a reliable and refined prognostic assessment strategy. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the workflow of the hepatocellular carcinoma vascular pattern analysis system.

[0021] Figure 2 This is a schematic diagram of the morphology of a vascular subtype.

[0022] Figure 3 This is a schematic diagram of the recognition performance of Example 1 on the validation set of Sun Yat-sen University Cancer Center; where A is the percentage normalized confusion matrix of the tumor region detection model; B is the ROC curve of the tumor region detection model; C is a schematic diagram of the segmentation results of the blood vessel segmentation model; D is the percentage normalized confusion matrix of the blood vessel subtype classification model; and E is the ROC curve of the blood vessel subtype classification model.

[0023] Figure 4 The following are the vascular pattern recognition results and risk stratification survival curves for Example 5 in cohorts 1, 2, and 3; where: A is a heatmap of the vascular subtype density normalization results in each cohort; B is the Kaplan-Meier survival curve of recurrence-free survival for patients with different vascular patterns.

[0024] Figure 5 The survival curves for risk stratification of each vascular pattern in queues 1, 2, and 3, combined with the other three vascular patterns, are shown in Example 5. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments in this application specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection claimed in this application.

[0026] In the description of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," indicating orientation or positional relationships, are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this application. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0027] Unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0028] The raw materials used in this invention are not particularly restricted in their source; they can be purchased on the market or prepared using conventional methods known to those skilled in the art.

[0029] This application provides a hepatocellular carcinoma vascular pattern analysis system, including an image acquisition module, a tumor region detection module, a vascular segmentation module, a vascular subtype classification module, and a vascular pattern analysis module; The image acquisition module is used to identify tissue sections and prepare full-view pathological section images; The tumor region detection module is used to identify tumor regions in full-view pathological slide images and output tumor region mask images; The blood vessel segmentation module is used to segment the blood vessel structure within the tumor region mask image and generate a binary mask image of blood vessels. The vessel subtype classification module is used to identify vessel subtypes in binary mask images of vessels; The vascular pattern analysis module is used to calculate the microvessel density of vascular subtypes. It uses a robust normalization method to standardize the microvessel density, classifies the abundance level of vascular subtypes and identifies vascular patterns based on preset thresholds, and performs hepatocellular carcinoma classification and prognostic assessment based on vascular patterns.

[0030] It should be noted that, Figure 1 This is a schematic diagram illustrating the workflow of the hepatocellular carcinoma vascular pattern analysis system of the present invention. From... Figure 1As can be seen, this application achieves fully automated analysis of the hepatocellular carcinoma vascular structure, from pathological image acquisition and vascular segmentation to vascular pattern recognition, effectively avoiding subjective bias caused by manual interpretation and significantly improving the objectivity and consistency of vascular assessment results. It should also be noted that in some embodiments, the hepatocellular carcinoma vascular pattern analysis system provided by this invention is a computer system, which is an artificial intelligence-based computer system.

[0031] In some embodiments, vascular subtypes include linear vessels, branching vessels, dilated vessels, and tumor-encircled vessels. See [link to morphological descriptions of each vascular subtype] for details. Figure 2 Linear vessels are slender, capillary-like structures without obvious branches; branched vessels are vascular structures with at least one branch point; dilated vessels are vascular structures with a maximum lumen diameter greater than 20 μm, that is, in dilated vessels, where the lumen exists, its diameter is greater than 20 μm; tumor-encircled vessels are network vascular structures formed by surrounding tumor cell clusters.

[0032] It should be noted that the technical solution provided in this application breaks through the limitations of judging by a single vascular feature. By identifying and quantifying four types of vascular subtypes in pathological images—linear vessels, branching vessels, dilated vessels, and tumor clusters surrounding vessels—multi-dimensional vascular-related parameters are extracted. After joint analysis by a computer program, liver cancer is effectively classified into four types with different recurrence risks. Each subtype is determined by a combination of multiple vascular features, rather than by a single indicator threshold, which greatly improves the scientificity and pertinence of the subtype.

[0033] In some embodiments, the vascular pattern analysis module has a preset threshold of 0.5, and the abundance level of vascular subtypes is divided according to the preset threshold; wherein, a microvessel density value greater than or equal to 0.5 obtained by standardization is considered high abundance, and a microvessel density value less than 0.5 obtained by standardization is considered low abundance.

[0034] In some embodiments, liver cancer is classified into types 1, 2, 3, and 4 based on vascular patterns. When the vascular pattern shows low abundance of all four vascular subtypes, the liver cancer is classified as type 1. When the vascular pattern shows high abundance of linear and / or branched vessels, and low abundance of dilated vessels and tumor-encased vessels, the liver cancer is classified as type 2. When the vascular pattern shows high abundance of dilated vessels and low abundance of tumor-encased vessels, the liver cancer is classified as type 3. When the vascular pattern shows high abundance of tumor-encased vessels, the liver cancer is classified as type 4.

[0035] It should be noted that the vascular patterns described in this application can be divided into four categories. Based on each vascular pattern, liver cancer can be classified into the above four types. Specifically, when all four vascular subtypes are of low abundance, it can be classified as the first vascular pattern, and liver cancer is classified as type 1 liver cancer based on this vascular pattern. When linear vessels and / or branched vessels are of high abundance, and dilated vessels and tumor-encased vessels are of low abundance, it can be classified as the second vascular pattern, and liver cancer is classified as type 2 liver cancer based on this vascular pattern. When dilated vessels are of high abundance and tumor-encased vessels are of low abundance, it can be classified as the third vascular pattern, and liver cancer is classified as type 3 liver cancer based on this vascular pattern. When tumor-encased vessels are of high abundance, it can be classified as the fourth vascular pattern, and liver cancer is classified as type 4 liver cancer based on this vascular pattern.

[0036] In some embodiments, the hepatocellular carcinoma vascular pattern analysis system provided by the present invention also includes a prognostic prediction module; the prognostic prediction module is used to classify patients with type 1 or type 2 hepatocellular carcinoma into a low-risk group and patients with type 3 or type 4 hepatocellular carcinoma into a high-risk group; the low-risk group corresponds to a longer recurrence-free survival time, and the high-risk group corresponds to a shorter recurrence-free survival time.

[0037] In some embodiments, the prognostic prediction module classifies hepatocellular carcinoma type 2 into the lowest risk subgroup, hepatocellular carcinoma type 1 into the low-to-medium risk subgroup, hepatocellular carcinoma type 3 into the medium-to-high risk subgroup, and hepatocellular carcinoma type 4 into the highest risk subgroup.

[0038] It should be noted that this application proposes a prognostic prediction module based on vascular patterns, which can effectively distinguish hepatocellular carcinoma patients with different recurrence risks, providing new specific indicators for clinical prognostic assessment and having important clinical guiding significance for optimizing clinical treatment plans and improving patient prognosis.

[0039] In some embodiments, the tumor region detection module is signal-connected to the image acquisition module; the tumor region detection module has a built-in tumor region detection model based on the ResNet-50 backbone network. The tumor region detection model is used to receive full-view pathological slide images, enlarge and pixel-block process the full-view pathological slide images, identify tumor regions and output tumor region mask images.

[0040] In some embodiments, the blood vessel segmentation module is signal-connected to the tumor region detection module; the blood vessel segmentation module has a built-in blood vessel segmentation model based on the U-Net architecture combined with the ResNet-50 encoder. The blood vessel segmentation model is used to receive the tumor region mask image, enlarge the image, perform pixel block processing, segment the blood vessel structure within the tumor region, and generate a binary mask image of blood vessels.

[0041] In some embodiments, in the hepatocellular carcinoma vascular pattern analysis system, the vascular pattern analysis module and the vascular subtype classification module are signal-connected. The vascular pattern analysis module has a built-in statistical analysis unit adapted to R language, Python, and GraphPad Prism software. The statistical analysis unit is used to calculate the microvessel density of vascular subtypes, standardize the microvessel density using a robust normalization method, classify the abundance level of vascular subtypes according to preset thresholds and identify vascular patterns, and perform prognostic assessment of hepatocellular carcinoma based on vascular patterns.

[0042] This application provides a hepatocellular carcinoma prognostic assessment device, including a hepatocellular carcinoma vascular pattern analysis system.

[0043] The applicant further provides the following specific embodiments to describe the present invention. It should be noted that these embodiments are merely descriptive and do not limit the present invention in any way.

[0044] HCC tumor samples were collected from the human biobanks of four medical institutions: Sun Yat-sen University Cancer Center (SYSUCC), the First Affiliated Hospital of Sun Yat-sen University (SYSUFH), the Third Affiliated Hospital of Sun Yat-sen University (SYSUTH), and Guangzhou Medical University Cancer Center (GMUCC). The diagnosis of HCC was confirmed by histological examination. All patients signed written informed consent forms, the research protocol was approved by the institution's research ethics committee, and the principles of the Declaration of Helsinki were followed.

[0045] Inclusion criteria include: (1) no history of systemic anticancer therapy (such as sorafenib or other drugs) or local area therapy (such as transcatheter arterial chemoembolization or radiofrequency ablation) before recurrence; (2) Barcelona Clinical Hepatocellular Carcinoma (BCLC) stage AC; (3) complete follow-up data; and (4) no recurrence or HCC-related death within 1 month after surgery.

[0046] Example 1

[0047] This embodiment provides a method for constructing and classifying hepatocellular carcinoma vascular patterns, including the following steps: Step S1, Pathological image acquisition: We collected hepatocellular carcinoma (HCC) tissue sections and corresponding clinical data from multiple medical centers, including tissue samples from 971 HCC patients. All tissue sections underwent CD34 immunohistochemical staining, and whole-slide images (WSI) were prepared. All patients were divided into three study cohorts: cohort 1 (388 patients) and cohort 2 (384 patients) were randomly assigned from Sun Yat-sen University Cancer Center; and cohort 3 (199 patients) was comprised of patients from the First Affiliated Hospital of Sun Yat-sen University, the Third Affiliated Hospital of Sun Yat-sen University, and Guangzhou Medical University Cancer Hospital. The obtained CD34-stained WSI images were used for subsequent deep learning model construction and vascular pattern analysis.

[0048] Step S2, Tumor Region Detection: 216 WSI images were manually labeled to distinguish tumor regions from normal tissue regions. The WSI images were then segmented into 512-pixel blocks at a 10x magnification, constructing a tumor region detection dataset containing 2,361,436 tumor image blocks and 3,000,738 normal tissue image blocks. This dataset was randomly divided into a training dataset and an internal validation dataset in an 8:2 ratio, with 80% used for model training and 20% for internal model validation. The tumor region detection model was built based on a ResNet-50 backbone network. The model was trained using a 512×512 pixel image patch at a magnification factor of 100% as input. The initial learning rate was set to 0.001. During training, the learning rate was dynamically adjusted using a cosine annealing strategy. At the same time, a cross-entropy loss function based on class weights was used to alleviate the class imbalance problem. The model training cycle was controlled within 100 epochs. If the validation set loss did not decrease for 15 consecutive epochs, an early termination mechanism was triggered. The checkpoint with the lowest cross-entropy loss on the validation set was selected as the optimal model. After training, the model was applied to all WSI images to automatically and accurately identify tumor regions.

[0049] Step S3, vessel segmentation: Tumor image patches with dimensions of 1024 pixels were randomly selected from 274 WSI images at a magnification of 10x. The vascular structures in these patches were manually annotated by professionals, forming a vascular segmentation dataset containing 1378 vascular mask images and corresponding tumor patches. The dataset was randomly divided into a training dataset and an internal validation dataset in an 8:2 ratio. The vascular segmentation model adopted the U-Net architecture combined with a ResNet-50 encoder and was trained based on the annotated 1024×1024 pixel vascular masks. The initial learning rate was set to 0.01, and the stochastic gradient descent (SGD) optimizer was selected with a momentum parameter of 0.9 and a weight decay coefficient of 0.0005. The learning rate was adjusted using a cosine annealing strategy, and the checkpoint with the smallest Dice loss value was selected as the optimal model. This model was used to automatically segment the vascular structures in the tumor region, resulting in binary mask images of the vascular region. Step S4, vascular subtype classification: Based on 337 WSI images, individual vessels were extracted to construct a hepatocellular carcinoma vascular dataset containing 61,699 single vessel images. This dataset includes 22,882 linear vessels (LV), 13,146 branched vessels (BV), 15,055 dilated vessels (DV), and vessels encapsulating tumors. The VETC clusters (10616 clusters) were divided into training and internal validation sets in an 8:2 ratio. The vascular type classifier was built based on the ResNet-50 backbone network, with the last layer configured as a four-classifier to distinguish four vascular subtypes. The four vascular subtypes were strictly defined according to morphological criteria: linear vessels are slender capillary-like structures without obvious branches; branched vessels have at least one branch point; dilated vessels are vascular structures with significantly enlarged lumens and a maximum lumen diameter greater than 20 μm; and tumor-encircled vessels are reticular vascular structures formed by clusters of tumor cells. The initial learning rate of the classification model was set to 0.001, and the same optimizer and learning rate adjustment strategy were used, along with a class weight cross-entropy loss function to solve the sample imbalance problem, thus completing the automatic identification of vascular subtypes.

[0050] Step S5, Vascular pattern analysis: After obtaining information on each vascular subtype, vascular pattern recognition and risk stratification were carried out. All statistical analyses were performed using R 4.2.1, Python 3.8.5, and GraphPad Prism 10.2.3 software. Model performance was comprehensively evaluated using classification indicators including accuracy, sensitivity, specificity, positive predictive value, and negative predictive value, as well as segmentation indicators including average pixel accuracy and average intersection-union ratio. All indicators were calculated based on the confusion matrix. See Equation (1) to calculate the microvascular density (MVDi) of each vascular subtype, i.e., the vascular density per unit area (mm²), by the tumor area weighted average. Equation (1); Where K refers to the number of tissue samples a patient has, and NK is the number of tumor plaques on the k-th tissue sample (k=1, 2, ..., K); n i，k This represents the vessel count for each vessel subtype (i) in the k-th slice, where i represents LV, BV, DV, or VETC, and TA represents the tumor area (1.16 mm²) in each tumor image patch. 2 ).

[0051] To standardize the density of vascular subtypes and reduce the influence of outliers, a robust scaling method is used for normalization, as shown in Equation (2): MVDi is converted into scaling values ​​using the median Q2 and the interquartile range IQR (Q3-Q1, where Q1 and Q3 are the 25th and 75th percentiles, respectively). The interquartile values ​​of the density of each vascular subtype are shown in Table 1.

[0052] Equation (2); Among them, MVD i It is the unit area (mm²) 2 )Vascular density, Q2 is the median, and IQR is the difference between the upper quartile Q3 and the lower quartile Q1; After screening and setting a scaling threshold of 0.5, vascular subtypes with a scaling value ≥ 0.5 were classified into the high abundance group, and vascular subtypes with a scaling value < 0.5 were classified into the low abundance group. Vascular patterns were then classified according to abundance levels, and hepatocellular carcinoma (HCC) was classified into four types accordingly. When all four vascular subtypes were low abundance, it was defined as HCC-1 (hereinafter referred to as HCC-1). When linear or branching vessels were high abundance, and dilated vessels and VETC were low abundance, it was defined as HCC-2 (hereinafter referred to as HCC-2). When dilated vessels were high abundance and VETC were low abundance, regardless of the levels of other subtypes, it was defined as HCC-3 (hereinafter referred to as HCC-3). When VETC was high abundance, it was directly defined as HCC-4 (hereinafter referred to as HCC-4), regardless of the levels of other subtypes.

[0053] Table 1. 25th, 50th, and 75th percentiles of vascular subtype density

[0054] Test Example 1

[0055] To evaluate the performance of the hepatocellular carcinoma vascular pattern analysis system, this application systematically tested the tumor region detection module, vascular segmentation module, and vascular subtype classification module on an independent validation set containing 20% ​​of the samples in the aforementioned dataset.

[0056] Test results are as follows Figure 3 As shown, the positive predictive value of the tumor region detection model was 0.9823, the negative predictive value was 0.9858, the accuracy was 0.9802, and the area under the ROC curve reached 0.9901 (see [reference]). Figure 3 (A and B in the original text). The average cross-union ratio of the blood vessel segmentation network is 0.8454, and the average pixel accuracy (mPA) is 0.9232 (see A and B in the original text). Figure 3 (C in the text). The overall accuracy of the vascular subtype classification model is 0.9519 (see [reference]). Figure 3 The classification accuracy for the four vascular subtypes LV, BV, DV, and VETC was 0.9720, 0.9702, 0.9783, and 0.9844, respectively, with corresponding AUC values ​​of 0.9812, 0.9780, 0.9861, and 0.9673 (see D in the original text). Figure 3 (E in the figure). The above results fully demonstrate that the system exhibits excellent performance in key tasks such as tumor region identification, vascular structure segmentation, and accurate classification of vascular subtypes.

[0057] Test Example 2: Multicenter cohort studies have validated the predictive value of vascular patterns for the prognosis of HCC patients.

[0058] Test results are as follows Figures 4-5 As shown, Figure 4 In the figure, A represents the heatmap of vascular subtype characteristics, and four vascular patterns were stably identified in all three independent cohorts. Figure 4 B in the figure represents Kaplan-Meier survival curve analysis, which shows that the classification of hepatocellular carcinoma based on different vascular patterns is significantly associated with postoperative recurrence-free survival. HCC-2 had the longest TTR, HCC-4 had the shortest TTR, and HCC-1 and HCC-3 had intermediate prognoses. Risk stratification analysis was performed by combining each vascular pattern with the other three patterns, and the results are as follows... Figure 5As shown, compared to HCC-3 and HCC-4, HCC-1 and HCC-2 have longer TTRs, allowing HCC-3 and HCC-4 to be classified as high-risk groups, and HCC-1 and HCC-2 as low-risk groups. Furthermore, HCC-2 exhibited significantly longer TTRs in all three cohorts (HR < 0.55, P < 0.005), suggesting it is the lowest recurrence risk subtype; HCC-4 was associated with significantly shorter TTRs in all three cohorts (HR > 1.86, P < 0.0004), suggesting it is the highest recurrence risk subtype. Therefore, HCC-2 in the low-risk group can be further classified as the lowest-risk subgroup, HCC-1 as the low-to-intermediate-risk subgroup, HCC-3 in the high-risk group as the intermediate-to-high-risk subgroup, and HCC-4 as the highest-risk subgroup.

[0059] Example 2

[0060] This embodiment provides a hepatocellular carcinoma vascular pattern analysis system, which includes an image acquisition module, a tumor region detection module, a vascular segmentation module, a vascular subtype classification module, a vascular pattern analysis module, and a prognosis prediction module. The image acquisition module serves as the system's input, used to acquire tissue sections from hepatocellular carcinoma patients, perform CD34 immunohistochemical staining on the tissue sections, and prepare WSI images. Simultaneously, it acquires the patient's corresponding clinical data. After standardized preprocessing, the WSI images and clinical data are transmitted to the tumor region detection module. The tumor region detection module is connected to the image acquisition module by signal. It has a built-in tumor region detection model based on the ResNet-50 backbone network. It receives WSI images transmitted from the image acquisition module, performs 10x magnification and 512×512 pixel block processing on the images, and automatically identifies tumor regions and normal tissue regions through the trained model. It outputs a tumor region mask image and transmits it synchronously to the blood vessel segmentation module. The blood vessel segmentation module is connected to the tumor region detection module. It uses a U-Net architecture combined with a ResNet-50 encoder to build a blood vessel segmentation model. It receives the tumor region mask image, performs 10x magnification and 1024×1024 pixel block processing on the image, automatically segments the blood vessel structure in the tumor region mask image, generates a binary blood vessel mask image, and transmits it to the blood vessel subtype classification module. The vessel subtype classification module is signal-connected to the vessel segmentation module. It has a built-in four-class classification model based on the ResNet-50 backbone network. It receives binary mask images of vessels and extracts individual vessel images. It identifies the subtype of each vessel and outputs the classification results of LV, BV, DV, and VETC, which are synchronously transmitted to the vessel pattern analysis module. The vessel subtype classification is based on the following criteria: LV is a slender capillary-like structure without obvious branches; BV has at least one branch point; DV is a vessel structure with a significantly enlarged lumen and a maximum lumen diameter greater than 20μm; and VETC is a network of vessels formed around clusters of tumor cells. The vascular pattern analysis module is signal-connected to the vascular subtype classification module and has a built-in statistical analysis unit (compatible with R4.2.1, Python 3.8.5, and Graph Pad Prism 10.2.3 software). It receives the vascular subtype classification results, calculates the MVDi of each vascular subtype, standardizes the MVDi using a robust normalization method, and classifies the abundance level of vascular subtypes based on a preset threshold of 0.5. Then, it identifies four types of liver cancer based on vascular patterns: HCC-1, HCC-2, HCC-3, and HCC-4, according to established rules, and transmits the vascular pattern classification results to the prognosis prediction module. The four hepatocellular carcinoma (HCC) types based on vascular patterns are as follows: HCC-1 is defined when all four vascular subtypes are low in abundance; HCC-2 is defined when linear or branching vessels are high in abundance and dilated vessels and VETC are low in abundance; HCC-3 is defined when dilated vessels are high in abundance and VETC are low in abundance, regardless of the levels of other subtypes; and HCC-4 is defined when VETC are high in abundance, unaffected by the levels of other subtypes.

[0061] Example 3

[0062] This embodiment provides a hepatocellular carcinoma vascular pattern analysis system. The analysis system is based on the analysis system provided in Embodiment 2 and also includes a prognostic prediction module. The prognostic prediction module is used to classify patients classified as HCC-1 or HCC-2 based on vascular pattern into a low-risk group and patients classified as HCC-3 or HCC-4 based on vascular pattern into a high-risk group. The low-risk group corresponds to a longer recurrence-free survival time, and the high-risk group corresponds to a shorter recurrence-free survival time.

[0063] Example 4

[0064] This embodiment provides a hepatocellular carcinoma vascular pattern analysis system. The analysis system is based on the analysis system provided in Embodiment 2 and also includes a prognostic prediction module. The prognostic prediction module divides HCC-1 in the low-risk group into a low-to-medium risk subgroup and HCC-2 into the lowest risk subgroup; it divides HCC-3 in the high-risk group into a medium-to-high risk subgroup and HCC-4 into the highest risk subgroup.

[0065] Test Example 3

[0066] This embodiment, based on the hepatocellular carcinoma vascular pattern analysis system provided in Embodiment 3, performs prognostic risk stratification analysis on hepatocellular carcinoma patients. The specific implementation process is as follows: After classifying the vascular patterns of all included patients, patients classified as HCC-1 or HCC-2 based on vascular patterns are divided into a low-risk group, and further divided into the lowest-risk subgroup (HCC-2) and the low-to-intermediate-risk subgroup (HCC-1); patients classified as HCC-3 or HCC-4 based on vascular patterns are divided into a high-risk group, and further divided into the highest-risk subgroup (HCC-4) and the intermediate-to-high-risk subgroup (HCC-3). Long-term follow-up data of the two groups of patients are collected, and the recurrence-free survival time of patients in different risk groups is statistically analyzed.

[0067] Test results showed that patients in the low-risk group had relatively longer recurrence-free survival and a lower risk of recurrence, especially the lowest-risk group; while patients in the high-risk group had relatively shorter recurrence-free survival and a significantly increased risk of recurrence, especially the highest-risk group. This indicates that hepatocellular carcinoma (HCC) classification based on vascular patterns can be effectively used for prognostic assessment of HCC patients. The analysis systems provided in Examples 2 and 3 can quickly and accurately complete classification and risk stratification, providing reliable technical support for clinicians to conduct patient risk assessments and formulate individualized treatment decisions.

[0068] Application Example 1

[0069] Using the hepatocellular carcinoma (HCC) vascular pattern analysis system provided in Example 3, a CD34-stained tissue sample from an HCC patient was analyzed. The normalized values ​​of the four vascular subtypes (LV, BV, DV, and VETC) were calculated to be 0.72, 0.93, -0.36, and 1.91, respectively. According to the established vascular pattern judgment rules, the normalized value of VETC in this patient was 1.91, ≥0.5, belonging to the high abundance group. According to the aforementioned rules, this patient can be classified as HCC-4. Combining the risk stratification criteria provided by the prognostic prediction module, this patient belongs to the highest risk group.

[0070] The above are merely preferred embodiments of this application and are not intended to limit the present invention. Although this application has been described in detail with reference to examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. However, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A system for analyzing vascular patterns in hepatocellular carcinoma, characterized in that, It includes an image acquisition module, a tumor region detection module, a blood vessel segmentation module, a blood vessel subtype classification module, and a blood vessel pattern analysis module; The image acquisition module is used to identify tissue sections and prepare full-view pathological section images; The tumor region detection module is used to identify tumor regions in the full-view pathological slice image and output a tumor region mask image. The blood vessel segmentation module is used to segment the blood vessel structure within the tumor region mask image and generate a binary blood vessel mask image. The blood vessel subtype classification module is used to identify blood vessel subtypes in the binary blood vessel mask image; The vascular pattern analysis module is used to calculate the microvessel density of the vascular subtype, standardize the microvessel density using a robust normalization method, classify the abundance level of the vascular subtype and identify the vascular pattern according to a preset threshold, and perform hepatocellular carcinoma classification and prognostic assessment based on the vascular pattern.

2. The hepatocellular carcinoma vascular pattern analysis system according to claim 1, characterized in that, The vascular subtypes include linear vessels, branching vessels, dilated vessels, and tumor-encircled vessels; Wherein, the linear blood vessel is a slender capillary-like structure without obvious branches; the branched blood vessel is a blood vessel structure with at least one branch point; the dilated blood vessel is a blood vessel structure with a maximum lumen diameter exceeding 20 μm; and the tumor-encircling blood vessel is a network of blood vessels formed by surrounding tumor cell clusters.

3. The hepatocellular carcinoma vascular pattern analysis system according to claim 2, characterized in that, The preset threshold in the vascular pattern analysis module is 0.5; Among them, a microvessel density value greater than or equal to 0.5 obtained by standardization is considered high abundance, and a microvessel density value less than 0.5 obtained by standardization is considered low abundance.

4. The hepatocellular carcinoma vascular pattern analysis system according to claim 3, characterized in that, Based on the aforementioned vascular patterns, liver cancer is classified into liver cancer type 1, liver cancer type 2, liver cancer type 3, and liver cancer type 4. When the vascular pattern is low in all four vascular subtypes, the liver cancer is classified as liver cancer type 1. When the vascular pattern is characterized by high abundance of linear vessels and / or branching vessels, and low abundance of dilated vessels and tumor-encircling vessels, the liver cancer is classified as liver cancer type 2. When the vascular pattern is characterized by high abundance of dilated vessels and low abundance of tumor-encircling vessels, the liver cancer is classified as type 3 liver cancer. When the vascular pattern is characterized by high abundance of tumor-encircling vessels, the liver cancer is classified as type 4 liver cancer.

5. The hepatocellular carcinoma vascular pattern analysis system according to claim 4, characterized in that, It also includes a prognostic prediction module; The prognosis prediction module is used to classify patients with type 1 or type 2 liver cancer into a low-risk group and patients with type 3 or type 4 liver cancer into a high-risk group; the low-risk group corresponds to a longer recurrence-free survival time, and the high-risk group corresponds to a shorter recurrence-free survival time.

6. The hepatocellular carcinoma vascular pattern analysis system according to claim 5, characterized in that, The prognosis prediction module classifies type 2 liver cancer into the lowest risk subgroup, type 1 liver cancer into the low-to-medium risk subgroup, type 3 liver cancer into the medium-to-high risk subgroup, and type 4 liver cancer into the highest risk subgroup.

7. The hepatocellular carcinoma vascular pattern analysis system according to claim 1, characterized in that: The tumor region detection module is signal-connected to the image acquisition module; the tumor region detection module has a built-in tumor region detection model based on the ResNet-50 backbone network. The tumor region detection model is used to receive the full-view pathological slide image, magnify and pixel-block the full-view pathological slide image, identify the tumor region and output a tumor region mask map.

8. The hepatocellular carcinoma vascular pattern analysis system according to claim 1, characterized in that: The blood vessel segmentation module is signal-connected to the tumor region detection module. The blood vessel segmentation module has a built-in blood vessel segmentation model based on the U-Net architecture combined with the ResNet-50 encoder. The blood vessel segmentation model is used to receive the tumor region mask image, enlarge the image, perform pixel block processing, segment the blood vessel structure within the tumor region, and generate a binary blood vessel mask image.

9. The hepatocellular carcinoma vascular pattern analysis system according to claim 1, characterized in that, The vascular pattern analysis module is signal-connected to the vascular subtype classification module; the vascular pattern analysis module has a built-in statistical analysis unit adapted to R language, Python, and GraphPad Prism software.

10. A device for prognostic assessment of hepatocellular carcinoma, characterized in that, The system includes the hepatocellular carcinoma vascular pattern analysis system according to any one of claims 1 to 9.

Images