Microvascular density analysis method, system and equipment based on digital image and medium

Abstract

The invention discloses a microvessel density analysis method, system and device based on a digital image and a medium. The method comprises the steps that the digital image of a target tissue area is input into a first recognition model to obtain a first recognition area; inputting the first channel and the fourth channel in the first identification area into a second identification model to obtain a second identification area; inputting a third channel in the second identification area and the first identification area into a third identification model to obtain a third identification area; performing spatial constraint on a second channel in the first recognition region based on the third recognition region to obtain a blood vessel feature sub-graph, and inputting the blood vessel feature sub-graph and a fourth channel in the third recognition region into a fourth recognition model to obtain a fourth recognition region; the number of the microvessels in the fourth recognition area and the area of the third recognition area are counted to obtain a microvessel density analysis result of the target tissue area, end-to-end analysis from the macroscopic tissue to the microscopic target area can be achieved based on multi-channel information fusion and model training, and the analysis efficiency and consistency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image analysis, in particular to a microvessel density analysis method, system, device and medium based on digital images. BACKGROUND

[0002] In the field of tumor biology and clinical pathology, the characteristics of tumor microenvironment (TME) are key indicators for evaluating tumor invasiveness, prognosis and treatment response. Hypoxia is a significant feature of the microenvironment of solid tumors, and is closely related to tumor malignant progression, radiotherapy and chemotherapy resistance, and poor prognosis. Hypoxic regions often induce the formation of abnormal microvascular networks, so accurate quantitative analysis of microvessel density (MVD) in hypoxic regions in tumor tissue sections has important clinical value and scientific significance.

[0003] The existing technology is a manual evaluation after immunohistochemistry (IHC) staining. This evaluation method requires a pathologist to identify the area with the highest vascular density under a microscope, and manually count the stained vascular endothelial cells or vascular lumens in the area with the highest vascular density, and then convert the count value to MVD per unit area. Therefore, the existing technology not only requires a physician with specific pathology expertise, has poor generalization ability and limited accuracy, but also lacks the ability to recognize and segment complex tissue structures, resulting in low accuracy of co-localization analysis of hypoxic regions and blood vessels, and is difficult to meet the needs of clinical diagnosis and high-throughput scientific research for the accuracy, robustness and efficiency of the analysis results. SUMMARY

[0004] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0005] The main purpose of the embodiments of the present disclosure is to provide a microvessel density analysis method, system, device and storage medium based on digital images, which can realize end-to-end analysis from macroscopic tissue to microscopic target area based on multi-channel information fusion and model training, and improve the analysis efficiency and consistency.

[0006] The first aspect of the embodiments of the present application provides a microvessel density analysis method based on digital images, for a central controller, the method comprising: obtaining a digital image of a target tissue region, the digital image being a multi-channel fusion image of the same region, and comprising a first channel, a second channel, a third channel and a fourth channel, the first channel comprising a tumor marker, the second channel comprising a blood vessel marker, the third channel comprising a hypoxia marker, and the fourth channel comprising a cell nucleus marker; inputting the digital image into a first recognition model to obtain a first recognition region output by the first recognition model, the first recognition model being trained based on a historical tissue cell training set, and the first recognition region being an image region in the digital image having a tissue cell; inputting a first channel and a fourth channel in the first recognition region into a second recognition model to obtain a second recognition region output by the second recognition model, the second recognition model being trained based on a historical tumor cell training set, and the second recognition region being an image region in the first recognition region having a tumor cell; inputting the second recognition region and a third channel in the first recognition region into a third recognition model to obtain a third recognition region output by the third recognition model, the third recognition model being trained based on a historical tumor hypoxic cell training set, and the third recognition region being an image region in the second recognition region having tumor cell hypoxia; based on the third recognition region, performing spatial constraint on a second channel in the first recognition region to obtain a blood vessel feature subgraph of the target tissue region, and inputting the blood vessel feature subgraph and a fourth channel in the third recognition region into a fourth recognition model to obtain a fourth recognition region output by the fourth recognition model, the fourth recognition model being trained based on a historical microvessel training set, and the fourth recognition region being each microvessel in the third recognition region; counting a number of microvessels in the fourth recognition region and an area of the third recognition region, and obtaining a microvessel density analysis result of the target tissue region according to the number of microvessels in the fourth recognition region and the area of the third recognition region.

[0007] In some embodiments of the present application, the digital image of the target tissue region is obtained by: performing staining treatment on the target tissue region with tumor markers, blood vessel markers, hypoxic markers and cell nucleus markers in sequence, and performing synchronous scanning on the target tissue region after the staining treatment to obtain single-channel images corresponding to each marker; performing image registration on each single-channel image to obtain the digital image of the target tissue region.

[0008] In some embodiments of the present application, the first recognition model is trained by the following steps: obtaining the historical tissue cell training set, the historical tissue cell training set being composed of image samples of different tissue types, and each image sample carrying a tissue identifier, the tissue identifier being used to distinguish tissue cells and non-tissue cells; constructing a first initial recognition model; The first initial recognition model is trained by using the image samples in the historical tissue cell training set as input to the first initial recognition model and the tissue identifiers corresponding to the image samples in the historical tissue cell training set as output to the first initial recognition model.

[0009] In some embodiments of this application, the second recognition model is trained through the following steps: The historical tumor cell training set is obtained, which consists of image samples of different cell types, and each image sample carries a pixel-level identifier, which is used to distinguish between tumor regions and stroma regions. Construct a second initial recognition model; The image samples from the historical tumor cell training set are used as input to the second initial recognition model, and the pixel-level identifiers corresponding to the image samples from the historical tumor cell training set are used as output to train the second initial recognition model, thereby obtaining the second recognition model.

[0010] In some embodiments of this application, the third recognition model is trained through the following steps: The historical tumor hypoxic cell training set is obtained, which consists of image samples of different tumor cell types, and each image sample carries a hypoxia identifier, which is used to distinguish between tumor hypoxic regions and non-tumor hypoxic regions. Construct a third initial recognition model; The third initial recognition model is trained by using the image samples from the historical tumor hypoxic cell training set as input and the hypoxia identifiers corresponding to the image samples from the historical tumor hypoxic cell training set as output, thereby obtaining the third recognition model.

[0011] In some embodiments of this application, the fourth recognition model is trained through the following steps: The historical microvascular training set is obtained, which consists of image samples of different blood vessel types, and each image sample carries a blood vessel identifier, which is used to distinguish between microvascular regions and non-vascular regions. Based on the blood vessel identifiers corresponding to the image samples in the historical microvascular training set, the corresponding blood vessel sub-images are extracted from the corresponding image samples to form a blood vessel sub-image training set. Construct an initial image recognition network and an initial image segmentation network; The initial image recognition network is trained by using the image samples from the historical microvessel training set as input to the initial image recognition network and the corresponding blood vessel identifiers of the image samples from the historical microvessel training set as output to the initial image recognition network, thereby obtaining the target recognition network. The initial image segmentation network is trained by using the blood vessel identifiers corresponding to the image samples in the historical microvessel training set as the input of the initial image segmentation network and the blood vessel sub-image training set as the output of the initial image segmentation network, thereby obtaining the target image segmentation network. The target recognition network and the target image segmentation network are used as the fourth recognition model.

[0012] In some embodiments of this application, obtaining the microvessel density analysis result of the target tissue region based on the number of microvessels in the fourth identification region and the area of ​​the third identification region includes: Divide the number of microvessels in the fourth identification region by the area of ​​the third identification region to obtain the number of microvessels per unit area. Based on the number of microvessels per unit area and a preset microvessel number standard, the microvessel density analysis results of the target tissue region are generated.

[0013] To achieve the above objectives, a second aspect of the present invention provides a microvessel density analysis system based on digital images, the system comprising: The acquisition module is used to acquire a digital image of a target tissue region. The digital image is a multi-channel fused image of the same region and includes a first channel, a second channel, a third channel, and a fourth channel. The first channel includes tumor markers, the second channel includes vascular markers, the third channel includes hypoxia markers, and the fourth channel includes nuclear markers. The first module is used to input the digital image into the first recognition model to obtain the first recognition region output by the first recognition model. The first recognition model is trained based on a historical tissue cell training set, and the first recognition region is the image region in the digital image that contains tissue cells. The second module is used to input the first channel and the fourth channel of the first recognition region into the second recognition model to obtain the second recognition region output by the second recognition model. The second recognition model is trained based on a historical tumor cell training set, and the second recognition region is the image region with tumor cells in the first recognition region. The third module is used to input the second recognition region and the third channel of the first recognition region into the third recognition model to obtain the third recognition region output by the third recognition model. The third recognition model is trained based on a historical tumor hypoxic cell training set, and the third recognition region is the image region in the second recognition region that has tumor cell hypoxia. The fourth module is used to spatially constrain the second channel in the first identification region based on the third identification region to obtain a vascular feature sub-map of the target tissue region, and input the vascular feature sub-map and the fourth channel in the third identification region into the fourth identification model to obtain the fourth identification region output by the fourth identification model. The fourth identification model is trained based on a historical microvascular training set, and the fourth identification region is each microvascular in the third identification region. The statistical module is used to determine the number of microvessels in the fourth identification region and the area of ​​the third identification region, so as to obtain the microvessel density analysis result of the target tissue region based on the number of microvessels in the fourth identification region and the area of ​​the third identification region.

[0014] To achieve the above objectives, a third aspect of the present invention provides an electronic device, comprising: at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, the instructions being executed by the at least one control processor to enable the at least one control processor to perform the above-described method for analyzing microvessel density based on digital images.

[0015] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium storing computer-executable instructions for causing a computer to perform the above-described method for microvascular density analysis based on digital images.

[0016] This application provides a method for microvascular density analysis based on digital images. The method involves inputting a digital image of a target tissue region into a first recognition model to obtain a first recognition region; inputting a first channel and a fourth channel from the first recognition region into a second recognition model to obtain a second recognition region; inputting the second recognition region and a third channel from the first recognition region into a third recognition model to obtain a third recognition region; spatially constraining the second channel from the first recognition region based on the third recognition region to obtain a vascular feature sub-image, which, along with the fourth channel from the third recognition region, is input into a fourth recognition model to obtain a fourth recognition region; and statistically analyzing the number of microvessels in the fourth recognition region and the area of ​​the third recognition region to obtain the microvascular density analysis result of the target tissue region. This method enables end-to-end analysis from macroscopic tissue to microscopic target area based on multi-channel information fusion and model training, improving analysis efficiency and consistency.

[0017] It is understood that the beneficial effects of the second to fourth aspects compared with the related technologies are the same as the beneficial effects of the first aspect compared with the related technologies. Please refer to the relevant description in the first aspect above, which will not be repeated here. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which: Figure 1 This is a schematic flowchart of a microvessel density analysis method based on digital images provided in an embodiment of this application; Figure 2 These are schematic diagrams of digital images provided in embodiments of this application; Figure 3 This is a schematic diagram provided in an embodiment of this application; Figure 4 This is another schematic diagram provided in the embodiments of this application; Figure 5 This is yet another schematic diagram provided in the embodiments of this application; Figure 6 This is another schematic diagram provided in the embodiments of this application; Figure 7 This is another schematic diagram provided in the embodiments of this application; Figure 8 This is a schematic diagram of the structure of a microvessel density analysis system based on digital images provided in an embodiment of this application; Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0020] In the description of this application, the use of terms such as "first," "second," etc., is for the purpose of distinguishing technical features only and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated.

[0021] In the description of this application, it should be understood that the orientation descriptions, such as up, down, etc., are based on the orientation or positional relationship shown in the accompanying drawings, and 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 of this application.

[0022] In the description of this application, it should be noted that, unless otherwise explicitly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution.

[0023] In the fields of tumor biology and clinicopathology, the characteristics of the tumor microenvironment (TME) are key indicators for assessing tumor invasiveness, prognosis, and treatment response. Among these, hypoxia is a prominent feature of the solid tumor microenvironment and is closely related to malignant tumor progression, radiotherapy and chemotherapy resistance, and poor prognosis. Hypoxia often induces the formation of abnormal microvascular networks; therefore, accurate quantitative analysis of microvessel density (MVD) in hypoxic areas of tumor tissue sections has significant clinical value and research implications.

[0024] Current techniques involve manual assessment following immunohistochemical (IHC) staining. This method requires pathologists to identify the areas with the highest vascular density under a microscope and manually count the stained vascular endothelial cells or vascular lumens within these areas, then convert the counts into mean vascular density (MVD) per unit area. Therefore, current techniques not only require specialized pathological expertise, resulting in poor generalization and limited accuracy, but also lack the ability to identify and segment complex tissue structures. This leads to low precision in the co-localization analysis of hypoxic regions and blood vessels, failing to meet the demands of clinical diagnosis and high-throughput research for accuracy, robustness, and efficiency.

[0025] Based on this, embodiments of this application provide a method, system, electronic device, and medium for microvascular density analysis based on digital images, aiming to achieve end-to-end analysis from macroscopic tissues to microscopic target areas based on multi-channel information fusion and model training, thereby improving analysis efficiency and consistency.

[0026] The microvascular density analysis method, system, electronic device and medium based on digital images provided in this application are specifically described through the following embodiments. First, the microvascular density analysis method based on digital images in this application embodiment is described.

[0027] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0028] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0029] The microvascular density analysis method based on digital images provided in this application relates to the field of medical image analysis technology. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the microvascular density analysis method based on digital images, but is not limited to the above forms.

[0030] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0031] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0032] Therefore, referring to Figure 1 This application provides a method for analyzing microvessel density based on digital images. This method is applied to a central controller, which can be a server, an electronic device, or a mobile terminal, etc. There are no specific limitations here. The method includes the following steps S110 to S140.

[0033] Step S110: Obtain a digital image of the target tissue region. The digital image is a multi-channel fusion image of the same region and includes a first channel, a second channel, a third channel and a fourth channel. The first channel includes tumor markers, the second channel includes vascular markers, the third channel includes hypoxia markers and the fourth channel includes nuclear markers. Step S120: Input the digital image into the first recognition model to obtain the first recognition region output by the first recognition model. The first recognition model is trained based on the historical tissue cell training set, and the first recognition region is the image region with tissue cells in the digital image. Step S130: Input the first channel and the fourth channel in the first recognition region into the second recognition model to obtain the second recognition region output by the second recognition model. The second recognition model is trained based on the historical tumor cell training set, and the second recognition region is the image region with tumor cells in the first recognition region. Step S140: Input the third channel of the second recognition region and the first recognition region into the third recognition model to obtain the third recognition region output by the third recognition model. The third recognition model is trained based on the historical tumor hypoxic cell training set, and the third recognition region is the image region with tumor cell hypoxia in the second recognition region. Step S150: Based on the third recognition region, spatial constraints are applied to the second channel in the first recognition region to obtain the vascular feature sub-map of the target tissue region. The vascular feature sub-map and the fourth channel in the third recognition region are then input into the fourth recognition model to obtain the fourth recognition region output by the fourth recognition model. The fourth recognition model is trained based on the historical microvascular training set, and the fourth recognition region is each microvascular in the third recognition region. Step S160: Count the number of microvessels in the fourth identification region and the area of ​​the third identification region, so as to obtain the microvessel density analysis results of the target tissue region based on the number of microvessels in the fourth identification region and the area of ​​the third identification region.

[0034] In this step, a digital image of the target tissue region is first acquired. This digital image is a multi-channel fusion image of the same tissue region, preferably obtained through multi-channel fluorescence scanning or scanning after multiple immunofluorescence stainings to ensure that the channels are naturally aligned in space.

[0035] Specifically, the digital image includes four key channels: the first channel is the tumor marker channel, for example, using cytokeratin CR8 / 18 to label hepatocellular carcinoma cells and CK20 to label colorectal cancer cells; the second channel is the vascular marker channel, for example, using CD31 or CD34 to label vascular endothelium; the third channel is the hypoxia marker channel, for example, using HIF-1α or CA-IX to label hypoxic regions; and the fourth channel is the nuclear marker channel, for example, using DAPI to label cell nuclei. These four key channels collectively provide multidimensional information about tissue at the cellular level, tumor specificity, vascular distribution, and hypoxia status, laying a data foundation for subsequent analysis.

[0036] Furthermore, the digital image is input into the first recognition model to identify regions containing tissue cells in the image, i.e., the first recognition region. Specifically, the first recognition model is a semantic segmentation model trained on a historical tissue cell training set. Preferably, it adopts a convolutional neural network (U-Net) architecture designed for biomedical image segmentation. Its function is to distinguish regions with biological tissue from background regions without tissue in the entire scanned image, and output a binary tissue mask, thereby limiting the effective range for subsequent analysis. The historical tissue cell training set can consist of images labeled by pathologists to distinguish tissue regions or background, or it can be obtained by identification and labeling through other large models.

[0037] Furthermore, the first channel (tumor marker channel) and the fourth channel (nuclear marker channel) of the first identification region are input into the second identification model to identify the image region containing tumor cells within the first identification region, i.e., the second identification region. Since the second identification region is a part of the first identification region, it is also a multi-channel fused image. Specifically, the second identification model is a semantic segmentation model trained on a historical tumor cell training set, preferably using the U-Net architecture. This allows the model to learn and extract features of specific types of tumor cells, achieving accurate segmentation of tumor regions. The historical tumor cell training set can be an image composition where pathologists annotate pixels within the tissue region based on the signal expression of tumor marker channels and cell morphology, dividing it into tumor and stroma regions. Alternatively, it can be obtained through identification and annotation using other large models.

[0038] Furthermore, the second identification region and the third channel (hypoxia marker channel) of the first identification region are input into the third identification model. Based on the multi-channels (first, second, third, and fourth channels) of the second identification region and the third channel (hypoxia marker channel) of the first identification region, the image region with tumor cell hypoxia in the second identification region is identified, which is the third identification region. Since the third identification region is a part of the second identification region, the third identification region and the second identification region are also multi-channel fused images. Specifically, the third identification model is a semantic segmentation model trained on a historical tumor hypoxic cell training set, preferably using the U-Net architecture, and accepting multi-channel input. In the process of the third identification model identifying the second identification region, it mainly relies on the first channel, the fourth channel of the second identification region, and the third channel of the first identification region. The annotation criteria of the historical tumor hypoxic cell training set are strictly limited to regions within the tumor area that simultaneously have significant hypoxia marker expression, thereby ensuring that the identified hypoxic regions have clear biological co-localization characteristics.

[0039] Furthermore, based on the third recognition region, the second channel (vascular marker channel) in the first recognition region is spatially constrained, and a vascular feature sub-map located only in the tumor hypoxia area is extracted. This vascular feature sub-map and the fourth channel (nuclear marker channel) in the third recognition region are then input into the fourth recognition model to identify each microvessel in the third recognition region, i.e., the fourth recognition region.

[0040] Specifically, the fourth identification model is an instance segmentation model trained on a historical microvessel training set. It preferably adopts the U-Net architecture, which enables the model to accurately identify and segment each independent microvessel contour, thereby achieving the differentiation and localization of individual blood vessels.

[0041] Furthermore, the number of microvessels (N) identified in the fourth identification region is counted, and the area (A) of the third identification region is calculated in square millimeters (mm²). Based on the formula MVD=N / A, the microvessel density (MVD) of the target tissue region within the tumor hypoxic zone is calculated. The microvessel density analysis results of the target tissue region are then obtained based on the microvessel density to assess the degree of vascularization of the tumor microenvironment and its correlation with hypoxia, providing a reliable basis for clinical diagnosis and scientific research analysis.

[0042] In one embodiment, such as Figure 2 As shown, digital images of multiple channels of the same tissue section were acquired using a multi-channel fluorescence scanner or by scanning after multiple rounds of immunofluorescence staining. These image samples were specifically labeled, including: Channel 1 ( Figure 2The yellow fluorescence in the channel indicates a tumor marker channel, such as using cytokeratin CK8 / 18 to label hepatocellular carcinoma tumor cells; channel two ( Figure 2 The green fluorescence in the channel indicates a vascular marker channel, such as using CD31 (a vascular endothelial marker) to label microvessels; channel three ( Figure 2 The orange fluorescence in the image indicates a hypoxia marker channel, such as HIF-1α (hypoxia marker) used to label hypoxic areas of the tumor; channel four (DAPI) is a nuclear marker channel, used to label cell nuclei. Because labeling and scanning are performed on the same slice, the images of these four channels are naturally aligned (registered) spatially, ensuring the accuracy of subsequent colocalization analysis.

[0043] Furthermore, the acquired multi-channel digital images are imported into a computer system pre-installed or deployed with the AI ​​algorithm models and image processing libraries required for subsequent steps. This system then decodes and standardizes the images (e.g., format conversion, resolution normalization) for later analysis. The AI ​​algorithm models required for these subsequent steps include a first AI algorithm model, a second AI tumor region identification algorithm model, a third AI algorithm model, and a fourth AI algorithm model.

[0044] Specifically, the first AI algorithm model is used to identify tissue regions in slides to separate areas with biological tissue from background areas without tissue (such as glass blanks). The specific construction process is as follows: First, a first AI initial model based on U-Net (Default) is built. Then, the first AI initial model is trained by using a large number of (tissue region / background) images annotated by pathologists to obtain the first AI model.

[0045] Specifically, the second AI tumor region identification algorithm model is used to accurately distinguish between tumor parenchyma and tumor stroma, in order to identify tumor regions within a tissue area and segment them into tumor regions and stroma regions. The specific construction process is as follows: First, a second AI initial model based on U-Net (Default) is built. Then, using tumor marker channel images obtained by pathologists, pixel-level annotations are performed within the total tissue area based on the signal expression and cell morphology of these channels, dividing the region into tumor regions and stroma regions. This training set is used to train the second AI initial model, resulting in the second AI tumor region identification algorithm model. The tumor region refers to the parenchymal region composed of tumor cells, while the stroma region refers to the region composed of non-tumor cell components such as fibroblasts, immune cells, blood vessels, and extracellular matrix.

[0046] Specifically, the third AI algorithm model is a semantic segmentation model based on U-Net, used to identify "tumor hypoxic regions" to accurately define hypoxic areas occurring within the tumor region. Its construction process is as follows: First, a third AI initial model based on U-Net (Default) is built. Then, the third AI initial model is trained using a historical tumor hypoxic cell training set to obtain the third AI algorithm model. The labeling criteria for the historical tumor hypoxic cell training set are: only regions within the tumor region that simultaneously exhibit significant hypoxia marker (HIF-1α) expression are labeled as "tumor hypoxic regions."

[0047] Specifically, the fourth AI algorithm model is an instance segmentation model used to identify blood vessels and distinguish each individual blood vessel to identify blood vessels and calculate MVD in the hypoxic area of ​​a tumor. The specific construction process is as follows: First, a fourth AI initial model based on U-Net (Default) is built. Then, the fourth AI initial model is trained using a training set in which the precise contours of each microvessel are annotated by pathologists on the vascular marker channel images to obtain the fourth AI algorithm model.

[0048] Furthermore, the entire digital image is first input into the first AI algorithm model for processing, resulting in a binary tissue region mask output by the first AI algorithm model, which is the first recognition region. For example... Figure 3 As shown, the pink area (foreground) represents the organized area, and the black area (background) represents the unorganized area.

[0049] Furthermore, channels one and four of the first identification region are input into the second AI tumor region identification algorithm model. The second AI tumor region identification algorithm model then performs predictions within the tissue mask area to obtain the tumor-stromal segmentation mask output by the second AI tumor region identification algorithm model, i.e., the second identification region. For example... Figure 4 As shown, the red area represents the tumor area, and the green area represents the stroma area. Together, they constitute the total tissue area.

[0050] Furthermore, channel one (tumor markers) in the second recognition region and channel three (hypoxia markers) in the first recognition region are input into the third AI algorithm model through image fusion or dual-channel input to obtain a binary tumor hypoxia region mask output by the third AI algorithm model, i.e., the third recognition region, in order to capture the spatial co-expression relationship between tumor markers and hypoxia markers. Figure 5 As shown, Figure 5 The red area represents the tumor area. Figure 5 The yellow area represents the hypoxic zone identified within the tumor area. Figure 5 The red area is Figure 4A subset of the tumor region ensures biological specificity. (Combined) Figure 6 As shown, Figure 6 The dark blue area on the left represents the cell nucleus in this region, the yellow area on the right represents the tumor hypoxic area, and the cyan area within the yellow area represents the identified microvessels, as shown in Figure 7. Figure 7 for Figure 6 The image shows a magnified view of a portion of the area, including identified microvessels (cyan area) and tumor hypoxia area (yellow area).

[0051] Furthermore, based on the third recognition region, the second channel in the first recognition region is spatially constrained to obtain a vascular feature sub-map. Then, the vascular feature sub-map and the fourth channel in the third recognition region are input into the fourth AI algorithm model to obtain the instance contour of each identified microvessel in the region output by the fourth AI algorithm model.

[0052] Furthermore, the number (N) of these identified microvessels was counted. Simultaneously, the area (A) of the tumor hypoxic zone mask was calculated in millimeters (mm²). Finally, the microvessel density (MVD) of the tumor hypoxic zone was calculated using the formula: MVD = N / A. Thus, by using the microvessel density of the tumor hypoxic zone as the core quantitative indicator, the analysis results of the microvessel density in the tumor hypoxic zone were obtained.

[0053] Therefore, this embodiment achieves full automation from tissue identification to final MVD calculation through the cascading of four dedicated AI models, greatly improving analysis efficiency and consistency, avoiding subjectivity and fatigue errors of manual operation, and accurately defining the "tumor hypoxia zone" through multi-channel information fusion and co-localization analysis, ensuring that the analyzed microvessels are biologically located in the most relevant functional microenvironment, significantly improving the pathological significance and clinical relevance of the results.

[0054] In some embodiments, the process of acquiring a digital image of the target tissue region in step S110 includes the following steps S210 to S220: Step S210: The target tissue area is stained sequentially with tumor markers, vascular markers, hypoxia markers and nuclear markers, so that the stained target tissue area is scanned synchronously to obtain single-channel images corresponding to each marker. Step S220: Perform image registration on each single-channel image to obtain a digital image of the target tissue region.

[0055] In this embodiment, the target tissue region is first subjected to sequential staining and simultaneous scanning. Specifically, the target tissue region (usually a paraffin-embedded or frozen tissue section) is stained sequentially with four specific markers: first, tumor markers (such as cytokeratin CK8 / 18) are used to specifically mark tumor cells; then, vascular markers (such as CD31 or CD34) are used to mark vascular endothelial structures; next, hypoxia markers (such as HIF-1α or CA-IX) are used to identify hypoxic areas of the tumor; and finally, nuclear markers (such as DAPI) are used to clearly show the location of the nuclei of all cells.

[0056] Furthermore, after completing the aforementioned multiple rounds of staining, the stained whole tissue section is placed in a multi-channel fluorescence scanner for simultaneous scanning. Through this scanning process, single-channel digital images corresponding to each marker can be acquired simultaneously, namely, images of tumor marker channels, vascular marker channels, hypoxia marker channels, and nuclear marker channels.

[0057] Furthermore, the obtained single-channel images are image registered to generate the final digital image of the target tissue region for analysis. Since the images originate from simultaneous scans of the same tissue slice, they have a natural spatial correspondence, but minor translational or deformation errors may still exist. Therefore, an image registration algorithm (such as feature point matching or phase correlation-based registration methods) is used to spatially align these four single-channel images, achieving precise pixel-level overlap.

[0058] Furthermore, after registration, pixels at the same spatial location in each channel image represent the expression of the tissue location under different markers, forming a spatially precisely aligned multi-channel fused digital image. This multi-channel fused digital image fully contains four-dimensional information of tumor, blood vessels, hypoxia, and cell nucleus, ensuring that the preparation of multi-channel image data has both biological specificity and spatial accuracy, thus laying a key technical foundation for subsequent automated and high-precision microvascular density analysis.

[0059] In some embodiments, the first recognition model is trained in step S120 through the following steps S310 to S330: Step S310: Obtain a historical tissue cell training set. The historical tissue cell training set consists of image samples of different tissue types, and each image sample carries a tissue identifier, which is used to distinguish tissue cells from non-tissue cells. Step S320: Construct the first initial recognition model; Step S330: Use the image samples in the historical tissue cell training set as the input of the first initial recognition model, and the tissue identifiers corresponding to the image samples in the historical tissue cell training set as the output of the first initial recognition model, to train the first initial recognition model and obtain the first recognition model.

[0060] In this embodiment, a historical tissue cell training set is first acquired. This training set consists of a large number of digital pathological image samples of different tissue types, covering the expected range of tissue morphological variations. For example, it may include tissue slide images from different organs, different pathological grades, or different preparation batches. Each image sample is precisely labeled by a pathologist based on morphological knowledge and carries a corresponding tissue identifier. This tissue identifier is a pixel-level binary mask used to clearly distinguish biologically structured tissue cell regions from tissueless background regions (such as blank areas on the glass substrate) in the image.

[0061] Furthermore, a convolutional neural network architecture suitable for biomedical image segmentation is preferably used to construct the first initial recognition model, such as the U-Net structure. The U-Net structure, characterized by its encoder-decoder format and skip connections, can effectively fuse contextual information and detailed features of the image, making it particularly suitable for tasks involving the accurate segmentation of tissue regions from complex backgrounds. The first initial recognition model is configured with randomized weight parameters during initialization, and corresponding loss functions (such as Dice Loss or cross-entropy loss) and optimizers (such as Adam) are preset to prepare the algorithm for model training.

[0062] Furthermore, the first initial identification model is trained under supervision using a historical tissue cell training set. Specifically, image samples from the historical tissue cell training set are used as input to the model, and their corresponding tissue identifiers are used as the target output for model training. The difference between the model's predictions and the actual annotations (i.e., the loss value) is calculated through forward propagation. The model's weight parameters are iteratively updated using a backpropagation algorithm and an optimizer. After multiple rounds of training, the first initial identification model can automatically learn and extract deep features related to tissue regions from complex pathological images.

[0063] Furthermore, when the model reaches the predetermined performance metrics (such as segmentation accuracy and intersection-union ratio) on an independent validation set, the training terminates, and the resulting first initial recognition model is the first recognition model that has the ability to perform high-precision and automated recognition of tissue regions in new samples and can be used for practical analysis.

[0064] Therefore, the first recognition model trained in this embodiment has strong generalization ability and robustness, and can accurately separate the tissue region from the background in digital pathological images, providing a reliable spatial range limit for subsequent tumor region recognition, hypoxia zone delineation and microvascular analysis.

[0065] In some embodiments, the second recognition model is trained in step S130 through the following steps S410 to S430: Step S410: Obtain a historical tumor cell training set. The historical tumor cell training set consists of image samples of different cell types, and each image sample carries a pixel-level identifier. The pixel-level identifier is used to distinguish between tumor regions and stroma regions. Step S420: Construct the second initial recognition model; Step S430: Use the image samples from the historical tumor cell training set as the input of the second initial recognition model, and the pixel-level identifiers corresponding to the image samples from the historical tumor cell training set as the output of the second initial recognition model, to train the second initial recognition model and obtain the second recognition model.

[0066] In this embodiment, a historical tumor cell training set is first acquired. This training set consists of a large number of digital pathological image samples containing different cell types, covering tissue sections of various tumor types and different degrees of differentiation, to ensure the model's generalization ability. Each image sample is finely annotated by experienced pathologists and carries pixel-level classification labels. These labels clearly distinguish each pixel in the image into two categories: tumor cell regions (i.e., tumor parenchyma) and stromal cell regions (the latter including non-tumor cell components such as fibroblasts, immune cells, blood vessels, and extracellular matrix). This fine pixel-level annotation provides the model with precise supervisory signals for learning the morphological differences between tumor cells and stromal cells.

[0067] Furthermore, a semantic segmentation network based on an encoder-decoder structure is preferably used to construct the second initial recognition model. For example, the U-Net architecture can capture multi-scale contextual features of the image through its encoding path and recover spatial details through the decoding path and skip connections, thereby achieving accurate segmentation of the boundary between the tumor region and the stroma region. The second initial recognition model is configured with untrained convolutional neural network weights during initialization and pre-sets a loss function suitable for multi-class pixel classification (such as weighted cross-entropy loss or Dice loss function) and corresponding optimization algorithms.

[0068] Furthermore, the second initial identification model is trained end-to-end using a historical tumor cell training set. Specifically, image samples from the training set (typically fused or stitched data of tumor marker channels and cell nuclear channels) are used as input to the model, and their corresponding pixel-level labeling maps are used as the target output for model training. The difference between the segmentation results predicted by the model and the actual labels is calculated through forward propagation, and the model parameters are iteratively optimized using the backpropagation algorithm to minimize the preset loss function.

[0069] Furthermore, through sufficient iterative training, the second initial recognition model is able to automatically learn and extract deep features that distinguish tumor cells from stromal cells. When the model exhibits stable high-precision segmentation performance on the reserved validation set (e.g., achieving a satisfactory cross-union ratio or pixel accuracy), the training process terminates, resulting in the optimized second initial recognition model, which is the second recognition model that can be used for practical analysis.

[0070] Therefore, the second recognition model trained in this embodiment has the ability to accurately identify and segment tumor regions in complex tissue microenvironments, laying a reliable anatomical foundation for the subsequent accurate definition of hypoxic regions occurring inside tumors.

[0071] In some embodiments, the third recognition model is trained in step S140 through the following steps S510 to S530: Step S510: Obtain a historical tumor hypoxic cell training set. The historical tumor hypoxic cell training set consists of image samples of different tumor cell types, and each image sample carries a hypoxia marker. The hypoxia marker is used to distinguish between hypoxic tumor regions and hypoxic non-tumor regions. Step S520: Construct the third initial recognition model; Step S530: Use the image samples from the historical tumor hypoxic cell training set as the input of the third initial recognition model, and use the hypoxia identifiers corresponding to the image samples from the historical tumor hypoxic cell training set as the output of the third initial recognition model to train the third initial recognition model and obtain the third recognition model.

[0072] In this embodiment, a historical tumor hypoxic cell training set is first obtained. This training set consists of image samples of different tumor cell types, each carrying an identifier for hypoxic regions. The hypoxia identifier is defined based on the co-expression relationship between tumor markers and hypoxia markers, and is strictly labeled by pathologists according to the following criteria: only image regions within a confirmed tumor area that simultaneously exhibit significant hypoxia marker signal expression are labeled as "tumor hypoxia regions," thereby ensuring that the identifier can accurately distinguish between hypoxic tumor cells and non-hypoxic tumor cells, or non-hypoxic tumor cells.

[0073] Furthermore, a semantic segmentation network with an encoder-decoder structure is preferably used to construct the third initial recognition model. For example, the U-Net architecture is suitable for pixel-level classification tasks of complex biological tissue regions. Simultaneously, the third initial recognition model is designed to accept multi-channel input, such as fusing or stitching tumor marker channel images with hypoxia marker channel images as input, thereby simultaneously learning the features of tumor localization and hypoxia expression.

[0074] Furthermore, the third initial identification model is trained under supervision using a historical tumor hypoxic cell training set. Specifically, image samples from the historical tumor hypoxic cell training set and their corresponding tumor hypoxic region identifiers are used together as training data.

[0075] Specifically, image samples are used as input to the third initial recognition model, and the corresponding hypoxia labels are used as the target output for model training. The difference between the predicted results and the ground truth labels is calculated through forward propagation, and the model parameters are iteratively optimized using the backpropagation algorithm to minimize a pre-defined loss function. During this process, the third initial recognition model learns and integrates deep features from tumor channels and hypoxia channels, ultimately mastering the ability to accurately define "tumor hypoxic regions." Training terminates when the model achieves satisfactory segmentation performance on the validation set, and the resulting optimized model becomes the third recognition model that can be used to accurately define hypoxic regions within tumors.

[0076] Therefore, the third recognition model trained in this embodiment can comprehensively utilize multi-channel information to achieve high specificity and high sensitivity recognition of hypoxic regions inside the tumor, providing a key prerequisite for subsequent calculation of microvessel density in the precise target area.

[0077] In some embodiments, the fourth recognition model is trained in step S150 through the following steps S610 to S660: Step S610: Obtain the historical microvessel training set. The historical microvessel training set consists of image samples of different blood vessel types, and each image sample carries a blood vessel identifier. The blood vessel identifier is used to distinguish between microvessel regions and non-blood vessel regions. Step S620: Based on the blood vessel identifiers corresponding to the image samples in the historical microvascular training set, extract the corresponding blood vessel sub-images from the corresponding image samples to form a blood vessel sub-image training set; Step S630: Construct the initial image recognition network and the initial image segmentation network; Step S640: Use the image samples from the historical microvessel training set as the input to the initial image recognition network, and the corresponding blood vessel identifiers from the image samples in the historical microvessel training set as the output of the initial image recognition network to train the initial image recognition network and obtain the target recognition network. Step S650: Use the blood vessel identifiers corresponding to the image samples in the historical microvascular training set as the input of the initial image segmentation network, and the blood vessel sub-image training set as the output of the initial image segmentation network to train the initial image segmentation network and obtain the target image segmentation network. Step S660: Use the target recognition network and the target image segmentation network as the fourth recognition model.

[0078] In this embodiment, a historical microvascular training set is first acquired. This training set consists of a large number of digital pathological image samples containing different vascular morphologies, covering various microvascular types commonly found in normal and tumor microenvironments, including intact, fragmented, adhered, or irregular vascular structures. Each image sample is meticulously annotated by a pathologist based on the signals of vascular marker channels (such as CD31 staining) and carries a dedicated vascular identifier. This identifier is a pixel-level precise contour annotation used to clearly distinguish microvascular areas from non-vascular background tissue areas in the image.

[0079] Furthermore, corresponding vascular sub-images are extracted from the original image samples based on vascular identifiers to form a vascular sub-image training set. Specifically, based on the labeled vascular identifiers, local image patches containing only microvascular regions are cropped from each original image, i.e., vascular sub-images. This effectively enhances the saliency of vascular targets in the training data by extracting vascular sub-images and provides high-quality, high-resolution input data for subsequent segmentation tasks focused on vascular instance contours.

[0080] Furthermore, two parallel initial network models are constructed, including an initial image recognition network and an initial image segmentation network. The initial image recognition network can employ an efficient convolutional neural network classification architecture to initially identify and locate regions that may contain blood vessels from the entire image. The initial image segmentation network preferably adopts a U-Net-like architecture suitable for instance segmentation to perform pixel-level fine segmentation of the identified blood vessel regions, thereby outputting the complete contour of each individual blood vessel.

[0081] Furthermore, complete image samples from the historical microvascular training set are used as input to the initial image recognition network, and their corresponding binarized vascular identifiers (indicating which areas in the image contain blood vessels) are used as supervision signals to train the initial image recognition network to complete the task of recognizing vascular regions, resulting in a target recognition network that can automatically and accurately locate vascular regions from complex tissue backgrounds.

[0082] Furthermore, the vascular identifiers (i.e. vascular region masks) corresponding to the image samples in the historical microvascular training set are used as the input of the initial image segmentation network, while the vascular sub-images extracted from the original image and corresponding to them are used as the expected output of the initial image segmentation network. This enables the initial image segmentation network to learn the inverse transformation capability from a coarse vascular region to a high-resolution, fine vascular contour, resulting in a target image segmentation network capable of pixel-level refinement of vascular regions and separation of adherent vascular instances.

[0083] Furthermore, the trained target recognition network and target image segmentation network are integrated to form the final fourth recognition model. In practical applications, the target recognition network first locates the approximate blood vessel region in the blood vessel marker channel image, and then the target image segmentation network performs fine processing on these regions, outputting the instance segmentation result of each individual microvessel.

[0084] Therefore, the fourth recognition model trained in this embodiment combines efficient vascular region localization capability with accurate vascular instance contour segmentation capability. It is particularly good at handling microvascular structures that are irregular in shape, densely distributed, or mutually adhered in the hypoxic area of ​​tumors, thus providing an accurate quantitative basis for the final calculation of microvascular density.

[0085] In some embodiments, step S160, based on the number of microvessels in the fourth identification region and the area of ​​the third identification region, yields the microvessel density analysis result of the target tissue region, including the following steps S710 to S720: Step S710: Divide the number of microvessels in the fourth identification region by the area of ​​the third identification region to obtain the number of microvessels per unit area. Step S720: Generate microvessel density analysis results for the target tissue region based on the number of microvessels per unit area and the preset microvessel number standard.

[0086] In this embodiment, the total number of microvessel instances contained in the fourth identification region output by the fourth identification model is first counted. This total number of microvessel instances represents the number of independent microvessels accurately identified within the tumor hypoxic region. Simultaneously, the area of ​​the third identification region output by the third identification model is calculated. This area value is the actual area of ​​the tumor hypoxic region on the two-dimensional image, expressed in square millimeters (mm²).

[0087] Furthermore, the number of microvessels is divided by the area of ​​the hypoxic tumor region to calculate the number of microvessels per unit area, which is a direct quantitative indicator of the degree of vascularization in the hypoxic tumor region. Based on this quantitative indicator, the final microvessel density analysis result is generated.

[0088] Specifically, the calculated number of microvessels per unit area is compared and interpreted with a preset standard for microvessel count. This preset standard is a reference range or threshold established based on extensive clinical research data. For example, it may include a threshold distinguishing between "high vascular density" and "low vascular density," or grading standards associated with tumor staging and prognostic assessment. Based on the comparison results, a structured analysis report is generated, containing the original microvessel density values ​​and a classification conclusion (e.g., "high MVD in hypoxic areas") or risk level assessment based on the preset standard. This transforms raw quantitative image data into analytical conclusions with clear clinicopathological significance, providing direct decision support for physicians in assessing tumor invasiveness, treatment resistance risk, and patient prognosis.

[0089] Thus, this embodiment completes a fully automated conversion process from microscopic image features to macroscopic clinical indicators, ensuring that the analysis results not only have computational accuracy, but also clear biological interpretive value and clinical applicability, significantly improving the objectivity and efficiency of pathological diagnosis.

[0090] like Figure 8 As shown in some embodiments of this application, a microvascular density analysis system based on digital images is provided. The system includes an acquisition module 810, a first module 820, a second module 830, a third module 840, a fourth module 850, and a statistics module 860. Specifically: The acquisition module 810 is used to acquire a digital image of the target tissue region. The digital image is a multi-channel fusion image of the same region and includes a first channel, a second channel, a third channel and a fourth channel. The first channel includes tumor markers, the second channel includes vascular markers, the third channel includes hypoxia markers and the fourth channel includes nuclear markers. The first module 820 is used to input a digital image into a first recognition model to obtain a first recognition region output by the first recognition model. The first recognition model is trained based on a historical tissue cell training set, and the first recognition region is an image region in the digital image that contains tissue cells. The second module 830 is used to input the first channel and the fourth channel of the first recognition region into the second recognition model to obtain the second recognition region output by the second recognition model. The second recognition model is trained based on a historical tumor cell training set, and the second recognition region is the image region with tumor cells in the first recognition region. The third module 840 is used to input the third channel of the second recognition region and the first recognition region into the third recognition model to obtain the third recognition region output by the third recognition model. The third recognition model is trained based on the historical tumor hypoxic cell training set, and the third recognition region is the image region with tumor cell hypoxia in the second recognition region. The fourth module 850 is used to spatially constrain the second channel in the first recognition region based on the third recognition region to obtain the vascular feature sub-map of the target tissue region. The vascular feature sub-map and the fourth channel in the third recognition region are input into the fourth recognition model to obtain the fourth recognition region output by the fourth recognition model. The fourth recognition model is trained based on the historical microvascular training set, and the fourth recognition region is each microvascular in the third recognition region. The statistics module 860 is used to determine the number of microvessels in the fourth identification region and the area of ​​the third identification region, so as to obtain the microvessel density analysis results of the target tissue region based on the number of microvessels in the fourth identification region and the area of ​​the third identification region.

[0091] It should be noted that the microvascular density analysis system based on digital images provided in this embodiment is based on the same inventive concept as the aforementioned microvascular density analysis method based on digital images. Therefore, the relevant content of the aforementioned microvascular density analysis method based on digital images is also applicable to the content of the microvascular density analysis system based on digital images. Therefore, it will not be repeated here.

[0092] To this end, the system inputs a digital image of the target tissue region into a first recognition model to obtain a first recognition region; inputs the first and fourth channels of the first recognition region into a second recognition model to obtain a second recognition region; inputs the second recognition region and the third channel of the first recognition region into a third recognition model to obtain a third recognition region; based on the third recognition region, spatial constraints are applied to the second channel of the first recognition region to obtain a vascular feature sub-image, which, along with the fourth channel of the third recognition region, is input into a fourth recognition model to obtain a fourth recognition region; the number of microvessels in the fourth recognition region and the area of ​​the third recognition region are counted to obtain the microvessel density analysis results of the target tissue region. In this way, end-to-end analysis from macroscopic tissue to microscopic target area can be achieved based on multi-channel information fusion and model training, improving analysis efficiency and consistency.

[0093] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described microvascular density analysis method based on digital images.

[0094] like Figure 9 , Figure 9 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. The electronic device includes: At least one battery; At least one memory; At least one processor; At least one program; The program is stored in memory, and the processor executes at least one program to implement the above-described method for microvascular density analysis based on digital images, as described in this disclosure.

[0095] This electronic device can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0096] The electronic devices according to embodiments of this application will now be described in detail.

[0097] The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this disclosure. The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1700 and called and executed by the processor 1600 to perform a microvascular density analysis method based on digital images according to an embodiment of this disclosure.

[0098] The input / output interface 1800 is used to implement information input and output. The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900); The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.

[0099] This disclosure also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the aforementioned microvascular density analysis method based on digital images.

[0100] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0101] The embodiments described in this disclosure are for the purpose of more clearly illustrating the technical solutions of this disclosure and do not constitute a limitation on the technical solutions provided by this disclosure. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by this disclosure are also applicable to similar technical problems.

[0102] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this disclosure, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0103] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0104] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0105] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any related variations, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0106] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0107] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0108] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0109] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0110] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0111] The above is a detailed description of the preferred embodiments of this application. However, the embodiments of this application are not limited to the above-described implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the embodiments of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of the embodiments of this application.

[0112] The embodiments of this application have been described in detail above with reference to the accompanying drawings. However, this application is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of this application.

Claims

1. A method for microvessel density analysis based on digital images, characterized by, The method comprises: obtaining a digital image of a target tissue region, the digital image being a multi-channel fusion image of the same region and comprising a first channel, a second channel, a third channel and a fourth channel, the first channel comprising a tumor marker, the second channel comprising a blood vessel marker, the third channel comprising a hypoxia marker, and the fourth channel comprising a cell nucleus marker; inputting the digital image into a first recognition model to obtain a first recognition region output by the first recognition model, the first recognition model being trained based on a historical tissue cell training set, and the first recognition region being an image region in the digital image having tissue cells; inputting the first channel and the fourth channel in the first recognition region into a second recognition model to obtain a second recognition region output by the second recognition model, the second recognition model being trained based on a historical tumor cell training set, and the second recognition region being an image region in the first recognition region having tumor cells; inputting the second recognition region and the third channel in the first recognition region into a third recognition model to obtain a third recognition region output by the third recognition model, the third recognition model being trained based on a historical tumor hypoxic cell training set, and the third recognition region being an image region in the second recognition region having tumor cell hypoxia; based on the third recognition region, performing spatial constraint on the second channel in the first recognition region to obtain a blood vessel feature subgraph of the target tissue region, and inputting the blood vessel feature subgraph and the fourth channel in the third recognition region into a fourth recognition model to obtain a fourth recognition region output by the fourth recognition model, the fourth recognition model being trained based on a historical microvessel training set, and the fourth recognition region being each microvessel in the third recognition region; counting the number of microvessels in the fourth recognition region and the area of the third recognition region, and obtaining a microvessel density analysis result of the target tissue region according to the number of microvessels in the fourth recognition region and the area of the third recognition region.

2. The digital image-based microvessel density analysis method according to claim 1, wherein, The method comprises: performing staining treatment on the target tissue region in sequence with a tumor marker, a blood vessel marker, a hypoxia marker and a cell nucleus marker, and performing synchronous scanning on the target tissue region after the staining treatment to obtain single-channel images corresponding to each marker; performing image registration on each single-channel image to obtain the digital image of the target tissue region.

3. The digital image-based microvessel density analysis method of claim 1, wherein, The first recognition model is trained by the following steps: obtaining the historical tissue cell training set, the historical tissue cell training set being composed of image samples of different tissue types, and each image sample carrying a tissue identifier for distinguishing tissue cells and non-tissue cells; constructing a first initial recognition model; training the first initial recognition model by taking the image samples in the historical tissue cell training set as inputs of the first initial recognition model and taking the tissue identifiers corresponding to the image samples in the historical tissue cell training set as outputs of the first initial recognition model, to obtain the first recognition model.

4. The digital image-based microvessel density analysis method of claim 1, wherein, The second identification model is obtained by the following steps: obtaining the historical tumor cell training set, the historical tumor cell training set being composed of image samples of different cell types, and each of the image samples carrying a pixel-level label for distinguishing tumor regions and interstitial regions; constructing a second initial identification model; training the second initial identification model by taking the image samples in the historical tumor cell training set as inputs of the second initial identification model and taking the pixel-level labels corresponding to the image samples in the historical tumor cell training set as outputs of the second initial identification model, to obtain the second identification model.

5. The digital image-based microvessel density analysis method according to claim 4, wherein, The third identification model is obtained by the following steps: obtaining the historical tumor hypoxic cell training set, the historical tumor hypoxic cell training set being composed of image samples of different tumor cell types, and each of the image samples carrying a hypoxic label for distinguishing tumor hypoxic regions and non-tumor hypoxic regions; constructing a third initial identification model; training the third initial identification model by taking the image samples in the historical tumor hypoxic cell training set as inputs of the third initial identification model and taking the hypoxic labels corresponding to the image samples in the historical tumor hypoxic cell training set as outputs of the third initial identification model, to obtain the third identification model.

6. The digital image-based microvessel density analysis method of claim 1, wherein, The fourth identification model is obtained by the following steps: obtaining the historical microvessel training set, the historical microvessel training set being composed of image samples of different vessel types, and each of the image samples carrying a vessel label for distinguishing microvessel regions and non-vessel regions; extracting corresponding vessel subgraphs from corresponding image samples according to the vessel labels corresponding to the image samples in the historical microvessel training set, to form a vessel subgraph training set; constructing an initial image recognition network and an initial image segmentation network; training the initial image recognition network by taking the image samples in the historical microvessel training set as inputs of the initial image recognition network and taking the vessel labels corresponding to the image samples in the historical microvessel training set as outputs of the initial image recognition network, to obtain a target recognition network; training the initial image segmentation network by taking the vessel labels corresponding to the image samples in the historical microvessel training set as inputs of the initial image segmentation network and taking the vessel subgraph training set as outputs of the initial image segmentation network, to obtain a target image segmentation network; taking the target recognition network and the target image segmentation network as the fourth identification model.

7. The digital image-based microvessel density analysis method of claim 1, wherein, The microvessel density analysis result of the target tissue region is obtained according to the number of microvessels in the fourth identification region and the area of the third identification region, and includes: dividing the number of microvessels in the fourth identification region by the area of the third identification region to obtain the number of microvessels per unit area; generating the microvessel density analysis result of the target tissue region according to the number of microvessels per unit area and a preset microvessel number standard.

8. A digital image-based microvessel density analysis system, characterized by, The system includes: An acquisition module is configured to acquire a digital image of a target tissue region, the digital image being a multi-channel fusion image of the same region and including a first channel, a second channel, a third channel, and a fourth channel, the first channel including a tumor marker, the second channel including a blood vessel marker, the third channel including a hypoxia marker, and the fourth channel including a cell nucleus marker; A first module is configured to input the digital image into a first recognition model to obtain a first recognition region output by the first recognition model, the first recognition model being trained based on a historical tissue cell training set, and the first recognition region being an image region of the digital image having a tissue cell; A second module is configured to input the first channel and the fourth channel in the first recognition region into a second recognition model to obtain a second recognition region output by the second recognition model, the second recognition model being trained based on a historical tumor cell training set, and the second recognition region being an image region of the first recognition region having a tumor cell; A third module is configured to input the second recognition region and the third channel in the first recognition region into a third recognition model to obtain a third recognition region output by the third recognition model, the third recognition model being trained based on a historical tumor hypoxia cell training set, and the third recognition region being an image region of the second recognition region having tumor cell hypoxia; A fourth module is configured to perform spatial constraint on the second channel in the first recognition region based on the third recognition region to obtain a blood vessel feature sub-image of the target tissue region, and input the blood vessel feature sub-image and the fourth channel in the third recognition region into a fourth recognition model to obtain a fourth recognition region output by the fourth recognition model, the fourth recognition model being trained based on a historical microvessel training set, and the fourth recognition region being each microvessel in the third recognition region; A statistical module is configured to count a number of microvessels in the fourth recognition region and an area of the third recognition region, and obtain a microvessel density analysis result of the target tissue region according to the number of microvessels in the fourth recognition region and the area of the third recognition region.

9. An electronic device, comprising: The memory is configured to store instructions executable by the at least one control processor, and the at least one control processor is configured to execute the instructions to enable the at least one control processor to perform the microvessel density analysis method based on a digital image according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is configured to store computer-executable instructions for causing a computer to perform the microvessel density analysis method based on a digital image according to any one of claims 1 to 7.

Images