A macrophage identification method and system based on IV-OCT images

Abstract

The present invention discloses a macrophage recognition method and system based on IV-OCT images, the method comprising multimodal preprocessing, dynamic feature enhancement, hybrid deep learning, adaptive threshold optimization, and macrophage recognition. The present invention relates to the field of macrophage recognition technology, and specifically to a macrophage recognition method and system based on IV-OCT images, which achieves more stable and accurate automatic recognition and positioning of macrophages, and improves the consistency and efficiency of recognition; adopts a multimodal image enhancement method combining adaptive contrast enhancement and a dual-channel improved generative adversarial network for multimodal preprocessing; adopts a normalized standard deviation ratio calculation method combined with partitioned sampling enhancement for dynamic feature enhancement; and adopts a branch feature extraction network with cross-attention feature fusion, combined with an enhanced discriminant generative adversarial network, for hybrid deep learning macrophage image segmentation.

Description

Technical Field

[0001] The present invention relates to the field of macrophage identification technology, and in particular to a macrophage identification method and system based on IV-OCT images. Background Art

[0002] The IV-OCT image-based macrophage identification method and system is a medical assistance tool that combines optical coherence tomography (IV-OCT) technology with artificial intelligence analysis. It uses a computer to automatically identify macrophage infiltration areas in coronary atherosclerotic plaques. As an important development in the field of intravascular imaging, this method can assess plaque vulnerability in real time and non-invasively during interventional procedures, providing clinicians with objective indicators of inflammatory activity to assist in formulating precise treatment plans. It can also be used for dynamic monitoring of the efficacy of anti-atherosclerotic drugs.

[0003] However, existing macrophage identification methods based on IV-OCT images have technical problems such as uneven image quality, blurred cell structure, and significant differences in tissue performance between different individuals, which often lead to recognition results relying on manual experience and poor repeatability. In existing basic image data processing, IV-OCT images are affected by factors such as blood flow and tissue reflection during acquisition, resulting in low image contrast and unclear structural hierarchy, which affects the accuracy of subsequent analysis. In existing image feature optimization, the complexity of tissue structure and the local distribution characteristics of macrophages make traditional methods unable to effectively capture local change trends, affecting the sensitive identification of tiny cell areas. In existing macrophage segmentation, unclear image hierarchy, blurred tissue boundaries, and high heterogeneity often affect cell boundaries, making it difficult to accurately segment cell boundaries and easily confusing adjacent structures. In existing segmentation threshold adjustment, the traditional fixed threshold segmentation method is prone to over- or under-identification due to the large individual differences in images and the wide range of cell activity variations, and the manual intervention cost is high and highly subjective. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a macrophage identification method and system based on IV-OCT images. In view of the technical problems in the existing macrophage identification methods based on IV-OCT images, there are uneven image quality, blurred cell structure and obvious differences in tissue performance between different individuals, which often lead to the identification results relying on manual experience and poor repeatability. This solution creatively starts from the significant differences of the identification objects themselves and creatively introduces a structure-guided feature separation mechanism, which can more clearly distinguish the differences between the macrophage area and the background tissue, and realizes more stable and accurate automatic identification and positioning of macrophages, significantly reduces the necessity of manual intervention, and improves the identification efficiency. In view of the technical problems in the existing basic processing of image data, such as low image contrast and unclear structural hierarchy due to the influence of factors such as blood flow and tissue reflection during the acquisition process of IV-OCT images, which affect the accuracy of subsequent analysis, this solution creatively adopts a multimodal image enhancement method that combines adaptive contrast enhancement and dual-channel improved generative adversarial network for multimodal preprocessing, which realizes the intelligent improvement of image quality and the true restoration of regional details, providing a clearer and more biologically meaningful image basis for subsequent recognition and analysis; In view of the technical problems in the existing image feature optimization process ... Local change trends affect the technical problem of sensitive identification of tiny cell areas. This solution creatively starts from the characteristics of obvious local structural changes and uneven distribution of macrophages, and adopts the normalized standard deviation ratio calculation method combined with partition sampling enhancement to perform dynamic feature enhancement, thereby achieving accurate capture of tiny cell changes and improved recognition sensitivity. In view of the technical problems in the existing macrophage segmentation process, which are often affected by unclear image layers, blurred tissue boundaries and large heterogeneity, resulting in difficulty in accurate segmentation of cell boundaries and easy confusion of adjacent structures, this solution creatively starts from the dual needs of image structure feature expression and cell tissue discrimination ability, and adopts a branch feature extraction network with cross-attention feature fusion. Combined with the enhanced discriminative generative adversarial network, hybrid deep learning macrophage image segmentation is performed, which achieves high-resolution segmentation of macrophage areas and stronger boundary discrimination capabilities, effectively improving the recognition accuracy and segmentation integrity under complex backgrounds; in the existing segmentation threshold adjustment process, there are technical problems such as large individual differences in images and wide range of cell activity variations, the traditional fixed threshold segmentation method is prone to over- or under-recognition, and the manual intervention cost is high and highly subjective. This scheme creatively adopts the differentiable maximum inter-class variance threshold algorithm combined with morphological correction to perform adaptive segmentation threshold optimization, realizing a more flexible, efficient and robust image segmentation process, effectively improving the stability and practicality of automatic recognition.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a macrophage identification method based on IV-OCT images, which includes the following steps:

[0006] Step S1: multimodal preprocessing;

[0007] Step S2: dynamic feature enhancement;

[0008] Step S3: Hybrid deep learning;

[0009] Step S4: adaptive threshold optimization;

[0010] Step S5: Macrophage identification.

[0011] Furthermore, in step S1, the multimodal preprocessing is used to collect data and enhance macrophage-related signal features in the image processing stage, specifically collecting IV-OCT images and catheter position information, and using a multimodal image enhancement method combining adaptive contrast enhancement and a dual-channel improved generative adversarial network to perform multimodal preprocessing to obtain macrophage feature-enhanced image data;

[0012] The step of performing multimodal preprocessing by using a multimodal image enhancement method combining adaptive contrast enhancement and a dual-channel improved generative adversarial network to obtain macrophage feature enhanced image data includes:

[0013] Step S11: constructing a dual-channel improved generative adversarial network, specifically by constructing a dual-channel U-shaped network as the generator structure of the adversarial generative network, and performing dual-channel setting and discriminator setting of the generator structure to obtain the dual-channel improved generative adversarial network;

[0014] The generator structure has a dual-channel setting, including a dual-channel input setting and an output setting; the dual-channel input setting specifically uses the IV-OCT image as a first input channel and the catheter position mask information as a second input channel; the output setting specifically outputs the artifact-removed image as generator data;

[0015] Step S12: multi-scale loss setting, specifically combining pixel-level L1 loss, adversarial loss, and frequency domain consistency loss to perform multi-scale loss weighting setting to obtain a multi-scale loss function;

[0016] Step S13: Adaptive contrast enhancement, specifically, performing generative adversarial training using the dual-channel improved generative adversarial network and the multi-scale loss setting, and outputting a macrophage-enhanced image, dividing the image in the macrophage-enhanced image into 16×16 blocks, and performing a contrast-limited adaptive histogram equalization operation on each block to obtain a contrast-optimized image;

[0017] Step S14: global intensity normalization, specifically, controlling the grayscale value distribution of the entire macrophage enhanced image by using a dynamic range compression method during the contrast-limited adaptive histogram equalization operation to obtain adaptive contrast-enhanced image data;

[0018] Step S15: multimodal preprocessing, specifically, obtaining a macrophage enhanced image through the dual-channel improved generative adversarial network and the multi-scale loss setting, and performing multimodal data preprocessing by combining the adaptive contrast enhancement and the global intensity normalization to obtain macrophage feature enhanced image data;

[0019] The macrophage feature-enhanced image data refers to macrophage image data after catheter artifact removal, structure clarity, and contrast optimization.

[0020] Furthermore, in step S2, the dynamic feature enhancement is used to construct a biomarker feature sensitive to the macrophage infiltration area. Specifically, based on the macrophage feature enhanced image data, a normalized standard deviation ratio calculation method combined with partition sampling enhancement is used to perform dynamic feature enhancement to obtain biomarker mapping data, including the following steps:

[0021] Step S21: Sampling by region, specifically, extracting the vertical distance from the vascular endothelial surface to the pixel point from the macrophage feature-enhanced image data, and dividing the image in the macrophage feature-enhanced image data into a shallow region and a deep region based on the vertical distance, and setting sampling region window sizes for the shallow region and the deep region respectively, and obtaining the macrophage heterogeneous distribution characteristics through radial sampling and axial sampling;

[0022] Step S22: defining a normalized standard deviation ratio calculation, specifically, performing normalized standard deviation calculation based on the first shallow window, the second shallow window, the first deep window, and the second deep window to obtain a reference normalized standard deviation value for the macrophage region;

[0023] The macrophage region is referenced to a normalized standard deviation value, which is used to reflect the degree of local texture fluctuation in the region and to indicate the probability of being a macrophage region;

[0024] Step S23: Dynamically weighting the normalized standard deviation ratio, specifically introducing a dynamic attenuation coefficient on the basis of the normalized standard deviation calculation, performing dynamic weighted calculation, and obtaining a noise-resistant optimized macrophage region reference value;

[0025] Step S24: biomarker map fusion, specifically, fusing the pixel values of the macrophage region reference normalized standard deviation value and the anti-noise optimized macrophage region reference value to obtain biomarker map fusion feature data;

[0026] Step S25: Dynamic feature enhancement, specifically performing dynamic sampling based on blood vessel location through the region-based sampling, and enhancing the boundary structure and texture changes of the cell image by combining the normalized standard deviation ratio calculation definition, the normalized standard deviation ratio dynamic weighting, and the biomarker map fusion to obtain biomarker map data;

[0027] Each pixel value of the biomarker mapping data is used to represent a dynamic weighted fusion feature value of a normalized standard deviation, and the dynamic weighted fusion feature value of the normalized standard deviation is used to reflect the activity level of macrophage infiltration.

[0028] Furthermore, in step S3, the hybrid deep learning is used to fuse local texture and global features and realize macrophage segmentation. Specifically, based on the macrophage feature enhanced image data and the biomarker map data, a branch feature extraction network with cross-attention feature fusion is used, combined with an enhanced discriminant generative adversarial network, to perform hybrid deep learning macrophage image segmentation to obtain macrophage segmentation feature data, including the following steps:

[0029] Step S31: constructing a dual-branch feature extraction network, specifically performing feature graph segmentation on the macrophage feature-enhanced image data and the biomarker mapping data, and constructing a dual-branch feature extraction network to extract basic feature data of macrophages;

[0030] The dual-branch feature extraction network includes a local texture extraction branch and a global semantic feature extraction branch;

[0031] Step S32: cross-attention feature fusion, specifically, fusing the local texture feature map data and the global feature map data through a gated weighting mechanism to obtain cross-attention fusion feature data;

[0032] Step S33: Discriminant enhancement adversarial training optimization, specifically, constructing a cross-dual-branch image segmentation subnet by fusing the dual-branch feature extraction network and the cross-attention feature, and using the cross-dual-branch image segmentation subnet as a generator to construct a three-scale enhanced discriminator, perform adversarial generative training, and obtain a discriminant enhancement generative adversarial subnet;

[0033] The three-scale enhanced discriminator includes a local discriminant scale, a mesoscale discriminant scale, and a global discriminant scale;

[0034] Step S34: Hybrid deep learning model training, specifically, constructing a cross-branch image segmentation subnet by constructing a dual-branch feature extraction network and the cross-attention feature fusion, and constructing a discriminant enhancement generation adversarial subnet through the discriminant enhancement adversarial training optimization, and performing subnet integration and hybrid deep learning model training based on the cross-branch image segmentation subnet and the discriminant enhancement generation adversarial subnet to obtain a macrophage segmentation model Model SEG ;

[0035] Step S35: Macrophage segmentation, specifically using the macrophage segmentation model Model based on the biomarker mapping data SEG , perform macrophage segmentation and obtain macrophage segmentation feature data.

[0036] Furthermore, in step S4, the adaptive threshold optimization is used to dynamically improve the optimal segmentation threshold of macrophages. Specifically, based on the macrophage segmentation feature data, a differentiable maximum inter-class variance threshold algorithm combined with morphological correction is used to perform adaptive segmentation threshold optimization to obtain adaptively corrected recognition data, including the following steps:

[0037] Step S41: Construct a differentiable maximum inter-class variance threshold algorithm, specifically by converting the standard maximum inter-class variance threshold algorithm into a differentiable form to obtain a differentiable maximum inter-class variance threshold algorithm, and embedding the differentiable maximum inter-class variance threshold algorithm into the macrophage segmentation model Model in step S3. SEG At the end of , the optimal initial segmentation threshold is calculated;

[0038] Step S42: constructing a morphological correction network, specifically by constructing a lightweight U-shaped network including a three-layer codec pair as the morphological correction network, predicting the segmentation mask gradient value according to the optimal initial segmentation threshold, and performing morphological correction by constructing an optimized correction loss function to obtain morphologically corrected segmentation threshold data;

[0039] Step S43: constructing a dynamic threshold feedback mechanism, specifically by counting the number of connected domains and the average area of the segmented area and setting a dynamic detection threshold for isolated areas, performing over-segmentation judgment, and dynamically adjusting the morphologically corrected segmentation threshold data to obtain dynamic feedback optimized segmentation threshold data;

[0040] Step S44: adaptive threshold optimization, specifically, dynamically adjusting the segmentation threshold of the macrophage segmentation feature data according to the dynamic feedback optimization segmentation threshold data to obtain adaptively corrected recognition data.

[0041] Furthermore, in step S5, the macrophage recognition is used to comprehensively output multiple features to output the final macrophage infiltration area to realize macrophage artificial intelligence recognition, specifically combining the macrophage segmentation feature data and the adaptive correction recognition data to perform macrophage multi-feature adaptive recognition to obtain macrophage comprehensive recognition data.

[0042] The present invention provides a macrophage recognition system based on IV-OCT images, comprising an input processing module, a dynamic enhancement module, a hybrid learning module, a threshold optimization module and a cell recognition module;

[0043] The input processing module is used for multimodal preprocessing, obtaining macrophage feature-enhanced image data through multimodal preprocessing, and sending the macrophage feature-enhanced image data to the dynamic enhancement module and the hybrid learning module;

[0044] The dynamic enhancement module is used for dynamic feature enhancement, obtains biomarker mapping data through dynamic feature enhancement, and sends the biomarker mapping data to the hybrid learning module;

[0045] The hybrid learning module is used for hybrid deep learning, obtains macrophage segmentation feature data through hybrid deep learning, and sends the macrophage segmentation feature data to the threshold optimization module and the cell recognition module;

[0046] The threshold optimization module is used for adaptive threshold optimization, obtains adaptively corrected recognition data through adaptive threshold optimization, and sends the adaptively corrected recognition data to the cell recognition module;

[0047] The cell identification module is used for macrophage identification, and obtains comprehensive macrophage identification data through macrophage identification.

[0048] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0049] (1) In view of the technical problems in existing macrophage recognition methods based on IV-OCT images, which often result in the recognition results relying on manual experience and poor repeatability due to uneven image quality, blurred cell structure and obvious differences in tissue performance between different individuals, this scheme creatively introduces a structure-guided feature separation mechanism based on the significant differences of the recognition objects themselves, which can more clearly distinguish the difference between the macrophage area and the background tissue, and achieve more stable and accurate automatic recognition and positioning of macrophages, significantly reducing the need for manual intervention and improving the consistency and efficiency of recognition;

[0050] (2) In the existing basic processing of image data, there are technical problems such as low image contrast and unclear structural hierarchy due to the influence of factors such as blood flow and tissue reflection during the acquisition process of IV-OCT images, which affects the accuracy of subsequent analysis. This solution creatively adopts a multimodal image enhancement method that combines adaptive contrast enhancement and a dual-channel improved generative adversarial network for multimodal preprocessing, achieving intelligent improvement of image quality and true restoration of regional details, providing a clearer and more biologically meaningful image basis for subsequent recognition and analysis;

[0051] (3) In the existing image feature optimization process, due to the complexity of tissue structure and the local distribution characteristics of macrophages, traditional methods cannot effectively capture local change trends, which affects the sensitive identification of small cell areas. This solution creatively takes into account the obvious changes and uneven distribution of local macrophage structures, and adopts the normalized standard deviation ratio calculation method combined with partition sampling enhancement to perform dynamic feature enhancement, thereby achieving accurate capture of small cell changes and improved recognition sensitivity.

[0052] (4) In view of the technical problems in the existing macrophage segmentation process, which are often affected by unclear image layers, blurred tissue boundaries and large heterogeneity, resulting in difficulty in accurately segmenting cell boundaries and easy confusion of adjacent structures, this solution creatively starts from the dual needs of image structure feature expression and cell tissue discrimination ability improvement, adopts a branch feature extraction network with cross-attention feature fusion, combined with an enhanced discriminant generative adversarial network, and performs hybrid deep learning macrophage image segmentation, achieving high-resolution segmentation of macrophage areas and stronger boundary discrimination ability, effectively improving the recognition accuracy and segmentation integrity under complex backgrounds;

[0053] (5) In view of the technical problems in the existing segmentation threshold adjustment process, the traditional fixed threshold segmentation method is prone to over- or under-recognition due to the large individual differences in images and the wide range of cell activity changes, and the manual intervention cost is high and highly subjective. This solution creatively adopts the differentiable maximum inter-class variance threshold algorithm combined with morphological correction to perform adaptive segmentation threshold optimization, realizing a more flexible, efficient and robust image segmentation process, and effectively improving the stability and practicality of automatic recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of a process for macrophage identification based on IV-OCT images provided by the present invention;

[0055] Figure 2 A schematic diagram of a macrophage identification system based on IV-OCT images provided by the present invention;

[0056] Figure 3 Schematic diagram of the multimodal preprocessing process in step S1;

[0057] Figure 4 This is a schematic diagram of the process of dynamic feature enhancement in step S2;

[0058] Figure 5 Schematic diagram of the process of hybrid deep learning in step S3;

[0059] Figure 6 Schematic diagram of the process of adaptive threshold optimization in step S4.

[0060] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0062] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0063] Example 1, see Figure 1 The present invention provides a macrophage identification method based on IV-OCT images, which comprises the following steps:

[0064] Step S1: multimodal preprocessing;

[0065] Step S2: dynamic feature enhancement;

[0066] Step S3: Hybrid deep learning;

[0067] Step S4: adaptive threshold optimization;

[0068] Step S5: Macrophage identification.

[0069] By performing the above operations, we address the technical issues in existing macrophage identification methods based on IV-OCT images, such as uneven image quality, blurred cell structure, and significant differences in tissue performance between different individuals, which often lead to identification results relying on manual experience and poor repeatability. This solution creatively introduces a structure-guided feature separation mechanism based on the significant differences in the identification objects themselves, which can more clearly distinguish the differences between macrophage areas and background tissues, achieve more stable and accurate automatic identification and positioning of macrophages, significantly reduce the need for manual intervention, and improve the consistency and efficiency of identification.

[0070] Example 2, see Figure 1 、 Figure 2 and Figure 3 In step S1, the multimodal preprocessing is used to collect data and enhance macrophage-related signal features in the image processing stage, specifically collecting IV-OCT images and catheter position information, and using a multimodal image enhancement method combining adaptive contrast enhancement and a dual-channel improved generative adversarial network to perform multimodal preprocessing to obtain macrophage feature-enhanced image data;

[0071] The IV-OCT image refers to a single-channel grayscale image with a size of 512×512;

[0072] The step of performing multimodal preprocessing by using a multimodal image enhancement method combining adaptive contrast enhancement and a dual-channel improved generative adversarial network to obtain macrophage feature enhanced image data includes:

[0073] Step S11: constructing a dual-channel improved generative adversarial network, specifically by constructing a dual-channel U-shaped network as the generator structure of the adversarial generative network, and performing dual-channel setting and discriminator setting of the generator structure to obtain the dual-channel improved generative adversarial network;

[0074] The generator structure has a dual-channel setting, including a dual-channel input setting and an output setting; the dual-channel input setting specifically uses the IV-OCT image as a first input channel and the catheter position mask information as a second input channel; the output setting specifically outputs the artifact-removed image as generator data;

[0075] The discriminator is set up using the standard PatchGAN architecture.

[0076] Step S12: multi-scale loss setting, specifically combining pixel-level L1 loss, adversarial loss, and frequency domain consistency loss to perform multi-scale loss weighting setting to obtain a multi-scale loss function;

[0077] The pixel-level L1 loss is used to preserve image details;

[0078] The adversarial loss is used to improve the realism of the generated image;

[0079] The frequency domain consistency loss is used to perform frequency domain constraints on the frequency domain extracted after wavelet transform processing of the generated image and the real image, and to retain the strong scattered points related to macrophages and the smooth features of the blood vessel wall;

[0080] The calculation formula of the multi-scale loss function is:

[0081] ;

[0082] Where, is the multi-scale loss function, is the pixel-level L1 loss weight, is the pixel-level L1 loss function, is the adversarial loss weight, is the adversarial loss function, is the frequency domain consistency loss weight, K is the number of frequency domain decomposition layers processed by wavelet transform, k is the frequency domain decomposition level index, is the frequency domain transformation function of the kth layer, I gen is the generated image, I real is a real image, ||·||2 is the L2 norm operator;

[0083] Step S13: Adaptive contrast enhancement, specifically, performing generative adversarial training using the dual-channel improved generative adversarial network and the multi-scale loss setting, and outputting a macrophage-enhanced image, dividing the image in the macrophage-enhanced image into 16×16 blocks, and performing a contrast-limited adaptive histogram equalization operation on each block to obtain a contrast-optimized image;

[0084] Step S14: global intensity normalization, specifically, controlling the grayscale value distribution of the entire macrophage enhanced image by using a dynamic range compression method during the contrast-limited adaptive histogram equalization operation to obtain adaptive contrast-enhanced image data;

[0085] Step S15: multimodal preprocessing, specifically, obtaining a macrophage enhanced image through the dual-channel improved generative adversarial network and the multi-scale loss setting, and performing multimodal data preprocessing by combining the adaptive contrast enhancement and the global intensity normalization to obtain macrophage feature enhanced image data;

[0086] The macrophage feature-enhanced image data refers to macrophage image data after catheter artifact removal, structure clarity, and contrast optimization.

[0087] By performing the above operations, we address the technical issues in existing basic image data processing, such as low image contrast and unclear structural hierarchy, which affect the accuracy of subsequent analysis due to the influence of factors such as blood flow and tissue reflection during the acquisition process of IV-OCT images. This solution creatively adopts a multimodal image enhancement method that combines adaptive contrast enhancement and a dual-channel improved generative adversarial network for multimodal preprocessing, achieving intelligent improvement in image quality and true restoration of regional details, providing a clearer and more biologically meaningful image foundation for subsequent recognition and analysis.

[0088] Example 3, see Figure 1 、 Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S2, the dynamic feature enhancement is used to construct a biomarker feature that is sensitive to the macrophage infiltration area. Specifically, based on the macrophage feature enhanced image data, a normalized standard deviation ratio calculation method combined with partition sampling enhancement is used to perform dynamic feature enhancement to obtain biomarker mapping data, including the following steps:

[0089] Step S21: Sampling by region, specifically, extracting the vertical distance from the vascular endothelial surface to the pixel point from the macrophage feature-enhanced image data, and dividing the image in the macrophage feature-enhanced image data into a shallow region and a deep region based on the vertical distance, and setting sampling region window sizes for the shallow region and the deep region respectively, and obtaining the macrophage heterogeneous distribution characteristics through radial sampling and axial sampling;

[0090] The shallow area specifically refers to an area with a vertical distance less than or equal to 1 mm. The sampling area window of the shallow area includes a first shallow window and a second shallow window; the size of the first shallow window is 7×7, and the size of the second shallow window is 5×5;

[0091] The deep region specifically refers to a region with a vertical distance greater than 1 mm. The sampling region window of the deep region includes a first deep window and a second deep window. The size of the first deep window is 5×5, and the size of the second deep window is 3×3.

[0092] The vertical distance from the vascular endothelium surface to the pixel point is extracted, specifically using the distance calculated from the center point of the blood vessel to the outward normal to estimate the vertical distance;

[0093] Step S22: defining a normalized standard deviation ratio calculation, specifically, performing normalized standard deviation calculation based on the first shallow window, the second shallow window, the first deep window, and the second deep window to obtain a reference normalized standard deviation value for the macrophage region;

[0094] The macrophage region is referenced to a normalized standard deviation value, which is used to reflect the degree of local texture fluctuation in the region and to indicate the probability of being a macrophage region;

[0095] Step S23: Dynamically weighting the normalized standard deviation ratio, specifically introducing a dynamic attenuation coefficient on the basis of the normalized standard deviation calculation, performing dynamic weighted calculation, and obtaining a noise-resistant optimized macrophage region reference value;

[0096] The calculation formula of the dynamic attenuation coefficient is:

[0097] ;

[0098] Where, is the dynamic attenuation coefficient, c is the slope parameter, NSD background is the normalized standard deviation ratio of the vascular wall background area, is the attenuation threshold;

[0099] Step S24: biomarker map fusion, specifically, fusing the pixel values of the macrophage region reference normalized standard deviation value and the anti-noise optimized macrophage region reference value to obtain biomarker map fusion feature data;

[0100] Step S25: Dynamic feature enhancement, specifically performing dynamic sampling based on blood vessel location through the region-based sampling, and enhancing the boundary structure and texture changes of the cell image by combining the normalized standard deviation ratio calculation definition, the normalized standard deviation ratio dynamic weighting, and the biomarker map fusion to obtain biomarker map data;

[0101] Each pixel value of the biomarker mapping data is used to represent a dynamic weighted fusion feature value of a normalized standard deviation, and the dynamic weighted fusion feature value of the normalized standard deviation is used to reflect the activity level of macrophage infiltration.

[0102] By performing the above operations, we address the technical problem in the existing image feature optimization process that, due to the complexity of tissue structure and the local distribution characteristics of macrophages, traditional methods are unable to effectively capture local change trends, affecting the sensitive identification of small cell areas. This solution creatively starts from the characteristics of obvious local structural changes and uneven distribution of macrophages, and adopts the normalized standard deviation ratio calculation method combined with partitioned sampling enhancement to perform dynamic feature enhancement, thereby achieving accurate capture of small cell changes and improved recognition sensitivity.

[0103] Example 4, see Figure 1 、 Figure 2 and Figure 5This embodiment is based on the above embodiment. In step S3, the hybrid deep learning is used to fuse local texture and global features and implement macrophage segmentation. Specifically, based on the macrophage feature-enhanced image data and the biomarker mapping data, a branch feature extraction network with cross-attention feature fusion is used in combination with an enhanced discriminant generative adversarial network to perform hybrid deep learning macrophage image segmentation to obtain macrophage segmentation feature data, including the following steps:

[0104] Step S31: constructing a dual-branch feature extraction network, specifically performing feature graph segmentation on the macrophage feature-enhanced image data and the biomarker mapping data, and constructing a dual-branch feature extraction network to extract basic feature data of macrophages;

[0105] The dual-branch feature extraction network includes a local texture extraction branch and a global semantic feature extraction branch;

[0106] The local texture extraction branch specifically adopts a lightweight 3D ResNet-18 network, replaces the standard convolution layer with a three-dimensional convolution layer, and groups the macrophage feature-enhanced image data after feature map segmentation into three frames as data input for the local texture extraction branch to extract local texture feature map data;

[0107] The global semantic feature extraction branch specifically adopts a Swin transformer structure, sets the segmentation input to a fixed window of 7×7, and performs global feature extraction based on the segmented biomarker map data to obtain global feature map data;

[0108] The feature image cutting specifically refers to cutting the images in the macrophage feature enhanced image data and the biomarker mapping image data into a size of 224×224;

[0109] Step S32: cross-attention feature fusion, specifically, fusing the local texture feature map data and the global feature map data through a gated weighting mechanism to obtain cross-attention fusion feature data;

[0110] The calculation formula of the cross attention fusion feature data is:

[0111] ;

[0112] Where, F fused is the cross-attention fusion feature data, is the S-type activation function, W1 is the local branch weight, W2 is the global semantic branch weight, F 3D is the local texture feature map data, F Transformer is the global feature map data, is the element-wise addition operator;

[0113] Step S33: Discriminant enhancement adversarial training optimization, specifically, constructing a cross-dual-branch image segmentation subnet by fusing the dual-branch feature extraction network and the cross-attention feature, and using the cross-dual-branch image segmentation subnet as a generator to construct a three-scale enhanced discriminator, perform adversarial generative training, and obtain a discriminant enhancement generative adversarial subnet;

[0114] The three-scale enhanced discriminator includes a local discriminant scale, a mesoscale discriminant scale, and a global discriminant scale;

[0115] The local discrimination scale specifically sets the receptive range to 70×70; the mesoscale discrimination scale specifically sets the receptive range to 140×140; and the global discrimination scale specifically sets the receptive range to 224×224.

[0116] The discriminative enhanced generative adversarial subnetwork specifically adopts a multi-scale adversarial loss function for optimization training;

[0117] The multi-scale adversarial loss function includes a basic segmentation loss and a scale loss; the scale loss includes a local discriminant scale loss, a mid-scale discriminant scale loss, and a global discriminant scale loss; the basic segmentation loss specifically adopts a combination of a standard Dice loss and a standard cross entropy loss; the scale loss specifically adopts a standard adversarial loss;

[0118] The multi-scale adversarial loss function is obtained by weighted combination of the basic segmentation loss and the scale loss;

[0119] Step S34: Hybrid deep learning model training, specifically, constructing a cross-branch image segmentation subnet by constructing a dual-branch feature extraction network and the cross-attention feature fusion, and constructing a discriminant enhancement generation adversarial subnet through the discriminant enhancement adversarial training optimization, and performing subnet integration and hybrid deep learning model training based on the cross-branch image segmentation subnet and the discriminant enhancement generation adversarial subnet to obtain a macrophage segmentation model Model SEG ;

[0120] Step S35: Macrophage segmentation, specifically using the macrophage segmentation model Model based on the biomarker mapping data SEG , perform macrophage segmentation and obtain macrophage segmentation feature data.

[0121] By performing the above operations, we address the technical problems in the existing macrophage segmentation process, which are often affected by unclear image layers, blurred tissue boundaries and high heterogeneity, resulting in difficulty in accurately segmenting cell boundaries and easy confusion of adjacent structures. This solution creatively starts from the dual needs of expressing image structure features and improving cell tissue discrimination capabilities. It adopts a branch feature extraction network with cross-attention feature fusion, combined with an enhanced discriminant generative adversarial network, to perform hybrid deep learning macrophage image segmentation, achieving high-resolution segmentation of macrophage areas and stronger boundary discrimination capabilities, effectively improving recognition accuracy and segmentation integrity in complex backgrounds.

[0122] Example 5, see Figure 1 、 Figure 2 and Figure 6 This embodiment is based on the above embodiment. In step S4, the adaptive threshold optimization is used to dynamically improve the optimal segmentation threshold of macrophages. Specifically, based on the macrophage segmentation feature data, a differentiable maximum inter-class variance threshold algorithm combined with morphological correction is used to perform adaptive segmentation threshold optimization to obtain adaptively corrected recognition data, including the following steps:

[0123] Step S41: Construct a differentiable maximum inter-class variance threshold algorithm, specifically by converting the standard maximum inter-class variance threshold algorithm into a differentiable form to obtain a differentiable maximum inter-class variance threshold algorithm, and embedding the differentiable maximum inter-class variance threshold algorithm into the macrophage segmentation model Model in step S3. SEG At the end of , the optimal initial segmentation threshold is calculated;

[0124] Step S42: constructing a morphological correction network, specifically by constructing a lightweight U-shaped network including a three-layer codec pair as the morphological correction network, predicting the segmentation mask gradient value according to the optimal initial segmentation threshold, and performing morphological correction by constructing an optimized correction loss function to obtain morphologically corrected segmentation threshold data;

[0125] The optimized correction loss function includes a standard segmentation loss and a boundary smoothing loss; the standard segmentation loss specifically adopts a standard Dice loss, and the boundary smoothing loss specifically adopts an L1 norm loss of the segmentation mask gradient value;

[0126] The calculation formula of the optimized correction loss function is:

[0127] ;

[0128] Where, is the optimization correction loss function, Dice(·) is the Dice similarity coefficient loss function, Y pred is the predicted segmentation result, Ypt is the true segmentation annotation, is the regularization weight coefficient, the specific value is 0.5, is the gradient operator, ||·||1 is the L1 norm operator;

[0129] Step S43: constructing a dynamic threshold feedback mechanism, specifically by counting the number of connected domains and the average area of the segmented area and setting a dynamic detection threshold for isolated areas, performing over-segmentation judgment, and dynamically adjusting the morphologically corrected segmentation threshold data to obtain dynamic feedback optimized segmentation threshold data;

[0130] Step S44: adaptive threshold optimization, specifically, dynamically adjusting the segmentation threshold of the macrophage segmentation feature data according to the dynamic feedback optimization segmentation threshold data to obtain adaptively corrected recognition data.

[0131] By performing the above operations, we address the technical issues in the existing segmentation threshold adjustment process, such as the large individual differences in images and the wide range of cell activity variations, which make the traditional fixed threshold segmentation method prone to over- or under-recognition, and the high cost and strong subjectivity of manual intervention. This solution creatively adopts the differentiable maximum inter-class variance threshold algorithm combined with morphological correction to perform adaptive segmentation threshold optimization, achieving a more flexible, efficient and robust image segmentation process, and effectively improving the stability and practicality of automatic recognition.

[0132] Example 6, see Figure 1 、 Figure 2 This embodiment is based on the above embodiment. In step S5, the macrophage recognition is used to comprehensively output multiple features to achieve macrophage artificial intelligence recognition of the final macrophage infiltration area. Specifically, the macrophage segmentation feature data and the adaptive correction recognition data are combined to perform macrophage multi-feature adaptive recognition to obtain macrophage comprehensive recognition data.

[0133] Example 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment. The present invention provides a macrophage recognition system based on IV-OCT images, including an input processing module, a dynamic enhancement module, a hybrid learning module, a threshold optimization module and a cell recognition module;

[0134] The input processing module is used for multimodal preprocessing, obtaining macrophage feature-enhanced image data through multimodal preprocessing, and sending the macrophage feature-enhanced image data to the dynamic enhancement module and the hybrid learning module;

[0135] The dynamic enhancement module is used for dynamic feature enhancement, obtains biomarker mapping data through dynamic feature enhancement, and sends the biomarker mapping data to the hybrid learning module;

[0136] The hybrid learning module is used for hybrid deep learning, obtains macrophage segmentation feature data through hybrid deep learning, and sends the macrophage segmentation feature data to the threshold optimization module and the cell recognition module;

[0137] The threshold optimization module is used for adaptive threshold optimization, obtains adaptively corrected recognition data through adaptive threshold optimization, and sends the adaptively corrected recognition data to the cell recognition module;

[0138] The cell identification module is used for macrophage identification, and obtains comprehensive macrophage identification data through macrophage identification.

[0139] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0140] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0141] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A macrophage identification method based on IV-OCT images, characterized by: The method comprises the following steps: Step S1: multimodal preprocessing, using a multimodal image enhancement method combining adaptive contrast enhancement and a dual-channel improved generative adversarial network to perform multimodal preprocessing to obtain macrophage feature-enhanced image data; Step S2: Dynamic feature enhancement: Based on the macrophage feature enhancement image data, a normalized standard deviation ratio calculation method combined with partition sampling enhancement is used to perform dynamic feature enhancement to obtain biomarker mapping data; Step S3: Hybrid deep learning, based on the macrophage feature enhanced image data and the biomarker mapping data, a branch feature extraction network with cross-attention feature fusion is used, combined with an enhanced discriminant generation adversarial network, to perform hybrid deep learning macrophage image segmentation to obtain macrophage segmentation feature data, including the following steps: Step S31: Constructing a dual-branch feature extraction network; Step S32: Cross-attention feature fusion, feature fusion of local texture feature map data and global feature map data through a gated weighting mechanism; Step S33: Discriminant enhancement adversarial training optimization, specifically, through the dual-branch feature extraction network and the cross-attention feature fusion, a cross dual-branch image segmentation subnet is constructed, and the cross dual-branch image segmentation subnet is used as a generator to construct a three-scale enhanced discriminator, perform adversarial generation training, and obtain a discriminant enhancement generation adversarial subnet; the three-scale enhanced discriminator includes local discriminant scale, medium scale, and high-resolution image segmentation subnet. Scale discrimination scale and global discrimination scale; the discrimination enhancement generation adversarial subnet specifically adopts a multi-scale adversarial loss function for optimization training; the multi-scale adversarial loss function includes basic segmentation loss and scale loss; the scale loss includes local discrimination scale loss, mid-scale discrimination scale loss and global discrimination scale loss; the basic segmentation loss specifically adopts a combination of standard Dice loss and standard cross entropy loss; the scale loss specifically adopts standard adversarial loss; step S34: hybrid deep learning model training, specifically through the construction of the dual-branch feature extraction network and the cross-attention feature fusion, a cross dual-branch image segmentation subnet is constructed, and through the discrimination enhancement adversarial training optimization, a discrimination enhancement generation adversarial subnet is constructed, and subnet integration and hybrid deep learning model training are performed based on the cross dual-branch image segmentation subnet and the discrimination enhancement generation adversarial subnet to obtain the macrophage segmentation model Model SEG ; Step S35: macrophage segmentation, specifically using the macrophage segmentation model Model based on the biomarker mapping data SEG , perform macrophage segmentation and obtain macrophage segmentation feature data; Step S4: Adaptive threshold optimization: Based on the macrophage segmentation feature data, a differentiable maximum inter-class variance threshold algorithm combined with morphological correction is used to perform adaptive segmentation threshold optimization to obtain adaptively corrected recognition data; Step S5: Identify macrophages and obtain comprehensive macrophage identification data.

2. The macrophage identification method based on IV-OCT images according to claim 1, characterized in that: In step S1, the multimodal preprocessing is used to collect data and enhance macrophage-related signal features in the image processing stage. Specifically, IV-OCT images and catheter position information are collected, and a multimodal image enhancement method combining adaptive contrast enhancement and a dual-channel improved generative adversarial network is used to perform multimodal preprocessing to obtain macrophage feature-enhanced image data.

3. The macrophage identification method based on IV-OCT images according to claim 2, characterized in that: In step S1, the multimodal image enhancement method combining adaptive contrast enhancement and a dual-channel improved generative adversarial network is used to perform multimodal preprocessing to obtain macrophage feature enhanced image data, including: Step S11: constructing a dual-channel improved generative adversarial network, specifically by constructing a dual-channel U-shaped network as the generator structure of the adversarial generative network, and performing dual-channel setting and discriminator setting of the generator structure to obtain the dual-channel improved generative adversarial network; The generator structure has a dual-channel setting, including a dual-channel input setting and an output setting; the dual-channel input setting specifically uses the IV-OCT image as a first input channel and the catheter position mask information as a second input channel; the output setting specifically outputs the artifact-removed image as generator data; Step S12: multi-scale loss setting, specifically combining pixel-level L1 loss, adversarial loss, and frequency domain consistency loss to perform multi-scale loss weighting setting to obtain a multi-scale loss function; The calculation formula of the multi-scale loss function is: Where, is the multi-scale loss function, λ1 is the pixel-level L1 loss weight, is the pixel-level L1 loss function, λ2 is the adversarial loss weight, is the adversarial loss function, λ3 is the frequency domain consistency loss weight, K is the number of frequency domain decomposition layers processed by wavelet transform, k is the frequency domain decomposition level index, is the frequency domain transformation function of the kth layer, I gen is the generated image, I real is a real image, ||·||2 is the L2 norm operator; Step S13: Adaptive contrast enhancement, specifically, performing generative adversarial training using the dual-channel improved generative adversarial network and the multi-scale loss setting, and outputting a macrophage-enhanced image, dividing the image in the macrophage-enhanced image into 16×16 blocks, and performing a contrast-limited adaptive histogram equalization operation on each block to obtain a contrast-optimized image; Step S14: global intensity normalization, specifically, controlling the grayscale value distribution of the entire macrophage enhanced image by using a dynamic range compression method during the contrast-limited adaptive histogram equalization operation to obtain adaptive contrast-enhanced image data; Step S15: multimodal preprocessing, specifically, obtaining a macrophage enhanced image through the dual-channel improved generative adversarial network and the multi-scale loss setting, and performing multimodal data preprocessing by combining the adaptive contrast enhancement and the global intensity normalization to obtain macrophage feature enhanced image data; The macrophage feature-enhanced image data refers to macrophage image data after catheter artifact removal, structure clarity, and contrast optimization.

4. The macrophage identification method based on IV-OCT images according to claim 3, characterized in that: In step S2, the dynamic feature enhancement is used to construct a biomarker feature sensitive to the macrophage infiltration area. Specifically, based on the macrophage feature enhanced image data, a normalized standard deviation ratio calculation method combined with partition sampling enhancement is used to perform dynamic feature enhancement to obtain biomarker mapping data, including the following steps: Step S21: Sampling by region, specifically, extracting the vertical distance from the vascular endothelial surface to the pixel point from the macrophage feature-enhanced image data, and dividing the image in the macrophage feature-enhanced image data into a shallow region and a deep region based on the vertical distance, and setting sampling region window sizes for the shallow region and the deep region respectively, and obtaining the macrophage heterogeneous distribution characteristics through radial sampling and axial sampling; The shallow area specifically refers to an area with a vertical distance less than or equal to 1 mm. The sampling area window of the shallow area includes a first shallow window and a second shallow window; the size of the first shallow window is 7×7, and the size of the second shallow window is 5×5; The deep region specifically refers to a region with a vertical distance greater than 1 mm. The sampling region window of the deep region includes a first deep window and a second deep window. The size of the first deep window is 5×5, and the size of the second deep window is 3×3. Step S22: defining a normalized standard deviation ratio calculation, specifically, performing normalized standard deviation calculation based on the first shallow window, the second shallow window, the first deep window, and the second deep window to obtain a reference normalized standard deviation value for the macrophage region; The macrophage region is referenced to a normalized standard deviation value, which is used to reflect the degree of local texture fluctuation in the region and to indicate the probability of being a macrophage region; Step S23: Dynamically weighting the normalized standard deviation ratio, specifically introducing a dynamic attenuation coefficient on the basis of the normalized standard deviation calculation, performing dynamic weighted calculation, and obtaining a noise-resistant optimized macrophage region reference value; The calculation formula of the dynamic attenuation coefficient is: Where α is the dynamic attenuation coefficient, c is the slope parameter, and NSD background is the normalized standard deviation ratio of the vascular wall background area, τ is the attenuation threshold; Step S24: biomarker map fusion, specifically, fusing the pixel values of the macrophage region reference normalized standard deviation value and the anti-noise optimized macrophage region reference value to obtain biomarker map fusion feature data; Step S25: Dynamic feature enhancement, specifically performing dynamic sampling based on blood vessel location through the region-based sampling, and enhancing the boundary structure and texture changes of the cell image by combining the normalized standard deviation ratio calculation definition, the normalized standard deviation ratio dynamic weighting, and the biomarker map fusion to obtain biomarker map data; Each pixel value of the biomarker mapping data is used to represent a dynamic weighted fusion feature value of a normalized standard deviation, and the dynamic weighted fusion feature value of the normalized standard deviation is used to reflect the activity level of macrophage infiltration.

5. The macrophage identification method based on IV-OCT images according to claim 4, characterized in that: In step S3, the hybrid deep learning is used to fuse local texture and global features and achieve macrophage segmentation. Specifically, based on the macrophage feature-enhanced image data and the biomarker map data, a branch feature extraction network with cross-attention feature fusion is used in combination with an enhanced discriminant generative adversarial network to perform hybrid deep learning macrophage image segmentation to obtain macrophage segmentation feature data, including the following steps: Step S31: constructing a dual-branch feature extraction network, specifically performing feature graph segmentation on the macrophage feature-enhanced image data and the biomarker mapping data, and constructing a dual-branch feature extraction network to extract basic feature data of macrophages; The dual-branch feature extraction network includes a local texture extraction branch and a global semantic feature extraction branch; Step S32: cross-attention feature fusion, specifically, fusing the local texture feature map data and the global feature map data through a gated weighting mechanism to obtain cross-attention fusion feature data; The calculation formula of the cross attention fusion feature data is: Where, F fused is the cross-attention fusion feature data, σ(·) is the S-type activation function, W1 is the local branch weight, W2 is the global semantic branch weight, F 3D is the local texture feature map data, F Transformer is the global feature map data, is the element-wise addition operator.

6. The macrophage identification method based on IV-OCT images according to claim 5, characterized in that: In step S4, the adaptive threshold optimization is used to dynamically improve the optimal segmentation threshold of macrophages. Specifically, based on the macrophage segmentation feature data, a differentiable maximum inter-class variance threshold algorithm combined with morphological correction is used to perform adaptive segmentation threshold optimization to obtain adaptively corrected recognition data, including the following steps: Step S41: Construct a differentiable maximum inter-class variance threshold algorithm, specifically by converting the standard maximum inter-class variance threshold algorithm into a differentiable form to obtain a differentiable maximum inter-class variance threshold algorithm, and embedding the differentiable maximum inter-class variance threshold algorithm into the macrophage segmentation model Model in step S3. SEG At the end of , the optimal initial segmentation threshold is calculated; Step S42: constructing a morphological correction network, specifically by constructing a lightweight U-shaped network including a three-layer codec pair as the morphological correction network, predicting the segmentation mask gradient value according to the optimal initial segmentation threshold, and performing morphological correction by constructing an optimized correction loss function to obtain morphologically corrected segmentation threshold data; Step S43: constructing a dynamic threshold feedback mechanism, specifically by counting the number of connected domains and the average area of the segmented area and setting a dynamic detection threshold for isolated areas, performing over-segmentation judgment, and dynamically adjusting the morphologically corrected segmentation threshold data to obtain dynamic feedback optimized segmentation threshold data; Step S44: adaptive threshold optimization, specifically, dynamically adjusting the segmentation threshold of the macrophage segmentation feature data according to the dynamic feedback optimization segmentation threshold data to obtain adaptively corrected recognition data.

7. The macrophage identification method based on IV-OCT images according to claim 6, characterized in that: In step S5, the macrophage recognition is used to output the final macrophage infiltration area by integrating multiple features to realize macrophage artificial intelligence recognition, specifically combining the macrophage segmentation feature data and the adaptive correction recognition data to perform macrophage multi-feature adaptive recognition to obtain macrophage comprehensive recognition data.

8. A macrophage identification system based on IV-OCT images, for implementing the macrophage identification method based on IV-OCT images according to any one of claims 1 to 7, characterized in that: It includes input processing module, dynamic enhancement module, hybrid learning module, threshold optimization module and cell recognition module.

9. The macrophage identification system based on IV-OCT images according to claim 8, characterized in that: The input processing module is used for multimodal preprocessing, and obtains macrophage feature enhanced image data through multimodal preprocessing, and sends the macrophage feature enhanced image data to the dynamic enhancement module and the hybrid learning module. The dynamic enhancement module is used for dynamic feature enhancement, obtains biomarker mapping data through dynamic feature enhancement, and sends the biomarker mapping data to the hybrid learning module; The hybrid learning module is used for hybrid deep learning, obtains macrophage segmentation feature data through hybrid deep learning, and sends the macrophage segmentation feature data to the threshold optimization module and the cell recognition module; The threshold optimization module is used for adaptive threshold optimization, obtains adaptively corrected recognition data through adaptive threshold optimization, and sends the adaptively corrected recognition data to the cell recognition module; The cell identification module is used for macrophage identification, and obtains comprehensive macrophage identification data through macrophage identification.

Images