AI-based anemia risk analysis method and system, terminal and storage medium
By combining local and global feature extraction networks with a self-attention mechanism, the problem of modeling contextual dependencies in red blood cell populations was solved, enabling efficient anemia risk analysis and improving the automatic analysis capability of blood morphology images.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-18
- Publication Date
- 2026-03-13
AI Technical Summary
Existing technologies are insufficient for effectively modeling the contextual dependencies of red blood cell populations and integrating multicellular distribution features, thus failing to meet the needs of automated anemia analysis for large-scale blood morphology images.
A combination of local and global feature extraction networks and a self-attention mechanism is used to calculate the attention weights of red blood cells, perform weighted aggregation of local features, output global features, and use a prediction network to predict anemia.
It achieves efficient encoding and temporal modeling of thousands of red blood cell images, integrates local features and global relationships, improves the accuracy and efficiency of anemia prediction, and provides auxiliary judgment basis for fine identification and risk assessment.
Smart Images
Figure CN121662393A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical image analysis technology, and in particular to an AI-based method, system, terminal, and storage medium for anemia risk analysis. Background Technology
[0002] Traditional methods of diagnosing anemia rely primarily on doctors' clinical experience for initial assessment, followed by a series of complex blood tests. This approach is not only cumbersome and time-consuming, but also prone to inefficiency due to human error or testing delays. Therefore, developing intelligent auxiliary diagnostic technologies for anemia is of paramount importance in medical practice. By incorporating artificial intelligence and big data analytics, the accuracy and efficiency of diagnosis can be significantly improved, assisting doctors in quickly developing treatment plans, reducing the burden on the healthcare system, and providing patients with more timely and personalized medical services.
[0003] Currently, in the field of erythrocyte morphology image analysis, mainstream methods mostly employ convolutional neural networks (CNNs) for single-cell image classification or region segmentation. However, these methods face difficulties in modeling the contextual dependencies between multiple cells, struggle to fully represent the distribution characteristics of cells within a population structure, and suffer from problems such as isolated single-cell discrimination and poor information integration capabilities, failing to meet the requirements of automated anemia analysis for large-scale blood morphology images.
[0004] Therefore, how to construct an analytical model that can effectively model the contextual dependencies of red blood cell populations and integrate multi-cell distribution characteristics to meet the needs of automated anemia analysis of large-scale blood morphology images is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] To address the aforementioned technical problems, this application provides an AI-based anemia risk analysis method that can effectively model the contextual dependencies of red blood cell populations, integrate multi-cell distribution features, and meet the needs of automated anemia analysis of large-scale blood morphology images. This application also provides an AI-based anemia risk analysis system, terminal, and storage medium, achieving the same technical effects.
[0006] The first objective of this application is to provide an AI-based method for anemia risk analysis.
[0007] The aforementioned objective of this application is achieved through the following technical solution: An AI-based method for anemia risk analysis includes: Obtain the red blood cell image dataset to be analyzed, wherein the red blood cell image dataset includes multiple individual red blood cell images; Each of the individual red blood cell images is input into a pre-constructed local feature extraction network to obtain the local features of each individual red blood cell image; A feature matrix is constructed based on the local features of each individual red blood cell image; A pre-constructed risk analysis model is obtained, wherein the risk analysis model includes a global feature extraction network and a prediction network. The global feature extraction network is used to calculate the attention weight of a single red blood cell based on the feature matrix through a self-attention mechanism, and to perform weighted aggregation of local features based on the attention weight to output global features. The prediction network is used to output the anemia prediction probability based on the global features. The feature matrix is input into the risk analysis model to obtain the anemia prediction probability.
[0008] Preferably, in the AI-based anemia risk analysis method, the local feature extraction network adopts one of MobileNetV3Small network, ResNet network and DenseNet network.
[0009] Preferably, in the AI-based anemia risk analysis method, the global feature extraction network adopts one of the Performer model, Transformer model, and Linformer model.
[0010] Preferably, the AI-based anemia risk analysis method further includes: Based on the attention weight of a single red blood cell, statistical calculations are performed to obtain a risk indicator, wherein the risk indicator includes the cumulative weight percentage and / or weight variance. Based on the risk indicators and preset thresholds, the risk assessment results are obtained.
[0011] Preferably, in the AI-based anemia risk analysis method, the cumulative weight percentage is calculated as follows: Based on the attention weight of individual red blood cells, all individual red blood cells are sorted in descending order of their attention weights. The attention weights of the top n% of individual red blood cells after sorting are added together to obtain a weight sum. Based on the weight sum and the total of the attention weights, the cumulative weight percentage is calculated, where n is a preset parameter.
[0012] Preferably, the AI-based anemia risk analysis method further includes: Using the Grad-CAM algorithm, a local heatmap is generated by weighted averaging of the activation map of the local feature extraction network through the gradient of the convolutional layer.
[0013] Preferably, in the AI-based anemia risk analysis method, acquiring the red blood cell image dataset to be analyzed includes: Acquire multiple peripheral blood cell images to be analyzed; Using a preset target detection algorithm, target detection is performed on multiple peripheral blood cell images to obtain multiple individual red blood cell images; A red blood cell image dataset is constructed based on multiple individual red blood cell images.
[0014] The second objective of this application is to provide an AI-based anemia risk analysis system.
[0015] The second objective of this application is achieved through the following technical solution: An AI-based anemia risk analysis system includes: The first acquisition unit is used to acquire a red blood cell image dataset to be analyzed, wherein the red blood cell image dataset includes multiple individual red blood cell images; An extraction unit is used to input each of the individual red blood cell images into a pre-constructed local feature extraction network to obtain the local features of each individual red blood cell image; A construction unit is used to construct a feature matrix based on the local features of each individual red blood cell image; The second acquisition unit is used to acquire a pre-constructed risk analysis model, wherein the risk analysis model includes a global feature extraction network and a prediction network. The global feature extraction network is used to calculate the attention weight of a single red blood cell based on the feature matrix through a self-attention mechanism, and to perform weighted aggregation of local features based on the attention weight to output global features. The prediction network is used to output the anemia prediction probability based on the global features. The analysis unit is used to input the feature matrix into the risk analysis model to obtain the anemia prediction probability.
[0016] The third objective of this application is to provide an AI-based anemia risk analysis terminal.
[0017] The aforementioned objective three of this application is achieved through the following technical solution: An AI-based anemia risk analysis terminal includes: a storage medium and a processor; The storage medium stores computer-executed instructions. The processor executes computer execution instructions stored in the storage medium to implement the AI-based anemia risk analysis method described above.
[0018] The fourth objective of this application is to provide a computer-readable storage medium.
[0019] The fourth objective of this application is achieved through the following technical solution: A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the AI-based anemia risk analysis method described above.
[0020] The aforementioned technical solution utilizes a pre-constructed local feature extraction network to perform local feature extraction on each individual red blood cell image in the dataset to be analyzed, obtaining the local features of each individual red blood cell image. Next, a feature matrix is constructed based on the local features of each individual red blood cell image. Subsequently, through a pre-constructed risk analysis model, the global feature extraction network of this model is used to calculate the attention weight of each red blood cell based on the feature matrix and a self-attention mechanism. Based on this attention weight, the local features are weighted and aggregated to output the global features. This solution can efficiently encode and temporally model thousands of red blood cell images, achieving efficient fusion of local features and global relationships, thus effectively compensating for the inability to perform global semantic modeling in traditional blood image analysis. Then, through the model's prediction network, the anemia prediction probability is output based on the global features, meeting the needs of large-scale automatic analysis of blood morphology images, enabling precise identification and risk assessment of anemia-related red blood cell images, and providing doctors with auxiliary diagnostic criteria for anemia.
[0021] In summary, the above technical solutions can effectively model the contextual dependencies of red blood cell populations, integrate multi-cell distribution features, and meet the needs of automated anemia analysis of large-scale blood morphology images. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0023] Figure 1 This is a flowchart illustrating an AI-based anemia risk analysis method in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of an AI-based anemia risk analysis system in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an AI-based anemia risk analysis terminal in an embodiment of this application. Detailed Implementation
[0024] To enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0025] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described below are merely illustrative. For example, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple modules or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, and can be electrical, mechanical, or other forms.
[0026] It should be understood that the use of terms such as "system," "device," "unit," and / or "module" in this application is merely one method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.
[0027] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "a plurality of" or "several" means two or more, unless otherwise explicitly specified.
[0028] If a flowchart is used in this application, it is used to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.
[0029] It should also be noted that, in this document, terms such as “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the article or device that includes the aforementioned element.
[0030] The embodiments in this application are written in a progressive manner.
[0031] like Figure 1 As shown in the embodiment of this application, an AI-based anemia risk analysis method is provided, including: S101. Obtain the red blood cell image dataset to be analyzed, wherein the red blood cell image dataset includes multiple individual red blood cell images; In S101, specifically, blood image samples of the person to be analyzed can be collected, and then image processing technology can be used to obtain multiple individual red blood cell images from the blood image samples as a red blood cell image dataset. The number of individual red blood cell images can be determined based on actual needs, for example, the number can be 1000, but this application is not limited to this.
[0032] In some embodiments, one implementation of this step specifically includes: S1011. Acquire multiple peripheral blood cell images to be analyzed; S1012. Using a preset target detection algorithm, target detection is performed on multiple peripheral blood cell images to obtain multiple individual red blood cell images; S1013. Construct a red blood cell image dataset based on multiple individual red blood cell images.
[0033] Specifically, an AVE-261 image reader can be used to acquire oil immersion peripheral blood cell image samples, obtaining multiple peripheral blood cell images. Then, using existing pre-defined target detection algorithms specifically designed for blood images, target detection is performed on the multiple peripheral blood cell images to identify the red blood cell locations. After filtering and cropping, multiple individual red blood cell images are obtained as a red blood cell image dataset. Other methods can also be used to obtain the red blood cell image dataset, and this application is not limited to these.
[0034] S102. Input each individual red blood cell image into a pre-constructed local feature extraction network to obtain the local features of each individual red blood cell image; In S102, specifically, a local feature extraction network can be pre-built through pre-training or self-tuning, and then the local feature extraction network can be used to extract the local features of all individual red blood cell images in the red blood cell image dataset.
[0035] In some embodiments, the local feature extraction network employs one of the following: MobileNetV3Small, ResNet, and DenseNet. Preferably, the local feature extraction network uses MobileNetV3Small. Compared to traditional CNNs, MobileNetV3Small has fewer parameters, is faster, and effectively captures features such as red blood cell size, edge texture, and staining intensity. Its design significantly reduces the number of parameters and computational resource consumption while maintaining high representational capability through depthwise separable convolutions and Squeeze-and-Excitation (SE) modules. In the local feature extraction stage, each 50×50 pixel single red blood cell image, after processing by the MobileNetV3Small network, generates a 576-dimensional feature vector, serving as the local feature of a single red blood cell image, reflecting information such as cell size, edge texture, and staining concentration.
[0036] S103. Construct a feature matrix based on the local features of each individual red blood cell image; In S103, specifically, a feature matrix is constructed based on the local features of all individual red blood cell images for subsequent risk analysis.
[0037] S104. Obtain a pre-built risk analysis model; In S104, specifically, the risk analysis model includes a global feature extraction network and a prediction network. The global feature extraction network is used to calculate the attention weight of a single red blood cell based on the feature matrix through a self-attention mechanism, and to perform weighted aggregation of local features based on the attention weight to output global features. The prediction network is used to output the anemia prediction probability based on the global features.
[0038] The risk analysis model can be pre-trained using existing model training methods by acquiring feature matrix samples; this application does not impose specific limitations on this. Utilizing the model's global feature extraction network, based on the feature matrix and employing a self-attention mechanism, the attention weight of a single red blood cell is calculated. Local features are then weighted and aggregated based on these attention weights to output global features. Calculating the attention weight of a single red blood cell through the self-attention mechanism allows for weighted aggregation of local features from different red blood cells, effectively capturing the contextual dependencies of red blood cell populations and achieving efficient fusion of local features and global relationships. This overcomes the limitation of traditional blood image analysis in its inability to perform global semantic modeling. Furthermore, the self-attention mechanism allows the model to automatically allocate attention weights, focusing on red blood cell features more valuable for anemia prediction, reducing interference from irrelevant information, and improving the accuracy of subsequent anemia prediction. Through the model's prediction network, based on global features, the anemia prediction probability is output, meeting the needs of large-scale automatic analysis of blood morphology images. This enables precise identification and risk assessment of anemia-related red blood cell images, providing doctors with auxiliary diagnostic criteria for anemia.
[0039] In some embodiments, the global feature extraction network employs one of the Performer, Transformer, and Linformer models. Preferably, the global feature extraction network employs the Performer model, which offers better computational efficiency and stronger global feature fusion capabilities compared to the Transformer and Linformer models. The Performer model is used to replace the traditional Transformer model to obtain the attention weights for each individual red blood cell. The global features are obtained by weighting the local features of all individual red blood cells, as calculated by the following formula: ; In the formula, The number of red blood cells. ≈1000, As a global feature, For the local features of the i-th single red blood cell, Let be the attention weight for the i-th single red blood cell.
[0040] In other embodiments, the prediction network employs fully connected layers (FC). These FC layers map the output an anemia prediction probability from 0 to 1 based on global features using a sigmoid function. The sigmoid function maps the output value to the range of 0 to 1, making it easy to interpret as the likelihood of anemia occurring. By learning the complex nonlinear relationship between global features and anemia risk, the FC layers can achieve accurate predictions of anemia risk, providing strong support for clinical anemia diagnosis.
[0041] S105. Input the feature matrix into the risk analysis model to obtain the anemia prediction probability.
[0042] In step S105, specifically, the feature matrix obtained in step S103 is input into a pre-constructed risk analysis model. The feature matrix is processed by the global feature extraction network of the risk analysis model to output global features. The global features are processed by the prediction network of the risk analysis model to output the anemia prediction probability. This can meet the needs of large-scale automatic analysis of blood morphology images, realize the fine recognition and risk assessment of anemia-related red blood cell images, and provide doctors with auxiliary judgment basis for anemia.
[0043] Currently, in the field of erythrocyte morphology image analysis, mainstream methods mostly employ convolutional neural networks for single-cell image classification or region segmentation. However, these methods face difficulties in modeling the contextual dependencies between multiple cells, struggle to fully represent the distribution characteristics of cells within a population structure, and suffer from problems such as isolated single-cell discrimination and poor information integration capabilities, failing to meet the requirements of automated anemia analysis for large-scale blood morphology images.
[0044] In the above embodiment, a pre-constructed local feature extraction network is used to perform local feature extraction on each individual red blood cell image in the red blood cell image dataset to be analyzed, obtaining the local features of each individual red blood cell image. Then, a feature matrix is constructed based on the local features of each individual red blood cell image. Subsequently, using a pre-constructed risk analysis model and its global feature extraction network, the attention weight of each individual red blood cell is calculated based on the feature matrix using a self-attention mechanism. Based on this attention weight, the local features are weighted and aggregated to output the global features. This scheme can efficiently encode and temporally model thousands of red blood cell images, achieving efficient fusion of local features and global relationships, thus effectively compensating for the inability to perform global semantic modeling in traditional blood image analysis. Afterwards, the prediction network of this model outputs the anemia prediction probability based on the global features, meeting the needs of large-scale automatic analysis of blood morphology images, realizing precise identification and risk assessment of anemia-related red blood cell images, and providing doctors with auxiliary judgment criteria for anemia. In summary, the above embodiments can effectively model the contextual dependencies of red blood cell populations, integrate multi-cell distribution features, and meet the needs of automated anemia analysis of large-scale blood morphology images.
[0045] In other embodiments of this application, the above-described AI-based anemia risk analysis method further includes: S201. Based on the attention weight of a single red blood cell, perform statistical calculations to obtain risk indicators, wherein the risk indicators include the cumulative weight percentage and / or weight variance; S202. Based on risk indicators and preset thresholds, obtain the risk assessment results.
[0046] Specifically, the cumulative weight percentage is calculated as follows: based on the attention weight of individual red blood cells, all individual red blood cells are sorted in descending order of attention weight. The attention weights of the top n% of individual red blood cells after sorting are summed to obtain a weight sum. This weight sum is then divided by the total sum of attention weights to calculate the cumulative weight percentage. Here, n is a preset parameter; for example, n can be 5, and this application does not impose specific restrictions on it. By calculating the cumulative weight percentage, the skewness of the model's attention distribution among cells can be statistically analyzed, and whether key abnormal cells show a local concentration trend can be quantitatively assessed. The calculated cumulative weight percentage is compared with its corresponding preset threshold. If the cumulative weight percentage exceeds the preset threshold, the patient's condition can be considered severe, serving as a risk assessment result and providing additional support for clinical decision-making. The weight variance is the variance of the attention weights of all individual red blood cells. The weight variance reflects the dispersion of the model's attention allocation to the red blood cell population. If the weight variance is large, it indicates that the model pays significant attention to different red blood cells, and some abnormal red blood cells may be given special attention; if the weight variance is small, it indicates that the model pays more even attention to red blood cells. The calculated weighted variance is compared with its corresponding preset threshold. If the weighted variance exceeds the preset threshold, the risk level of the analyzed individual can be comprehensively judged by combining the cumulative weight percentage, thus obtaining a risk assessment result. This method of obtaining risk assessment results based on risk indicators and preset thresholds can assess the anemia risk reflected by red blood cell image data from different perspectives, evaluate the local concentration trend of abnormal red blood cells in the analyzed individual's sample, provide doctors with more comprehensive and accurate auxiliary judgment information, and help improve the accuracy and reliability of anemia diagnosis.
[0047] In other embodiments of this application, the above-described AI-based anemia risk analysis method also includes... S301. Using the Grad-CAM algorithm, a local heatmap is generated by weighting the activation map of the local feature extraction network through the gradient of the convolutional layer.
[0048] Specifically, the Grad-CAM (Gradient-weighted Class Activation Mapping) algorithm is a tool for visualizing deep learning models, helping to understand the decision-making process in convolutional neural networks. It generates heatmaps by utilizing gradient information of the target class to highlight the regions in the input image that contribute most to the prediction results. Taking the MobileNetV3Small network as an example for local feature extraction, the Grad-CAM algorithm generates local heatmaps on the MobileNetV3Small convolutional activation map through gradient backpropagation and weighted averaging. This visually displays the model's focus on abnormal red blood cell regions, locating cell regions of high model interest, thus providing doctors with visual references and improving the interpretability of assisted diagnosis. In some embodiments, local heatmaps can also be overlaid on the original microscopic image, with red / yellow areas representing model focus points (such as cell edge defects or central pale areas), assisting doctors in identifying potential abnormal regions and improving the model's interpretability and clinical adaptability.
[0049] like Figure 2 As shown, in another embodiment of this application, an AI-based anemia risk analysis system is provided, comprising: The first acquisition unit 10 is used to acquire the red blood cell image dataset to be analyzed, wherein the red blood cell image dataset includes multiple individual red blood cell images; Extraction unit 11 is used to input each individual red blood cell image into a pre-constructed local feature extraction network to obtain the local features of each individual red blood cell image; Construction unit 12 is used to construct a feature matrix based on the local features of each individual red blood cell image; The second acquisition unit 13 is used to acquire a pre-constructed risk analysis model, wherein the risk analysis model includes a global feature extraction network and a prediction network. The global feature extraction network is used to calculate the attention weight of a single red blood cell based on the feature matrix through a self-attention mechanism, and to perform weighted aggregation of local features based on the attention weight to output global features. The prediction network is used to output the anemia prediction probability based on the global features. Analysis unit 14 is used to input the feature matrix into the risk analysis model to obtain the anemia prediction probability.
[0050] like Figure 3 As shown, in another embodiment of this application, an AI-based anemia risk analysis terminal is also provided, including: a storage medium 20 and a processor 21; Storage medium 20 stores computer-executed instructions; The processor 21 executes computer execution instructions stored in the storage medium 20 to implement the AI-based anemia risk analysis method described above.
[0051] The processor 21 may include one or more processing cores. The processor 21 executes instructions, programs, code sets, or instruction sets stored in the storage medium 20, and calls data stored in the storage medium 20 to perform various functions and process data as described in this application. The processor 21 may be at least one of a specific application-specific integrated circuit, digital signal processor, digital signal processing device, programmable logic device, field-programmable gate array, central processing unit, controller, microcontroller, and microprocessor. It is understood that, for different devices, the electronic devices used to implement the functions of the processor 21 may also be other types.
[0052] The storage medium 20 can be used to store instructions, programs, code, code sets, or instruction sets. The storage medium 20 may include a stored program area and a stored data area. The stored program area may store instructions for implementing an operating system, instructions for at least one function, and instructions for implementing the aforementioned AI-based anemia risk analysis method; the stored data area may store data involved in the aforementioned AI-based anemia risk analysis method.
[0053] In another embodiment of this application, a computer-readable storage medium is also provided, which stores computer-executable instructions that, when executed by a processor, are used to implement the AI-based anemia risk analysis method described above.
[0054] The computer-readable storage medium can be any medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory, random access memory, or optical disc.
[0055] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An AI-based method for anemia risk analysis, characterized in that, include: Obtain the red blood cell image dataset to be analyzed, wherein the red blood cell image dataset includes multiple individual red blood cell images; Each of the individual red blood cell images is input into a pre-constructed local feature extraction network to obtain the local features of each individual red blood cell image; A feature matrix is constructed based on the local features of each individual red blood cell image; A pre-constructed risk analysis model is obtained, wherein the risk analysis model includes a global feature extraction network and a prediction network. The global feature extraction network is used to calculate the attention weight of a single red blood cell based on the feature matrix through a self-attention mechanism, and to perform weighted aggregation of local features based on the attention weight to output global features. The prediction network is used to output the anemia prediction probability based on the global features. The feature matrix is input into the risk analysis model to obtain the anemia prediction probability.
2. The method as described in claim 1, characterized in that, The local feature extraction network adopts one of the following: MobileNetV3Small network, ResNet network, and DenseNet network.
3. The method as described in claim 1, characterized in that, The global feature extraction network employs one of the Performer, Transformer, and Linformer models.
4. The method as described in claim 1, characterized in that, Also includes: Based on the attention weight of a single red blood cell, statistical calculations are performed to obtain a risk indicator, wherein the risk indicator includes the cumulative weight percentage and / or weight variance. Based on the risk indicators and preset thresholds, the risk assessment results are obtained.
5. The method as described in claim 4, characterized in that, The cumulative weight percentage is calculated as follows: Based on the attention weight of individual red blood cells, all individual red blood cells are sorted in descending order of their attention weights. The attention weights of the top n% of individual red blood cells after sorting are added together to obtain a weight sum. Based on the weight sum and the total of the attention weights, the cumulative weight percentage is calculated, where n is a preset parameter.
6. The method as described in claim 1, characterized in that, Also includes: Using the Grad-CAM algorithm, a local heatmap is generated by weighted averaging of the activation map of the local feature extraction network through the gradient of the convolutional layer.
7. The method as described in claim 1, characterized in that, The process of acquiring the red blood cell image dataset to be analyzed includes: Acquire multiple peripheral blood cell images to be analyzed; Using a preset target detection algorithm, target detection is performed on multiple peripheral blood cell images to obtain multiple individual red blood cell images; A red blood cell image dataset is constructed based on multiple individual red blood cell images.
8. An AI-based anemia risk analysis system, characterized in that, include: The first acquisition unit is used to acquire a red blood cell image dataset to be analyzed, wherein the red blood cell image dataset includes multiple individual red blood cell images; The extraction unit is used to input each of the individual red blood cell images into a pre-constructed local feature extraction network to obtain the local features of each individual red blood cell image; A construction unit is used to construct a feature matrix based on the local features of each individual red blood cell image; The second acquisition unit is used to acquire a pre-constructed risk analysis model, wherein the risk analysis model includes a global feature extraction network and a prediction network. The global feature extraction network is used to calculate the attention weight of a single red blood cell based on the feature matrix through a self-attention mechanism, and to perform weighted aggregation of local features based on the attention weight to output global features. The prediction network is used to output the anemia prediction probability based on the global features. The analysis unit is used to input the feature matrix into the risk analysis model to obtain the anemia prediction probability.
9. An AI-based anemia risk analysis terminal, characterized in that, include: Storage media and processor; The storage medium stores computer-executed instructions. The processor executes computer execution instructions stored in the storage medium to implement the method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1 to 7.