Plasmodium infection stage detection method and system

Abstract

The invention provides a plasmodium infection stage detection method and system, and the method comprises the steps: obtaining a plasmodium image, carrying out the marking of the plasmodium image, determining a corresponding infection stage, constructing a preliminary data set through the marked plasmodium image, and carrying out the data processing of the preliminary data set, and determining a plasmodium infection stage image data set; constructing a ResNet-based plasmodium infection stage detection framework, and designing a convolution block attention module in a feature fusion module of the ResNet framework; constructing a mixed loss function to optimize the detection architecture; and training the constructed detection architecture according to the plasmodium infection stage image data set to determine a target plasmodium infection stage detection architecture. The method solves the problem that in the prior art, a plasmodium infection stage detection method for accurately and rapidly identifying irregular forms of plasmodium in different infection stages is lacked.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image recognition, in particular to a method and system for detecting the infection stage of Plasmodium. BACKGROUND

[0002] Plasmodium is the core pathogen that causes malaria. The detection method in clinical practice is usually based on the results of blood smear microscopy for diagnosis. However, this method is usually inefficient and requires a professional doctor to stay beside the microscope to perform complex operations, which cannot meet the needs of timely diagnosis in clinical practice. Therefore, how to accurately and timely detect Plasmodium infection and the infection stage of Plasmodium is an urgent diagnostic requirement in clinical practice.

[0003] Although the deep learning detection architecture based on artificial intelligence has made important progress in many medical scenarios, it is less directly applied to the detection of Plasmodium infection. In the prior art, although there is a detection model for Plasmodium infection, the model is usually limited to detecting the presence or absence of Plasmodium, and ignores the infection stage of Plasmodium.

[0004] However, in fact, when Plasmodium is in different infection stages, such as ring, schizont, trophozoite and gametocyte, the morphology and corresponding patient infection conditions differ significantly, and different infection stages require appropriate treatment. Different stages of treatment also have significant differences. SUMMARY

[0005] Therefore, the purpose of the present application is to provide a method and system for detecting the infection stage of Plasmodium, which aims to solve the problem that the prior art lacks a method for detecting the infection stage of Plasmodium that accurately and quickly identifies the irregular morphology of Plasmodium in different infection stages.

[0006] According to an embodiment of the present application, a method for detecting the infection stage of Plasmodium, the method comprising: obtaining a Plasmodium image and labeling the Plasmodium image to determine the corresponding infection stage, so as to form a preliminary data set of the labeled Plasmodium image, and performing data processing on the preliminary data set to determine a Plasmodium infection stage image data set; constructing a Plasmodium infection stage detection architecture based on ResNet, and designing a convolution block attention module in the feature fusion module of the ResNet architecture; constructing a hybrid loss function to optimize the detection architecture, wherein the hybrid loss function at least simultaneously combines the features of the mean square error loss function, the focal loss function and the Dice loss function; training the constructed detection architecture according to the Plasmodium infection stage image data set to determine a target Plasmodium infection stage detection architecture.

[0007] In addition, the method for detecting the infection stage of Plasmodium according to the above-mentioned embodiment of the present application can further have the following additional technical features: Further, the step of performing data processing on the preliminary data set to determine the image data set of the infection stage of Plasmodium comprises: a wavelet transform method is designed to decompose the Plasmodium image into low-frequency components and high-frequency components to construct a frequency domain expression matrix; Gaussian normalization method is used to standardize the low-frequency components and high-frequency components respectively, and the standardized high-frequency components and low-frequency components are stacked to form a frequency domain expression matrix with fused multi-channel input; Transform is used to filter key frequency band features and remove noise frequency bands in the preliminary data set to form the image data set of the infection stage of Plasmodium; wherein the function corresponding to the wavelet transform method is a Haar wavelet basis function, and the expression of the Haar wavelet basis function is

[0008] wherein, is a scale function value, is a pixel information value of an input image, is a scale factor, is a translation parameter.

[0009] Further, the step of using Transform to filter key frequency band features and remove noise frequency bands in the preliminary data set to form the image data set of the infection stage of Plasmodium comprises: flattening each channel component of the frequency domain expression matrix to form a frequency domain sequence; constructing a linear layer to perform Transform mapping on the frequency domain sequence, and adding position encoding to the mapped frequency domain sequence; constructing a multi-head self-attention mechanism to extract the dependency relationship across frequency bands in the mapped frequency domain sequence, and extracting features for global context information; based on attention weight analysis, filtering out key frequency band features and removing noise frequency band features in the Plasmodium image to construct the image data set of the infection stage of Plasmodium.

[0010] Further, the detection architecture of the infection stage of Plasmodium based on ResNet is constructed, which specifically comprises four modules: an input module, a residual connection module, a feature fusion module, and a classification head module, as follows: In the input module, a convolution kernel with a size of 7*7 is used to perform convolution operation on the input image, and a maximum pooling operation is performed on the feature image after the convolution operation to obtain a preliminary feature extraction image; In the residual connection module, three convolution kernels of different sizes are sequentially used to extract features from the preliminary feature extraction graph, and the residual information of the previous step is gradually fused during the feature extraction process to match the irregular scale characteristics of the Plasmodium image. In the feature fusion module, the outputs of the three scales in the residual connection module are spliced in the channel dimension to obtain a comprehensive feature graph, and a convolution block attention module is designed to improve the attention to irregular shapes. In the classification head module, a fully connected layer and an activation function are designed to obtain score values of the Plasmodium image in the infection stage, and a detection result is obtained based on the maximum score.

[0011] Further, in the feature fusion module of the ResNet architecture, a convolution block attention module is designed to improve the attention to irregular shapes, and the specific steps are as follows: A convolution block is designed to perform convolution operation on the comprehensive feature graph to obtain a feature graph after convolution; A channel attention mechanism is designed to enhance the feature graph after convolution in the channel direction; A spatial attention mechanism is designed to enhance the feature graph after channel enhancement in the spatial direction.

[0012] Further, the hybrid loss function is

[0013] wherein, and are coefficient factors corresponding to the loss function, is a mean square error loss function, is a focal loss function, is a loss function.

[0014] Further, the steps of training the constructed detection architecture according to the Plasmodium infection stage image dataset to determine the target Plasmodium infection stage detection architecture include: The Plasmodium infection stage image dataset is divided into a training set, a test set and a validation set according to a proportion; The constructed detection architecture is trained on the training set to determine the optimal training parameters; The trained detection architecture is tested and verified on the test set and the validation set, the detection result is obtained, and it is judged whether the detection result meets the preset requirement; If yes, the trained detection architecture is output as the target Plasmodium infection stage detection architecture.

[0015] Another object of the present application is a Plasmodium infection stage detection system, which comprises: a data determination module configured to obtain a malaria image, determine an infection stage of the malaria image, construct a preliminary data set by labeling the malaria image, and process the preliminary data set to determine a malaria infection stage image data set; an architecture construction module configured to construct a ResNet-based malaria infection stage detection architecture, and design a convolution block attention module in a feature fusion module of the ResNet architecture; a loss function determination module configured to construct a hybrid loss function to optimize the detection architecture, the hybrid loss function at least simultaneously fusing features of a mean square error loss function, a focal loss function, and a Dice loss function; a target architecture determination module configured to train the constructed detection architecture according to the malaria infection stage image data set to determine a target malaria infection stage detection architecture.

[0016] Another object of the embodiments of the present application is to provide a storage medium having a computer program stored thereon, the program being executed by a processor to implement the steps of the malaria infection stage detection method described above.

[0017] Another object of the embodiments of the present application is to provide an electronic device including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the malaria infection stage detection method described above when executing the program.

[0018] The present application constructs a fusion image quality improvement architecture and a ResNet-based detection architecture to accurately and timely detect the infection stage of malaria parasites, providing an effective reference model for rapid diagnosis of malaria. Based on the characteristics of the diverse and complex noise of the malaria image in different infection stages, such as ring-shaped bodies, schizonts, trophozoites, and gametocytes, an image enhancement method that fuses wavelet transform and Transformer is designed to filter the key frequency characteristics in the malaria image and remove noise frequency features, achieving high-quality construction of the malaria infection stage data set. Based on the characteristics of the diverse morphological features of malaria parasites in different infection stages, a ResNet detection structure that fuses convolution block attention modules and hybrid loss functions is designed to improve the accuracy and robustness of the detection of diverse morphological features of malaria parasites in different infection stages. In addition, the present application can also be extended to accurate detection of other bacterial infection stages, which is conducive to large-scale application in clinical detection equipment. Therefore, the present application solves the problem of the lack of accurate and rapid identification of irregular morphological features of malaria parasites in different infection stages in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 Flow chart of the method for detecting the stage of Plasmodium infection in the first embodiment of the present application; Figure 2 Result diagram of the system for detecting the stage of Plasmodium infection in the second embodiment of the present application; Figure 3 Structural diagram of the electronic device in the third embodiment of the present application; The following detailed description will further describe the present application in combination with the above-mentioned drawings. DETAILED DESCRIPTION

[0020] In order to facilitate the understanding of the present application, the present application will be described more fully below with reference to the accompanying drawings. The drawings show several embodiments of the present application. However, the present application can be realized in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing the specific embodiments and are not intended to limit the present application. The term "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0022] Embodiment one Please refer to Figure 1 , which shows the method for detecting the stage of Plasmodium infection in the first embodiment of the present application, which specifically includes steps S01-S04.

[0023] S01, obtaining a Plasmodium image and labeling the Plasmodium image to determine the corresponding stage of infection, constructing a preliminary data set of the labeled Plasmodium image, and processing the preliminary data set to determine a Plasmodium infection stage image data set.

[0024] Specifically, images of Plasmodium at different infection stages in clinic are recorded, and four infection stages (i.e., ring, schizont, trophozoite and gametocyte) of Plasmodium in the Plasmodium images are labeled based on image characteristics. The Plasmodium images labeled by the database management software are used to construct a preliminary data set. A wavelet transform method is designed to decompose the Plasmodium images into low-frequency components and high-frequency components to construct a frequency domain expression matrix. A Gaussian normalization method is used to standardize the low-frequency components and the high-frequency components, and the standardized high-frequency components and the low-frequency components are stacked to form a frequency domain expression matrix with fused multi-channel input. A Transformer is used to filter key frequency band features and remove noise frequency bands in the preliminary data set to form a Plasmodium infection stage image data set. The function corresponding to the wavelet transform method is a Haar wavelet basis function, and the expression of the Haar wavelet basis function is

[0025] wherein, is a scale function value, is a pixel information value of an input image, is a scale factor, is a translation parameter.

[0026] Further, the step of using the Transformer to filter key frequency band features and remove noise frequency bands in the preliminary data set to form the Plasmodium infection stage image data set includes: flattening each channel component of the frequency domain expression matrix to form a frequency domain sequence; constructing a linear layer to map the frequency domain sequence by the Transformer, adding position encoding to the mapped frequency domain sequence; constructing a multi-head self-attention mechanism to extract the dependence relationship across frequency bands in the mapped frequency domain sequence, and extracting features for global context information; filtering out key frequency band features based on attention weight analysis, and removing noise frequency band features in the Plasmodium images to construct the Plasmodium infection stage image data set.

[0027] In specific implementation, four infection stages are labeled based on clinical images to form a preliminary data set, which provides a basic supervision signal for model training. Then, the image is decomposed into low-frequency components and high-frequency components by using the Haar wavelet basis function, and the frequency domain components are standardized by Gaussian normalization to avoid the influence of magnitude difference on subsequent processing, so as to convert the spatial domain image into a frequency domain expression matrix, highlight the morphological features of Plasmodium, and suppress part of the noise. The frequency domain sequence is input into the Transformer to capture the cross-frequency band dependence relationship through position encoding and multi-head self-attention mechanism. Key frequency bands such as Plasmodium edges and internal structural features are selected based on attention weight, and noise frequency bands such as background interference are removed. Thus, a high-quality Plasmodium infection stage image data set is constructed, and the input quality of the subsequent model is improved.

[0028] S02, configure a ResNet-based malaria infection stage detection architecture, and in a feature fusion module of the ResNet architecture, design a convolution block attention module.

[0029] Specifically, the ResNet-based malaria infection stage detection architecture is configured and specifically includes four modules: an input module, a residual connection module, a feature fusion module, and a classification head module, which are specifically as follows: in the input module, a 7x7 convolution kernel is used to perform convolution operation on the input image, and a maximum pooling operation is performed on the feature image after the convolution operation to obtain a preliminary feature extraction image; in the residual connection module, three convolution kernels of different sizes (i.e., 1x1, 3x3, and 5x5) are sequentially used to perform feature extraction on the preliminary feature extraction image, and the residual information of the previous step is gradually fused during the feature extraction process to match the irregular scale characteristics of the malaria image; in the feature fusion module, the outputs of the three scales in the residual connection module are spliced in the channel dimension to obtain a comprehensive feature map, and a convolution block attention module is designed to improve the attention to irregular shapes; in the classification head module, a fully connected layer and an activation function are designed to obtain the score value of the malaria image in which infection stage, and the detection result is obtained based on the maximum score.

[0030] In specific implementation, the input module uses a 7x7 convolution kernel and a maximum pooling to capture preliminary features, so that the 7x7 convolution can quickly expand the receptive field at the early stage of the network and capture some more macro and rough features in the image. The pooling operation is used for dimension reduction, reduces the amount of calculation, and provides certain translation invariance. The residual connection module uses 1x1, 3x3, and 5x5 multi-scale convolution kernels to adapt to the irregular shape and multi-scale shape characteristics of malaria, such as the circular profile of the ring-shaped body or the irregular splitting shape of the schizont, to ensure that targets of different sizes can be effectively captured, and to alleviate the gradient vanishing problem in deep network training through residual connection, so that the network can be designed very deep, thus being more powerful and easy to train, while preserving the morphological characteristics of different scales. The feature fusion module splices the outputs of the three different scales in the residual connection module in the channel dimension to form a “comprehensive feature map”. This feature map contains information extracted from different scales and different abstraction levels. Thus, multi-scale information is integrated, so that the subsequent classifier can utilize both detailed features and global features to make a comprehensive judgment. The classification head module inputs the fused comprehensive feature map into one or more fully connected layers, and finally outputs a four-dimensional vector through an activation function to map the rich features extracted before to the final classification result. The four values of the output vector respectively represent the probability scores of the input image belonging to the “ring-shaped body”, “schizont”, “nutrient body”, and “gametocyte” categories, and the highest score is taken as the final detection result.

[0031] Further, in the feature fusion module of the ResNet architecture, a convolution block attention module is designed to improve the attention to irregular shapes, and the specific steps are as follows: a convolution block is designed to perform convolution operation on the integrated feature map to obtain a feature map after convolution; a channel attention mechanism is designed to enhance the feature map in the channel direction; and a spatial attention mechanism is designed to enhance the feature map after channel enhancement in the spatial direction. In specific implementation, even if the image is preprocessed, it may still contain some information irrelevant to classification. The network needs to learn to automatically focus on the most critical area of the image for classification, such as the Plasmodium body, and ignore irrelevant background areas, so as to further improve the accuracy. Therefore, the channel attention mechanism is used to weight the feature map channels, enhance the channels related to the infection stage, such as specific color or texture channels, and combine the spatial attention mechanism to focus on the key spatial area of Plasmodium, such as the nucleus position, and suppress irrelevant backgrounds. Thus, the attention to irregular shapes is significantly improved, and false detection is reduced.

[0032] S03, a hybrid loss function is constructed to optimize the detection architecture, which at least simultaneously combines the features of mean square error loss function, focal loss function and Dice loss function.

[0033] Specifically, the hybrid loss function is

[0034] wherein, and are the coefficient factors of the corresponding loss functions, is the mean square error loss function, is the focal loss function, is loss function. In specific implementation, in order to ensure that the training direction of the Plasmodium infection stage detection architecture based on ResNet matches the characteristics of images in different Plasmodium infection stages, a hybrid loss function that can simultaneously combine the characteristics of L2 loss function, Focal loss function and Dice loss function is designed to optimize the parameters of the training process. More specifically, L2 loss is used as a regularization term to constrain the model complexity, improve the stability and generalization ability of the training, and effectively prevent overfitting; Focal loss reduces the weight of easy-to-classify samples through a dynamic modulation mechanism, forcing the model to focus on difficult-to-classify samples and minority classes, and fundamentally alleviating the class imbalance problem in Plasmodium images; Dice loss is introduced from the field of image segmentation to optimize the shape perception, and by maximizing the overlap between the prediction and the true label, the recognition accuracy of the model for the irregular shape and fine boundary of Plasmodium is significantly improved.

[0035] S04, training the constructed detection architecture according to the image dataset of the malaria infection stage to determine a target malaria infection stage detection architecture.

[0036] Specifically, the image dataset of the malaria infection stage is divided into a training set, a test set and a validation set in a ratio of The constructed detection architecture is trained on the training set to determine optimal training parameters. The trained detection architecture is tested and verified on the test set and the validation set to obtain a detection result, and it is determined whether the detection result meets a preset requirement. If yes, the trained detection architecture is output as the target malaria infection stage detection architecture. The required detection model, i.e., the target malaria infection stage detection architecture, is determined through iterative training of the model.

[0037] To further illustrate the present embodiment, the method of the present embodiment is tested by using a malaria infection stage image database. The malaria infection stage image database contains two types of images. The first type is malaria images in which malaria parasites are in four different infection stages, i.e., ring stage, schizont stage, trophozoite stage and gametocyte stage, and a total of 200 images. The second type is cryptococcus images similar to malaria, and a total of 40 images. The number ratio of the first type of images to the second type of images is 5:1. The average precision AP 0.5 As an evaluation index, the evaluation index refers to the area under the Precision-Recall curve calculated when the intersection over union (IoU) threshold is 0.5. When the intersection over union (IoU) of the model prediction box and the real box is ≥0.5, it is considered that the model prediction is correct. The experimental results are shown in Table 1.

[0038] Table 1 Image recognition results of the present application on the present embodiment

[0039] As can be seen from the results in Table 1, the method of the present embodiment has obtained a high detection accuracy, which reflects that the method of the present embodiment has strong detection capability for determining the infection stage of malaria parasites in patients infected with malaria. It can be considered that the method of the present embodiment has good performance on the problem of malaria infection stage detection.

[0040] To sum up, the malaria infection stage detection method in the above embodiments of the application accurately and timely detects the infection stages of malaria parasites by constructing a fusion image quality improvement architecture and a ResNet-based detection architecture, providing an effective reference model for rapid diagnosis of malaria. Moreover, based on the characteristics of the diversified and complex noise of the malaria image in different infection stages, such as ring-shaped bodies, schizonts, trophozoites, and gametocytes, an image enhancement method that fuses wavelet transform and Transformer is designed to filter the key frequency band characteristics in the malaria image and remove noise frequency band features, achieving high-quality construction of the malaria infection stage dataset. Furthermore, based on the characteristics of the diversified morphological features of malaria parasites in different infection stages, a ResNet detection structure that fuses convolution block attention modules and hybrid loss functions is designed to improve the accuracy and robustness of the detection of diversified morphological features of malaria parasites in different infection stages. In addition, the application can also be extended to accurate detection of other bacterial infection stages, which is conducive to large-scale application in clinical detection equipment. Therefore, the application solves the problem of the lack of accurate and rapid identification of irregular morphological malaria parasites in different infection stages in the prior art.

[0041] Embodiment Two Referring to Figure 2 The structure block diagram of the malaria infection stage detection system 200 proposed in the second embodiment of the application is shown in the figure, and the malaria infection stage detection system 200 comprises a data determination module 21, an architecture construction module 22, a loss function determination module 23, and a target architecture determination module 24, wherein: The data determination module 21 is used to obtain a malaria image and determine the corresponding infection stage by labeling the malaria image, so as to form a preliminary dataset by labeling the malaria image and determine a malaria infection stage image dataset by processing the preliminary dataset; The architecture construction module 22 is used to construct a ResNet-based malaria infection stage detection architecture, and design a convolution block attention module in the feature fusion module of the ResNet architecture; The loss function determination module 23 is used to construct a hybrid loss function to optimize the detection architecture, wherein the hybrid loss function at least simultaneously fuses the features of the mean square error loss function, the focal loss function, and the Dice loss function; The target architecture determination module 24 is used to train the constructed detection architecture according to the malaria infection stage image dataset to determine a target malaria infection stage detection architecture.

[0042] To sum up, the malaria infection stage detection system in the above embodiments of the present application can accurately and timely detect the infection stage of malaria by constructing a fusion image quality improvement architecture and a ResNet-based detection architecture, thereby providing an effective reference model for rapid diagnosis of malaria. In addition, based on the characteristics of the diversified and complex noise of the malaria image in different infection stages, such as ring, schizont, trophozoite and gametocyte, an image enhancement method combining wavelet transform and Transformer is designed to filter the key frequency characteristics of the malaria image and remove the noise frequency characteristics, thereby realizing high-quality construction of the malaria infection stage dataset. Furthermore, based on the characteristics of the diversified morphological features of the malaria image in different infection stages, a ResNet detection structure combining a convolution block attention module and a hybrid loss function is designed to improve the accuracy and robustness of the detection of diversified morphological features of the malaria infection stage. In addition, the present application can also be extended to accurate detection of other bacterial infection stages, which is conducive to large-scale popularization and application in clinical detection equipment. Therefore, the present application solves the problem of the lack of accurate and rapid identification of the irregular morphology of the malaria infection stage presented by the malaria in different infection stages in the prior art Embodiment three In another aspect, the present application also provides an electronic device, please refer to Figure 3 , which is a schematic diagram of the electronic device in the third embodiment of the present application, comprising a memory 20, a processor 10 and a computer program 30 stored in the memory and executable on the processor, wherein the processor 10 implements the above-mentioned malaria infection stage detection method when executing the computer program 30.

[0043] In some embodiments, the processor 10 can be a central processing unit (CPU), a controller, a microcontroller, a microprocessor or other data processing chip, which is used to run the program code or process data stored in the memory 20, such as executing access restriction programs.

[0044] The memory 20 includes at least one type of readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. The memory 20 can be an internal storage unit of the electronic device in some embodiments, such as a hard disk of the electronic device. The memory 20 can also be an external storage device of the electronic device in other embodiments, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device. Further, the memory 20 can include both the internal storage unit and the external storage device of the electronic device. The memory 20 can be used not only to store application software and various data of the electronic device, but also to temporarily store data that has been output or will be output.

[0045] It is noted that, Figure 3 The illustrated structure does not limit the electronic device, and in other embodiments, the electronic device can include fewer or more components than illustrated, or combine certain components, or arrange the components differently.

[0046] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program. The program is executed by a processor to implement the method for detecting the infection stage of Plasmodium as described above.

[0047] Those skilled in the art can understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a list of executable instructions for implementing the logic function, which can be specifically implemented in any computer readable medium for use by or in conjunction with an instruction execution system, device or apparatus, such as a computer-based system, a system including a processor or other system that can fetch and execute instructions from the instruction execution system, device or apparatus. For the present specification, the "computer readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by or in conjunction with the instruction execution system, device or apparatus, or in conjunction with these instruction execution systems, devices or apparatus.

[0048] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that is then suitable for use by the computer. Program code embodied on a computer-readable medium can be transmitted using any apparatus adapted to transmit such a program code, including a modem, or other fetal communication apparatus.

[0049] It should be understood that aspects of the application can be implemented in hardware, software, firmware, or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, by software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following techniques, which are well known in the art, can be used to implement the application: a hybrid of the above techniques, discrete logic circuit(s) having logic gates for implementing logic functions upon request pins of the discrete logic circuit(s), programmable logic array(s) (PLAs), field programmable gate array(s) (FPGAs), etc.

[0050] In the description of the specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the application. In the specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in one or more embodiments or examples.

[0051] The above embodiments only express several implementation manners of the application, and the description is more specific and detailed, but it should not be understood as a limitation on the patent scope of the application. It should be noted that for those skilled in the art, without departing from the concept of the application, a number of modifications and improvements can be made, which are all within the protection scope of the application. Therefore, the patent protection scope of the application should be subject to the appended claims.

Claims

1. A method for detecting the infection stage of Plasmodium, characterized in that, The method includes: Images of Plasmodium are acquired and labeled to determine the corresponding infection stage, so as to form a preliminary dataset of labeled Plasmodium images, and the preliminary dataset is processed to determine the image dataset of Plasmodium infection stage. We construct a ResNet-based architecture for detecting the malaria infection stage and design a convolutional block attention module in the feature fusion module of the ResNet architecture. A hybrid loss function is constructed to optimize the detection architecture, wherein the hybrid loss function at least simultaneously integrates features of the mean squared error loss function, the focus loss function, and the Dice loss function; The constructed detection architecture is trained based on the image dataset of the malaria infection stage to determine the target malaria infection stage detection architecture.

2. The method for detecting the infection stage of Plasmodium according to claim 1, characterized in that, The steps of processing the preliminary dataset to determine the image dataset of the malaria parasite infection stage include: A wavelet transform method is designed to decompose the malaria parasite image into low-frequency and high-frequency components to construct a frequency domain representation matrix. The low-frequency component and the high-frequency component are standardized by Gaussian normalization, and the standardized high-frequency component and the low-frequency component are stacked to form a frequency domain representation matrix that integrates multi-channel input. The Transformer was used to filter key frequency band features and remove noisy frequency bands from the initial dataset to form an image dataset of the malaria infection stage. Wherein, the function corresponding to the wavelet transform method is the Haar wavelet basis function, and the expression of the Haar wavelet basis function is: in, The value of the scaling function. The pixel information values ​​of the input image. It is a scale factor. These are the translation parameters.

3. The method for detecting the infection stage of Plasmodium according to claim 2, characterized in that, The steps for using Transformer to filter key frequency band features and remove noisy frequency bands from the initial dataset to form an image dataset of the malaria infection stage include: Flatten each channel component of the frequency domain representation matrix to form a frequency domain sequence; A linear layer is constructed to perform Transformer mapping on the frequency domain sequence, and position encoding is added to the mapped frequency domain sequence; A multi-head self-attention mechanism is constructed to extract cross-frequency band dependencies in the mapped frequency domain sequence, and feature extraction is performed based on global context information; Key frequency band features were selected based on attention weight analysis, and noisy frequency band features were removed from the malaria parasite images to construct an image dataset of the malaria parasite infection stage.

4. The method for detecting the infection stage of Plasmodium according to claim 1, characterized in that, A ResNet-based architecture for detecting the malaria parasite infection stage is constructed, comprising four modules: an input module, a residual connection module, a feature fusion module, and a classification head module, as detailed below: In the input module, a 7*7 kernel is used to perform a convolution operation on the input image, and max pooling is performed on the feature image after the convolution operation to obtain a preliminary feature extraction map. In the residual connection module, three convolutional kernels of different sizes are executed sequentially to extract features from the preliminary feature extraction map. During the feature extraction process, the residual information from the previous step is gradually fused to match the irregular scale characteristics of the malaria parasite image. In the feature fusion module, the outputs of the three scales in the residual connection module are concatenated along the channel dimension to obtain a comprehensive feature map. A convolutional block attention module is designed to improve attention to irregular shapes. In the classification head module, a fully connected layer and activation function are designed to obtain a score value indicating the infection stage of the malaria parasite image, and the detection result is obtained based on the maximum score.

5. The method for detecting the infection stage of Plasmodium according to claim 4, characterized in that, In the feature fusion module of the ResNet architecture, a convolutional block attention module is designed to improve attention to irregular shapes. The specific steps are as follows: Design convolutional blocks to perform convolution operations on the comprehensive feature maps to obtain convolutional feature maps; Design a channel attention mechanism to enhance the convolutional feature map in the channel direction; The design incorporates a spatial attention mechanism to enhance the feature map after channel enhancement in the spatial direction.

6. The method for detecting the infection stage of Plasmodium according to claim 1, characterized in that, The hybrid loss function is: in, and These are the coefficient factors corresponding to the loss function. Let the mean squared error loss function be . For focus loss function, for Loss function.

7. The method for detecting the infection stage of Plasmodium according to claim 3, characterized in that, The steps for training the constructed detection architecture based on the image dataset of the malaria infection stage to determine the target malaria infection stage detection architecture include: The image dataset of the malaria parasite infection stage is divided into training set, test set and validation set according to the proportion; The constructed detection architecture is trained on the training set to determine the optimal training parameters; The trained detection architecture is tested and validated on the test set and the validation set to obtain detection results and determine whether the detection results meet the preset requirements. If so, the trained detection architecture is output as the detection architecture for the target malaria parasite infection stage.

8. A system for detecting the infection stage of Plasmodium, characterized in that, For implementing the method for detecting the stage of malaria infection as described in any one of claims 1 to 7, the system comprises: The data determination module is used to acquire images of Plasmodium and label the images to determine the corresponding infection stage, so as to construct a preliminary dataset from the labeled images of Plasmodium and process the preliminary dataset to determine the image dataset of the infection stage of Plasmodium. An architecture building module is used to construct a ResNet-based architecture for detecting the malaria infection stage, and a convolutional block attention module is designed in the feature fusion module of the ResNet architecture. The loss function determination module is used to construct a hybrid loss function to optimize the detection architecture, wherein the hybrid loss function at least simultaneously integrates the features of the mean squared error loss function, the focus loss function, and the Dice loss function; The target architecture determination module is used to train the constructed detection architecture based on the image dataset of the malaria infection stage to determine the target malaria infection stage detection architecture.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method for detecting the stage of malaria infection as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for detecting the stage of malaria infection as described in any one of claims 1-7.

Images