Disease detection method based on brain iron deposition, computing device and computer storage medium

Abstract

The embodiment of the invention provides a disease detection method based on brain iron deposition, computing equipment, a computer storage medium and a computer program product. The disease detection method based on brain iron deposition comprises the following steps: acquiring to-be-detected target brain medical image data; and inputting the target brain medical image data into a detection model generated by pre-training, so that the detection model outputs a detection result determined based on iron deposition information extracted from the target brain medical image data. According to the technical scheme provided by the embodiment of the invention, the accurate detection method for the chronic kidney disease is provided.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a disease detection method, computing device, computer storage medium, and computer program product based on brain iron deposition. Background Technology

[0002] Chronic kidney disease (CKD) is a global public health problem with a rising incidence rate, posing a serious threat to human health. Because CKD often lacks obvious symptoms in its early stages, early diagnosis is extremely difficult, and patients are often diagnosed only at later stages of disease progression, missing the optimal treatment window.

[0003] Therefore, how to provide an accurate detection method for chronic kidney disease has become an urgent technical problem to be solved. Summary of the Invention

[0004] This application provides a disease detection method, computing device, computer storage medium, and computer program product based on brain iron deposition.

[0005] In a first aspect, embodiments of this application provide a disease detection method based on brain iron deposition, comprising:

[0006] Acquire medical image data of the target brain to be detected;

[0007] The target brain medical image data is input into a pre-trained detection model, so that the detection model outputs a detection result based on iron deposition information extracted from the target brain medical image data, the iron deposition information including iron deposition information of the target region of the target brain medical image data; wherein, the detection model includes alternately stacked convolutional layers and a spatial attention module, the convolutional layers are used to extract features from the target brain medical image data to generate feature data, and the feature data is input into the spatial attention module, the spatial attention module is used to determine the target region in the target brain medical image data based on the feature data, and to determine the receptive field matching the target region, and to use the receptive field to obtain the iron deposition information of the target region.

[0008] Secondly, this application provides a disease detection device based on brain iron deposition, comprising:

[0009] The first acquisition module is used to acquire medical image data of the target brain to be detected;

[0010] A detection module is used to input the target brain medical image data into a pre-trained detection model, so that the detection model outputs a detection result based on iron deposition information extracted from the target brain medical image data, wherein the iron deposition information includes iron deposition information of the target region of the target brain medical image data; wherein, the detection model includes alternately stacked convolutional layers and a spatial attention module, the convolutional layers are used to extract features from the target brain medical image data, generate feature data, and input the feature data into the spatial attention module, the spatial attention module is used to determine the target region in the target brain medical image data based on the feature data, and determine the receptive field matching the target region, and use the receptive field to obtain the iron deposition information of the target region.

[0011] Thirdly, this application provides a computing device, including a processing component and a storage component;

[0012] The storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the disease detection method provided in the embodiments of this application.

[0013] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements the disease detection method provided in this application.

[0014] Fifthly, this application provides a computer program product, which includes computer program code. When the computer program code is executed by a computer, it implements the disease detection method provided in this application.

[0015] In the embodiments of this application, a detection model consisting of alternating stacks of convolutional layers and spatial attention modules is employed to detect target brain medical image data. The convolutional layers extract features from the image, acting like a filter to extract valuable features such as edges, textures, and shapes from the original image, providing rich information for subsequent analysis. The spatial attention module, based on the feature data extracted by the convolutional layers, identifies the target region in the image, automatically focusing on potentially problematic areas, improving detection accuracy and efficiency. It can also determine the receptive field matching the target region, more precisely focusing on specific details of the target region to obtain iron deposition information. This technical solution can more accurately detect subtle changes in iron content in the brain, while also considering individual differences, providing more accurate and interpretable diagnostic results.

[0016] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0017] 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of a disease detection method based on brain iron deposition according to an embodiment of this application is shown;

[0019] Figure 2 A block diagram of a disease detection device based on brain iron deposition according to an embodiment of this application is shown;

[0020] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of this application is shown. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0022] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0023] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0024] Chronic kidney disease (CKD) is a global public health problem with a rising incidence rate, posing a serious threat to human health. Because CKD often lacks obvious symptoms in its early stages, early diagnosis is extremely difficult, and patients are often diagnosed only at a later stage of disease progression, missing the optimal treatment window.

[0025] Traditional medical imaging diagnostic methods have played a crucial role in the diagnosis of chronic kidney disease (CKD), but they have also revealed some insurmountable problems when facing the need for early CKD diagnosis and a deeper understanding of its pathological mechanisms. In terms of accuracy, traditional methods may fail to accurately identify subtle pathological changes in the brains of early-stage CKD patients, leading to misdiagnosis or missed diagnosis. These methods may only detect relatively obvious lesions, while failing to detect more subtle changes in the early stages, thus affecting the accuracy of early diagnosis.

[0026] Traditional medical imaging diagnostic methods also have limitations in terms of interpretability. When using these methods for diagnosis, doctors often find it difficult to intuitively understand the pathological information reflected in the images, and also struggle to clearly explain the basis of the diagnostic results to patients. This not only affects doctors' accurate judgment of the disease and the formulation of treatment plans, but may also lead to patients' distrust and confusion regarding the diagnostic results. Therefore, the limitations of traditional medical imaging diagnostic methods in terms of accuracy and interpretability make it difficult for them to meet the urgent clinical needs for early diagnosis of chronic kidney disease and in-depth exploration of its pathological mechanisms.

[0027] In the embodiments of this application, a detection model consisting of alternating stacks of convolutional layers and spatial attention modules is used to detect target brain medical image data. The convolutional layers can extract features from the image, acting like a filter to extract valuable features such as edges, textures, and shapes from the original image, providing rich information for subsequent analysis. The spatial attention module determines the target region in the image based on the feature data extracted by the convolutional layers, automatically focusing on areas that may have problems, improving detection accuracy and efficiency. It can also determine the receptive field that matches the target region, more precisely focusing on specific details of the target region to obtain iron deposition information. This technical solution can more accurately detect subtle changes in iron content in the brain, while taking into account individual differences, providing more accurate and interpretable diagnostic results.

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] The implementation details of the technical solutions in the embodiments of this application are described in detail below.

[0030] Figure 1 The flowchart shown is a disease detection method based on brain iron deposition according to an embodiment of this application. Figure 1 As shown, this disease detection method based on brain iron deposition may specifically include the following steps:

[0031] 101. Acquire medical image data of the target brain to be detected;

[0032] 102. The target brain medical image data is input into a pre-trained detection model so that the detection model outputs a detection result based on iron deposition information extracted from the target brain medical image data, wherein the iron deposition information includes iron deposition information of the target region of the target brain medical image data; wherein the detection model includes alternately stacked convolutional layers and a spatial attention module, the convolutional layers are used to extract features from the target brain medical image data to generate feature data, and the feature data is input into the spatial attention module, the spatial attention module is used to determine the target region in the target brain medical image data based on the feature data, and to determine the receptive field matching the target region, and to use the receptive field to obtain the iron deposition information of the target region.

[0033] Target brain medical image data can be acquired through medical imaging equipment, such as magnetic resonance imaging (MRI) and quantitative susceptibility imaging (QSM). These devices can generate high-resolution brain images that display the brain's structure and tissue characteristics. For example, in hospitals or research institutions, the brain imaging data obtained after a patient undergoes a specific scanning procedure can be used as the target brain medical image data for testing.

[0034] In one embodiment of this application, image data acquired through medical imaging equipment can be preprocessed to generate target brain medical image data to ensure data quality and consistency. Preprocessing operations may include, for example, image denoising, normalization, and cropping. Image denoising removes noise from the image, improving image clarity. Normalization adjusts the values ​​of the image data to a specific range, facilitating subsequent processing and analysis. Cropping removes irrelevant background information, focusing attention on the brain region.

[0035] In embodiments of this application, the detection model consists of alternating stacked convolutional layers and a spatial attention module. The convolutional layers are used to extract features from the target brain medical image data to generate feature data. The spatial attention module then uses this feature data to determine the target region in the image and obtains iron deposition information in that region.

[0036] Convolutional layers perform convolution operations on target brain medical image data, extracting features such as edges, textures, and shapes. Stacking multiple convolutional layers allows for the progressive extraction of higher-level features. Feature extraction is achieved by sliding convolutional kernels across the image to perform convolution operations. A convolutional kernel can be viewed as a filter capable of detecting specific patterns in an image. Different kernels extract different types of features, and stacking multiple convolutional layers progressively extracts more advanced and abstract features. For example, in brain medical images, convolutional layers can extract edge features of brain iron deposition areas and brightness differences between different brain regions; these features will serve as input to subsequent spatial attention modules.

[0037] The spatial attention module can adaptively focus on the distribution of iron in different regions of the brain based on the feature data output by the convolutional layer, thereby identifying the target region in the target brain medical image data. The target region may include key areas related to iron deposition in the brain, which may be a specific brain region or lesion site.

[0038] The spatial attention module can dynamically adjust its receptive field to match the size and characteristics of the target region, i.e., the size of the area of ​​focus, and then use this receptive field to acquire iron deposition information in the target region. For example, for some small iron deposition areas or early lesions, the spatial attention module can narrow its receptive field to focus on these local areas.

[0039] By adjusting the size and location of the receptive field, the spatial attention module can more accurately capture the features of these fine structures, thereby improving the model's sensitivity to changes in fine structures.

[0040] Iron deposition information may include, for example, iron deposition intensity information and distribution information.

[0041] In the embodiments of this application, to enable the detection model to accurately detect iron deposition information in the brain, a large amount of labeled data can be used for training. Labeled data may include brain medical images, corresponding iron deposition region annotations, and disease diagnosis results. During training, the model continuously adjusts its parameters to make the output detection results as close as possible to the actual labeled results. The training algorithm can employ backpropagation, calculating the gradient of the loss function with respect to the model parameters, and then using optimization algorithms (such as stochastic gradient descent, Adam, etc.) to update the parameters. A fully trained detection model can learn the patterns and rules for extracting iron deposition information from brain medical images, thereby enabling accurate detection of new target brain medical image data.

[0042] In this embodiment, when target brain medical image data is input into the detection model, the model can output a detection result based on iron deposition information extracted from the data. The detection result can be a classification result, such as determining whether a patient suffers from a disease related to brain iron deposition; or it can be a quantitative result, such as the degree or distribution of iron deposition. The specific output format depends on the model design and application scenario. For example, if used for disease diagnosis, the output might be "has disease" or "does not have disease"; if used to study the distribution of brain iron deposition, the output might be a three-dimensional iron deposition heatmap showing the intensity of iron deposition in different brain regions.

[0043] In the embodiments of this application, data augmentation techniques can be used when training the detection model to increase the diversity of data, so that the detection model can better adapt to different situations during the training process and improve the robustness of the detection model.

[0044] Data augmentation techniques can include, for example, random flipping, random cropping, adding noise, and color transformation. Random rotation allows the image to be rotated randomly within a certain angle range, enabling the model to learn brain morphology from different angles. Random cropping randomly selects different regions from the original image, increasing the model's ability to learn different local features. Adding noise simulates noise conditions in real-world data, improving the model's robustness to interference. Color transformation adjusts the image's color channels, increasing the model's adaptability to different color features.

[0045] Data augmentation techniques can increase the number of training data samples without increasing data acquisition costs, providing more learning data for the detection model. Furthermore, different brain medical images may vary due to individual patient differences, different scanning equipment, variations in scanning angles, and other factors. Data augmentation techniques can simulate these differences, allowing the model to encounter more diverse scenarios during training, thereby improving the robustness of the detection model.

[0046] In the embodiments of this application, a detection model consisting of alternating stacks of convolutional layers and spatial attention modules is employed to detect target brain medical image data. The convolutional layers extract features from the image, acting like a filter to extract valuable features such as edges, textures, and shapes from the original image, providing rich information for subsequent analysis. The spatial attention module, based on the feature data extracted by the convolutional layers, identifies the target region in the image, automatically focusing on potentially problematic areas, improving detection accuracy and efficiency. It can also determine the receptive field matching the target region, more precisely focusing on specific details of the target region to obtain iron deposition information. This technical solution can more accurately detect subtle changes in iron content in the brain, while also considering individual differences, providing more accurate and interpretable diagnostic results.

[0047] In some embodiments, acquiring the target brain medical image data to be detected can specifically be implemented as follows:

[0048] Acquire first-order medical brain image data and brain region segmentation templates;

[0049] The original brain medical image data is corrected using the brain region segmentation template to generate the target brain medical image data.

[0050] The first brain medical image data can be the initial image data acquired through medical imaging equipment. This initial image data may contain problems such as noise, distortion, and offset. These problems may affect the quality and accuracy of the image, thus affecting subsequent analysis and diagnosis. To improve the quality and accuracy of the image, brain region segmentation templates can be used to correct the first brain medical image data.

[0051] In embodiments of this application, the brain region segmentation template can be a tool for dividing brain images into different regions. Based on anatomical knowledge and clinical experience, the brain region segmentation template can divide the brain into multiple specific regions, such as the frontal lobe, parietal lobe, temporal lobe, occipital lobe, and basal ganglia.

[0052] Brain region segmentation templates can be obtained in various ways. In one possible implementation of this application, existing standard templates can be used, such as the brain templates provided by the Montreal Neuroscience Institute (MNI) and AAL (Automated Anatomical Labeling). These templates have undergone extensive research and validation, demonstrating high accuracy and reliability. In another possible implementation of this application, brain region segmentation templates can be constructed independently based on specific research needs. This can be achieved through manual annotation, image segmentation algorithms, or machine learning methods.

[0053] When correcting primary brain medical image data using a brain region segmentation template, the first step is to register the template with the original brain medical image data. Registration is the process of aligning two or more images so that they correspond in spatial location. This can be achieved using image registration algorithms, such as mutual information-based registration algorithms or feature point-based registration algorithms.

[0054] After registration, the original brain medical image data can be corrected based on the structure and features of the templates divided into brain regions. Correction methods can include image smoothing, noise reduction, and distortion correction. For example, if the original brain medical image data contains noise, image smoothing algorithms can be used to remove the noise; if distortion exists, distortion correction algorithms can be used to restore the normal shape of the image.

[0055] During the correction process, the raw brain medical image data can be standardized based on the information from the brain region division template. Standardization can adjust the brain images of different patients to the same spatial coordinate system and grayscale range, facilitating comparison and analysis.

[0056] By utilizing brain region segmentation templates for correction, more accurate and higher-quality target brain medical image data can be obtained.

[0057] In some embodiments, the step of using the brain region segmentation template to correct the first brain medical image data and generate the target brain medical image data can be specifically implemented as follows:

[0058] The original brain medical image data is corrected using the brain region segmentation template to generate second brain medical image data;

[0059] The second brain medical image data is cropped according to a preset cropping size to generate the third brain medical image data;

[0060] The third brain medical image data is normalized to generate the target brain medical image data.

[0061] In the embodiments of this application, the preset cropping size can be determined according to specific application requirements and data analysis objectives. In brain medical image analysis, the selection of the cropping size may take into account multiple factors, such as the size of brain structures, the location of lesions, and limitations of computing resources. For example, if the focus of the study is a specific brain region, the cropping size can be determined based on the size and location of that brain region to ensure that the cropped image only contains the region of interest. If computing resources are limited, a smaller cropping size can also be selected to reduce the time and space complexity of data processing.

[0062] In one possible embodiment of this application, the mode size can be obtained by analyzing the non-background portion of a large amount of brain image data, and then the mode size can be determined as the cropping size, thereby improving the processing efficiency and analysis accuracy of brain image data.

[0063] In the embodiments of this application, in brain images, the background portion typically refers to areas in the image unrelated to brain structures, such as image edges or blank areas. The non-background portion refers to areas containing brain structures and tissues, which are of significant importance for the diagnosis and research of brain diseases. To determine the non-background portion, image segmentation techniques can be used to divide the brain image into different regions. Then, based on anatomical knowledge and clinical experience, it can be determined which regions belong to the non-background portion. For example, image segmentation techniques such as threshold-based segmentation methods, region growing algorithms, and watershed algorithms can be used to divide the brain image into different tissue regions such as gray matter, white matter, and cerebrospinal fluid. Then, gray matter and white matter regions are identified as non-background portions.

[0064] For large datasets of brain images, image processing software or programming languages ​​can be used to analyze the dimensions of the non-background regions. Specific analysis methods can include calculating dimensional parameters such as volume, surface area, length, width, and height of the non-background regions, or using image feature extraction algorithms to extract features such as shape, texture, and color of the non-background regions, and then calculating the dimensional parameters based on these features. For example, image processing libraries in Python (such as OpenCV and Scikit-image) can be used to process brain images and calculate the dimensional parameters of the non-background regions. Alternatively, image feature extraction algorithms in deep learning (such as convolutional neural networks and autoencoders) can be used to extract features of the non-background regions, and then the dimensional parameters can be calculated based on these features.

[0065] Analyzing the dimensions of the non-background portions of a large dataset of brain images yields a size distribution. This distribution may contain multiple distinct size values, with the most frequent value being the mode size. To calculate the mode size, statistical methods can be used to analyze the size distribution. For example, frequency distribution tables, histograms, box plots, and other statistical charts can be used to visualize the size distribution, and then the mode size can be determined based on these charts. Alternatively, statistical software can be used to analyze the size data and calculate the mode size.

[0066] The mode size represents the size of the non-background portion of most brain image data. Therefore, using the mode size as the cropping size can preserve the most useful information in brain images while reducing data volume and computational complexity.

[0067] Once the mode size is determined as the cropping size, this size can be used to crop new brain image data to ensure that the cropped images have the same size and shape, facilitating subsequent analysis and processing. For example, image processing software or programming languages ​​can be used to crop new brain images, adjusting the size of the non-background portions to the mode size.

[0068] In the embodiments of this application, normalization is a data preprocessing technique aimed at adjusting data values ​​to a specific range to improve data quality and model performance. After normalization, the values ​​of the third-party brain medical image data can be adjusted to a specific range to obtain the target brain medical image data. The target brain medical image data has better data quality and model performance, making it more suitable for subsequent analysis and diagnosis.

[0069] In some embodiments, the step of inputting the target brain medical image data into a pre-trained detection model so that the detection model outputs a detection result based on iron deposition information extracted from the target brain medical image data can be specifically implemented as follows:

[0070] The target brain medical image data is input into the first convolutional layer, and the first feature data is output.

[0071] The first feature data is input into the first spatial attention module so that the first spatial attention module can determine the target region from multiple regions of the target brain medical image data and output the fourth brain medical image data.

[0072] The fourth brain medical image data is output to the second convolutional layer, which outputs the second feature data.

[0073] The second feature data is input into the second spatial attention module so that the second spatial attention module determines the receptive field that matches the target region based on the second feature data, obtains the iron deposition information of the target region using the receptive field, and outputs the detection result based on the iron deposition information.

[0074] In the embodiments of this application, convolutional layers and spatial attention modules are alternately stacked to form the detection model. "Alternate stacking" can refer to a structural method for constructing a deep learning model. In the detection model provided in this application, it means that convolutional layers and spatial attention modules are arranged one after another in a certain order, i.e., a convolutional layer is followed by a spatial attention module, then another convolutional layer, and so on. This arrangement allows data to pass through different types of processing layers sequentially in the model, thereby gradually extracting and refining information.

[0075] When the target brain medical image data is input into the model, it first enters the first convolutional layer. The first convolutional layer can extract features from the target brain medical image data and output preliminary feature data. This feature data is like passing through a filter, extracting basic features such as edges, textures, and shapes from the image.

[0076] These preliminary feature data are then fed into the first spatial attention module. The first spatial attention module uses these features to determine key regions in the image. It learns the importance weight of each location, focusing attention on regions relevant to the target, such as specific brain regions associated with brain diseases or areas where iron deposits may be present, and outputs attention-filtered feature data.

[0077] Next, this data enters the second convolutional layer. The second convolutional layer can perform more in-depth feature extraction based on the features that have already been focused by the spatial attention module. Because the key regions have been highlighted before, this convolutional layer can better mine high-level features of these key regions, such as more complex texture combinations and shape changes, and output new feature data again.

[0078] Subsequently, the new feature data is sent to the second spatial attention module to further optimize the focus and filtering of key regions. This cycle continues, with data flowing alternately between the convolutional layer and the spatial attention module.

[0079] In the embodiments of this application, the convolutional layer is responsible for extracting features, while the spatial attention module filters and enhances these extracted features. By stacking them alternately, the features after each pass through the spatial attention module are used as more targeted inputs by the next convolutional layer, thereby extracting high-level features that better match the key regions; then these high-level features are further focused and enhanced by the next spatial attention module, thus forming a loop that continuously improves the ability to extract key regions and key features.

[0080] As data flows through an alternating, stacked structure, the detection model's understanding of brain medical images deepens progressively from basic features to more refined and complex features and regions. It moves from broad areas potentially containing useful information to more precise localization of specific small regions and complex features highly relevant to the target (such as brain iron deposition), shifting from a holistic observation to a focus on detail. This structure allows the model to adapt to the complex data of brain medical images, effectively uncovering hidden information relevant to goals such as disease diagnosis.

[0081] In deep neural networks, as the number of network layers increases, the gradient may gradually decrease during backpropagation, leading to the problem of vanishing gradients, or the gradient may become very large, causing training instability and thus the problem of exploding gradients.

[0082] In one possible implementation of this application, in the multiple convolutional layers and spatial attention modules of the detection model, the output of each attention module can be added to the output of the previous convolutional layer to form a residual connection.

[0083] Residual connections allow information to be directly passed from lower layers (the previous convolutional layer) to higher layers (the layers following the current spatial attention module). This ensures that even in deep networks, important details extracted by lower convolutional layers can be successfully transmitted to higher layers without being lost due to complex transformations in intermediate layers. For example, in brain medical images, lower convolutional layers may extract subtle brain tissue structural features; through residual connections, this information can be quickly passed to subsequent layers for more accurate diagnosis.

[0084] Furthermore, through residual connections, when training the detection model to process brain medical images, it can learn patterns that extract information such as brain iron deposition from image features more quickly. This helps to alleviate the problems of gradient vanishing and gradient exploding, improves the stability of the model training process, and accelerates the convergence speed.

[0085] Because the model can simultaneously learn the residual (the difference between the output of the spatial attention module and the output of the convolutional layer) and the original input (the output of the convolutional layer), it can better fit complex functional relationships. For complex data such as brain medical images, it can more accurately capture the features of brain iron deposition and their relationship with brain diseases, thereby improving the accuracy of the detection model.

[0086] In some embodiments, the step of inputting the first feature data into a first spatial attention module so that the first spatial attention module can determine the target region from multiple regions of the target brain medical image data and output the fourth brain medical image data can be specifically implemented as follows:

[0087] The first feature data is input into the first spatial attention module, so that the first spatial attention module assigns attention weights to multiple regions of the target brain medical image data based on the first feature data, and determines the target region from the multiple regions based on the attention weights, and outputs the fourth brain medical image data.

[0088] After receiving the first feature data output from the first convolutional layer, the first spatial attention module analyzes this feature data. It assesses the importance of each region to the task by calculating the similarity between the features of each region and known feature patterns related to brain iron deposition, and then calculates the attention weight for each region. The weight is typically in the range [0,1], representing the relative importance of the region. Based on these weights, the target region is determined from multiple regions in the target brain medical image data; that is, regions with higher attention weights. For example, a brain region with a significantly higher weight than other regions may be identified as a target region related to brain iron deposition. Finally, a fourth brain medical image data is output. This data highlights the target region. It may have the same spatial dimensions as the original image, but the feature representation is more focused on the target region. For example, the pixel values ​​of the target region are enhanced, while the pixel values ​​of other regions are weakened to emphasize its importance.

[0089] In some embodiments, the disease detection method based on brain iron deposition may further include:

[0090] Determine the detection result and the gradient information of the convolutional layer;

[0091] A brain attention heatmap is generated based on the gradient information and the target brain medical image data. The brain attention heatmap is used to display the attention of each region in the target brain medical image data.

[0092] Output the aforementioned brain attention heatmap.

[0093] In the embodiments of this application, gradients can be used to reflect the rate of change of a function with respect to various variables. For detection models, by calculating the gradient of the detection result relative to the feature map of the convolutional layer, it is possible to understand the degree of influence of features at various locations in the convolutional layer on the detection result. This gradient information helps to understand how the model makes detection decisions based on the input brain medical image data.

[0094] In the embodiments of this application, after calculating and generating gradient information, the gradient information is used as weights and combined with each location in the target brain medical image data to generate a brain attention heatmap.

[0095] In one possible implementation of this application, a brain attention heatmap can be generated by multiplying the absolute value of the gradient or a processed value with the image data value at the corresponding location as a weight. In the brain attention heatmap, the intensity value of each location represents the importance of that location in the model's decision-making process, i.e., the attention level. Regions with high intensity values ​​indicate that the model pays more attention to these regions when making detection results, which may be closely related to disease features such as brain iron deposition.

[0096] Finally, a brain attention heatmap is output. This heatmap visually displays the distribution of attention across different regions within a target brain medical image dataset. Doctors and researchers can quickly identify key areas of focus for the model by observing the heatmap, providing strong evidence for further analysis and interpretation of the model's decision-making process. Simultaneously, the brain attention heatmap can also help doctors and researchers better understand the distribution of disease-related features in brain medical images, providing valuable information for disease diagnosis, monitoring, and treatment.

[0097] In some embodiments, determining the detection result and the gradient information of the convolutional layer can specifically be implemented as follows:

[0098] Determine the detection result and the first gradient information of the first convolutional layer, and the detection result and the second gradient information of the second convolutional layer;

[0099] The generation of a brain attention heatmap based on the gradient information and the target brain medical image data includes:

[0100] The first gradient information and the second gradient information are weighted and summed along the channel dimension to generate the brain attention heatmap.

[0101] In the embodiments of this application, the channel dimension in the feature map of the convolutional layer can represent different feature types or patterns. By weighted summing of the two gradient information along the channel dimension, the contributions of different convolutional layers can be comprehensively considered, resulting in a more comprehensive heatmap reflecting the attention of various brain regions.

[0102] When weighting and summing the first gradient information and the second gradient information along the channel dimension, different weights can be assigned to the first gradient information and the second gradient information according to different needs and prior knowledge, so as to highlight the importance of a specific convolutional layer or balance the influence of different layers.

[0103] Figure 2 A block diagram of a disease detection device based on brain iron deposition provided in an embodiment of this application is shown, such as... Figure 2 As shown, a disease detection device based on brain iron deposition may include:

[0104] The first acquisition module 201 is used to acquire medical image data of the target brain to be detected;

[0105] The detection module 202 is used to input the target brain medical image data into a pre-trained detection model, so that the detection model outputs a detection result based on iron deposition information extracted from the target brain medical image data, wherein the iron deposition information includes iron deposition information of the target region of the target brain medical image data; wherein, the detection model includes alternately stacked convolutional layers and a spatial attention module, the convolutional layers are used to extract features from the target brain medical image data, generate feature data, and input the feature data into the spatial attention module, the spatial attention module is used to determine the target region in the target brain medical image data based on the feature data, and determine the receptive field matching the target region, and use the receptive field to obtain the iron deposition information of the target region.

[0106] In some embodiments, the first acquisition module 201 is specifically used for:

[0107] Acquire first-order medical brain image data and brain region segmentation templates;

[0108] The original brain medical image data is corrected using the brain region segmentation template to generate the target brain medical image data.

[0109] In some embodiments, the first acquisition module 201 is specifically used to correct the original brain medical image data using the brain region segmentation template to generate second brain medical image data;

[0110] The second brain medical image data is cropped according to a preset cropping size to generate the third brain medical image data;

[0111] The third brain medical image data is normalized to generate the target brain medical image data.

[0112] In some embodiments, the detection module 202 is specifically used for:

[0113] The target brain medical image data is input into the first convolutional layer, and the first feature data is output.

[0114] The first feature data is input into the first spatial attention module so that the first spatial attention module can determine the target region from multiple regions of the target brain medical image data and output the fourth brain medical image data.

[0115] The fourth brain medical image data is output to the second convolutional layer, which outputs the second feature data.

[0116] The second feature data is input into the second spatial attention module so that the second spatial attention module determines the receptive field that matches the target region based on the second feature data, obtains the iron deposition information of the target region using the receptive field, and outputs the detection result based on the iron deposition information.

[0117] In some embodiments, the detection module 202 is specifically used for:

[0118] The first feature data is input into the first spatial attention module, so that the first spatial attention module assigns attention weights to multiple regions of the target brain medical image data based on the first feature data, and determines the target region from the multiple regions based on the attention weights, and outputs the fourth brain medical image data.

[0119] In some embodiments, the disease detection device based on brain iron deposition further includes:

[0120] A gradient information determination module is used to determine the detection result and the gradient information of the convolutional layer;

[0121] A heatmap generation module is used to generate a brain attention heatmap based on the gradient information and the target brain medical image data. The brain attention heatmap is used to display the attention of each region in the target brain medical image data.

[0122] The heatmap output module is used to output the brain attention heatmap.

[0123] In some embodiments, the gradient information determination module is specifically used for:

[0124] Determine the detection result and the first gradient information of the first convolutional layer, and the detection result and the second gradient information of the second convolutional layer;

[0125] The generation of a brain attention heatmap based on the gradient information and the target brain medical image data includes:

[0126] The first gradient information and the second gradient information are weighted and summed along the channel dimension to generate the brain attention heatmap.

[0127] Figure 2 The aforementioned disease detection device based on brain iron deposition can perform... Figure 1 The implementation principle and technical effects of the disease detection method based on brain iron deposition described in the illustrated embodiment will not be repeated here. The specific operation methods of each module and unit in the disease detection device based on brain iron deposition in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0128] In one possible design, the disease detection device based on brain iron deposition provided in this application embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 301 and a processing component 302;

[0129] Storage component 301 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by processing component 302 to implement the disease detection method based on brain iron deposition provided in the embodiments of this application.

[0130] Of course, computing devices may also include other components, such as input / output interfaces and communication components. Input / output interfaces provide an interface between processing components and peripheral interface modules, which can be output devices, input devices, etc. Communication components are configured to facilitate wired or wireless communication between the computing device and other devices.

[0131] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0132] When the computing device is a physical device, it can be implemented as a distributed cluster consisting of multiple servers or terminal devices, or as a single server or a single terminal device.

[0133] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a computer, can implement the disease detection method based on brain iron deposition provided in this application.

[0134] This application also provides a computer program product, including a computer program that, when executed by a computer, can implement the disease detection method based on brain iron deposition provided in this application.

[0135] The processing component in the corresponding embodiments described above may include one or more processors to execute computer instructions to complete all or part of the steps in the method described above. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the method described above.

[0136] Storage components are configured to store various types of data to support operation within the device. Storage components can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0138] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0139] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A disease detection method based on brain iron deposition, characterized in that, include: Acquire medical image data of the target brain to be detected; The target brain medical image data is input into a pre-trained detection model, so that the detection model outputs a detection result based on iron deposition information extracted from the target brain medical image data, the iron deposition information including iron deposition information of the target region of the target brain medical image data; wherein, the detection model includes alternately stacked convolutional layers and a spatial attention module, the convolutional layers are used to extract features from the target brain medical image data to generate feature data, and the feature data is input into the spatial attention module, the spatial attention module is used to determine the target region in the target brain medical image data based on the feature data, and to determine the receptive field matching the target region, and to use the receptive field to obtain the iron deposition information of the target region.

2. The method according to claim 1, characterized in that, The acquisition of the target brain medical image data to be detected includes: Acquire first-order medical brain image data and brain region segmentation templates; The original brain medical image data is corrected using the brain region segmentation template to generate the target brain medical image data.

3. The method according to claim 2, characterized in that, The step of correcting the first brain medical image data using the brain region segmentation template to generate the target brain medical image data includes: The original brain medical image data is corrected using the brain region segmentation template to generate second brain medical image data; The second brain medical image data is cropped according to a preset cropping size to generate the third brain medical image data; The third brain medical image data is normalized to generate the target brain medical image data.

4. The method according to claim 2, characterized in that, The step of inputting the target brain medical image data into a pre-trained detection model, so that the detection model outputs a detection result based on iron deposition information extracted from the target brain medical image data, includes: The target brain medical image data is input into the first convolutional layer, and the first feature data is output. The first feature data is input into the first spatial attention module so that the first spatial attention module can determine the target region from multiple regions of the target brain medical image data and output the fourth brain medical image data. The fourth brain medical image data is output to the second convolutional layer, which outputs the second feature data. The second feature data is input into the second spatial attention module so that the second spatial attention module determines the receptive field that matches the target region based on the second feature data, obtains the iron deposition information of the target region using the receptive field, and outputs the detection result based on the iron deposition information.

5. The method according to claim 4, characterized in that, The step of inputting the first feature data into the first spatial attention module, so that the first spatial attention module can determine the target region from multiple regions of the target brain medical image data, and output the fourth brain medical image data includes: The first feature data is input into the first spatial attention module, so that the first spatial attention module assigns attention weights to multiple regions of the target brain medical image data based on the first feature data, and determines the target region from the multiple regions based on the attention weights, and outputs the fourth brain medical image data.

6. The method according to claim 4, characterized in that, The method further includes: Determine the detection result and the gradient information of the convolutional layer; A brain attention heatmap is generated based on the gradient information and the target brain medical image data. The brain attention heatmap is used to display the attention of each region in the target brain medical image data. Output the aforementioned brain attention heatmap.

7. The method according to claim 6, characterized in that, Determining the detection result and the gradient information of the convolutional layer includes: Determine the detection result and the first gradient information of the first convolutional layer, and the detection result and the second gradient information of the second convolutional layer; The generation of a brain attention heatmap based on the gradient information and the target brain medical image data includes: The first gradient information and the second gradient information are weighted and summed along the channel dimension to generate the brain attention heatmap.

8. A computing device, characterized in that, This includes processing components and storage components; The storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement the disease detection method based on brain iron deposition as described in any one of claims 1 to 7.

9. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements the disease detection method based on brain iron deposition as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes computer program code, which, when executed by a computer, implements the disease detection method based on brain iron deposition as described in any one of claims 1 to 7.

Images