Ball tube life evaluation method and device based on attention mechanism, equipment and medium

Abstract

This application provides a method, device, equipment, and medium for assessing the lifespan of X-ray tubes based on an attention mechanism, relating to the field of medical device technology. This method effectively mines degradation information in the temporal dimension of data through a multi-scale feature extraction network and spatial dimension through a two-dimensional convolutional network, achieving multi-dimensional data mining and providing more comprehensive degradation information. A dual-branch attention module is employed to further extract key features from the fused features. This module uses a multi-scale parallel structure to capture local and long-term dependencies in the data, enabling the dual-branch attention module to learn richer feature representations. Using the trained target prediction model to predict the lifespan of the X-ray tube improves prediction accuracy.

Description

Technical Field

[0001] This application relates to the field of medical device technology, specifically to a method, device, equipment, and medium for evaluating the lifespan of an X-ray tube based on an attention mechanism. Background Technology

[0002] In the field of medical imaging today, X-ray computed tomography (X-ray CT) is an important diagnostic tool, crucial for examining various parts of the human body, including the brain, spine, chest, abdomen, pelvis, and limbs. CT equipment plays a significant role not only in accurate diagnosis, guiding surgery, evaluating treatment efficacy, and disease research. However, the X-ray tube, one of the core components of CT equipment, is a high-value consumable, and its performance and reliability directly affect image quality, imaging speed, and the stability of equipment operation. The CT X-ray tube system involves multiple subsystems, including mechanical and electronic ones, and its operational data exhibits high dimensionality and complex coupling relationships. These data show significant correlations in time series and nonlinear correlations in spatial dimensions. This spatial correlation is determined by the interactions of the internal structures of the CT X-ray tube, and the interdependence between various operating parameters forms a complex network in space.

[0003] For CT tube lifespan prediction, current methods mainly include expert knowledge-based, model-based, and data-driven approaches. Expert knowledge-based methods rely on experts in the field of medical imaging equipment, using their expertise to build expert systems and then forming lifespan prediction models. However, this approach faces the challenge of acquiring and translating expert knowledge into code, typically requiring extensive collaborative development between experts and knowledge engineers. Model-based methods focus on building accurate physical models and determining whether equipment malfunctions by comparing the residuals between model predictions and actual measurements. Traditional time series prediction networks primarily focus on single-dimensional or single-scale features, rarely addressing multi-scale, multi-dimensional features simultaneously. Furthermore, traditional lifespan prediction models cannot capture local and long-term dependencies in the data. Summary of the Invention

[0004] This application provides a method, apparatus, device, and medium for evaluating the lifespan of X-ray tubes based on an attention mechanism, which can improve the accuracy of X-ray tube lifespan evaluation based on an attention mechanism.

[0005] This application provides a method for evaluating the lifespan of X-ray tubes based on an attention mechanism, comprising:

[0006] Obtain a training sample dataset, which includes multiple training samples, each training sample corresponding to a training label, the training label representing the remaining lifespan of the X-ray tube corresponding to the training sample, and the training samples including time series corresponding to multiple feature parameters of the X-ray tube;

[0007] The training samples are input into the multi-scale feature extraction network of the initial prediction model, and the multi-scale feature extraction network extracts the temporal features of the training samples to obtain multiple temporal features.

[0008] The training samples are input into the two-dimensional convolutional network of the initial prediction model, and the spatial features of the training samples are extracted by the two-dimensional convolutional network to obtain spatial features.

[0009] The multiple temporal and spatial features are input into the feature fusion layer of the initial prediction model to obtain fused features;

[0010] The fused features are input into the dual-branch attention module of the initial prediction model. The attention weights and recombination features are determined by the dual-branch attention module. The output of the attention module is determined based on the attention weights and recombination features.

[0011] The output of the attention module is input into the fully connected network of the initial prediction model to obtain the lifetime prediction label. Based on the training label and the lifetime prediction label, the initial prediction model is iteratively trained, and the iteratively trained initial prediction model is used as the target prediction model for lifetime prediction of the X-ray tube to be predicted.

[0012] In one embodiment of this application, the multi-scale feature extraction network includes multiple temporal convolutional networks, each with different convolutional kernels. Multiple temporal features are extracted from the training samples using the multi-scale feature extraction network, including:

[0013] The training samples are processed by various temporal convolutional networks to extract temporal features, resulting in multiple temporal features.

[0014] In one embodiment of this application, multiple temporal features and spatial features are input into the feature fusion layer of the initial prediction model to obtain fused features, including:

[0015] The feature fusion layer stitches together and fuses multiple temporal and spatial features to obtain the fused feature.

[0016] In one embodiment of this application, the fused features are input into the dual-branch attention module of the initial prediction model, attention weights and reconstructed features are determined through the dual-branch attention module, and the output of the attention module is determined based on the attention weights and the reconstructed features, including:

[0017] The fusion features are grouped according to a preset number of groups and the number of channels of the fusion features to obtain multiple grouped features, and a recombined feature is obtained based on the multiple grouped features;

[0018] The recombined features are subjected to channel information interoperation to obtain channel interoperation features;

[0019] Local spatial features are extracted from the recombined features to obtain local spatial features;

[0020] A first weight is determined based on the channel interaction characteristics, and a second weight is determined based on the local spatial characteristics.

[0021] The first interaction result is determined based on the channel interaction characteristics and the second weight;

[0022] The second interaction result is determined based on the local spatial features and the first weight;

[0023] The sum of the first interaction result and the second interaction result is determined as the attention weight;

[0024] The attention module output is obtained based on the attention weights and the reorganization features.

[0025] In one embodiment of this application, channel information interleaving is performed on the recombined features to obtain channel interleaving features, including:

[0026] The recombined features are subjected to adaptive pooling to obtain a pooled feature map;

[0027] The channel interaction features are determined based on the pooling feature map and the recombination features.

[0028] In one embodiment of this application, local spatial feature extraction is performed on the recombined features to obtain local spatial features, including:

[0029] The recombined features are processed by convolution to extract local spatial features, thus obtaining the local spatial features.

[0030] In one embodiment of this application, the characteristic parameters of the training samples include the cumulative number of scans, the initial value of tube current, the maximum value of tube current, the nominal value of tube current, the final value of dose, the initial value of filament current, the final value of filament current, the nominal value of filament current, the motor drive current, and the actual tube voltage.

[0031] To achieve the above and other related objectives, this application provides an attention-based tube life assessment device, comprising:

[0032] The data acquisition module is used to acquire a training sample dataset, which includes multiple training samples, each training sample corresponding to a training label, the training label representing the remaining lifespan of the X-ray tube corresponding to the training sample, and the training samples including time series corresponding to multiple feature parameters of the X-ray tube.

[0033] The first feature extraction module is used to input the training samples into the multi-scale feature extraction network of the initial prediction model, and extract multiple time features from the training samples through the multi-scale feature extraction network.

[0034] The second feature extraction module is used to input the training samples into the two-dimensional convolutional network of the initial prediction model, and extract the spatial features of the training samples through the two-dimensional convolutional network to obtain spatial features.

[0035] The feature fusion module is used to input multiple temporal features and spatial features into the feature fusion layer of the initial prediction model to obtain fused features;

[0036] The output determination module is used to input the fused features into the dual-branch group attention module of the initial prediction model, determine the attention weights and recombination features through the dual-branch group attention module, and determine the output of the attention module based on the attention weights and recombination features.

[0037] The model training module is used to input the output of the attention module into the fully connected network of the initial prediction model to obtain the lifetime prediction label, and to iteratively train the initial prediction model based on the training label and the lifetime prediction label, and to use the iteratively trained initial prediction model as the target prediction model for predicting the lifetime of the X-ray tube to be predicted.

[0038] To achieve the above and other related objectives, this application also provides an electronic device, the electronic device comprising:

[0039] One or more processors;

[0040] Memory used to store the executable program code of the processor;

[0041] The processor is configured to execute the program code to implement the above-described X-ray tube life assessment method based on the attention mechanism.

[0042] To achieve the above and other related objectives, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a computer's processor, causes the computer to perform one or more of the aforementioned attention-based tube life assessment methods.

[0043] As described above, the X-ray tube life assessment method, apparatus, device, and medium based on attention mechanism provided in this application have the following beneficial effects:

[0044] This application presents an attention-based method for evaluating X-ray tube lifespan. This method effectively mines degradation information in the temporal dimension of data through a multi-scale feature extraction network and spatial dimension through a two-dimensional convolutional network, achieving multi-dimensional data mining and providing more comprehensive degradation information. A dual-branch attention module further extracts key features from the fused features. This module employs a multi-scale parallel structure to capture local and long-term dependencies in the data, enabling the dual-branch attention module to learn richer feature representations. Using the trained target prediction model to predict X-ray tube lifespan improves prediction accuracy.

[0045] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0046] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0047] Figure 1 This is a flowchart illustrating an attention mechanism-based X-ray tube life assessment method, as shown in an exemplary embodiment of this application.

[0048] Figure 2 This is a schematic diagram illustrating a sliding window process in an exemplary embodiment of this application;

[0049] Figure 3 This is a schematic diagram of the structure of a temporal convolutional network provided in an embodiment of this application;

[0050] Figure 4 This is a schematic diagram illustrating the structure of a dual-branch group attention module, as shown in an exemplary embodiment of this application.

[0051] Figure 5 This is a flowchart illustrating the output of the attention module in an exemplary embodiment of this application;

[0052] Figure 6 This is a schematic diagram of the model architecture of the initial prediction model shown in one embodiment of this application;

[0053] Figure 7 This is a structural block diagram of an attention-based tube life assessment device, as illustrated in an exemplary embodiment of this application. Detailed Implementation

[0054] The embodiments of this application will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be understood that the preferred embodiments are only for illustrating this application and are not intended to limit the scope of protection of this application.

[0055] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. Therefore, the drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0056] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the present application. However, it will be apparent to those skilled in the art that embodiments of the present application may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the present application.

[0057] Please see Figure 1 , Figure 1 This is a flowchart illustrating an attention-based tube life assessment method according to an exemplary embodiment of this application. (Reference) Figure 1 It can be seen that this attention-based method for evaluating X-ray tube lifespan can include:

[0058] Step S110: Obtain the training sample dataset.

[0059] The training sample dataset includes multiple training samples, each corresponding to a training label. The training label represents the remaining lifespan of the X-ray tube corresponding to the training sample. Each training sample includes time series corresponding to multiple feature parameters of the X-ray tube.

[0060] In one embodiment of this application, a training sample dataset can be obtained. The training sample dataset includes multiple training samples, each containing a time series corresponding to multiple feature parameters of the X-ray tube. The training samples consist of a series of time points, each corresponding to multiple feature parameters of the X-ray tube. In this embodiment, the X-ray tube represents the CT X-ray tube. The CT (Computed Tomography) X-ray tube is one of the core components of a CT scanner; it is responsible for generating X-rays, which are captured by detectors and converted into images after passing through the human body or object. The quality and performance of the CT X-ray tube directly affect the image quality, the reliability of the equipment, and the patient's radiation safety.

[0061] In one embodiment, the characteristic parameters of the training samples include the cumulative number of scans, the initial value of the tube current, the maximum value of the tube current, the nominal value of the tube current, the final value of the dose, the initial value of the filament current, the final value of the filament current, the nominal value of the filament current, the motor drive current, and the actual tube voltage.

[0062] In one possible implementation, the process of obtaining the training sample dataset may include:

[0063] Step 1: Operational data of 33 parameters were collected during the operation of the CT tube. This data includes full lifecycle data, comprising four sets of operational data for different tubes. Before analyzing the full lifecycle operational data of the CT tube, preprocessing is necessary to ensure the accuracy and reliability of the analysis results. Specifically, blank and outlier values ​​in the dataset need to be identified and processed. Blank values ​​may be caused by errors during data acquisition or equipment malfunctions, while outliers may be caused by measurement errors or data entry errors. Deleting these outliers reduces noise in the data and improves the accuracy of subsequent analysis. Subsequently, feature selection was performed based on expert knowledge. Ten parameters were selected as feature parameters: cumulative scan count, initial tube current, maximum tube current, nominal tube current, final dose, initial filament current, final filament current, nominal filament current, motor drive current, and actual tube voltage. This yields multiple datasets. , Indicates the first One tube, Indicates the first The first X-ray tube corresponds to the first A dataset, This represents 10 characteristic parameters.

[0064] Step 2: In the process of labeling CT tube operation data, a time-series-based reverse labeling strategy is adopted. This strategy first sorts each data set according to its time series to ensure the temporal sequence of the data. Then, the labels are generated by incrementing daily based on the timestamp from the last record in the dataset to the first record. The label value of the last record is set to 0. The last record represents the end of the CT tube's lifespan. Each record includes the values ​​of 10 feature parameters at that time point, resulting in the labels. ,in Indicates the first This method ensures the consistency and traceability of label values ​​for each dataset over time. Furthermore, to highlight the data degradation characteristics near the end of the X-ray tube's lifespan, all label values ​​greater than 15 can be uniformly adjusted to 15 to emphasize the data degradation features as the lifespan nears its end. This label generation method provides a standardized and ordered label system for subsequent data analysis and model training. Labels are then added to the data at each time point of the four datasets cleaned in step one.

[0065] Step 3: Data Standardization. To eliminate the problem of different units between different operating parameters, this embodiment uses max-min standardization to normalize the data cleaned in Step 1. The four standardized datasets are represented as follows: The calculation process is as follows:

[0066] ;

[0067] in Represents one of the ten characteristic parameters. Represents the first in the dataset Data entries. Indicates the first The minimum value of each feature parameter. Indicates the first The maximum value of each feature parameter. This represents standardized data. Each data point corresponds to a time point (i.e., a timestamp).

[0068] Step 4: Sliding window processing and data noise reduction. Please refer to [link / reference]. Figure 2 This is a schematic diagram illustrating sliding window processing in an exemplary embodiment of this application. The total data length of the four datasets is calculated. The total number of time points in the four datasets (When dividing time points by days, the total number of days across four datasets can be calculated.) Sliding window width Step length For example, the total data length of the four datasets. The total number of time points in the four datasets Sliding window width Step length The window data after applying sliding window processing to the data in each dataset is as follows:

[0069] ;

[0070] in, express Time of the first The values ​​of each feature parameter.

[0071] The form data exhibits significant fluctuations, failing to highlight any degradation trends. Therefore, noise reduction is achieved through data downsampling and smoothing operations. The sampling interval for data downsampling is... The downsampled window data is as follows:

[0072] ;

[0073] The data smoothing operation uses the exponential moving average (EMA), and its iterative formula is as follows:

[0074] ;

[0075] in, To downsample the first window The first feature parameter Smoothed data, To downsample the first window The first feature parameter Data points, smoothing factor The calculation formula is as follows:

[0076] ;

[0077] Considering the long-term and localized nature of data degradation, The smooth window length is described in the embodiments of this application. The smoothed dataset is represented as follows: .

[0078] It should be noted that the exponential moving average (EMA) smoothing technique is used to reduce noise in the data. By weighting the most recent points, it makes the data more influential on future data points, while gradually reducing the influence on historical data points. When the data changes drastically, the EMA can respond quickly and adjust its value to follow the data trend, unlike the traditional moving average which lags behind the data.

[0079] Step 5: Dataset Splitting. The four datasets obtained from the smoothing and noise reduction in Step 4 are divided into training and test sets. Three X-ray tube data points are selected as the training set, i.e., the training sample dataset. The data from another X-ray tube was used as the test set. .

[0080] It should be noted that the steps in the embodiments of this application can be executed by devices such as servers, server clusters, and terminals.

[0081] Step S120: Input the training samples into the multi-scale feature extraction network of the initial prediction model, and extract multiple time features from the training samples through the multi-scale feature extraction network.

[0082] In one embodiment of this application, training samples can be input into the multi-scale feature extraction network of the initial prediction model to obtain multiple temporal features. Three temporal features can be extracted.

[0083] In one embodiment, the multi-scale feature extraction network includes multiple temporal convolutional networks, each with different convolutional kernels. The multi-scale feature extraction network extracts temporal features from the training samples to obtain multiple temporal features, including:

[0084] Multiple temporal features are obtained by extracting temporal features from the training samples through various temporal convolutional networks.

[0085] Training samples are input into a multi-scale, multi-dimensional feature extraction network to obtain temporal features at multiple different scales. The multi-scale feature extraction network includes multiple Temporal Convolutional Networks (TCNs). The core components of a TCN include causal convolution and dilated convolution. These two mechanisms allow the TCN to expand its receptive field without sacrificing sequence length. See also... Figure 3 This is a schematic diagram of the structure of a temporal convolutional network provided in an embodiment of this application. A TCN typically consists of multiple causal convolutional layers, which may have different dilation rates to capture patterns at different time scales. Residual connections can be added between each layer to promote gradient flow and avoid the vanishing gradient problem in deep network training.

[0086] Exemplary multi-scale temporal feature representation:

[0087] ;

[0088] ;

[0089] ;

[0090] in, TCN branches with convolutional kernels of different sizes are used to extract temporal features at different scales. Indicates the characteristics of the first time. Indicates the second time feature. This is the third temporal feature. The size of the convolution kernel in the TCN branch can be selected by the operator according to the actual situation. For example, the kernel size can be 2, 3, or 5. Figure 3 The temporal convolutional network shown has 3 convolutional kernels.

[0091] It should be noted that multi-scale feature extraction networks can effectively mine data degradation information in the time dimension. Compared with traditional time series prediction networks that only focus on single-dimensional or single-scale features, the method provided in this application embodiment can perform data mining in multiple dimensions and provide more comprehensive degradation information.

[0092] Step S130: Input the training samples into the two-dimensional convolutional network of the initial prediction model, and extract the spatial features of the training samples through the two-dimensional convolutional network to obtain spatial features.

[0093] In one embodiment of this application, training samples can be input into a two-dimensional convolutional network of an initial prediction model to obtain spatial features.

[0094] For example, spatial feature representation:

[0095] ;

[0096] in, Represents a two-dimensional convolutional network. Represents spatial characteristics.

[0097] It should be noted that two-dimensional convolutional networks are excellent at uncovering degradation information in the spatial dimension of data.

[0098] Step S140: Input multiple temporal and spatial features into the feature fusion layer of the initial prediction model to obtain fused features.

[0099] In one embodiment of this application, multiple temporal and spatial features can be input into the feature fusion layer of the initial prediction model to obtain fused features.

[0100] In one embodiment, multiple temporal and spatial features are input into the feature fusion layer of the initial prediction model to obtain fused features, including: splicing and fusing multiple temporal and spatial features through the feature fusion layer to obtain fused features.

[0101] For example, the extracted multi-scale temporal and spatial features are fused. In this embodiment, a splicing fusion strategy is adopted, as shown in the following expression:

[0102] ;

[0103] in, This indicates the fusion feature.

[0104] Step S150: Input the fused features into the dual-branch attention module of the initial prediction model, determine the attention weights and reconstructed features through the dual-branch attention module, and determine the output of the attention module based on the attention weights and reconstructed features.

[0105] In one embodiment of this application, the fused features can be input into the dual-branch attention module of the initial prediction model, the attention weights and reconstructed features can be determined based on the fused features, and the output of the attention module can be determined based on the attention weights and reconstructed features.

[0106] For example, the dual-branch attention model employs a multi-scale parallel structure to capture both local and long-term dependencies in the data. This design allows the model to learn richer feature representations at different scales. Simultaneously, channel-wise learning avoids information loss associated with channel dimensionality reduction. The fused features (feature maps) ) as input, Indicates the batch processing size. Indicates the number of channels. This represents the width of the feature map. Therefore, the output of the attention module can be expressed as:

[0107] ;

[0108] in This indicates a dual-branch attention module. This represents the output of the attention module.

[0109] Please see Figure 4 This is a schematic diagram illustrating the structure of a dual-branch attention module, as shown in an exemplary embodiment of this application. Please refer to... Figure 5 This is a flowchart illustrating the output of the attention module in an exemplary embodiment of this application.

[0110] In one embodiment, step S150 involves inputting the fused features into the dual-branch attention module of the initial prediction model, determining the attention weights and reconstructed features through the dual-branch attention module, and determining the output of the attention module based on the attention weights and reconstructed features. This process includes:

[0111] Step S151: Group the fusion features according to the preset number of groups and the number of channels of the fusion features to obtain multiple grouped features, and obtain recombined features based on the multiple grouped features.

[0112] For example, fusion features Divided into Group, then recombination characteristics are , The number of preset groups is represented.

[0113] Step S152: Perform channel information inter-intersection on the recombined features to obtain channel inter-intersection features.

[0114] In one embodiment, channel information interoperation is performed on the recombined features to obtain channel interoperation features, including: performing adaptive pooling processing on the recombined features to obtain a pooled feature map; and determining the channel interoperation features based on the pooled feature map and the recombined features.

[0115] For example, channel information exchange is achieved through 1×1 convolutions to enhance information flow between different channels. This allows the model to focus on the dependencies between different feature parameters. The recombined features are then processed through adaptive pooling, 1×1 convolutions, activation functions, and group normalization to obtain the channel exchange features.

[0116] ;

[0117] in, Represents the pooling feature map. This indicates an adaptive pooling operation;

[0118] ;

[0119] in, Indicates the characteristics of channel interaction. Indicates group normalization, This represents a 1x1 convolution operation. This represents the Sigmoid activation function.

[0120] Step S153: Extract local spatial features from the recombined features to obtain local spatial features.

[0121] In one embodiment of this application, convolution processing can be used to extract local spatial features from the recombined features to obtain local spatial features.

[0122] For example, local spatial feature extraction uses a 3x1 convolutional operation to extract local spatial features, focusing on local region information in the width direction. By using a larger convolutional kernel, the model can capture more complex patterns in the width direction. The local spatial feature extraction is as follows:

[0123] ;

[0124] in Representing local spatial features, This represents a 3*1 convolution operation.

[0125] Step S154: Determine the first weight based on the channel interaction characteristics, and determine the second weight based on the local spatial characteristics.

[0126] In one embodiment of this application, the first weight can be obtained by performing adaptive pooling on the transpose matrix of the channel intersecting features and then using the Softmax activation function. Similarly, the second weight can be obtained by performing adaptive pooling on the transpose matrix of the local spatial features and then using the Softmax activation function.

[0127] For example, the first weight can be represented as:

[0128] ;

[0129] in, Indicates the first weight. This indicates transpose.

[0130] For example, the second weight can be expressed as:

[0131] ;

[0132] in, This indicates the second weight.

[0133] Step S155: Determine the first cross-interaction result based on the channel cross-interaction characteristics and the second weight.

[0134] In one embodiment of this application, the product of the channel interoperability feature and the second weight can be determined as the first interoperability result.

[0135] For example, the first cross-intersection result can be represented as:

[0136] ;

[0137] in, This indicates the result of the first cross-intersection.

[0138] Step S156: Determine the second cross-interaction result based on local spatial features and the first weight.

[0139] In one embodiment of this application, the product of local spatial features and a first weight can be determined as the second cross-interaction result.

[0140] For example, the second cross-interaction result can be represented as:

[0141] ;

[0142] in, This indicates the result of the second cross-intersection.

[0143] Step S157: The sum of the first and second cross-interaction results is determined as the attention weight.

[0144] In one embodiment of this application, the sum of the first interaction result and the second interaction result can be determined as the attention weight.

[0145] For example, attention weights can be represented as:

[0146] ;

[0147] in, This represents the attention weight.

[0148] Step S158: Obtain the attention module output based on the attention weights and reorganization features.

[0149] In one embodiment of this application, the Sigmoid activation weights can be used to weight the inputs to obtain the output of the attention module.

[0150] For example, the output of the attention module can be represented as:

[0151] ;

[0152] in, This represents the output of the attention module.

[0153] It's worth noting that a dual-branch attention module is used to further extract key features from the fused features. This module employs a multi-scale parallel structure to capture both local and long-term dependencies in the data. This design allows the model to learn richer feature representations at different scales. Simultaneously, the module learns channels in batches, avoiding information loss caused by channel dimensionality reduction. Traditional attention mechanisms, on the other hand, do not explicitly distinguish between features at different scales and do not particularly emphasize the independent processing capabilities between channels.

[0154] Step S160: Input the output of the attention module into the fully connected network of the initial prediction model to obtain the lifetime prediction label. Then, based on the training label and the lifetime prediction label, iteratively train the initial prediction model and use the iteratively trained initial prediction model as the target prediction model for the lifetime prediction of the X-ray tube to be predicted.

[0155] In one embodiment of this application, the output of the attention module can be input into a fully connected network of the initial prediction model to obtain lifetime prediction labels. The difference between the lifetime prediction labels and the training labels can be measured using a loss function, and the model parameters of the initial prediction model can be updated based on the difference and a gradient update strategy. The target prediction model is obtained when the number of iterations meets a preset number. The operator can select an appropriate loss function, batch size, number of iterations, and gradient update strategy.

[0156] For example, a lifespan prediction label can be represented as:

[0157] ;

[0158] in, Indicates life expectancy prediction label, This indicates a fully connected network.

[0159] The loss function for iterative training can be a piecewise weighted loss function, which can include:

[0160] ;

[0161] in, Indicates the number of training labels. Indicates the training labels.

[0162] It should be noted that a piecewise weighted loss function is used, with different weights applied to the loss function at different stages. The penalty coefficient is smaller in the steady stage and larger in the degradation stage, making the model pay more attention to the data in the degradation stage.

[0163] The trained target prediction model can be tested using a test set. The mean-square error (MSE) can be used as the evaluation metric, and its calculation formula is as follows:

[0164] .

[0165] It should be noted that when using the target prediction model to perform X-ray tube lifetime assessment based on the attention mechanism, the prediction data of the X-ray tube to be predicted can be obtained, and the prediction data can be input into the target prediction model to obtain the remaining lifetime of the X-ray tube.

[0166] Please see Figure 6 This is a schematic diagram of the model architecture of the initial prediction model shown in one embodiment of this application.

[0167] Figure 7 This is a block diagram illustrating an attention-based tube life assessment device, as shown in an exemplary embodiment of this application. Figure 7 As shown, the exemplary attention-based tube life assessment device 700 includes:

[0168] The data acquisition module 710 is used to acquire the training sample dataset, which includes multiple training samples, each training sample corresponding to a training label. The training label represents the remaining lifespan of the X-ray tube corresponding to the training sample. The training samples include time series corresponding to multiple feature parameters of the X-ray tube.

[0169] The first feature extraction module 720 is used to input the training samples into the multi-scale feature extraction network of the initial prediction model, and extract multiple time features from the training samples through the multi-scale feature extraction network.

[0170] The second feature extraction module 730 is used to input training samples into the two-dimensional convolutional network of the initial prediction model, and extract spatial features from the training samples through the two-dimensional convolutional network.

[0171] The feature fusion module 740 is used to input multiple temporal and spatial features into the feature fusion layer of the initial prediction model to obtain fused features.

[0172] The output determination module 750 is used to input the fused features into the dual-branch group attention module of the initial prediction model, determine the attention weights and reconstructed features through the dual-branch group attention module, and determine the output of the attention module based on the attention weights and reconstructed features.

[0173] The model training module 760 is used to input the output of the attention module into the fully connected network of the initial prediction model to obtain the lifetime prediction label. Based on the training label and the lifetime prediction label, the initial prediction model is iteratively trained and used as the target prediction model for lifetime prediction of the X-ray tube to be predicted.

[0174] It should be noted that the X-ray tube life assessment device based on the attention mechanism provided in the above embodiments and the X-ray tube life assessment method based on the attention mechanism provided in the above embodiments belong to the same concept. The specific way in which each module and unit performs its operation has been described in detail in the method embodiments, and will not be repeated here. In practical applications, the X-ray tube life assessment device based on the attention mechanism provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the system can be divided into different functional modules to complete all or part of the functions described above, and this is not a limitation here.

[0175] Embodiments of this application also provide an electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by one or more processors, cause the electronic device to implement the X-ray tube life assessment method based on the attention mechanism provided in the above embodiments.

[0176] Another aspect of this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a computer's processor, causes the computer to perform the X-ray tube lifetime evaluation method based on the attention mechanism provided in the various embodiments above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0177] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the attention-based X-ray tube lifetime assessment method provided in the various embodiments described above.

[0178] In the embodiments of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. The terms "comprising" and "including" as used throughout the specification and claims are open-ended terms and should therefore be interpreted as "comprising but not limited to".

[0179] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for evaluating the lifespan of an X-ray tube based on an attention mechanism, characterized in that, include: Obtain a training sample dataset, which includes multiple training samples, each training sample corresponding to a training label, the training label representing the remaining lifespan of the X-ray tube corresponding to the training sample, and the training samples including time series corresponding to multiple feature parameters of the X-ray tube; The training samples are input into the multi-scale feature extraction network of the initial prediction model, and the multi-scale feature extraction network extracts the temporal features of the training samples to obtain multiple temporal features. The training samples are input into the two-dimensional convolutional network of the initial prediction model, and the spatial features of the training samples are extracted by the two-dimensional convolutional network to obtain spatial features. The multiple temporal and spatial features are input into the feature fusion layer of the initial prediction model to obtain fused features; The fused features are input into the dual-branch attention module of the initial prediction model. The attention weights and recombination features are determined by the dual-branch attention module. The output of the attention module is determined based on the attention weights and recombination features. The output of the attention module is input into the fully connected network of the initial prediction model to obtain the lifetime prediction label. Based on the training label and the lifetime prediction label, the initial prediction model is iteratively trained, and the iteratively trained initial prediction model is used as the target prediction model for lifetime prediction of the X-ray tube to be predicted.

2. The X-ray tube life assessment method based on attention mechanism according to claim 1, characterized in that, The multi-scale feature extraction network includes multiple temporal convolutional networks, each with different convolutional kernels. The multi-scale feature extraction network extracts temporal features from the training samples to obtain multiple temporal features, including: The training samples are processed by various temporal convolutional networks to extract temporal features, resulting in multiple temporal features.

3. The X-ray tube life assessment method based on attention mechanism according to claim 1, characterized in that, The multiple temporal and spatial features are input into the feature fusion layer of the initial prediction model to obtain fused features, including: The feature fusion layer stitches together and fuses multiple temporal and spatial features to obtain the fused feature.

4. The X-ray tube life assessment method based on attention mechanism according to claim 1, characterized in that, The fused features are input into the dual-branch attention module of the initial prediction model. Attention weights and reconstructed features are determined through the dual-branch attention module. The output of the attention module is determined based on the attention weights and reconstructed features, including: The fusion features are grouped according to a preset number of groups and the number of channels of the fusion features to obtain multiple grouped features, and a recombined feature is obtained based on the multiple grouped features; The recombined features are subjected to channel information interoperation to obtain channel interoperation features; Local spatial features are extracted from the recombined features to obtain local spatial features; A first weight is determined based on the channel interaction characteristics, and a second weight is determined based on the local spatial characteristics. The first interaction result is determined based on the channel interaction characteristics and the second weight; The second interaction result is determined based on the local spatial features and the first weight; The sum of the first interaction result and the second interaction result is determined as the attention weight; The attention module output is obtained based on the attention weights and the reorganization features.

5. The X-ray tube life assessment method based on attention mechanism according to claim 4, characterized in that, The recombined features are subjected to channel information interoperation to obtain channel interoperation features, including: The recombined features are subjected to adaptive pooling to obtain a pooled feature map; The channel interaction features are determined based on the pooling feature map and the recombination features.

6. The X-ray tube life assessment method based on attention mechanism according to claim 4, characterized in that, Local spatial features are extracted from the recombined features to obtain local spatial features, including: The recombined features are processed by convolution to extract local spatial features, thus obtaining the local spatial features.

7. The X-ray tube life assessment method based on an attention mechanism according to any one of claims 1-6, characterized in that, The characteristic parameters of the training samples include the cumulative number of scans, initial tube current, maximum tube current, nominal tube current, final dose, initial filament current, final filament current, nominal filament current, motor drive current, and actual tube voltage.

8. A device for evaluating the lifespan of an X-ray tube based on an attention mechanism, characterized in that, include: The data acquisition module is used to acquire a training sample dataset, which includes multiple training samples, each training sample corresponding to a training label, the training label representing the remaining lifespan of the X-ray tube corresponding to the training sample, and the training samples including time series corresponding to multiple feature parameters of the X-ray tube. The first feature extraction module is used to input the training samples into the multi-scale feature extraction network of the initial prediction model, and extract multiple time features from the training samples through the multi-scale feature extraction network. The second feature extraction module is used to input the training samples into the two-dimensional convolutional network of the initial prediction model, and extract the spatial features of the training samples through the two-dimensional convolutional network to obtain spatial features. The feature fusion module is used to input multiple temporal features and spatial features into the feature fusion layer of the initial prediction model to obtain fused features; The output determination module is used to input the fused features into the dual-branch group attention module of the initial prediction model, determine the attention weights and recombination features through the dual-branch group attention module, and determine the output of the attention module based on the attention weights and recombination features. The model training module is used to input the output of the attention module into the fully connected network of the initial prediction model to obtain the lifetime prediction label, and to iteratively train the initial prediction model based on the training label and the lifetime prediction label, and to use the iteratively trained initial prediction model as the target prediction model for predicting the lifetime of the X-ray tube to be predicted.

9. An electronic device, characterized in that, The electronic device includes: One or more processors; Memory used to store the executable program code of the processor; The processor is configured to execute the program code to implement the X-ray tube life assessment method based on the attention mechanism as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by the computer's processor, causes the computer to perform the X-ray tube life assessment method based on the attention mechanism as described in any one of claims 1 to 7.

Images