Anomaly detection method, image reconstruction method, device and equipment of front-end detector
Patent Information
- Application Number
- CN202510533861.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-25
- Publication Date
- 2026-10-27
Smart Images
Figure CN122989818A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of detector technology, and in particular to an anomaly detection method, image reconstruction method, apparatus and device for a front-end detector. Background Technology
[0002] The front-end detector is an important component of the positron emission tomography (PET) system, and its performance directly determines the overall system efficiency.
[0003] During the manufacturing and operation of the front-end detector, numerous complex factors (such as fluctuations in component operating status, non-uniformity of radiation source spatial distribution, and deviations in operating procedures) can affect the generated 2D map, energy spectrum, and time-to-digital converter (TDC) signals. These anomalies may include abnormal brightness distribution, baseline drift, or inconsistent responses. These anomalies significantly impact the spatial resolution, sensitivity, and quantitative accuracy of the PET system, leading to quality issues such as artifacts and decreased contrast in the reconstructed images.
[0004] Therefore, there is an urgent need for a method to detect anomalies in front-end detectors. Summary of the Invention
[0005] Therefore, it is necessary to provide an anomaly detection method, image reconstruction method, apparatus, and device for a front-end detector to address the aforementioned technical problems, enabling anomaly detection of the front-end detector.
[0006] Firstly, this application provides an anomaly detection method for a front-end detector, including:
[0007] Acquire imaging test data of the test object collected by the front-end detector;
[0008] The imaging test data is input into the preset anomaly detection model, and the anomaly detection model is used to analyze the imaging test data to determine the anomaly detection result of the front-end detector.
[0009] The anomaly detection model is trained by fusing convolutional feature information and semantic information of sample imaging data extracted from multi-layer cascaded convolutional layers.
[0010] In one embodiment, the anomaly detection model includes a multi-layered cascaded encoder;
[0011] Anomaly detection models are used to analyze imaging test data to determine the anomaly detection results of the front-end detector, including:
[0012] The encoder extracts semantic information from the imaging test data and convolutional feature information from the multi-layered cascaded convolutional layers.
[0013] Based on the convolutional feature information and semantic information on the multi-layered cascaded convolutional layers, it is determined whether there are anomalies in the front-end detector.
[0014] In one embodiment, the anomaly detection model further includes a feature fusionist and a classifier;
[0015] Based on convolutional feature information and semantic information from multiple cascaded convolutional layers, it is determined whether there are anomalies in the front-end detector, including:
[0016] The feature fusion method fuses the convolutional feature information and semantic information of multiple cascaded convolutional layers to obtain fused information.
[0017] The fused information is classified using a classifier to determine whether there are any anomalies in the front-end detector.
[0018] In one embodiment, the feature fusion processor includes a feature extraction network;
[0019] Extract semantic information from imaging test data, including:
[0020] The convolutional feature information on multiple cascaded convolutional layers is fused using a feature fusion processor to obtain fused convolutional feature information.
[0021] Semantic information is extracted from fused convolutional feature information using a feature extraction network.
[0022] In one embodiment, the training process of the anomaly detection model includes:
[0023] Acquire training sample data; the training sample data includes imaging data of samples with normal labels and imaging data of samples with abnormal labels;
[0024] The initial anomaly detection model is trained using training sample data to obtain the label prediction results for the training sample data.
[0025] Based on the differences between the labels and label prediction results of the training sample data, the parameters in the initial anomaly detection model are adjusted until the label prediction results output by the initial anomaly detection model meet the preset conditions. The adjusted initial anomaly detection model is then determined as the anomaly detection model.
[0026] In one embodiment, an initial anomaly detection model is trained using training sample data to obtain label prediction results for the training sample data, including:
[0027] The initial anomaly detection model extracts the sample semantic information of the sample imaging data and the sample convolution feature information on the multi-layered cascaded convolutional layers.
[0028] By fusing sample semantic information and sample convolutional feature information, sample fusion information is obtained;
[0029] Adding noise to the fusion information of some samples resulted in label predictions indicating anomalies; classifying the fusion information of other samples resulted in label predictions indicating normality.
[0030] Secondly, this application also provides an image reconstruction method, the method comprising:
[0031] If the anomaly detection result of the front-end detector is abnormal, the parameters of the image reconstruction algorithm are adjusted; the anomaly detection result is obtained by analyzing the imaging test data of the test object collected by the front-end detector using a preset anomaly detection model.
[0032] The acquired imaging data is reconstructed using the adjusted parameters to obtain the reconstructed image.
[0033] Thirdly, this application also provides a detector detection system, which includes a data receiver and a processor;
[0034] The data receiver is used to receive the imaging test data of the test object collected by the front-end detector and forward the imaging test data to the processor;
[0035] The processor is used to input imaging test data into a preset anomaly detection model, and to analyze the imaging test data using the anomaly detection model to determine the anomaly detection result of the front-end detector. The anomaly detection model is obtained by fusing and training the convolutional feature information and semantic information of the sample imaging data extracted from multi-layer cascaded convolutional layers.
[0036] Fourthly, this application also provides an anomaly detection device for a front-end detector, the device comprising:
[0037] The acquisition module is used to acquire imaging test data of the test object collected by the front-end detector;
[0038] The detection module is used to input imaging test data into a preset anomaly detection model, and to analyze the imaging test data using the anomaly detection model to determine the anomaly detection results of the front-end detector.
[0039] The anomaly detection model is trained by fusing convolutional feature information and semantic information of sample imaging data extracted from multi-layer cascaded convolutional layers.
[0040] Fifthly, this application also provides an image reconstruction apparatus, which includes:
[0041] The adjustment module is used to adjust the parameters of the image reconstruction algorithm when the anomaly detection result of the front-end detector is abnormal; the anomaly detection result is obtained by analyzing the imaging test data of the test object collected by the front-end detector using a preset anomaly detection model.
[0042] The reconstruction module is used to reconstruct images from the acquired imaging data using adjusted parameters, thereby obtaining reconstructed images.
[0043] Sixthly, this application also provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement any one of the embodiments of the anomaly detection method for the front-end detector in the first aspect and / or the image reconstruction method in the second aspect.
[0044] In a seventh aspect, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any one of the embodiments of the anomaly detection method for the front-end detector in the first aspect and / or the image reconstruction method in the second aspect.
[0045] Eighthly, this application also provides a computer program product, including a computer program that, when executed by a processor, implements any one of the embodiments of the anomaly detection method for the front-end detector in the first aspect and / or the image reconstruction method in the second aspect.
[0046] The aforementioned anomaly detection method, image reconstruction method, apparatus, and equipment for front-end detectors acquire imaging test data of the test object collected by the front-end detector; input the imaging test data into a preset anomaly detection model, and use the anomaly detection model to analyze the imaging test data to determine the anomaly detection result of the front-end detector; wherein, the anomaly detection model is obtained by fusing and training convolutional feature information and semantic information of the sample imaging data extracted by multi-layer cascaded convolutional layers. In the anomaly detection model training process, this method extracts convolutional feature information of the sample imaging data through multi-layer cascaded convolutional layers and combines it with semantic information to obtain the anomaly detection model, which can more comprehensively and deeply mine the potential features and patterns in the imaging test data. Compared with traditional single feature analysis methods, this fusion method can extract deeper features, and the feature fusion process ensures more comprehensive features, thereby greatly improving the detection accuracy of the front-end detector's anomaly state and reducing false positives and false negatives. Attached Figure Description
[0047] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0048] Figure 1 This is an application environment diagram of an anomaly detection method for a front-end detector in one embodiment;
[0049] Figure 2 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0050] Figure 3 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0051] Figure 4 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0052] Figure 5 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0053] Figure 6 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0054] Figure 7 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0055] Figure 8 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0056] Figure 9 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0057] Figure 10 This is a schematic diagram of a reconstruction method under abnormal conditions in one embodiment;
[0058] Figure 11 This is a flowchart illustrating an anomaly detection method for a front-end detector in one embodiment;
[0059] Figure 12 This is a flowchart illustrating an image reconstruction method in one embodiment;
[0060] Figure 13 This is a structural block diagram of an anomaly detection device for a front-end detector in one embodiment. Detailed Implementation
[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0062] The anomaly detection method for the front-end detector provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown includes a computer device 101 and a front-end detector 102, which communicate with each other. The front-end detector 102 is used to collect imaging test data of the test object and transmit the imaging test data to the computer device 101. The computer device 101 is used to input the imaging test data into a preset anomaly detection model, analyze the imaging test data using the anomaly detection model, and determine the anomaly detection result of the front-end detector.
[0063] In one exemplary embodiment, such as Figure 2 As shown, an anomaly detection method for a front-end detector is provided, which is then applied to... Figure 1 The following steps, 201 to 202, are used as an example of computer equipment.
[0064] S201, acquire imaging test data of the test object collected by the front-end detector.
[0065] The imaging test data can be the raw data collected by the front-end detector, or it can be a 2D map, energy spectrum, and TDC signal generated based on the raw data.
[0066] In this embodiment, if the front-end detector has not yet imaged the test object, the computer device can send a data acquisition command to the front-end detector. Upon receiving the data acquisition command, the front-end detector images the test object to collect its imaging test data and forwards the collected imaging test data to the computer device. If the front-end detector has already imaged the test object and stored the collected imaging test data in the corresponding database, then the computer device can search the database for test data matching the identification information of the test object and use the matching test data as the imaging test data of the test object. This embodiment does not limit the method of obtaining the imaging test data of the test object collected by the front-end detector.
[0067] S202, the imaging test data is input into the preset anomaly detection model, and the anomaly detection model is used to analyze the imaging test data to determine the anomaly detection result of the front-end detector; wherein, the anomaly detection model is obtained by fusing the convolutional feature information and semantic information of the sample imaging data extracted by multi-layer cascaded convolutional layers.
[0068] Before using anomaly detection models for anomaly analysis, the models need to be trained. The training process is as follows: First, a large amount of sample imaging data, including normal and various anomalous states, is collected. This sample imaging data comes from a wide range of sources, covering data collected by front-end detectors under different usage durations and operating environments, ensuring the model has broad adaptability and generalization ability. Next, this imaging test data is input into the initial anomaly detection model, which uses multi-layered cascaded convolutional layers for feature extraction. Each convolutional layer has a specific kernel size and stride. Through continuous convolution operations, convolutional feature information from low to high levels, such as edges and textures, is gradually extracted from the sample imaging data. Simultaneously, semantic segmentation and other techniques are used to deeply mine the semantic information of the sample imaging data, clarifying the physical meaning represented by different regions in the data.
[0069] The extracted convolutional features are then fused with semantic information. This fusion process combines feature concatenation with an attention mechanism, enabling the initial anomaly detection model to more effectively focus on key information. Subsequently, the model is trained using the fused feature data. During training, the cross-entropy loss function is used as the optimization objective, and an optimizer iteratively updates the parameters of the initial anomaly detection model. After multiple rounds of training and validation, the final trained anomaly detection model is obtained.
[0070] Once the trained anomaly detection model is acquired, the computer device can input the collected imaging test data into the preset anomaly detection model. The anomaly detection model will then perform convolutional feature extraction and semantic information analysis on the data according to the training process, and compare the fused result with the normal and abnormal patterns learned during training. Through complex calculations and judgments, it determines whether the front-end detector is currently in an abnormal state and outputs the anomaly detection result.
[0071] It should be noted that, assuming the imaging test data is raw data collected by the front-end detector, or energy spectrum or TDC signal generated based on the raw data, the anomaly detection result includes whether the front-end detector is in an abnormal state or not. If the imaging test data is a 2D map generated from the raw data, the anomaly detection result is the label of the abnormal region on the 2D map.
[0072] In the aforementioned anomaly detection method for the front-end detector, imaging test data of the test object collected by the front-end detector is acquired. This imaging test data is then input into a pre-set anomaly detection model, which analyzes the data to determine the anomaly detection result of the front-end detector. The anomaly detection model is obtained by fusing convolutional feature information extracted from sample imaging data through multi-layered cascaded convolutional layers with semantic information from the sample imaging data. During the training process, this method extracts convolutional feature information from sample imaging data through multi-layered cascaded convolutional layers and combines it with semantic information to obtain the resulting anomaly detection model. This allows for a more comprehensive and in-depth exploration of potential features and patterns in the imaging test data. This fusion method extracts deeper features, and the feature fusion process ensures more comprehensive features, thereby significantly improving the accuracy of anomaly detection by the front-end detector and reducing false positives and false negatives.
[0073] Assuming the anomaly detection model includes multi-layered cascaded encoders, the following example illustrates the specific process of analyzing imaging test data using the anomaly detection model to determine the anomaly detection results of the front-end detector. Figure 3 As shown, the specific content includes:
[0074] S301 extracts semantic information from imaging test data and convolutional feature information from multi-layered cascaded convolutional layers through an encoder.
[0075] In the anomaly detection model, the encoder can be configured with multiple cascaded convolutional layers, which are composed of multiple convolutional layers connected sequentially.
[0076] In this embodiment, when the encoder extracts convolutional feature information from multiple cascaded convolutional layers, convolutional features of different scales and levels of abstraction are obtained after processing through multiple convolutional layers. For example, the cascaded convolutional layers include three convolutional layers, and the features extracted by the second convolutional layer can be used as one convolutional feature information. Furthermore, based on the features extracted by the second convolutional layer, a third convolutional layer is used to continue convolution, and the features extracted by the third convolutional layer are used as one convolutional feature information. In addition, the encoder can also extract semantic information from the imaging test data based on the multiple extracted convolutional feature information.
[0077] S302 determines whether there are anomalies in the front-end detector based on the convolutional feature information and semantic information on the multi-layered cascaded convolutional layers.
[0078] In this embodiment, the computer device can deeply fuse convolutional feature information and semantic information from multiple cascaded convolutional layers to obtain fused information. Specifically, feature concatenation can be used to combine the two types of information along a specific dimension to form fused information containing rich details and semantic meaning.
[0079] Next, based on the feature classification information learned by the anomaly detection model, the fused information is analyzed to determine the anomaly probability of the fused information. If the anomaly probability is greater than a preset probability threshold, it is determined that the front-end detector is abnormal; if the anomaly probability is less than or equal to the preset probability threshold, it is determined that the front-end detector is not abnormal.
[0080] In the aforementioned anomaly detection method for the front-end detector, semantic information and convolutional feature information from multi-layered cascaded convolutional layers are extracted from the imaging test data by an encoder. Based on the convolutional feature information and semantic information from the multi-layered cascaded convolutional layers, the presence of anomalies in the front-end detector is determined. This method extracts semantic information from the imaging test data through the encoder in the anomaly detection model, enabling a deeper understanding of the actual meaning represented by the data. The convolutional feature information extracted from the multi-layered cascaded convolutional layers enriches the data features at a low-level detail level. Combining these two approaches allows for comprehensive data analysis from different dimensions, significantly improving the accuracy of determining whether anomalies exist in the front-end detector and effectively avoiding misjudgments and missed judgments due to incomplete feature analysis.
[0081] Based on the above embodiments, assuming the anomaly detection model further includes a feature fusionist and a classifier, then in one embodiment, such as... Figure 4 As shown, the specific content of determining whether there is anomaly in the front-end detector based on the convolutional feature information and semantic information on the multi-layered cascaded convolutional layers includes:
[0082] S401 uses a feature fusion processor to fuse convolutional feature information and semantic information from multiple cascaded convolutional layers to obtain fused information.
[0083] In this embodiment, after the encoder completes the extraction of convolutional features and semantic information from the imaging test data across multiple cascaded convolutional layers, the computer device can input the extracted convolutional feature information and semantic information into a feature fusion unit. The feature fusion unit can concatenate the convolutional feature information and semantic information according to specific dimensions to form a fused information containing rich details and semantic meaning.
[0084] S402 uses a classifier to classify the fused information and determine whether there are any anomalies in the front-end detector.
[0085] In this embodiment, after obtaining the fused information, the computer device can input the fused information into the classifier of the anomaly detection model. The classifier performs feature transformation and abstraction on the fused information through a series of linear transformations and nonlinear activation functions, mapping the fused information to different feature spaces for analysis, and determining the probability that the imaging test data is anomalous based on the analysis results. Next, the probability of the anomalous data is compared with a preset probability threshold. If the probability of the anomalous data is greater than the preset probability threshold, it indicates that the probability of the imaging test data being anomalous is relatively high. In this case, it can be determined that the imaging test data is anomalous, and the front-end detector must also have an anomaly. If the probability of the anomalous data is less than or equal to the preset probability threshold, it indicates that the probability of the imaging test data being anomalous is relatively low. In this case, it can be determined that the imaging test data is not anomalous, and the front-end detector is in normal working condition. For example, assuming the preset probability threshold is 30%, when the probability of the imaging test data being anomalous is 10% or 20%, the front-end detector is in normal working condition; when the probability of the imaging test data being anomalous is 80%, the front-end detector has an anomaly.
[0086] In the aforementioned anomaly detection method for front-end detectors, a feature fusion unit fuses convolutional feature information and semantic information from multiple cascaded convolutional layers to obtain fused information. A classifier then classifies this fused information to determine if the front-end detector exhibits anomalies. This method utilizes a feature fusion unit to integrate convolutional feature information and semantic information from multiple cascaded convolutional layers, enabling comprehensive analysis of the front-end detector's imaging test data from multiple dimensions. Convolutional features reflect the underlying details of the image, while semantic information provides the data with practical meaning. Combining these two approaches overcomes the limitations of single-feature analysis, providing the classifier with more sufficient evidence to determine whether the front-end detector is abnormal, effectively reducing false positives and false negatives, and significantly improving anomaly detection accuracy.
[0087] Furthermore, the feature fusion unit includes a feature extraction network; therefore, in one embodiment, such as Figure 5 As shown, the specific content of the semantic information extracted from the imaging test data includes:
[0088] S501 uses a feature fusion unit to fuse convolutional feature information from multiple cascaded convolutional layers to obtain fused convolutional feature information.
[0089] In this embodiment, the convolutional feature information on multiple cascaded convolutional layers differs in both spatial and channel dimensions. Therefore, before fusion, the computer device can use upsampling or subsampling to unify the convolutional feature information across these different spatial and channel dimensions, thus aligning the convolutional feature information across the multiple cascaded convolutional layers. Next, the computer device can multiply each convolutional feature information with its corresponding weight and fuse the results to obtain the fused convolutional feature information.
[0090] S502 extracts semantic information from fused convolutional feature information through a feature extraction network.
[0091] In this embodiment of the application, after obtaining the fused convolutional feature information, the computer device can input the fused convolutional feature information into the feature extraction network, and analyze the fused convolutional feature through the feature extraction network to obtain semantic information that can indicate whether the front-end detector is in an abnormal working state.
[0092] In the aforementioned anomaly detection method for the front-end detector, a feature fusion processor fuses convolutional feature information from multiple cascaded convolutional layers to obtain fused convolutional feature information; a feature extraction network then extracts semantic information from the fused convolutional feature information. This method, by fusing convolutional feature information from multiple cascaded convolutional layers through a feature fusion processor, can integrate convolutional features at different depths, minimizing information omissions and providing a more comprehensive data foundation for subsequent analysis.
[0093] In one specific embodiment of this application, such as Figure 6 As shown, the anomaly detection method of the above-mentioned front-end detector includes the following steps:
[0094] S601, acquire the imaging test data of the test object collected by the front-end detector;
[0095] S602 extracts convolutional feature information of imaging test data on multi-layered cascaded convolutional layers through an encoder;
[0096] S603 uses a feature fusion processor to fuse convolutional feature information from multiple cascaded convolutional layers to obtain fused convolutional feature information.
[0097] S604 extracts semantic information from fused convolutional feature information through a feature extraction network;
[0098] S605 uses a feature fusion processor to fuse convolutional feature information and semantic information from multiple cascaded convolutional layers to obtain fused information.
[0099] S606 uses a classifier to classify the fused information and determine whether there are any anomalies in the front-end detector.
[0100] The above embodiments are specific illustrations of the practical application process of the anomaly detection model. The following embodiment will introduce the training process of the anomaly detection model, such as... Figure 7 As shown, the training process includes:
[0101] S701, Obtain training sample data; the training sample data includes imaging data of samples with normal labels and imaging data of samples with abnormal labels.
[0102] The training sample data and the imaging test data are both imaging data of the same modality. The imaging data of normal label samples are detected by the front-end detector that has not experienced any anomalies, while the imaging data of abnormal label samples are detected by the front-end detector that has experienced anomalies.
[0103] In this embodiment, if the sample imaging data is tagged and stored in a sample database, the computer device can search for tagged sample imaging data from the sample database. The tags of the searched sample imaging data include normal tags and abnormal tags, and the searched tagged sample imaging data is used as training sample data. If the sample imaging data is untagged, the computer device can output tagging information to the tagging personnel to tag the sample imaging data and obtain the tagged sample imaging data to obtain training sample data. This embodiment does not limit the method of obtaining training sample data.
[0104] S702 trains the initial anomaly detection model using training sample data to obtain the label prediction results of the training sample data.
[0105] In this embodiment, a computer device can input training sample data into an initial anomaly detection model, and use the initial anomaly detection model to extract sample convolutional feature information and sample semantic information from the training sample data. Then, the sample convolutional feature information and sample semantic information are fused, and based on the fused information, it is determined whether the training sample data is abnormal, thereby determining whether the front-end detector is abnormal, and using the abnormality of the front-end detector as the label prediction result of the training sample data.
[0106] S703, based on the difference between the labels of the training sample data and the label prediction results, adjust the parameters in the initial anomaly detection model until the label prediction results output by the initial anomaly detection model meet the preset conditions, and then determine the adjusted initial anomaly detection model as the anomaly detection model.
[0107] In this embodiment, after obtaining the label prediction result, the computer device can calculate the difference between the label of the training sample and the label prediction result. This difference represents the detection error of the initial anomaly detection model. Based on this difference, the parameters in the initial anomaly detection model are adjusted, and the labels of the training sample data are predicted again. Then, based on the difference between the prediction result and the true label, the parameters of the initial anomaly detection model are further adjusted. After multiple iterative adjustments, until the label prediction result output by the initial anomaly detection model meets the preset conditions, the training process is completed, and the adjusted initial anomaly detection model is determined as the trained anomaly detection model.
[0108] In the aforementioned anomaly detection method for the front-end detector, training sample data is acquired. This training sample data includes imaging data of samples with normal labels and imaging data of samples with abnormal labels. An initial anomaly detection model is trained using this training sample data to obtain the label prediction results for the training sample data. Based on the difference between the labels of the training sample data and the label prediction results, the parameters in the initial anomaly detection model are adjusted until the label prediction results output by the initial anomaly detection model meet preset conditions. The adjusted initial anomaly detection model is then determined as the anomaly detection model. This method uses imaging data containing both normal and abnormal labels to train the initial anomaly detection model, enabling the initial anomaly detection model to fully learn the data features under different states. Furthermore, during the training process, the parameters of the initial anomaly detection model are continuously adjusted based on the difference between the labels and the prediction results, making the prediction results of the initial anomaly detection model increasingly closer to the true labels. This greatly improves the accuracy of the anomaly detection model training process and enables the acquisition of a more accurate anomaly detection model.
[0109] In one embodiment, such as Figure 8 As shown, the specific process of training the initial anomaly detection model using training sample data to obtain the label prediction results of the training sample data is explained. This specific process includes:
[0110] S801 extracts sample semantic information from sample imaging data and sample convolution feature information from multi-layered cascaded convolutional layers through an initial anomaly detection model.
[0111] In this embodiment, the computer device can extract sample convolutional feature information from multiple cascaded convolutional layers using the encoder in the initial anomaly detection model. After processing through multiple convolutional layers, convolutional features of different scales and levels of abstraction are obtained, and these convolutional features are used as sample convolutional feature information from the multiple cascaded convolutional layers. Furthermore, the encoder can also extract sample semantic information from the imaging test data based on the extracted sample convolutional feature information.
[0112] S802, fuses sample semantic information and sample convolutional feature information to obtain sample fusion information.
[0113] In this embodiment, a computer device can deeply fuse sample convolutional feature information and sample semantic information from multiple cascaded convolutional layers to obtain sample fusion information. Specifically, feature concatenation can be used to combine the two types of information along a specific dimension to form sample fusion information containing rich details and semantic meaning.
[0114] S803: Noise is added to the fusion information of some samples, and the label prediction result is that there is an anomaly; the fusion information of other samples is classified, and the label prediction result is that it is normal.
[0115] In this embodiment, the computer device can split the fused sample information into two parts. For one part, noise is added to the fused sample information to generate a negative sample, i.e., the sample imaging data labeled as having anomalies. The other part of the fused sample information remains unchanged, and after classification, the label prediction result is normal.
[0116] In the aforementioned anomaly detection method for the front-end detector, the initial anomaly detection model extracts sample semantic information and sample convolutional feature information from the sample imaging data. It then fuses the sample semantic information and sample convolutional feature information to obtain sample fusion information. Noise is added to some of the sample fusion information, and the label prediction result is "anomaly present." Other sample fusion information is classified, and the label prediction result is "normal." This method introduces noise into some of the sample fusion information after feature fusion to generate negative samples. The introduction of noise increases the diversity of feature samples, enabling the anomaly detection model to learn more anomalies during training, thereby improving the initial anomaly detection model's ability to identify anomalies.
[0117] Figure 9 This diagram illustrates an anomaly detection method for a front-end detector. The execution flow of this method includes: initiating the anomaly detection process, acquiring imaging test data, and performing data preprocessing on the imaging test data (e.g., upsampling, denoising, normalization, etc.). Then, based on the data preprocessing, automatic anomaly detection is performed (i.e., anomaly analysis using an anomaly detection model) to determine the anomaly probability of the imaging test data. Finally, the data is post-processed to determine whether an anomaly has occurred in the front-end detector. If an anomaly has occurred, the anomaly is recorded.
[0118] Figure 10This diagram illustrates the reconstruction method under abnormal conditions. Optimized reconstruction consists of two parts: an anomaly detection process and an optimized reconstruction process. The anomaly detection process can be run every morning before use, and the results are stored in the computer after detection. After subsequent patient scans, the system's abnormal status is checked before reconstruction. If a system anomaly is detected, the reconstruction algorithm is adjusted; otherwise, reconstruction proceeds according to the normal process. Specifically, if an anomaly is determined in the front-end detector, the parameters in the front-end detector scanning protocol are adjusted, and scan data is re-acquired based on the adjustments. The re-acquired scan data is then assessed for anomalies. If anomalies are found, the reconstruction algorithm is adjusted to optimize the quality of the reconstructed image. If no anomalies are found, the scan data is considered normal, and the image reconstruction process can be completed using the normal reconstruction algorithm.
[0119] Figure 11 This diagram illustrates the training process of an anomaly detection model. After obtaining training sample data, this data is input into the model. An encoder extracts convolutional feature information from different cascaded convolutional layers, and this information is then fused to obtain fused feature information. A feature fusion processor extracts semantic information from this fused feature information. This semantic information is further fused with the fused feature information to obtain fused information. Noise is then added to a portion of the fused information to generate negative samples, while another portion generates positive samples. Simultaneously, a classifier classifies the fused information, predicts the labels for the training sample data, and compares the predicted labels with the true labels to determine the differences and further optimize the anomaly detection model.
[0120] In this model, the network structure of the encoder can be Encoder, the network of the feature fusion unit can be Fusion, the network structure of the classifier can be Recon, and the network for comparing the predicted label with the real label can be Cls. In other words, the anomaly detection model can be composed of four network structures: Encoder, Fusion, Recon, and Cls.
[0121] The Encoder primarily extracts data features and consists of the first three layers of a ResNet32 (pre-trained) network, concatenating the features from the second and third layers. Fusion fuses the output features of the Encoder network to extract high-level semantic information; it consists of the results of two Transform Encoders, which are then concatenated with the Encoder network's output features. The Recon structure mainly segments the output features from the two networks above to obtain an anomaly mask. It consists of a three-layer MLP structure, each consisting of a fully connected layer, a batch normalization layer, and a ReLU activation layer. It's important to note that the MLP structure and parameters are identical in both the training and testing parts. The Cls layer primarily outputs the results from the Recon structure and consists of a fully connected layer and a softmax classifier.
[0122] For the training process, a branch is added during training. Random noise is added to the Fusion output and input into the recon structure; this branch does not participate in prediction. Furthermore, the total loss function consists of classification loss (BCEloss), BCEloss, and Diceloss. That is, the total loss function is: total loss = Lc + Lo + Lg.
[0123] By applying the Transformer Encoder to a traditional convolutional neural network, TFEncoder utilizes the Transformer Encoder's encoding structure to fuse features from the convolutional feature maps. This feature fusion mechanism captures and integrates more complex contextual information, thereby enhancing the network's ability to understand high-level information. The Transformer's self-attention mechanism allows the network to better focus on important regions in the image, improving the globality and robustness of feature representation. Through feature fusion and the introduction of noise, the TFEncoder network not only better understands high-level information but also effectively identifies and processes anomalous data, significantly improving the network's overall performance and adaptability. This results in higher accuracy for anomaly detection models.
[0124] In one embodiment, such as Figure 12 As shown, an image reconstruction method is also provided, which can also be applied to... Figure 1 The computer device in the middle, the method includes:
[0125] S901, when the anomaly detection result of the front-end detector is abnormal, the parameters of the image reconstruction algorithm are adjusted; the anomaly detection result is obtained by analyzing the imaging test data of the test object collected by the front-end detector using a preset anomaly detection model.
[0126] The image reconstruction algorithm can include parameters such as filtering parameters, number of iterations, and weight coefficients.
[0127] In this embodiment, when the anomaly detection model determines that the front-end detector has an anomaly, the parameters of the image reconstruction algorithm need to be adjusted. Specifically, different anomaly types correspond to different target parameters, and the computer device can determine the target parameters corresponding to the anomaly type. Then, based on the difference between the current parameters and the target parameters, the image reconstruction parameters are adjusted according to this difference, and the adjusted image reconstruction parameters are the target parameters corresponding to the anomaly type.
[0128] S902 uses the adjusted parameters to reconstruct the image from the acquired imaging data, thus obtaining the reconstructed image.
[0129] In this embodiment of the application, after obtaining the adjusted image reconstruction parameters, the collected imaging data is reconstructed according to the adjusted parameters. After filtering analysis and multiple iterations, a high-quality reconstructed image is obtained.
[0130] In the aforementioned image reconstruction method, when the anomaly detection result of the front-end detector is abnormal, the parameters of the image reconstruction algorithm are adjusted. The anomaly detection result is obtained by analyzing the imaging test data of the test object collected by the front-end detector using a preset anomaly detection model. The adjusted parameters are then used to reconstruct the image from the collected imaging data, resulting in a reconstructed image. This method accurately identifies anomalies through the anomaly detection model and adjusts the image reconstruction algorithm parameters promptly after anomaly detection, effectively eliminating or mitigating the negative impact of front-end detector anomalies on image reconstruction and significantly improving the clarity, contrast, and detail integrity of the reconstructed image.
[0131] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0132] Based on the same inventive concept, this application also provides a detector detection system for implementing the above-described anomaly detection method for a front-end detector. The solution provided by this detector detection system is similar to the implementation described in the above method; therefore, the specific limitations in one or more detector detection system embodiments provided below can be found in the limitations of the anomaly detection method for the front-end detector described above, and will not be repeated here.
[0133] In one embodiment, a detector detection system is provided, the detector detection system including a data receiver and a processor;
[0134] The data receiver is used to receive the imaging test data of the test object collected by the front-end detector and forward the imaging test data to the processor;
[0135] The processor is used to input imaging test data into a preset anomaly detection model, and to analyze the imaging test data using the anomaly detection model to determine the anomaly detection result of the front-end detector. The anomaly detection model is obtained by fusing and training the convolutional feature information and semantic information of the sample imaging data extracted from multi-layer cascaded convolutional layers.
[0136] The aforementioned detector detection system includes a data receiver and a processor. The data receiver receives imaging test data of the test object collected by the front-end detector and forwards the imaging test data to the processor. The processor inputs the imaging test data into a preset anomaly detection model, analyzes the imaging test data using the anomaly detection model, and determines the anomaly detection result of the front-end detector. The anomaly detection model is obtained by fusing convolutional feature information extracted from sample imaging data through multi-layer cascaded convolutional layers with semantic information from the sample imaging data. Before using this detector detection system for anomaly detection, the anomaly detection model needs to be trained. During training, the anomaly detection model obtained by fusing convolutional feature information extracted from sample imaging data through multi-layer cascaded convolutional layers with semantic information can more comprehensively and deeply mine the potential features and patterns in the imaging test data. This fusion method can extract deeper features, and the feature fusion process ensures more comprehensive features, thereby greatly improving the detection accuracy of anomalies in the front-end detector and reducing false positives and false negatives.
[0137] Based on the same inventive concept, this application also provides an anomaly detection device for a front-end detector to implement the anomaly detection method for the front-end detector described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the anomaly detection device for the front-end detector provided below can be found in the limitations of the anomaly detection method for the front-end detector described above, and will not be repeated here.
[0138] In one exemplary embodiment, such as Figure 13 As shown, an anomaly detection device for a front-end detector is provided, comprising: an acquisition module 11 and a detection module 12, wherein:
[0139] The acquisition module 11 is used to acquire the imaging test data of the test object collected by the front-end detector;
[0140] The detection module 12 is used to input imaging test data into a preset anomaly detection model, analyze the imaging test data using the anomaly detection model, and determine the anomaly detection result of the front-end detector.
[0141] The anomaly detection model is trained by fusing convolutional feature information and semantic information of sample imaging data extracted from multi-layer cascaded convolutional layers.
[0142] In an exemplary embodiment, the detection module includes: an extraction unit and an anomaly determination unit, wherein:
[0143] The extraction unit is used to extract semantic information and convolutional feature information from the imaging test data and the multi-layer cascaded convolutional layers through the encoder.
[0144] The anomaly determination unit is used to determine whether there are anomalies in the front-end detector based on the convolutional feature information and semantic information on the multi-layered cascaded convolutional layers.
[0145] In an exemplary embodiment, the above-mentioned anomaly determination unit is further configured to fuse convolutional feature information and semantic information on multi-layer cascaded convolutional layers through a feature fusion processor to obtain fused information; and to classify the fused information using a classifier to determine whether there is an anomaly in the front-end detector.
[0146] In an exemplary embodiment, the above-mentioned anomaly determination unit is further configured to fuse convolutional feature information on multiple cascaded convolutional layers through a feature fusion processor to obtain fused convolutional feature information; and to extract semantic information from the fused convolutional feature information through a feature extraction network.
[0147] In an exemplary embodiment, the anomaly detection device of the aforementioned front-end detector includes: a sample acquisition module, a training module, and an adjustment module, wherein:
[0148] The sample acquisition module is used to acquire training sample data; the training sample data includes imaging data of samples with normal labels and imaging data of samples with abnormal labels.
[0149] The training module is used to train the initial anomaly detection model using training sample data to obtain the label prediction results of the training sample data.
[0150] The adjustment module is used to adjust the parameters in the initial anomaly detection model based on the difference between the labels and label prediction results of the training sample data, until the label prediction results output by the initial anomaly detection model meet the preset conditions, and then the adjusted initial anomaly detection model is determined as the anomaly detection model.
[0151] In an exemplary embodiment, the training module includes: a sample feature extraction unit, a sample feature fusion unit, and a noise addition unit, wherein:
[0152] The sample feature extraction unit is used to extract the sample semantic information and sample convolution feature information on the multi-layer cascaded convolutional layers from the initial anomaly detection model.
[0153] The sample feature fusion unit is used to fuse sample semantic information and sample convolutional feature information to obtain sample fusion information;
[0154] The noise addition unit is used to add noise to the fusion information of some samples, and the label prediction result is "abnormal"; for the fusion information of other samples, the label prediction result is "normal".
[0155] In one exemplary embodiment, an image reconstruction apparatus for a front-end detector is provided, comprising:
[0156] The adjustment module is used to adjust the parameters of the image reconstruction algorithm when the anomaly detection result of the front-end detector is abnormal; the anomaly detection result is obtained by analyzing the imaging test data of the test object collected by the front-end detector using a preset anomaly detection model.
[0157] The reconstruction module is used to reconstruct images from the acquired imaging data using adjusted parameters, thereby obtaining reconstructed images.
[0158] Each module in the anomaly detection device of the aforementioned front-end detector can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.
[0159] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the embodiments of the above-described anomaly detection method for the front-end detector.
[0160] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the content of any one embodiment of the anomaly detection method of the front-end detector described above.
[0161] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the content of any one embodiment of the anomaly detection method for the front-end detector described above.
[0162] 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, and the collection, use and processing of the relevant data must comply with relevant regulations.
[0163] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.
[0164] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.
[0165] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. An anomaly detection method for a front-end detector, characterized in that, The method includes: Acquire imaging test data of the test object collected by the front-end detector; The imaging test data is input into a preset anomaly detection model, and the anomaly detection model is used to analyze the imaging test data to determine the anomaly detection result of the front-end detector. The anomaly detection model is obtained by fusing convolutional feature information extracted from sample imaging data and semantic information of the sample imaging data through multi-layer cascaded convolutional layers.
2. The method according to claim 1, characterized in that, The anomaly detection model includes multi-layer cascaded encoders; The step of analyzing the imaging test data using the anomaly detection model to determine the anomaly detection result of the front-end detector includes: The encoder extracts semantic information and convolutional feature information from the imaging test data and the multi-layered cascaded convolutional layers. Based on the convolutional feature information on the multi-layered cascaded convolutional layers and the semantic information, it is determined whether the front-end detector has any anomalies.
3. The method according to claim 2, characterized in that, The anomaly detection model also includes a feature fusion unit and a classifier; The determination of whether the front-end detector has an anomaly based on the convolutional feature information on the multi-layered cascaded convolutional layers and the semantic information includes: The feature fusion processor fuses the convolutional feature information and the semantic information on the multi-layered cascaded convolutional layers to obtain fused information. The classifier is used to classify the fused information to determine whether there is an anomaly in the front-end detector.
4. The method according to claim 3, characterized in that, The feature fusion unit includes a feature extraction network; The extraction of semantic information from the imaging test data includes: The feature fusion processor fuses the convolutional feature information on the multi-layered cascaded convolutional layers to obtain fused convolutional feature information. The semantic information in the fused convolutional feature information is extracted through the feature extraction network.
5. The method according to any one of claims 1-4, characterized in that, The training process of the anomaly detection model includes: Acquire training sample data; the training sample data includes imaging data of samples with normal labels and imaging data of samples with abnormal labels; The initial anomaly detection model is trained using the training sample data to obtain the label prediction results of the training sample data; Based on the difference between the labels of the training sample data and the label prediction results, the parameters in the initial anomaly detection model are adjusted until the label prediction results output by the initial anomaly detection model meet the preset conditions, and the adjusted initial anomaly detection model is determined as the anomaly detection model.
6. The method according to claim 5, characterized in that, The step of training the initial anomaly detection model using the training sample data to obtain the label prediction results of the training sample data includes: The initial anomaly detection model extracts the sample semantic information and sample convolution feature information on the multi-layered cascaded convolutional layers from the sample imaging data. The sample semantic information and the sample convolutional feature information are fused to obtain sample fusion information; Noise is added to some of the sample fusion information, and the label prediction result is that there is an anomaly; other sample fusion information is classified, and the label prediction result is that it is normal.
7. An image reconstruction method, characterized in that, The method includes: If the anomaly detection result of the front-end detector is abnormal, the parameters of the image reconstruction algorithm are adjusted; the anomaly detection result is obtained by analyzing the imaging test data of the test object collected by the front-end detector using a preset anomaly detection model. The acquired imaging data is reconstructed using the adjusted parameters to obtain the reconstructed image.
8. A detector detection system, characterized in that, The detector detection system includes a data receiver and a processor; The data receiver is used to receive imaging test data of the test object collected by the front-end detector, and forward the imaging test data to the processor; The processor is used to input the imaging test data into a preset anomaly detection model, and use the anomaly detection model to analyze the imaging test data to determine the anomaly detection result of the front-end detector; wherein, the anomaly detection model is obtained by fusing and training the convolutional feature information of the sample imaging data extracted by multi-layer cascaded convolutional layers and the semantic information of the sample imaging data.
9. An anomaly detection device for a front-end detector, characterized in that, The device includes: The acquisition module is used to acquire imaging test data of the test object collected by the front-end detector; The detection module is used to input the imaging test data into a preset anomaly detection model, analyze the imaging test data using the anomaly detection model, and determine the anomaly detection result of the front-end detector. The anomaly detection model is obtained by fusing convolutional feature information extracted from sample imaging data and semantic information of the sample imaging data through multi-layer cascaded convolutional layers.
10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.