Radiographic inspection real-time quality inspection system and device based on artificial intelligence and cloud edge collaboration

Abstract

The invention provides a radiographic inspection real-time quality inspection system and device based on artificial intelligence and cloud edge collaboration, and relates to the technical field of radiographic inspection quality inspection.The system mainly comprises a data acquisition and preprocessing module, an AI analysis module and an early warning response and feedback module which are sequentially arranged; the data acquisition and preprocessing module comprises a medical image unit, a hospital information unit, an equipment parameter unit and a data association unit; the AI analysis module comprises a medical image recognition unit, an anatomical feature and left and right identifier analysis unit and a data fusion comparison and risk assessment unit. According to the scheme, abnormal conditions such as mismatching between the inspection application and the actual operation and image artifacts can be recognized and early warned in an extremely early stage, so that the probability of operation errors is remarkably reduced, medical quality accidents are effectively avoided, and the standardization and safety of radiation inspection are improved.

Description

Technical Field

[0001] The present invention relates to the field of radiological inspection quality inspection technology, and in particular to a real-time radiological inspection quality inspection system and device based on artificial intelligence and cloud-edge collaboration. Background Art

[0002] The workload of the hospital's radiology department is huge, and equipment operators are prone to various operational errors under high-load working conditions, such as confusion of examination parts, wrong examination parts, and incorrect equipment parameter settings.

[0003] Currently, traditional quality control methods that rely on post-event reviews are no longer sufficient to meet practical needs and are unable to effectively prevent medical issues such as misdiagnosis and delayed treatment caused by radiological examination errors. Traditional process management systems typically only record the operator's operations but are unable to effectively monitor and provide early warnings. Rule-based monitoring systems are not only inflexible but also prone to false alarms. Summary of the Invention

[0004] The purpose of the present invention is to provide a real-time quality inspection system and device for radiological inspection based on artificial intelligence and cloud-edge collaboration to solve at least one of the above-mentioned technical problems existing in the prior art.

[0005] First, in order to solve the above-mentioned technical problems, the present invention provides a real-time quality inspection system for radiological inspection based on artificial intelligence and cloud-edge collaboration, including a data acquisition and preprocessing module, an AI (Artificial Intelligence) analysis module, and an early warning response and feedback module arranged in sequence.

[0006] In a feasible implementation manner, the data acquisition and preprocessing module includes a medical imaging unit, a hospital information unit, an equipment parameter unit, and a data association unit.

[0007] In a feasible embodiment, the medical imaging unit is used to collect medical imaging data from a radiological examination device and extract DICOM information (DICOM header information) before performing image preprocessing; the medical imaging data includes early medical imaging and a complete sequence of medical imaging data; the early medical imaging includes a locator image (Localizer / ScoutImage) or a first-frame medical image; and the DICOM information includes a patient ID, an examination number, an equipment ID, and plan left and right marks.

[0008] Preferably, the medical imaging unit obtains the medical imaging data of the radiological examination equipment through a PACS (Picture Archiving and Communication System).

[0009] Preferably, the image preprocessing includes conventional methods such as noise reduction, window width and window position adjustment, image standardization and anonymization processing, so as to ensure efficient operation of subsequent processing.

[0010] In a feasible implementation, the hospital information unit is used to obtain planned inspection information from the inspection application form information and perform format conversion, verification and encryption processing; the planned inspection information includes planned inspection items and planned inspection parts, etc.

[0011] Preferably, the hospital information unit includes a natural language processing subunit for parsing unstructured or semi-structured text in the examination application form information to obtain the planned examination information.

[0012] Preferably, the hospital information unit encrypts data through the SSL / TLS protocol.

[0013] Preferably, the hospital information unit monitors and periodically polls the examination application form information in real time through incremental synchronization and local cache methods, thereby ensuring data integrity and real-time performance.

[0014] Preferably, the hospital information unit further comprises an examination dictionary to support standardization of examination descriptions from multiple sources.

[0015] In a feasible embodiment, the equipment parameter unit is used to obtain standard scanning parameters from the parameter standard database based on the planned inspection information, and monitor the actual scanning parameters of the radiological inspection equipment (when setting the formal scanning sequence and during the scanning process) so as to monitor the changes in the scanning parameters in real time and issue early warnings for abnormal values.

[0016] Preferably, the device parameter unit obtains actual scanning parameters through professional API interfaces such as DICOM MPPS (a service in the DICOM protocol used to track the status of the imaging device executing the inspection process) and Structured Report (an object in the DICOM protocol used to standardize the recording of medical imaging diagnostic information).

[0017] Preferably, the scanning parameters include device ID, kV (tube voltage), mA (tube current), layer thickness, scanning range, FOV (field of view), scanning sequence, etc.

[0018] In a feasible embodiment, the data association unit is used to perform three-dimensional association and timestamp synchronization processing on the (multimodal) data obtained by the medical imaging unit, the hospital information unit and the equipment parameter unit to ensure the temporal consistency of the (multimodal) data; and then perform format unification and structured processing to facilitate subsequent fusion and comparison of the (multimodal) data.

[0019] Preferably, the three-dimensional association refers to a data structure (such as a hash table, object, etc.) using the patient ID, examination number, and equipment ID as a composite key to aggregate related information.

[0020] Preferably, the format unification refers to converting the HL7 format or XML format into the JSON format.

[0021] Preferably, the structured processing refers to extracting key information from semi-structured text or free text and filling it into predefined structured fields.

[0022] In a feasible implementation, the AI ​​analysis module includes a medical image recognition unit, an anatomical feature and left and right identification analysis unit, and a data fusion comparison and risk assessment unit.

[0023] In a feasible embodiment, the medical image recognition unit is used to perform (preliminary) identification of body parts on early medical images through a deep learning model (and determine whether the scanning range and field of view are consistent with the planned examination part) to obtain a first recognition result; then based on the complete sequence of medical image data, (the body parts are classified into organ level) the (actual coverage) scanning range and field of view are calculated to obtain a second recognition result; (based on the first recognition result and the second recognition result) a matching score is calculated through a multimodal fusion method (combined with the scanning parameters and the examination application form information).

[0024] Preferably, the deep learning model includes a 3D convolutional neural network (such as 3D-ResNet) and an attention mechanism.

[0025] In a feasible embodiment, the anatomical feature and left-right identification analysis unit is used to extract anatomical features for different examination parts, determine the actual left-right markings (through the inherent asymmetry and relative position relationship of the organs), compare them with the planned left-right markings (in the DICOM information), and calculate the consistency score to identify potential left-right errors.

[0026] In a feasible embodiment, the AI ​​analysis module also includes an image artifact processing unit; the image artifact processing unit is used to perform artifact detection on medical imaging data and mark artifact information through a multi-scale feature extraction method of a deep convolutional neural network; the artifact information includes the artifact type and the artifact affected area, so as to identify image artifacts caused by patient movement, metal implants, equipment failure, etc., for the doctor's reference.

[0027] In a feasible embodiment, the data fusion comparison and risk assessment unit is used to output a comprehensive risk score through a fusion assessment model (cross-comparing the examination application form information, scanning parameters and medical imaging data) to assess the risk of operational errors.

[0028] In a feasible implementation, the specific output process of the comprehensive risk score includes: After obtaining early medical images, the comprehensive risk score is initially calculated by combining the first recognition results, planned examination items and standard scanning parameters; After obtaining the actual scanning parameters, the comprehensive risk score is calculated again in combination with the planned inspection items and standard scanning parameters; After obtaining the complete sequence of medical imaging data, the comprehensive risk score is finally calculated by combining the examination application information, actual scanning parameters, second recognition results, consistency score, and artifact information.

[0029] Preferably, the fusion evaluation model is a machine learning model including expert rules, integrated learning and attention mechanism.

[0030] Preferably, the numerical range of the comprehensive risk score is [0, 1]; wherein [0, 0.3) represents low risk, [0.3, 0.7) represents medium risk, and [0.7, 1.0] represents high risk.

[0031] In a feasible implementation manner, the early warning response and feedback module is used to execute a corresponding early warning plan and generate an early warning processing log according to the comprehensive risk score through an early warning response strategy.

[0032] Preferably, the early warning response strategy includes: For low-risk situations, the early warning solution is to output only a single-channel prompt message, and the system continues to operate; For medium-risk situations, the early warning plan is to immediately issue a system interruption command, output a single-channel prompt message, and only after receiving confirmation feedback from the radiology examination operator can the system cancel the interruption command be issued; For high-risk situations, the early warning plan is to immediately issue an interruption inspection instruction, send emergency information through multiple channels (device side, mobile terminal and quality control center background, etc.), and only after receiving manual cross-verification information can the system cancel the interruption instruction be issued.

[0033] Preferably, the warning processing log includes user confirmation information interacting with the outside and warning processing results; the user confirmation information includes warning confirmation (warning correct) information and warning rejection (warning error) information.

[0034] In a feasible implementation, the AI ​​analysis module also includes an online learning and adaptive optimization unit; the online learning and adaptive optimization unit is used to collect warning processing results and user confirmation information through a closed-loop feedback mechanism to build a continuous learning data set; through incremental learning and federated learning methods, the fusion evaluation model is periodically updated in the cloud, and low-latency inference is maintained for edge nodes to ensure adaptation to the different scenario requirements of each hospital; the warning processing log is received, and the adaptive optimization of the fusion evaluation model is completed to reduce the false alarm rate and missed alarm rate in the future.

[0035] In a feasible embodiment, the system further includes a system integration interface module, including a hospital system communication unit, a device communication unit, a mobile terminal communication unit, and a quality control center backend communication unit; The hospital system communication unit is used to integrate the system with the hospital's RIS system; The equipment communication unit is used to integrate the system with the hospital's radiological examination equipment; The mobile terminal communication unit is used to integrate the system with the mobile terminals of the hospital's radiological examination equipment operators at all levels to achieve multi-terminal information distribution; The quality control center backend communication unit is used to integrate this system with the hospital's quality control center backend to achieve data visualization, system logs and unified operation and maintenance monitoring functions.

[0036] In one feasible implementation, the system includes cloud-edge collaborative deployment methods, specifically including: local deployment, hybrid cloud deployment, and pure cloud deployment; Local deployment refers to building edge nodes only in hospitals and fully deploying the system on each edge node to complete real-time monitoring tasks, ensuring low-latency reasoning and data security; Hybrid cloud deployment involves separating model training, data storage, and system management tasks from the system and deploying them on servers in the cloud. Meanwhile, the remaining system tasks are deployed on edge nodes built in the hospital for real-time reasoning, enabling the two to work together. The pure cloud deployment mentioned above refers to deploying the system completely on the cloud to centrally manage models and data, which is particularly suitable for scenarios with loose real-time requirements.

[0037] On the second aspect, based on the same inventive concept, the present application also provides a real-time quality inspection device for radiological inspection based on artificial intelligence and cloud-edge collaboration, including a processor, a memory and a bus. The memory stores instructions and data that can be read by the processor. The processor is used to call the instructions and data in the memory to realize the real-time quality inspection system for radiological inspection based on artificial intelligence and cloud-edge collaboration as described above. The bus connects the functional components for transmitting information.

[0038] By adopting the above technical solution, the present invention has the following beneficial effects: The present invention provides a real-time quality inspection system and device for radiological examinations based on artificial intelligence and cloud-edge collaboration. This innovatively introduces instant AI analysis of early medical images and real-time monitoring of actual scanning parameters. This enables the early detection of potential operational errors (e.g., incorrect examination site, improper key parameter settings) before large amounts of imaging data are generated. This provides operators with valuable opportunities for immediate correction, effectively avoiding unnecessary repeated scans and reducing patients' radiation exposure and waiting time. This solution establishes a multi-level, progressive, real-time quality inspection process, from early medical image analysis and actual scanning parameter verification to in-depth analysis of the complete sequence of medical image data. Each stage can issue early warnings for different types of problems, achieving dynamic and refined quality control of the entire radiological examination process, significantly improving the sensitivity and timeliness of error detection. This solution not only compares text information and parameters, but also uses AI to intelligently analyze and judge the image content itself (such as the actual scanning range, FOV, left and right orientation, artifacts, etc.), making the quality inspection results closer to clinical practice and improving the accuracy and reliability of automated quality inspection. This solution can monitor abnormalities such as mismatches between the examination site, scanning parameters, and examination application form information in real time, and issue early warnings in a timely manner, thereby reducing the probability of operational errors and avoiding medical quality accidents. This solution can use medical imaging and anatomical knowledge to accurately identify left and right landmarks, avoiding recognition errors caused by organ reversal or image noise; This solution can integrate and compare multi-source data to reduce false positives and missed negatives; This solution can combine cloud-based model training with real-time inference at the edge node to build a continuously optimized feedback mechanism, thereby enabling online updates of the fusion evaluation model while ensuring data security and privacy. This solution can achieve seamless connection and collaborative operation with resources such as RIS systems and radiological examination equipment without changing the existing hospital hardware equipment and work processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1A diagram of a real-time quality inspection system for radiological inspection based on artificial intelligence and cloud-edge collaboration, provided by an embodiment of the present invention; Figure 2 This is a diagram of the internal architecture of the data acquisition and preprocessing module provided in an embodiment of the present invention; Figure 3 A diagram of another real-time quality inspection system for radiological examinations based on artificial intelligence and cloud-edge collaboration, provided by an embodiment of the present invention; Figure 4 This is a diagram of the internal architecture of the system integration interface module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0042] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0043] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0044] The present invention will be further explained below with reference to specific embodiments.

[0045] It should also be noted that the following specific embodiments or specific implementations are a series of optimized settings listed in the present invention to further explain the specific content of the invention, and these settings can be combined or used in association with each other.

[0046] Example 1: like Figure 1As shown, this embodiment provides a real-time quality inspection system for radiological inspection based on artificial intelligence and cloud-edge collaboration, including a data acquisition and preprocessing module, an AI analysis module, and an early warning response and feedback module that are arranged in sequence.

[0047] Furthermore, if Figure 2 As shown, the data acquisition and preprocessing module includes a medical imaging unit, a hospital information unit, an equipment parameter unit and a data association unit.

[0048] Furthermore, the medical imaging unit is used to collect medical imaging data from radiological examination equipment and extract DICOM information (DICOM header information) before performing image preprocessing; the medical imaging data includes early medical images and complete sequences of medical imaging data; the early medical images include localizer / scout images or first-frame medical images; the DICOM information includes patient ID, examination number, equipment ID, and plan left and right marks.

[0049] Preferably, the medical imaging unit obtains the medical imaging data of the radiological examination equipment through a PACS (Picture Archiving and Communication System).

[0050] Preferably, the image preprocessing includes conventional methods such as noise reduction, window width and window position adjustment, image standardization and anonymization processing, so as to ensure efficient operation of subsequent processing.

[0051] Furthermore, the hospital information unit is used to obtain the examination application form information in the RIS system (Radiology Information System) in real time according to the examination number, obtain the planned examination information therefrom and perform format conversion, verification and encryption processing; the planned examination information includes the examination number, patient ID, planned examination items, planned examination site, clinical diagnosis and medical order text, etc.

[0052] Preferably, the hospital information unit is connected to the RIS system via HL7 (Health Level Seven), FHIR (Fast Healthcare Interoperability Resources) protocol or ORM (Object-Relational Mapping) message.

[0053] Preferably, the hospital information unit includes a natural language processing subunit, which is used to parse the unstructured or semi-structured text in the examination application form information through a natural language processing method (such as Transformer architecture and Chinese medical NLP model) to obtain the planned examination information.

[0054] Preferably, the hospital information unit encrypts data through the SSL / TLS protocol.

[0055] Preferably, the hospital information unit monitors and periodically polls the examination application form information in real time through incremental synchronization and local cache methods, thereby ensuring data integrity and real-time performance.

[0056] Preferably, the hospital information unit further includes an examination dictionary to support standardization of examination descriptions from multiple sources.

[0057] Furthermore, the equipment parameter unit is used to obtain standard scanning parameters from the parameter standard database based on the planned inspection information, and through the API interface, monitor the actual scanning parameters of the radiological inspection equipment (CT machine / MRI machine) in real time during the formal scanning sequence setting and during the scanning process, so as to monitor the changes in scanning parameters in real time and issue early warnings for abnormal values.

[0058] Preferably, the device parameter unit obtains actual scanning parameters through professional API interfaces such as DICOM MPPS (a service in the DICOM protocol used to track the status of the imaging device executing the inspection process) and Structured Report (an object in the DICOM protocol used to standardize the recording of medical imaging diagnostic information).

[0059] Preferably, the scanning parameters include device ID, kV (tube voltage), mA (tube current), layer thickness, scanning range, FOV (field of view), scanning sequence, etc.

[0060] Furthermore, the data association unit is used to perform three-dimensional association and timestamp synchronization processing on the (multimodal) data obtained by the medical imaging unit, the hospital information unit and the equipment parameter unit to ensure the temporal consistency of the (multimodal) data; and then perform format unification and structured processing to facilitate subsequent fusion and comparison of the (multimodal) data.

[0061] Preferably, the three-dimensional association refers to a data structure (such as a hash table, object, etc.) using the patient ID, examination number, and equipment ID as a composite key to aggregate related information.

[0062] Preferably, the format unification refers to converting the HL7 format or XML format into the JSON format.

[0063] Preferably, the structured processing refers to extracting key information from semi-structured text or free text and filling it into predefined structured fields.

[0064] Furthermore, the AI ​​analysis module includes a medical image recognition unit, an anatomical feature and left and right identification analysis unit, an image artifact processing unit, a data fusion comparison and risk assessment unit, and an online learning and adaptive optimization unit.

[0065] Furthermore, the medical image recognition unit is used to perform preliminary identification of body parts in early medical images through a deep learning model, and determine whether the scanning range and field of view are consistent with the planned inspection parts to obtain a first recognition result; then, for the complete sequence of medical image data, the body parts are classified to the organ level, and the actual coverage of the scanning range and field of view are calculated to obtain a second recognition result; based on the first recognition result and the second recognition result, a multimodal fusion method is used, combined with the scanning parameters and the inspection application form information, to calculate the matching score.

[0066] Preferably, the matching score is based on a comprehensive evaluation of the consistency between the identified parts of the medical image data and the information on the examination application form, the consistency between the scanning parameters and the information on the examination application form, and the spatial overlap between the identified parts of the medical image data and the scanning parameters; the comprehensive evaluation adopts a weighted summation method; the specific calculation formula includes: ; in, Represents the matching score, usually normalized to interval, where higher values ​​indicate better matching between medical imaging data, scanning parameters, and examination application form information; Both represent weight coefficients, which are used to adjust the influence of each item in the formula on the matching score, and meet the requirements. , the specific value can be determined based on clinical experience, data statistics or model training. For example, if the matching degree between medical imaging data and examination application form information is the most important, then Relatively large; Represents the similarity score between the identified parts of medical image data and the parts extracted from the examination application form information, which is used to calculate the body part information identified by the 3D convolutional neural network and attention mechanism from the actual scanned medical image data The planned inspection location information extracted from the inspection application form information by the natural language processing subunit The degree of similarity or consistency between them can be calculated based on conventional methods such as exact matching of classification labels, semantic similarity (such as using medical ontology library to judge the relevance of "chest anteroposterior radiograph" and "chest") or recognition confidence. The output value is usually interval; Indicates the consistency score between the scan parameters and the inspection application information, which is used to evaluate the actual scan parameters Whether it complies with the planned inspection part information extracted from the inspection application form and information on planned examination items (such as CT plain scan, MRI enhanced scan, etc.) The requirements can be calculated based on conventional methods such as predefined rule bases (for example, checking the CT scan of a certain part to see whether its kV and mA are within the recommended range and whether the scan range covers the planned scanned body part) and machine learning models (judging the rationality of parameters based on historical data). The output value is usually also interval; Represents the spatial coincidence score between the identified parts of medical imaging data and the scanning parameters, which is used to evaluate Is it The image is calculated based on geometric calculations, such as the degree of overlap between the bounding box or segmentation mask of the identified body part and the spatial area defined by the scan range (e.g., volume intersection ratio), and other conventional methods. The output value is usually also interval.

[0067] Preferably, the deep learning model includes a conventional 3D convolutional neural network (such as 3D-ResNet) and an attention mechanism.

[0068] Furthermore, the anatomical feature and left-right identification analysis unit is used to extract anatomical features (for example, the heart is generally on the left) for different examination parts (such as the head and neck, chest, abdomen and limbs, etc.), determine the actual left and right markings through the inherent asymmetry and relative position relationship of the organs, and compare them with the planned left and right markings in the DICOM information to calculate the consistency score, so as to identify potential left-right errors.

[0069] Preferably, the specific calculation formula of the consistency score includes: ; in, A quantitative score representing the consistency score, reflecting the actual left-right labeling determined by anatomical features Plan left and right markers in DICOM information (usually from fields such as "Laterality") The degree of consistency between , a higher score indicates better consistency; The value rule is: +1 means it is identified as the anatomical "left side", -1 means it is identified as the anatomical "right side", and 0 means that the left and right directions cannot be determined or the examination site is not suitable for left and right distinction; The value rule is: +1 means the DICOM mark is "Left", -1 means the DICOM mark is "Right", 0 means the DICOM mark is missing / empty / "Both" / other situations where the unilateral orientation cannot be clearly indicated; Indicates the confidence of the actual left and right labels, used for The reliability of the judgment results is evaluated to reflect the clarity and specificity of the anatomical features, and the value range is ,The higher the value, the more reliable the judgment result; Indicates the default score, used when a valid identity comparison cannot be performed (e.g. Undetermined or unclear), the assigned score is usually preset to a neutral or low value (such as 0.5), indicating that consistency cannot be confirmed under the current circumstances and can be configured according to subsequent needs; In this way, and the same, and neither is indeterminate (i.e. and )hour, equal , which means that we can trust that the left and right markings are consistent, and the degree of confidence depends on the degree of confidence in the anatomical analysis; and When the opposite is true (e.g. one is +1 and the other is -1), it means there is a clear conflict between the left and right labels. Equal to 0, indicating that the left and right marks are seriously inconsistent; or When , it means that it is impossible to make an effective consistency judgment. equal ; Through the above formula, we can comprehensively consider the actual judgment, planned marking and the confidence of the judgment to obtain a quantitative consistency measure, which provides a reliable input basis for subsequent calculations.

[0070] Furthermore, the image artifact processing unit is used to perform artifact detection on medical imaging data and mark artifact information through a multi-scale feature extraction method of a deep convolutional neural network (such as a DCNN model); the artifact information includes the artifact type (such as motion artifact, metal artifact, beam hardening artifact, and undersampling artifact, etc.) and the artifact-affected area, so as to identify image artifacts caused by patient movement, metal implants, and equipment failure, etc., for the doctor's reference.

[0071] Preferably, the specific process of the artifact detection may include: ; in, Represents input medical imaging data, usually 2D slices or 3D volume data; Represents the artifact processing DCNN model, which is used to detect and classify artifacts through multi-scale feature extraction; Represents the model parameters of the DCNN model, which can be obtained by training on a standard dataset; represents the artifact probability map, and Same dimension, each element value is Between, that is, the probability of artifacts existing at the corresponding position; Represents the artifact type map, and Same dimension (or mask only in the affected area of ​​the artifact The marked area is defined), the element value is a predefined artifact type label (such as an integer or enumeration value); The affected area mask of the artifact , which can be achieved through thresholding. The specific expression is: ; in, represents a pixel or voxel location in the artifact probability map; Represents the probability threshold, used to generate The limit value is a hyperparameter with a value range of ; In this way, artifacts can be detected accurately and efficiently.

[0072] Furthermore, the data fusion comparison and risk assessment unit is used to cross-compare the examination application information, scanning parameters and medical imaging data through a fusion assessment model, and output a comprehensive risk score to assess the risk of operational errors; specifically, it includes: After obtaining early medical images, the comprehensive risk score is initially calculated by combining the first recognition results, planned examination items and standard scanning parameters; After obtaining the actual scanning parameters, the comprehensive risk score is calculated again in combination with the planned inspection items and standard scanning parameters; After obtaining the complete sequence of medical imaging data, the comprehensive risk score is finally calculated by combining the examination application information, actual scanning parameters, second recognition results, consistency score, and artifact information.

[0073] Preferably, the fusion assessment model is a conventional machine learning model that includes expert rules (logical judgment rules preset based on medical knowledge and clinical guidelines), ensemble learning (combining the prediction results of multiple basic learners) and an attention mechanism (dynamically and selectively focusing on the most important information parts for risk assessment when processing input features).

[0074] Preferably, the specific expression of the comprehensive risk score can be: ; in, Represents a comprehensive risk score, which is used to quantify the risk level of potential operational errors or mismatches assessed by all the information of the current examination (medical imaging data, scanning parameters, examination application information and analysis results of this system). The value range is , a higher score indicates a greater risk; represents the fusion evaluation model; Represents the information features of the inspection application form, that is, the key features extracted from the inspection application form information, including planned inspection items and planned inspection locations; represents the scanning parameter features, i.e., the key features extracted from the scanning parameters and their analysis results; Represents the characteristics of medical imaging data, that is, the key features analyzed from medical imaging data, including examination site, matching score, etc. Represents the left and right logo analysis features, including consistency scores, etc. Indicates the characteristics of image artifacts, including the presence or absence of artifacts, artifact types, and affected areas; The model parameters representing the fusion evaluation model (such as network weights, rule thresholds, and attention weights) can be obtained by learning from a large amount of labeled data.

[0075] Preferably, the numerical range of the comprehensive risk score is [0, 1]; wherein [0, 0.3) represents low risk, [0.3, 0.7) represents medium risk, and [0.7, 1.0] represents high risk.

[0076] Furthermore, the online learning and adaptive optimization unit is used to collect early warning processing results and user confirmation information through a closed-loop feedback mechanism to build a continuous learning data set; through conventional methods such as incremental learning and federated learning, the fusion evaluation model is periodically updated in the cloud, and low-latency reasoning is maintained for edge nodes to ensure adaptation to the different scenario requirements of each hospital.

[0077] Preferably, the specific construction method of the continuous learning dataset includes: recording the input features of the radiological inspection task, the prediction results and warning content of the fusion evaluation model; collecting user feedback information as labels; associating the input features, prediction results, warning content and labels to form a structured learning sample; and iteratively adding the newly generated learning samples to the continuous learning dataset.

[0078] Furthermore, the early warning response and feedback module is used to execute the corresponding early warning plan based on the comprehensive risk score through the early warning response strategy; generate an early warning processing log and feed it back to the online learning and adaptive optimization unit to complete the adaptive optimization of the fusion assessment model and reduce the false alarm rate and missed alarm rate in the future.

[0079] Preferably, the early warning response strategy includes: For low-risk situations, the early warning solution is to output only a single-channel prompt message, and the system continues to operate; For medium-risk situations, the early warning plan is to immediately issue a system interruption command, output a single-channel prompt message, and only after receiving confirmation feedback from the radiology examination operator can the system cancel the interruption command be issued; For high-risk situations, the early warning plan is to immediately issue an interruption inspection instruction, send emergency information through multiple channels (device side, mobile terminal and quality control center background, etc.), and only after receiving manual cross-verification information can the system cancel the interruption instruction be issued.

[0080] Preferably, the warning processing log includes user confirmation information interacting with the outside and warning processing results; the user confirmation information includes warning confirmation (warning correct) information and warning rejection (warning error) information.

[0081] Furthermore, if Figure 3 As shown, the system also includes a system integration interface module for unified external communication interaction; Figure 4 As shown, the system integration interface module specifically includes a hospital system communication unit, a device communication unit, a mobile terminal communication unit and a quality control center background communication unit; The hospital system communication unit is used to integrate the system with the hospital's RIS system; The equipment communication unit is used to integrate the system with the hospital's radiological examination equipment; The mobile terminal communication unit is used to integrate the system with the mobile terminals of the hospital's radiological examination equipment operators at all levels to achieve multi-terminal information distribution; The quality control center backend communication unit is used to integrate this system with the hospital's quality control center backend to achieve data visualization, system logs and unified operation and maintenance monitoring functions.

[0082] Furthermore, this system includes a cloud-edge collaborative deployment mode, and its specific implementation can include the following three deployment modes: Local deployment mode (edge ​​computing mode): In this deployment approach, the entire real-time radiology quality inspection system, including data acquisition and preprocessing modules, AI analysis modules, and early warning, response, and feedback modules, is fully deployed on edge computing nodes within the hospital (such as dedicated servers or embedded devices). Data flows are completely closed within the hospital firewall: the RIS system, PACS system, and radiology equipment interact with the system deployed on the edge nodes via the hospital's LAN. The inference process of each AI model in the system (deep learning models, fusion evaluation models, etc.) is completed in real time on the edge nodes, ensuring minimal response latency and maximum data security. No sensitive patient data ever leaves the hospital. This approach is suitable for medical institutions with extremely high requirements for data privacy and real-time performance.

[0083] Hybrid cloud deployment mode (cloud-edge collaboration mode): In this deployment method, system functions are rationally divided. Tasks with high real-time requirements are deployed on edge nodes, while non-real-time tasks that consume large amounts of computing resources are deployed in the cloud. Specifically, they include: Edge nodes (hospital side): Deploy the data collection and preprocessing module, the inference engine of the AI ​​analysis module, and the early warning response and feedback module. They are responsible for acquiring data in real time, performing rapid AI analysis (inference), and issuing immediate warnings.

[0084] Cloud servers: Deploy training and optimization engines for AI analysis modules (such as online learning and adaptive optimization units), large-scale data storage and archiving, and system management backends.

[0085] Data exchange: Edge nodes upload desensitized and anonymized non-sensitive data, such as warning logs and user feedback, to the cloud via encrypted channels (such as HTTPS / VPN). Cloud servers use this data to periodically perform incremental learning or federated learning on the AI ​​model and distribute the optimized new model parameters to each edge node for updating. This approach balances real-time response, data security, and continuous iterative optimization of the model.

[0086] Pure cloud deployment (SaaS model): In this deployment approach, the entire real-time quality inspection system is deployed entirely on cloud servers. The hospital uploads desensitized inspection request form information, equipment parameters, and early medical imaging data to the cloud in real time via a lightweight proxy program or a secure API gateway. After the cloud-based AI analysis module completes all calculations and analysis, it transmits the warning results back in real time via an encrypted connection to the hospital's equipment console or the operator's mobile device. This approach eliminates the need for complex hardware deployment within the hospital, facilitating rapid deployment and unified maintenance and upgrades. It is particularly suitable for medical institutions sensitive to initial investment costs or those seeking to adopt a Software as a Service (SaaS) model.

[0087] Example 2: This embodiment provides a real-time quality inspection device for radiological inspection based on artificial intelligence and cloud-edge collaboration, including a processor, a memory and a bus. The memory stores instructions and data that can be read by the processor. The processor is used to call the instructions and data in the memory to realize the real-time quality inspection system for radiological inspection based on artificial intelligence and cloud-edge collaboration as described above. The bus connects the functional components for transmitting information.

[0088] In another embodiment, this solution can also be implemented in the form of an integrated device, which can include corresponding modules for performing each or several steps in each of the above embodiments. The modules can be one or more hardware modules specifically configured to perform the corresponding steps, or implemented by a processor configured to perform the corresponding steps, or stored in a computer-readable medium for implementation by the processor, or implemented by some combination.

[0089] The processor performs the various methods and processes described above. For example, the method implementation in this solution can be implemented as a software program, which is tangibly contained in a machine-readable medium, such as a memory. In some embodiments, part or all of the software program can be loaded and / or installed via a memory and / or a communication interface. When the software program is loaded into the memory and executed by the processor, one or more steps in the method described above can be performed. Alternatively, in other embodiments, the processor can be configured to perform one of the above methods by any other appropriate means (e.g., by means of firmware).

[0090] The device can be implemented using a bus architecture. The bus architecture can include any number of interconnecting buses and bridges, depending on the specific application and overall design constraints of the hardware. The bus connects various circuits including one or more processors, memories, and / or hardware modules. The bus can also connect various other circuits such as peripherals, voltage regulators, power management circuits, external antennas, etc.

[0091] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Component (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A real-time quality inspection system for radiological inspection based on artificial intelligence and cloud-edge collaboration, characterized by: It includes the data collection and preprocessing module, AI analysis module and early warning response and feedback module set up in sequence; The data acquisition and preprocessing module includes a medical imaging unit, a hospital information unit, an equipment parameter unit, and a data association unit; Medical imaging unit, used to collect medical imaging data from radiological examination equipment and perform image preprocessing after extracting DICOM information; The hospital information unit is used to obtain the planned inspection information from the inspection application form information and perform format conversion, verification and encryption processing; the planned inspection information includes planned inspection items and planned inspection parts; An equipment parameter unit, used to obtain standard scanning parameters from a parameter standard database according to planned inspection information and monitor actual scanning parameters of the radiological inspection equipment; The data association unit is used to perform three-dimensional association and time stamp synchronization processing on the data obtained by the above units; and then perform format unification and structure processing; The AI ​​analysis module includes a medical image recognition unit, an anatomical feature and left-right identification analysis unit, and a data fusion comparison and risk assessment unit; The medical image recognition unit is used to identify body parts in early medical images using a deep learning model to obtain a first recognition result; then, based on the complete sequence of medical image data, it calculates the scanning range and field of view to obtain a second recognition result; and calculates a matching score using a multimodal fusion method; The anatomical feature and left-right marking analysis unit is used to extract anatomical features for different examination sites, determine the actual left-right markings, compare them with the planned left-right markings, and calculate the consistency score; Data fusion comparison and risk assessment unit, used to output comprehensive risk scores through fusion assessment models; The early warning response and feedback module is used to execute the early warning plan and generate the early warning processing log based on the comprehensive risk score and the early warning response strategy.

2. The system according to claim 1, wherein: The hospital information unit includes a natural language processing subunit, which is used to parse the unstructured text in the examination application form information and obtain the planned examination information.

3. The system according to claim 1, wherein: The image preprocessing includes noise reduction, window width and window position adjustment, image standardization and anonymization processing.

4. The system according to claim 1, wherein: The fusion evaluation model is a machine learning model that includes expert rules, integrated learning and attention mechanism.

5. The system according to claim 1, wherein: When the data fusion comparison and risk assessment unit calculates the matching score, it is based on a comprehensive assessment of the consistency between the medical image data identification part and the examination application form information, the consistency between the scanning parameters and the examination application form information, and the spatial overlap between the medical image data identification part and the scanning parameters.

6. The system according to claim 5, characterized in that The comprehensive evaluation specifically adopts a weighted summation method.

7. The system according to claim 6, characterized in that The specific calculation formula of the weighted summation is: ； in, represents the matching score; Both represent weight coefficients; Represents the similarity score between the identified part of the medical image data and the part extracted from the examination application form information; Indicates the consistency score between the scan parameters and the inspection request form information; Represents the spatial coincidence score between the identified parts of medical imaging data and the scanning parameters; Represents body part information identified from actual scanned medical imaging data; Indicates the planned inspection part information extracted from the inspection application information; Indicates the actual scanning parameters; Indicates the planned inspection item information extracted from the inspection application form information.

8. The system according to claim 1, wherein: The specific calculation formula of the consistency score includes: ； in, A quantitative score representing the consistency rating; Indicates actual left and right markings; Indicates planned left and right marks; Indicates the confidence of the actual left and right labels; Indicates the default rating.

9. The system according to claim 1, wherein: The value range of the comprehensive risk score is [0, 1]; wherein [0, 0.3) represents low risk, [0.3, 0.7) represents medium risk, and [0.7, 1.0] represents high risk.

10. The system according to claim 9, characterized in that The early warning response strategy includes: For low-risk situations, the early warning solution is to output only a single-channel prompt message, and the system continues to operate; For medium-risk situations, the early warning plan is to immediately issue a system interruption command, output a single-channel prompt message, and only after receiving confirmation feedback from the radiology examination operator can the system cancel the interruption command be issued; For high-risk situations, the early warning plan is to immediately issue an interruption inspection instruction, send out multi-channel emergency information, and only after receiving manual cross-verification information can the system issue a cancellation interruption instruction.

11. The system according to claim 1, wherein: The warning processing log includes user confirmation information interacting with the outside and warning processing results; the user confirmation information includes warning confirmation information and warning rejection information.

12. The system according to claim 1, wherein: This system includes cloud-edge collaborative deployment methods, including: local deployment, hybrid cloud deployment and pure cloud deployment; The local deployment refers to building edge nodes only in the hospital and fully deploying the system on each edge node; The hybrid cloud deployment refers to splitting the model training, data storage, and system management tasks in the system and deploying them on cloud servers, while deploying the remaining system tasks on edge nodes built in the hospital for real-time reasoning; The pure cloud deployment refers to deploying the system completely in the cloud.

13. A real-time quality inspection device for radiological inspection based on artificial intelligence and cloud-edge collaboration, characterized in that: It includes a processor, a memory and a bus, the memory stores instructions and data read by the processor, the processor is used to call the instructions and data in the memory to implement the system as described in any one of claims 1 to 12, and the bus connects the functional components to transmit information.

Images