Multi-level medical image quality control method and system

Abstract

The invention relates to the technical field of medical image processing, in particular to a multi-level medical image quality control method and system, and the method comprises the following steps: obtaining an image to be subjected to quality control, and determining a quality control rule set based on the image to be subjected to quality control and a pre-stored test database, the quality control rule set represents an image detection rule related to the image to be subjected to quality control; loading and analyzing the quality control rule set to obtain a quality control instruction set, and analyzing the image to be subjected to quality control to obtain image analysis data; and performing logic comparison on the image analysis data according to the quality control instruction set to generate a quality control detection report corresponding to the image to be subjected to quality control. According to the method, the quality control is ensured to have pertinence, and misinformation of irrelevant rules and resource waste are avoided.

Description

Technical Field

[0001] This application relates to the technical field of medical image processing, and in particular to a multi-level medical image quality control method and system. Background Technology

[0002] With the rapid development of precision medicine and personalized treatment, clinical trials are playing an increasingly prominent role in drug development and treatment optimization. Specifically, medical imaging (multimodal imaging technologies such as CT, MRI, and PET-CT) is a crucial objective evaluation tool in clinical trials and is known as "imaging biomarkers." However, quality issues remain a significant factor affecting diagnostic accuracy during the acquisition, processing, and use of different modalities of medical images. These issues include image noise, artifacts, insufficient resolution, and non-standard operating procedures. These problems not only increase the difficulty of diagnosis but may also lead to misdiagnosis or missed diagnosis.

[0003] Furthermore, image quality issues are more complex in multimodal and multi-scenario medical imaging. For example, modal differences between CT and MRI images can lead to significant variations in the appearance of the same lesion in different images; in ultrasound images, image clarity and the discernibility of lesion features also vary considerably due to differences in acquisition equipment and manipulation techniques. How to achieve automated quality assessment of images across different modalities and provide accurate and reliable quality control results is one of the key issues in current clinical imaging applications.

[0004] Existing medical image quality control methods primarily rely on manual inspection and subjective judgment, which are inefficient and limited by the experience levels of different operators, exhibiting significant subjectivity and variability. Furthermore, with the rapid increase in medical data volume, manual quality control processing is inefficient, lacks priority management, and cannot be dynamically adapted. Summary of the Invention

[0005] To improve the processing efficiency of quality control methods and realize a flexible and adaptable quality control verification system, this application provides a multi-level medical image quality control method and system.

[0006] Firstly, this application provides a multi-level medical image quality control method, which adopts the following technical solution:

[0007] A multi-level medical image quality control method includes the following steps:

[0008] Acquire the image to be quality controlled, and determine the set of quality control rules based on the image to be quality controlled and the pre-stored test database, wherein the set of quality control rules represents the image detection rules related to the image to be quality controlled;

[0009] Load and parse the quality control rule set to obtain a quality control instruction set, and simultaneously parse the image to be quality controlled to obtain image parsing data;

[0010] The image analysis data is logically compared according to the quality control instruction set to generate a quality control inspection report corresponding to the image to be quality controlled.

[0011] By adopting the above technical solution, the quality control rule set is determined based on the image to be quality controlled and the pre-stored test database. The system no longer blindly applies all rules, but intelligently filters out the most relevant rules from the image to be quality controlled and the pre-stored test database, ensuring targeted quality control and avoiding false alarms and resource waste caused by irrelevant rules. Furthermore, the quality control instruction set is equivalent to providing the CPU with a pre-compiled, efficiently executable instruction list, avoiding the performance overhead of repeatedly parsing complex configurations at runtime. Preprocessing of the image parsing data extracts all key features at once, allowing multiple rules to reuse them, avoiding repeated parsing of DICOM files and significantly improving computational efficiency. When new... When there are new scanning sequences or quality control requirements (such as new artifact detection algorithms), new rules can be configured for the corresponding projects in the "pre-stored trial database" without modifying the core system code. This perfectly supports the rapid iteration and innovation of clinical research protocols. Through the core technical path of "dynamic rule matching - parallel data parsing - automated logical comparison", an intelligent quality control system capable of coping with the complexity and variability of clinical trials has been successfully built, which greatly improves the efficiency and accuracy of quality control. Furthermore, through its flexible and scalable architecture, it lays a solid foundation for the continuous evolution and development of future medical image quality control, and ultimately effectively ensures the scientific rigor and reliability of clinical research with imaging as the key endpoint.

[0012] In one embodiment, determining a set of quality control rules based on the image to be quality controlled and a pre-stored test database includes the following steps:

[0013] Anatomical identification labels are generated based on the images to be quality controlled, and the anatomical identification labels are matched in the pre-stored test database to obtain the set of quality control rules.

[0014] In one embodiment, the quality control instruction set includes key matching instructions and rule verification instructions. The image parsing data is logically compared according to the quality control instruction set to generate a quality control inspection report, including the following steps:

[0015] Based on the key matching instructions, determine whether the image to be quality controlled meets the criteria for rule matching;

[0016] If the image to be quality controlled is determined to meet the rule matching, the image parsing data is controlled to perform rule comparison according to the preset verification rules based on the rule verification instruction, and a corresponding quality control inspection report is generated.

[0017] In one embodiment, the pre-stored test database includes a project configuration file. The quality control rule set is selected from the image to be quality controlled in the pre-stored test database. The generation method of the pre-stored test database includes the following steps:

[0018] If the image to be quality controlled is determined to be inconsistent with the rule matching, then the corresponding test plan is determined based on the image to be quality controlled.

[0019] Based on the proposed test scheme, the parameters to be changed are confirmed, and a parameter change signal is generated based on the changed parameters to update the project configuration file.

[0020] By adopting the above technical solutions, different clinical trial projects (such as oncology, neurology, and cardiovascular) have drastically different imaging requirements. This method achieves project-level scenario-based quality control by linking to a "pre-stored trial database." The same system can examine multiple phases of the liver for project A and examine cerebral perfusion for project B, with quality control standards perfectly matching the clinical protocol. The tedious work that originally relied on manual visual inspection and recording has been transformed into a fully automated pipeline operation. From data upload to report generation, no manual intervention is required, which greatly improves the speed and throughput of quality control, making it particularly suitable for multi-center, large-sample clinical trials. In addition, only the project configuration file needs to be updated. When a certain module needs to be modified or debugged, the impact is limited to the local area, preventing a "one-size-fits-all" approach. Complex quality control issues are decomposed into multiple levels and processed layer by layer, achieving separation of responsibilities and refined management.

[0021] In one embodiment, the key matching instruction includes a key rule instruction. Determining whether the image parsing data conforms to a rule-based matching function based on the key matching instruction includes the following steps:

[0022] Based on the key rule instructions, determine whether the sequence corresponding to the image to be quality controlled is continuous;

[0023] If the sequence corresponding to the image to be quality controlled is determined to be continuous, then the image to be quality controlled is determined to meet the rule matching, and a rule verification instruction is generated, and the quality control instruction set is updated based on the rule verification instruction;

[0024] If it is determined that the sequence corresponding to the image to be controlled is not continuous, then the image to be controlled is determined to not meet the rule matching, and a short-circuit return instruction is generated, and the rule verification is terminated based on the short-circuit return instruction.

[0025] In one embodiment, the quality control inspection report includes a coverage verification report, the preset verification rules include scan position rules, and the step of controlling the image parsing data to perform rule comparison according to the preset verification rules based on the rule matching instruction and generating a corresponding quality control inspection report includes the following steps:

[0026] When the scan position rule is executed, an anatomical coverage trigger signal is generated, and the set of rules to be verified and the area to be detected corresponding to the quality control image are obtained based on the anatomical coverage trigger signal.

[0027] A list of required regions is generated based on the region to be detected, and the list of required regions and the set of rules to be verified are used to determine whether the list of required regions is configured.

[0028] If it is determined that the required region list is configured, then the set of sequence regions corresponding to the subject is aggregated;

[0029] A set of missing regions is determined based on the set of sequence regions, and a coverage verification report is generated based on the set of missing regions.

[0030] In one embodiment, the quality control inspection report includes a numerical verification report, the preset verification rules include numerical range rules, and the step of controlling the image analysis data to perform rule comparison according to the preset verification rules based on the rule matching instruction and generating a corresponding quality control inspection report includes the following steps:

[0031] When the numerical range rule is executed, a numerical verification signal is generated, and the current value corresponding to the image to be quality controlled is obtained based on the numerical verification signal.

[0032] Based on the numerical verification signal, determine whether the numerical range rule is configured with verification parameters;

[0033] If so, the current value is compared with the configured verification parameters, and a value verification report is output.

[0034] In one embodiment, a set of missing regions is determined based on the set of sequence regions, and a coverage verification report is generated based on the set of missing regions. The coverage verification report update method includes the following steps:

[0035] Based on the coverage verification report, an organ details list is generated, and the organ details list is traversed to determine the organs for grading.

[0036] The actual coverage rate and the corresponding specific threshold of the organ for grading are obtained sequentially, and a grading evaluation is performed based on the actual coverage rate and the specific threshold to obtain the state coverage rate threshold.

[0037] The worst coverage rate is updated based on the state coverage threshold until all organ details are traversed, and a coverage verification report is determined based on the updated worst coverage rate.

[0038] Secondly, this application provides a multi-level medical image quality control system, which adopts the following technical solution:

[0039] A multi-level medical image quality control system, which executes the multi-level medical image quality control method described in the first aspect, includes a rule configuration manager, a rule execution engine module, a rule application module, and a problem recording module.

[0040] The rule execution engine module is used to acquire the image to be quality controlled and filter the set of quality control rules based on the image to be quality controlled in the pre-stored test database in the rule configuration manager. The set of quality control rules represents the image inspection rules related to the image to be quality controlled.

[0041] The rule execution engine module is also used to load and parse the quality control rule set to obtain a quality control instruction set, and at the same time parse the image to be quality controlled to obtain image parsing data;

[0042] The rule execution engine module calls the corresponding quality control instructions in the rule application module according to the quality control instruction set to generate a quality control inspection report for the image to be quality controlled.

[0043] The rule execution engine module stores the quality control test report in the problem record module.

[0044] In one embodiment, the rule configuration manager includes a basic rule layer, a project rule layer, a sequence rule layer, and an anatomy rule layer. The basic rule layer is used to store non-configurable mandatory rules; the project rule layer is used to configure project-level configurable rules; the sequence rule layer is used to configure sequence-level verification rules; and the anatomy rule layer is used to configure anatomy region-aware rules.

[0045] In summary, this application includes at least one of the following beneficial technical effects:

[0046] 1. Through the core technical path of "dynamic rule matching - parallel data parsing - automated logical comparison", an intelligent quality control system capable of coping with the complexity and variability of clinical trials has been successfully constructed, which greatly improves the efficiency and accuracy of quality control. Furthermore, through its flexible and scalable architecture, it lays a solid foundation for the continuous evolution and development of medical image quality control in the future, and ultimately effectively ensures the scientificity and reliability of clinical research with images as the key endpoint.

[0047] 2. Different clinical trial projects (such as oncology, neurology, and cardiovascular) have drastically different imaging requirements. This method achieves project-level scenario-based quality control by linking to a pre-stored trial database. The same system can examine multiple phases of the liver for project A and examine cerebral perfusion for project B, with quality control standards perfectly aligned with the clinical protocol. It transforms the tedious work that originally relied on manual visual inspection and recording into a fully automated pipeline operation. From data upload to report generation, no manual intervention is required, greatly improving the speed and throughput of quality control, which is particularly suitable for multi-center, large-sample clinical trials. In addition, only the project configuration file needs to be updated. When a module needs to be modified or debugged, the impact is limited to the local area. Complex quality control issues are decomposed into multiple levels and processed layer by layer, achieving separation of responsibilities and refined management. Attached Figure Description

[0048] Figure 1 This is a block diagram of the multi-level medical image quality control method provided in the embodiments of this application;

[0049] Figure 2 This is a flowchart of the rule verification process provided in an embodiment of this application;

[0050] Figure 3 This is a block diagram of the quality control and testing report generation method provided in the embodiments of this application;

[0051] Figure 4 This is a block diagram of the numerical verification report generation method provided in the embodiments of this application;

[0052] Figure 5 This is a flowchart of the numerical range rule verification provided in the embodiments of this application;

[0053] Figure 6 This is a flowchart of the coverage verification report provided in the embodiments of this application;

[0054] Figure 7 This is a structural block diagram of the multi-level medical image quality control system provided in this embodiment. Detailed Implementation

[0055] To better understand the purpose, technical solutions, and advantages of this application, it has been described and illustrated below with reference to the accompanying drawings and embodiments. However, those skilled in the art should understand that this application can be implemented without these details. In some cases, to avoid obscuring various aspects of this application due to unnecessary description, well-known methods, processes, systems, components, and / or circuits already described at a higher level will not be elaborated upon. It will be apparent to those skilled in the art that various modifications can be made to the embodiments disclosed in this application, and the general principles defined in this application can be applied to other embodiments and application scenarios without departing from the principles and scope of this application. Therefore, this application is not limited to the illustrated embodiments, but conforms to the broadest scope consistent with the scope of protection claimed in this application.

[0056] This application discloses a multi-level medical image quality control method, which is applied to a multi-level medical image quality control system. The system adopts a four-layer rule system architecture and sets up a corresponding rule engine, transforming quality control from a tedious, passive, and manual task into an automated, intelligent, and predictable process.

[0057] It's important to note that the four-layer rule system architecture specifically includes a basic rule layer, a project rule layer, a sequence rule layer, and an anatomical rule layer. The basic rule layer stores non-configurable mandatory rules, including image file integrity, sequence continuity checks, and patient ID consistency. The project rule layer configures project-level configurable rules, including project-level modality configuration, cross-visit viewpoint consistency rules, and device vendor restrictions. The sequence rule layer configures sequence-level validation rules, including slice thickness rules, slice spacing rules, slice interval rules, enhancement rules, scan phase rules, and custom rules. The anatomical rule layer configures anatomical region-aware rules, including scan range rules, dynamic thresholds based on anatomical regions, and organ coverage validation.

[0058] This system employs a layered architecture where layers communicate via interfaces, ensuring clear responsibilities. When a module needs modification or debugging, the impact is limited to the local area. Complex quality control issues are broken down into multiple layers for sequential processing, achieving separation of responsibilities and refined management. Specifically, the basic rules layer ensures the basic readability, integrity, and consistency of data, providing a reliable foundation for upper-level analysis. The basic rules layer consists of mandatory, non-configurable system-level rules. These are the minimum requirements for all image data. The project rules layer ensures that all image data acquisition schemes from participating centers comply with the project's research protocol. The sequence rules layer verifies that the acquisition parameters of each specific scan sequence fully meet the scheme requirements. The rule engine corresponding to the sequence rules layer automatically parses the sequence names and parameters in the DICOM header file and matches and verifies them against predefined rules. The rule engine in the anatomy rules layer can identify the major anatomical structures (such as liver, lungs, and brain) contained in the image and automatically trigger specific quality control rules for that region.

[0059] For example, when the system recognizes a liver image, it automatically checks whether it includes key phases such as the arterial phase, portal venous phase, and delayed phase. The rule engine in the anatomical rule layer uses image recognition technology to automatically determine whether the scan range covers the target organ (e.g., whether a lung scan extends from the apex of the lung to the diaphragm) and whether it includes areas that should not be present (e.g., whether a head scan includes the neck). Based on the dynamic adjustment threshold of the anatomical region, parameter standards are adjusted according to the characteristics of different organs. For example, the signal-to-noise ratio requirements for gray and white matter in the brain can be higher than those for muscle tissue. Organ coverage is validated, quantitatively assessing the integrity of the target lesion or organ coverage in the image to ensure that key structures used for efficacy evaluation are not truncated. This layer typically requires the integration of lightweight deep learning models (such as image classification and segmentation models) and is crucial for achieving automated and intelligent quality control.

[0060] It's important to note that the clear layers in the four-layer architecture provide a standard for large development teams, allowing different teams to focus on different layers, develop in parallel, and improve efficiency. The four-layer structure is not simply a hierarchical division, but a strategic design decision. It decomposes a complex quality control system into a series of modules with single responsibilities and clear collaboration, thereby achieving system flexibility, robustness, scalability, and maintainability. This forms the technological cornerstone supporting the entire engine in handling complex and ever-changing clinical trial scenarios. It not only significantly improves the quality and consistency of imaging data in multi-center clinical trials but also provides a solid technical foundation for image-driven new clinical trial designs (such as more complex quantitative imaging biomarker studies). It achieves comprehensive quality control from basic to advanced levels, and the rule prioritization and short-circuit mechanisms employed ensure quality control efficiency and accuracy.

[0061] like Figure 1As shown, a multi-level medical image quality control method includes the following steps:

[0062] S101, acquire the image to be quality controlled, and determine the set of quality control rules based on the image to be quality controlled and the pre-stored test database.

[0063] The images to be quality controlled refer to the DICOM images uploaded to the quality control system of this application for quality control. The preset trial database refers to a database containing a rule configuration manager. This database includes clinical trial projects and a set of preset quality control rules corresponding to each clinical trial project. The preset quality control rule set is a collection of quality control rules that all image examinations under the clinical trial project must meet. Unlike the preset quality control rule set, the quality control rule set represents the image detection rules related to the images to be quality controlled. The quality control rule set is specifically generated by filtering the images to be quality controlled in the pre-stored trial database to obtain the image detection rules corresponding to the images.

[0064] Specifically, first, the system receives the uploaded DICOM image file. This is not merely a file transfer; the system immediately parses the DICOM file header to extract key metadata, providing a basis for decision-making in subsequent steps. This metadata includes, but is not limited to, patient ID, study / sequence description, scan parameters, and equipment manufacturer and model. The patient ID is used to associate with subjects in clinical trials, the study / sequence description is used to initially understand the scan content and purpose, the equipment manufacturer and model are used for equipment compatibility rules, and scan parameters such as modality and slice thickness are the foundation for rule matching.

[0065] Next, the system uses information extracted from the images (primarily patient IDs) to perform correlation queries in a pre-stored trial database to determine the context of the images. The core information queried includes project identity, subject identity, and visit location information. Project identity refers to which clinical trial project the image belongs to. Subject identity refers to the specific subject from whom the image was taken. Different visit locations may have different scanning requirements; therefore, visit location information specifically indicates whether this image was taken during a baseline visit, a treatment visit, or an endpoint visit. Finally, based on the results of the first two steps, the system filters and assembles the final set of quality control rules from the rule base. This process perfectly demonstrates the system's hierarchical architecture and dynamic configuration capabilities.

[0066] S102, load and parse the quality control rule set to obtain the quality control instruction set, and at the same time parse the image to be quality controlled to obtain image analysis data.

[0067] The system reads the set of quality control rules (usually in JSON format) corresponding to the current project from a pre-stored test database and loads it into memory. Instead of simply reading the JSON, the system parses and compiles it to generate a structured set of quality control instructions. This process includes syntax validation, semantic mapping, parameter binding, and instruction generation. Syntax validation primarily checks the correctness of the JSON format and the existence of required fields. Semantic mapping maps key-value pairs in the configuration to specific validation functions and logic. For example, upon parsing the "layer_thickness" key, the system knows to call the `apply_thickness_rules` function. Parameters in the configuration (such as `maxValue: 5.0`) are bound to the corresponding validation functions. Finally, each rule is transformed into one or more clear, executable sets of quality control instructions. Internally, these sets may be represented as a list, where each element is an execution instruction.

[0068] The system performs deep analysis of the DICOM file header, extracts all key labels, and calls an integrated AI model (such as TotalSegmentator) to segment and recognize the images, outputting a list of anatomical structures contained in the image sequence, such as ["liver", "kidney_right", "L1_vertebra"]. Scan range calculation: Based on the ImagePositionPatient of all slices, the actual range covered by the sequence in three-dimensional space (a three-dimensional bounding box) is calculated. Image feature extraction is performed on the quality control images, and the average pixel value (HU value) is calculated to help determine whether it is a flat scan or enhanced scan. The image texture of the quality control images is analyzed to prepare for subsequent artifact detection. It should be noted that existing technology is used for key label extraction, which will not be elaborated upon here.

[0069] S103, logically compare the image analysis data according to the quality control instruction set to generate a quality control inspection report corresponding to the image to be quality controlled.

[0070] Finally, the rules engine performs the core matching and decision-making steps, comparing each quality control instruction with the corresponding features in the image analysis data. For example, it uses the "THICKNESS_CHECK" instruction to compare the slice_thickness value and the "SCAN_LOCATION_CHECK" instruction to compare the anatomical_regions list. Based on the comparison results, it generates the final quality control inspection report.

[0071] It's important to note that the quality control test report includes a report header, execution summary, detailed issue list, coverage validation report, numerical validation report, and rule configuration traceability. The report header acts as the report's identifier, automatically extracting and encapsulating core identification information to ensure uniqueness and traceability. This includes basic report identifiers, project and image identifiers, subject information, and quality control execution information. The report header is generated using existing technology, so details will not be elaborated upon here. The execution summary provides an overview of the quality control results, such as the total number of image sequences to be controlled, the number of sequences that passed quality control, the number of sequences that failed, and the pass rate (accurate to two decimal places). The detailed issue list includes the issue ID, associated rule ID, rule name, severity level, and rule level. The coverage validation report specifies the required set of scan areas, the actual set of scan areas aggregated by the subject, and missing areas (including Chinese name mappings, such as "NECK→neck"). The numerical validation report specifies the actual values, allowable ranges / thresholds, and validation results (compliant / non-compliant) for parameters such as slice thickness, slice spacing, slice interval, and FOV. Rule configuration traceability refers to configuring the ID, version, creation time, last update time, and operator.

[0072] For example, the coverage verification report can achieve subject-level region aggregation, intelligent missing area prompts, and Chinese name mapping. Example of coverage verification report output: Case 1: A single site is missing, specifically mapped as "Incomplete scan range, missing neck area". Case 2: Multiple sites are missing, correspondingly mapped as "Incomplete scan range, missing neck and chest areas".

[0073] In one embodiment, step S101, determining the quality control rule set based on the image to be quality controlled and the pre-stored test database, includes the following steps:

[0074] S104: Generate anatomical recognition labels based on the images to be quality controlled, and match them with the pre-stored experimental database to obtain a set of quality control rules.

[0075] Specifically, DICOM pixel data is converted into a format suitable for AI model input, which may include grayscale normalization, resampling to a uniform resolution, and image cropping. Ensuring the quality and consistency of the input data improves the accuracy and stability of the AI ​​model's recognition. A pre-trained deep learning model (such as the TotalSegmentator model) is invoked to infer the preprocessed image. The model predicts the anatomical structure category (e.g., liver, right kidney, spleen, L1 vertebra) pixel by pixel. The model's output is then post-processed to generate structured label data, outputting an anatomical recognition label set. This set includes an existence map, coverage map, spatial boundaries, and confidence scores. The existence map is a list of all uniquely identified anatomical structure names, such as ["liver","kidney_right","L1","L2"], which is the most direct label.

[0076] The system matches the generated anatomical recognition labels with the rule configurations in the pre-stored experimental database. This matching is multi-layered and logic-driven. The specific implementation process is as follows: input image -> AI recognition generates labels -> labels match rules in the database -> output a highly customized set of quality control rules that perfectly matches the image content -> guide the engine to execute the most relevant and accurate quality control checks, thereby enabling the quality control rules to dynamically and accurately adapt to the actual content of the image.

[0077] In medical imaging quality control, sequence continuity monitoring is a crucial foundational examination, directly impacting the accuracy and reliability of all subsequent analyses and diagnoses. Sequence discontinuities mean that several image layers are missing within the scanned area. A small lesion (such as an early-stage tumor or a micrometastasis) might be located precisely between the missing layers, leading to a missed diagnosis. This is the most serious clinical risk. Sequence discontinuities are usually a sign of scanning errors or equipment malfunctions. Continuing to perform complex calculations such as slice thickness and coverage on a discontinuous sequence is meaningless and a waste of computational resources. This is precisely why... Figure 2 The reason for setting it as the highest priority and short-circuiting it is that once a discontinuity is detected, all subsequent quality control rules are immediately terminated, and a critical error is returned directly.

[0078] Combination Figure 2 In one embodiment, the quality control instruction set includes key matching instructions and rule verification instructions. The image analysis data is logically compared according to the quality control instruction set to generate a quality control inspection report, including the following steps:

[0079] S105, based on the key matching instruction, determine whether the image to be controlled meets the requirements for rule matching.

[0080] S106, if it is determined that the image to be controlled meets the rule matching, then the image parsing data is controlled to perform rule comparison according to the preset verification rules based on the rule verification instruction, and the corresponding quality control test report is generated.

[0081] The key matching instructions include item matching instructions and sequence continuity detection instructions. Figure 2 The flowchart represents two key diamond-shaped decision boxes: whether the item exists and whether the sequence is continuous. These are the highest priority key matching instructions. Only when the image to be controlled is determined to meet the rule matching will the system control the image parsing data to perform rule comparison according to the preset verification rules.

[0082] The system checks the validity of the project configuration retrieved from the database. This is the most fundamental eligibility check. If it doesn't meet the requirements, the process short-circuits and returns directly with "Error: No project". This indicates that the data lacks the necessary quality control context, making any meaningful rule matching impossible. If the project exists, the highest priority quality control check is performed to verify the integrity of the data foundation. If this also fails, the process short-circuits and returns directly with "Generate CRITICAL issue - short-circuit return". This indicates that the data itself has a fatal flaw, and further fine-grained rule matching is unnecessary. Only when both of these key matching instructions are determined to be "yes" does it mean that the image to be controlled meets the eligibility criteria for rule matching, and the process can continue. Figure 2 The flowchart's decision box checks for validation rules. The system verifies whether specific, executable business rules (such as layer thickness, scope, etc.) have been configured for this project. If not, it directly returns "Validation Passed." Because there are no rules, it is considered automatically compliant. If yes, it proceeds to the "Execute Rules Sequentially" box, which performs rule comparison processing based on the rule validation instructions.

[0083] It's important to note that the preset verification rules include scan position rules, slice thickness rules, phase rules, slice spacing rules, inter-slice thickness rules, and enhancement rules. At this point, the system begins iterating through the set of rule verification instructions loaded in the "Get Project Configuration and Rules" phase. For each instruction (such as slice thickness or phase rules), the corresponding parameters are located from the image analysis data based on the instruction type. The verification logic defined in the instruction is executed, comparing the actual parameter values ​​with the preset rule thresholds. The comparison results (pass or problem record) are generated and temporarily stored internally. The processing box "Summarize Problem List and Generate Quality Control Inspection Report" summarizes, categorizes, and formats all problems generated in the previous steps (including critical problems from short-circuit returns and various problems discovered by sequential rule execution). The process ends with "Return Result," which is a structured quality control inspection report containing an execution summary and a detailed problem list.

[0084] Pre-stored trial databases cannot cover all clinical trials. Therefore, to enable pre-stored trial databases to cover more clinical trials and improve the accuracy of quality control methods, one embodiment includes a project configuration file. The images to be quality controlled are filtered from the pre-stored trial database to obtain a set of quality control rules. The generation of the pre-stored trial database includes the following steps:

[0085] S107 If it is determined that the image to be controlled does not meet the rule matching, the corresponding test plan is determined based on the image to be controlled.

[0086] S108: Confirm the changed parameters based on the test plan, and generate a parameter change signal based on the changed parameters to update the project configuration file.

[0087] In one instance, the system failed to match a rule to an image (i.e., "the image to be quality controlled was determined not to meet the rule matching criteria"). Instead of simply marking it as an error and discarding it, the system initiated a diagnostic and learning process, with the ultimate goal of updating the project configuration file so that similar images in the future could be correctly quality controlled.

[0088] The system submits the trial protocol and related images to the project administrator or imaging expert. The expert uses a management interface to refer to the clinical protocol of the scanning protocol and clarify its quality control standards (e.g., slice thickness should be ≤1.5mm, must cover the entire heart, and must include end-diastolic and end-systolic phases, etc.). These are the parameters to be changed.

[0089] The system can automatically recommend an initial set of modified parameters for expert confirmation based on existing rules from similar projects or general standards for this type of scan. Parameter change signals trigger a write operation to the pre-stored experimental database, updating the corresponding "project configuration file." Specifically, this involves adding a quality control rule for this new scan sequence to the project's configuration (such as the `validation_rulesJSON` field in the `ProjectModalityConfig` table).

[0090] By employing steps S107-S108, the system will not crash or provide meaningless quality control results when encountering unknown scan types; instead, it will recognize them as new situations requiring learning. Through a closed loop of "quality control - missing data detection - expert definition - rule update," the quality control rule base can be continuously enriched and improved as clinical trial projects expand and evolve. It transforms rule maintenance from passive, batch code modification to proactive, on-demand configuration updates via a user-friendly interface, significantly reducing management costs and time.

[0091] Reference Figure 2In one embodiment, the key matching instruction includes a key rule instruction. Determining whether the image parsing data conforms to a rule for matching based on the key matching instruction includes the following steps:

[0092] S109, based on key rule instructions, determines whether the sequence corresponding to the image to be controlled is continuous.

[0093] S110, if the sequence corresponding to the image to be controlled is determined to be continuous, the image to be controlled is determined to meet the rule matching, and a rule verification instruction is generated, and the quality control instruction set is updated based on the rule verification instruction.

[0094] S111, if it is determined that the sequence corresponding to the image to be controlled is not continuous, then the image to be controlled does not meet the rule matching, and a short-circuit return instruction is generated, and the rule verification is ended based on the short-circuit return instruction.

[0095] Specifically, the system extracts the spatial location information of all slices from the image analysis data (such as ImagePositionPatient and SliceThickness in the DICOM header file), and calculates whether the slices are closely arranged in three-dimensional space, without missing or overlapping errors. This workflow focuses on the core link of "scan position rule execution → anatomical coverage verification → coverage verification report generation". Through a four-step closed loop of "trigger signal generation → rule configuration verification → region aggregation calculation → missing region determination", it ensures the accurate implementation of anatomical region coverage integrity verification of the images to be quality controlled, and finally outputs a coverage verification report that meets clinical and project requirements.

[0096] Sequence continuity is a fundamental prerequisite for meaningful quality control of data. Since this prerequisite is met, the system determines that the image qualifies for subsequent, more refined quality control processes. The system instantiates all static rules configured in the project and related to the image (such as slice thickness, phase, scan range, etc.) into executable "rule verification instructions." These newly generated rule verification instructions are added to the engine's current set of quality control instructions. The engine's task list is updated and enriched, including all necessary fine-tuning checks. The engine continues with the subsequent steps in S106, sequentially executing these instructions and comparing them with the image resolution data. Upon receiving this instruction, the engine no longer executes any rule verification instructions generated in S110 (e.g., no longer checks slice thickness, phase, etc.). The process jumps directly to the report generation stage. The generated quality control report will only contain one critical issue at the CRITICAL level: sequence discontinuity, clearly indicating that data verification failed. Immediate return is made upon discovering a fatal flaw to avoid wasting resources and ensure that the problem report directly addresses the core issue. By employing a short-circuit mechanism, a large amount of meaningless calculations on invalid data (discontinuous sequences) are avoided, significantly improving the system's processing efficiency. This ensures that the most serious and fundamental problems are presented to the user clearly and prioritized, preventing them from being overwhelmed by numerous secondary issues, greatly enhancing problem-solving susceptibility. S109-S111 embody the hierarchical nature of the quality control process, strictly adhering to the principle of "basic before advanced." Only after passing the basic integrity check is a process eligible for advanced parameter compliance checks.

[0097] Reference Figure 3 In one embodiment, the quality control inspection report includes a coverage verification report, and the preset verification rules include scan position rules. Based on the rule matching instructions, the image parsing data is controlled to perform rule comparison according to the preset verification rules, and a corresponding quality control inspection report is generated, including the following steps:

[0098] S201, when the scan position rule is executed, an anatomical coverage trigger signal is generated, and the set of rules to be verified and the area to be detected corresponding to the quality control image are obtained based on the anatomical coverage trigger signal.

[0099] S202, generate a list of required regions based on the region to be detected, and determine whether the list of required regions is configured based on the list of required regions and the set of rules to be verified.

[0100] S203, if the required region list is configured, then aggregate the sequence region set corresponding to the subject.

[0101] S204, determine the set of missing regions based on the set of sequence regions, and generate a coverage verification report based on the set of missing regions.

[0102] This workflow focuses on the core chain of "scan position rule execution → anatomical coverage verification → coverage verification report generation." Through a four-step closed loop of "trigger signal generation → rule configuration verification → region aggregation calculation → missing region determination," it ensures the accurate implementation of anatomical region coverage integrity verification for images under quality control, ultimately outputting a coverage verification report that meets clinical and project requirements. When the rule execution engine schedules to scan position rules, it triggers the anatomical coverage verification process and completes the data acquisition of the rules to be verified and the regions to be detected.

[0103] Specifically, S201-S204 represents an upgrade to the quality control paradigm, aggregating all sequences from a single subject's scan during a single visit to construct a complete panoramic view, and then determining whether this panoramic view meets the requirements of the trial protocol. This effectively avoids misjudgments caused by segmented scans. The rule engine executes quality control instructions sequentially, triggering when it comes to scan position rules. The coverage verification implemented by S201-S204 avoids false alarms and improves the intelligence level of quality control. Even if a single sequence only covers a portion of the body, as long as all sequences combined meet the requirements, the system will not report an error. This aligns with actual clinical scanning procedures, achieving true protocol compliance verification and directly answering the core question, "For this trial protocol, have we scanned all the necessary areas?" Clear reports of missing areas directly guide radiographers to perform supplementary scans, forming an efficient quality improvement closed loop.

[0104] When the rule execution engine schedules the scan location rule, it triggers the anatomical coverage validation process and completes data acquisition for the rule to be validated and the region to be detected. The `apply_scan_location_rules` function in the rule application module (validation / rules.py) automatically generates a standardized trigger signal (including project ID, image ID, and rule type identifier) ​​when the rule execution order reaches the scan location rule. The scan location rule configuration (e.g., allowedValues:["NECK","CHEST","ABDOMEN","PELVIS"]) is queried by project ID from the `validation_rules` field of the `ProjectModalityConfig` table in the database. The anatomical region label corresponding to the current image is extracted from the DICOM metadata (e.g., the `BodyPartExamined` field) of the image to be quality controlled or from the AI ​​anatomical segmentation results.

[0105] If the set of rules to be verified is empty, a rule configuration missing warning is triggered, a MINOR-level issue is recorded, and subsequent processes are executed according to the default rules (such as the project's general anatomical region requirements). If the extraction of the region to be detected fails (e.g., missing DICOM field, abnormal segmentation results), a MAJOR-level issue is generated, marked "Unknown region to be detected," the current verification is paused, and the process is restarted after manual completion of the information.

[0106] Based on the area to be detected, a "list of anatomical regions that must be covered" is determined according to the project requirements, and the existence of a valid configuration in this list is verified to avoid invalid verification without rules to follow. If the allowedValues ​​in the rule to be verified is a specific list of regions (such as ["NECK", "CHEST"]), it is directly deduplicated and used as the list of required regions. If the allowedValues ​​in the rule to be verified is a "dynamic matching" identifier (such as "DYNAMIC_MATCH_BODY_PART"), a list is generated based on the "location-required region mapping table" preset by the project associated with the area to be detected (e.g., when the area to be detected is "liver", ["ABDOMEN"] is generated).

[0107] For example, the set of rules to be verified is: {"scan_location":{"allowedValues":["NECK","CHEST","ABDOMEN"],"message":"Missing ${} part"}}, and the list of required regions is generated as: ["NECK","CHEST","ABDOMEN"].

[0108] Overcoming the limitations of "single-sequence local validation," this approach aggregates the anatomical regions corresponding to all valid sequences from the same subject, forming a complete "subject-level region set" to ensure comprehensive coverage validation. A MongoDB batch query interface (e.g., db.sequences.find({project_subject_id:xxx})) is used to reduce database interactions and improve aggregation efficiency. The aggregation results for the same subject's regions are cached for 30 minutes to avoid repeated queries and calculations within a short period. The format of region labels from different sources is standardized (e.g., converting "Neck" and "neck" in DICOM to uppercase "NECK") to ensure no duplicate labels during aggregation. For subdivided regions in the AI ​​segmentation results (e.g., "left lung," "right lung"), they are mapped to higher-level regions (e.g., "CHEST") using ANATOMICAL_REGIONS, avoiding aggregation dispersion caused by subdivided labels.

[0109] For example, the effective sequences of the subjects specifically include sequence 1, sequence 2, and sequence 3. Sequence 1 is a region label ["NECK","CHEST"], sequence 2 is a region label ["CHEST","ABDOMEN"], and sequence 3 is a region label ["PELVIS"]. The final aggregated result is {"NECK","CHEST","ABDOMEN","PELVIS"}.

[0110] Missing anatomical regions are identified by performing a difference operation between the "list of required regions" and the "set of subject sequence regions," generating a coverage validation report that includes details of the missing regions, severity grading, and recommended interventions. When there are ≥2 missing regions, a coherent description is generated through string concatenation (e.g., missing "NECK" and "CHEST" → "missing neck and chest regions"). The missing regions are then sorted by "clinical importance" in the report (e.g., abdomen > chest > neck > pelvis) to guide priority in scanning critical regions.

[0111] Basic information includes project ID, subject ID, image ID, and verification time. Verification results include coverage status (complete / incomplete) and severity rating (INFO / MAJOR / MINOR). The details list shows the missing region (code + Chinese characters), problem description, and suggested rescan plan. It also includes a diagram of the subject's sequence region aggregation (visually displaying covered / missing regions).

[0112] It's important to note that before region aggregation, the "patient ID consistency" of all sequences is verified (already validated at the basic rule layer) to avoid aggregation errors caused by cross-subject sequence mixing. A secondary validity check is performed on the anatomical region labels to ensure that only predefined labels from ANATOMICAL_REGIONS are included, eliminating invalid data interference. If region aggregation fails (e.g., database query timeout), it automatically retryes twice, with a 3-second interval between retryes. If it still fails, a fallback is triggered, using a "single-sequence region" to temporarily replace the "subject-level region set," and a MINOR-level problem is recorded. If the missing region determination logic is abnormal (e.g., difference operation error), it automatically switches to a backup algorithm (based on list traversal comparison) to ensure uninterrupted process flow.

[0113] The coverage verification report records the "data source chain" (such as rule configuration version, regional extraction source, and aggregation algorithm version) to facilitate subsequent issue tracing and auditing. Logs of all verification steps (such as trigger signal generation time and configuration verification results) are written to the system log database in real time and retained for one year for future reference.

[0114] In one embodiment, the quality control inspection report includes a numerical verification report, and the preset verification rules include numerical range rules. Based on the rule matching instruction, the image analysis data is controlled to perform rule comparison according to the preset verification rules, and a corresponding quality control inspection report is generated, including the following steps:

[0115] S301, when the numerical range rule is executed, a numerical verification signal is generated, and the current value corresponding to the image to be controlled is obtained based on the numerical verification signal.

[0116] S302, determine whether the numerical range rule is configured with verification parameters based on the numerical verification signal.

[0117] S303, if so, compare the current value with the configured verification parameters and output a value verification report.

[0118] The numerical verification signal is generated by the system when the numerical range rule is executed during the current rule verification. Based on this signal, the current value corresponding to the image to be controlled can be retrieved. The configuration verification parameters include the maximum and minimum values.

[0119] Specifically, numerical range rules are the most common and fundamental type of rule in quality control. The essence of the S301-S303 design lies in its abstraction of the verification logic for various numerical parameters into a unified, configurable process. This avoids writing repetitive code for different parameters such as slice thickness, slice spacing, and FOV, greatly improving the system's maintainability and scalability. When the rule engine executes quality control instructions sequentially, it triggers an instruction of type numerical range rule. Based on the specific rule type, the system extracts the actual value of the target parameter from the pre-prepared image analysis data.

[0120] Example: If it's a layer_thickness rule, extract the slice_thickness value (e.g., 2.5). If it's a layer_distance rule, extract the calculated slice_gap value (e.g., 0.5). If it's a FOV (field of view) rule, calculate the field_of_view value from the pixel size and matrix size.

[0121] If no validation parameters are configured (e.g., allowedValues, maxValue, and minValue are all absent), the rules are incomplete, and the process can end prematurely, possibly logging a warning or skipping directly. Discrete Value List: Checks if the current value is in the allowedValues ​​list.

[0122] Reference Figure 5 Example: The current layer thickness is 3.0, and the allowed list is [1.0, 1.25, 2.5, 5.0] -> Fail. When maximum value (maxValue) validation is required, it checks if the current value is less than 5.0. When minimum value (minValue) validation is required, it checks if the current value is greater than 1.0.

[0123] Specifically, it verifies whether the current value is within the range [minValue, maxValue]. If the comparison passes, it will not be reflected in the report, or a pass log will be recorded. If the comparison fails: the pre-defined message template in the rule is used. The current value and the required value are dynamically filled into the message to generate a clear problem description. The problem severity is set according to the severity defined in the rule. Finally, a specific numerical verification report is output as part of the overall quality control report. For example: "Layer thickness (6.0mm) exceeds the maximum allowable value (5.0mm)" or "Layer spacing (-0.2mm) exceeds the allowable range (0.0mm~5.0mm)".

[0124] The numerical validation workflow implemented in S301-S303 allows a single set of logic to serve the validation of all numerical parameters, including slice thickness, slice spacing, field of view (FOV), and any new parameters introduced in the future. Only rules need to be configured; no new code needs to be developed. By configuring different validation parameters (discrete values, maximum values, ranges), various clinical protocol requirements can be flexibly expressed, ranging from "standard slice thickness must be used" to "cannot exceed a certain threshold." All rules exist as configuration data; modifying thresholds or adding / removing rules requires no modification to the program code, achieving decoupling between business logic and system code, making the system easy to maintain and extend.

[0125] Reference Figure 6 Unlike the above embodiments, this embodiment introduces a graded evaluation mechanism, which no longer simply answers "whether this organ was scanned or not", but further answers "how well this organ was scanned and how complete the coverage is". This is crucial to ensuring the accuracy of subsequent quantitative analysis.

[0126] In one embodiment, a set of missing regions is determined based on a set of sequence regions, and a coverage verification report is generated based on the set of missing regions. The coverage verification report update method includes the following steps:

[0127] S401, Generate an organ details list based on the coverage verification report, and traverse the organ details list to determine the organs for grading.

[0128] S402, sequentially obtain the actual coverage rate and the corresponding specific threshold of the organ for graded judgment, and perform graded evaluation based on the actual coverage rate and the specific threshold to obtain the state coverage rate threshold.

[0129] S403, update the worst coverage rate based on the state coverage rate threshold until all organ details are traversed, and determine the coverage verification report based on the updated worst coverage rate.

[0130] The organ details list is the `coverage_findings` array in the aforementioned "coverage verification report." It is the input data list for the system's next refined processing. The graded organs refer to those organs that require different levels of quality control assessment. Essentially, it is each member of the "organ details list." The emphasis on "graded assessment" is to clarify that these organs are not simply judged as "present / absent," but rather require further refinement in the next step (S402) with grades (such as Critical, Warning, Pass).

[0131] Actual coverage rate refers to the actual coverage ratio calculated by AI for the organ currently being processed (such as the "liver"). It is a decimal between 0 and 1. Specific thresholds are pre-set standard lines for judging the quality of different organs from the system's "anatomical knowledge base".

[0132] The state coverage threshold contains a structured object of all judgment information. The worst coverage is the actual coverage with the smallest value among all the organs being evaluated. It is a global and summary indicator.

[0133] Specifically, the input for this step is the initial coverage verification report generated by S204. This report may only contain basic information about the missing areas. Based on the anatomical recognition results, the system generates a more detailed list of organ details. This list includes not only the organs that are required to be scanned, but also the organs that were actually scanned and identified by AI, along with their detailed information.

[0134] Example of organ details list content:

[0135] {"organ":"liver","actual_coverage":0.95} (Liver, actual coverage 95%)

[0136] {"organ":"spleen","actual_coverage":0.70} (Spleen, actual coverage 70%)

[0137] {"organ":"kidney_right","actual_coverage":0.50} (Right kidney, actual coverage 50%)

[0138] The system begins traversing the list of organ details, performing a refined evaluation of each organ. Each organ traversed becomes the organ for grading in the current loop. The system retrieves the quality control standard for the organ from its details (e.g., 0.95 for the liver in the example above). This value originates from the AI ​​model's image segmentation and calculation. The system then obtains the quality control standard for the organ from the configuration of the pre-stored experimental database. The system compares the actual coverage rate with a specific threshold, but not in a simple "pass / fail" manner; instead, it employs a grading strategy.

[0139] In another embodiment, the inferred grading logic includes states 1, 2, and 3. State 1 is Normal, indicating excellent organ coverage quality if the actual coverage rate is greater than or equal to a specific threshold. State 2 is Warning, indicating acceptable but flawed organ coverage quality if the specific threshold * attenuation factor (e.g., 0.8) * actual coverage rate * specific threshold. State 3 is Critical, indicating insufficient organ coverage and poor quality if the actual coverage rate * specific threshold * attenuation factor (e.g., 0.8). The output of this step is a state coverage threshold, which essentially represents the current organ's quality level (e.g., NORMAL, WARNING, CRITICAL). The system maintains a variable, such as worst_coverage_status, to record the worst quality level among all organs of the current subject. When iterating through each organ, the current organ's state coverage threshold is compared with the global worst_coverage_status. If the current state is worse, worst_coverage_status is updated.

[0140] Example: The initial status is worst_coverage_status=NORMAL; checking the liver (status: NORMAL) indicates no update. Checking the spleen (status: WARNING) updates worst_coverage_status=WARNING; checking the right kidney (status: CRITICAL) updates worst_coverage_status=CRITICAL. After iterating through all organs in the organ details list, the system generates or updates the final coverage verification report based on the final worst_coverage_status. The report not only lists the missing organs but also details the poorly covered organs and their coverage levels. The overall coverage verification conclusion is determined by the worst organ status. For example, even if only one organ is CRITICAL, the overall coverage verification report might still be "fail".

[0141] The refined coverage verification implemented in S401-S403 upgrades from binary judgment to spectral assessment, more accurately reflecting the actual quality of image data. Clearly defined "warning" statuses can prompt researchers to make improvements in future scans, while "serious" statuses may require immediate rescanning, achieving differentiated quality management. For studies relying on complete organ data for volumetric measurements or radiomics analysis, ensuring high-quality coverage of key organs is a prerequisite for obtaining reliable scientific conclusions.

[0142] In summary, this implementation, by introducing tiered assessment and the "worst-case" decision-making principle, transforms the coverage validation report from a simple integrity check tool into a powerful data quality insight tool, greatly enhancing the clinical usability and scientific rigor of the quality control system.

[0143] This application also discloses a multi-level medical image quality control system and a multi-level medical image quality control method.

[0144] like Figure 7 As shown, the multi-level medical image quality control system includes a rule configuration manager, a rule execution engine module, a rule application module, and a problem recording module. The rule execution engine module acquires the images to be quality controlled and filters a set of quality control rules based on the pre-stored test database of these images in the rule configuration manager. This set of quality control rules represents the image examination rules related to the images to be quality controlled. The rule execution engine module also loads and parses the set of quality control rules to obtain a set of quality control instructions, and simultaneously parses the images to be quality controlled to obtain image parsing data. The rule execution engine module then calls the corresponding quality control instructions in the rule application module according to the set of quality control instructions to generate a quality control inspection report for the images to be quality controlled. Finally, the rule execution engine module stores the quality control inspection report in the problem recording module.

[0145] It's important to note that the Rule Configuration Manager acts as the "knowledge base and legislative body" for quality control standards. This module is the cornerstone of the entire system, responsible for storing and managing all "quality standards." The Rule Configuration Manager is a dynamic, categorized database of quality control rules.

[0146] The rule execution engine module is responsible for scheduling and executing the entire quality control process. It acquires images to be quality controlled and filters the set of quality control rules based on the pre-stored experimental database of these images in the rule configuration manager, thus achieving data acquisition and filtering, and dynamically assembling rule sets for specific images. It loads and parses the quality control rule set to obtain a set of quality control instructions, and simultaneously parses the images to be quality controlled to obtain image parsing data for data parsing and compilation. The system then uses the image parsing data to call the corresponding quality control instructions in the rule application module according to the quality control instruction set to achieve scheduling and execution. This corresponds to step S106, where the engine precisely calls specific functions in downstream application modules based on the instruction type.

[0147] The rule execution engine is an intelligent scheduler responsible for managing processes, distributing tasks, and summarizing results. The rule application module is the "professional execution department" for quality control operations, consisting of a series of specialized, fine-grained function modules. The rule application module contains the algorithms and logic for implementing various specific quality control rules: `apply_scan_location_rules` (apply scan location rules), `apply_thickness_rules` (apply slice thickness rules), and `apply_phase_rules` (apply phase rules). When the rule execution engine needs to execute the "check slice thickness" instruction, it calls the `apply_thickness_rules` function, passing the slice thickness value from the image analysis data and relevant configuration parameters to this function. After completing the calculation and comparison, this function returns the result to the engine.

[0148] The issue logging module serves as an "archive and public notice board" for quality control results, responsible for persisting, managing, and querying these results. Issues are categorized into critical, major, and minor levels. Quality control reports are stored in a database to ensure traceability, providing a query interface for users and generating quality statistical reports for monitoring multi-center data quality. The issue logging module ensures transparency and auditability in the quality control process and provides data support for continuous improvement.

[0149] Reference Figure 7 In one embodiment, the rule configuration manager includes a base rule layer, a project rule layer, a sequence rule layer, and an anatomy rule layer. The base rule layer stores non-configurable mandatory rules. The project rule layer configures project-level configurable rules. The sequence rule layer configures sequence-level validation rules. The anatomy rule layer configures anatomy region-aware rules.

[0150] It should be noted here that, as Figure 7As shown, the basic rule layer stores non-configurable mandatory rules. These correspond to the highest priority checks in the previous workflow, such as DICOM file integrity, sequence continuity, and patient ID consistency. These rules are the passports for all data. The project rule layer stores project-level configurable rules, such as ProjectModalityConfig, cross-visit viewpoint consistency rules, and equipment vendor restrictions. It enables personalization for different projects. The sequence rule layer stores industry standards for specific scan sequences, such as layer thickness rules, layer distance rules, enhancement rules, and scan phase rules. This ensures quality control accuracy at the individual sequence level. The anatomy rule layer stores intelligent judgments based on image content, which are anatomical region-aware rules, such as scan location rules, organ coverage verification, and dynamic thresholds based on anatomical regions.

[0151] The implementation principle is as follows:

[0152] The rule execution engine receives a quality control image and queries the rule configuration manager to filter out relevant rule sets based on the image's context. The rule execution engine parses the rule set to generate a set of quality control instructions. Simultaneously, it parses the image to generate image parsing data. For each quality control instruction, the rule execution engine calls the corresponding specialized tool (function) from the rule application module to execute it. Each specialized tool returns its execution result (problem found or pass) to the rule execution engine. The rule execution engine summarizes all results to generate a complete quality control inspection report. Finally, the rule execution engine submits this report to the problem recording module for archiving and subsequent management.

[0153] The other functions executed in the rule configuration manager, rule execution engine module, rule application module, and problem recording module, as well as the technical details of each function, are the same as or similar to the corresponding features in the multi-level medical image quality control method described above, so they will not be repeated here.

[0154] It should be understood that although the steps in the flowcharts in the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless otherwise expressly stated herein, there is no strict order in which these steps are performed, and they may be performed in other orders.

[0155] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A multi-level medical image quality control method, characterized in that, Includes the following steps: Acquire the image to be quality controlled, and determine the set of quality control rules based on the image to be quality controlled and the pre-stored test database, wherein the set of quality control rules represents the image detection rules related to the image to be quality controlled; Load and parse the quality control rule set to obtain a quality control instruction set, and simultaneously parse the image to be quality controlled to obtain image parsing data; The image analysis data is logically compared according to the quality control instruction set to generate a quality control inspection report corresponding to the image to be quality controlled.

2. The multi-level medical image quality control method according to claim 1, characterized in that, The quality control rule set is determined based on the images to be quality controlled and the pre-stored test database, including the following steps: Anatomical identification labels are generated based on the images to be quality controlled, and the anatomical identification labels are matched in the pre-stored test database to obtain the set of quality control rules.

3. The multi-level medical image quality control method according to claim 1, characterized in that, The quality control instruction set includes key matching instructions and rule verification instructions. The image parsing data is logically compared according to the quality control instruction set to generate a quality control inspection report, including the following steps: Based on the key matching instructions, determine whether the image to be quality controlled meets the criteria for rule matching; If the image to be quality controlled is determined to meet the rule matching, the image parsing data is controlled to perform rule comparison according to the preset verification rules based on the rule verification instruction, and a corresponding quality control inspection report is generated.

4. The multi-level medical image quality control method according to claim 3, characterized in that, The pre-stored test database includes a project configuration file. The quality control rules set is selected from the image to be quality controlled in the pre-stored test database. The generation method of the pre-stored test database includes the following steps: If the image to be quality controlled is determined to be inconsistent with the rule matching, then the corresponding test plan is determined based on the image to be quality controlled. Based on the proposed test scheme, the parameters to be changed are confirmed, and a parameter change signal is generated based on the changed parameters to update the project configuration file.

5. The multi-level medical image quality control method according to claim 4, characterized in that, The key matching instruction includes key rule instructions. Determining whether the image parsing data meets the criteria for rule matching based on the key matching instructions includes the following steps: Based on the key rule instructions, determine whether the sequence corresponding to the image to be quality controlled is continuous; If the sequence corresponding to the image to be quality controlled is determined to be continuous, then the image to be quality controlled is determined to meet the rule matching, and a rule verification instruction is generated, and the quality control instruction set is updated based on the rule verification instruction; If it is determined that the sequence corresponding to the image to be controlled is not continuous, then the image to be controlled is determined to not meet the rule matching, and a short-circuit return instruction is generated, and the rule verification is terminated based on the short-circuit return instruction.

6. The multi-level medical image quality control method according to claim 3, characterized in that, The quality control inspection report includes a coverage verification report, and the preset verification rules include scan position rules. The step of controlling the image parsing data to perform rule comparison according to the preset verification rules based on the rule matching instruction and generating a corresponding quality control inspection report includes the following steps: When the scan position rule is executed, an anatomical coverage trigger signal is generated, and the set of rules to be verified and the area to be detected corresponding to the quality control image are obtained based on the anatomical coverage trigger signal. A list of required regions is generated based on the region to be detected, and the list of required regions and the set of rules to be verified are used to determine whether the list of required regions is configured. If it is determined that the required region list is configured, then the set of sequence regions corresponding to the subject is aggregated; A set of missing regions is determined based on the set of sequence regions, and a coverage verification report is generated based on the set of missing regions.

7. The multi-level medical image quality control method according to claim 3 or 6, characterized in that, The quality control test report includes a numerical verification report, and the preset verification rules include numerical range rules. The step of controlling the image analysis data to perform rule comparison according to the preset verification rules based on the rule matching instruction and generating a corresponding quality control test report includes the following steps: When the numerical range rule is executed, a numerical verification signal is generated, and the current value corresponding to the image to be quality controlled is obtained based on the numerical verification signal. Based on the numerical verification signal, determine whether the numerical range rule is configured with verification parameters; If so, the current value is compared with the configured verification parameters, and a value verification report is output.

8. The multi-level medical image quality control method according to claim 6, characterized in that, A set of missing regions is determined based on the set of sequence regions, and a coverage verification report is generated based on the set of missing regions. The update method of the coverage verification report includes the following steps: Based on the coverage verification report, an organ details list is generated, and the organ details list is traversed to determine the organs for grading. The actual coverage rate and the corresponding specific threshold of the organ for grading are obtained sequentially, and a grading evaluation is performed based on the actual coverage rate and the specific threshold to obtain the state coverage rate threshold. The worst coverage rate is updated based on the state coverage threshold until all organ details are traversed, and a coverage verification report is determined based on the updated worst coverage rate.

9. A multi-level medical image quality control system, characterized in that, The multi-level medical image quality control method according to any one of claims 1-8 includes a rule configuration manager, a rule execution engine module, a rule application module, and a problem recording module. The rule execution engine module is used to acquire the image to be quality controlled and filter the set of quality control rules based on the image to be quality controlled in the pre-stored test database in the rule configuration manager. The set of quality control rules represents the image inspection rules related to the image to be quality controlled. The rule execution engine module is also used to load and parse the quality control rule set to obtain a quality control instruction set, and at the same time parse the image to be quality controlled to obtain image parsing data; The rule execution engine module calls the corresponding quality control instructions in the rule application module according to the quality control instruction set to generate a quality control inspection report for the image to be quality controlled. The rule execution engine module stores the quality control test report in the problem record module.

10. The multi-level medical image quality control system according to claim 9, characterized in that, The rule configuration manager includes a basic rule layer, a project rule layer, a sequence rule layer, and an anatomy rule layer. The basic rule layer is used to store non-configurable mandatory rules; the project rule layer is used to configure project-level configurable rules; the sequence rule layer is used to configure sequence-level verification rules; and the anatomy rule layer is used to configure anatomy region-aware rules.

Images