Intelligent DR image quality control classification method

Abstract

The application discloses a kind of quality control classification methods of intelligent DR image, belong to medical image processing and artificial intelligence technical field.The application solves the problems that existing intelligent DR quality control technology is based on image appearance correlation judgment, cannot trace back the root cause of quality problem, poor cross-device generalization, lack of clinical relevance and insufficient explainability.The application is based on X-ray imaging physical process to construct causal model including device parameter node, imaging process node, image quality node and clinical influence node, determines the type, severity and root cause of image quality problem through causal inference algorithm, and generates quality control classification results and targeted improvement suggestions in combination with clinical correlation evaluation.The application can be used stably without retraining for different brands and models of DR equipment, effectively reduces unnecessary re-shooting, reduces patient radiation dose, and significantly improves the clinical trust and acceptance of AI quality control results.

Description

Technical Field

[0001] This invention relates to the fields of medical image processing and artificial intelligence technology, specifically to a quality control classification method for intelligent DR images. Background Technology

[0002] Digital radiography (DR) is one of the most widely used medical imaging methods in clinical practice, broadly covering disease screening and diagnosis in multiple parts of the body, including the chest, limbs, spine, and abdomen. The quality of its images directly determines the accuracy of disease diagnosis and the rationality of subsequent treatment plans. In recent years, with the rapid penetration of artificial intelligence technology into the medical field, intelligent DR image quality control technology based on deep learning has gradually become a research hotspot in the industry, aiming to replace traditional manual quality control methods and improve the efficiency, consistency, and standardization of quality control work.

[0003] However, existing intelligent DR image quality control technologies still have many shortcomings that make it difficult to meet actual clinical needs. For example, a DR image quality control method based on convolutional neural networks disclosed in Chinese patent application CN114596270A extracts image pixel features for binary classification of qualified / unqualified and detection of common artifacts. Essentially, this method is based on the correlation judgment of image appearance and cannot trace the root cause of quality problems, resulting in a significant performance drop of the model on different brands and models of DR equipment. At the same time, this method does not establish a correlation between image quality and its impact on clinical diagnosis, cannot distinguish between serious problems that affect diagnosis and minor problems that do not affect diagnosis, and only outputs simple classification results without providing clear causes of problems and targeted improvement suggestions, thus limiting its large-scale application in clinical practice. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a quality control and classification method for intelligent DR images to solve the aforementioned problems.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides a quality control classification method for intelligent DR images, comprising the following steps: S1. Construct a causal model of the physical chain of DR imaging. The causal model is a directed acyclic graph structure that includes device parameter nodes, imaging process nodes, image quality nodes, and clinical impact nodes. S2. Obtain the DICOM file of the DR image to be evaluated, and extract the DR image data and the corresponding DICOM header information from it; S3. Extract device parameters and imaging parameters from the DICOM header information and input them into the device parameter node of the causal model; extract image features from the DR image data and input them into the image quality node of the causal model; S4. Calculate the causal effects between nodes using a causal reasoning algorithm to determine the type, severity, and root cause of image quality problems; S5. Based on the causal reasoning results, calculate the degree of impact of quality problems on clinical diagnosis and generate a clinical severity score; S6. Generate quality control classification results based on clinical severity scores, and generate improvement suggestions based on root causes.

[0006] Preferably, the step of constructing the causal model of the physical chain for DR imaging includes: (1) Collect imaging data of DR equipment of different brands and models and build equipment feature library; (2) For each type of device, train a corresponding causal model submodule, wherein the submodule contains the inherent imaging characteristic parameters of the device; (3) During the inference phase, the device model is automatically identified based on the DICOM header information, and the corresponding causal model submodule is called for calculation.

[0007] Preferably, the inherent imaging characteristic parameters included in the device feature library include detector response curves, noise models, and contrast transfer functions.

[0008] Preferably, the step of calculating the impact of quality issues on clinical diagnosis includes: (1) Calculate the degree of impact of image quality issues on the display of key anatomical structures; (2) Calculate the degree of impact of abnormalities in key anatomical structures on the diagnosis of common diseases; (3) The clinical severity score is obtained by combining the two levels of influence mentioned above.

[0009] Preferably, the clinical severity score is divided into four levels: Level 0: No impact; does not affect the diagnosis of any disease. Level 1: Minor impact, does not affect the diagnosis of the main disease; Level 2: Moderate impact, may affect the diagnosis of some diseases, reshoot recommended; Level 3: Severe impact, diagnosis impossible, re-exposure required.

[0010] Preferably, the step of generating improvement suggestions includes: (1) Based on the causal reasoning results, the root causes of quality problems are classified as equipment problems, operational problems, or patient problems; (2) Generate a natural language interpretation report to describe the specific manifestations of the quality problem, its causes, and its impact on diagnosis; (3) Provide specific suggestions for parameter adjustment or operation improvement based on the root cause.

[0011] A quality control system for intelligent DR images, comprising: The data acquisition module is used to acquire DICOM files of DR images and extract DR image data and DICOM header information; The causal model storage module is used to store pre-built causal models of the physical chain of DR imaging and a library of device features; The causal reasoning module is used to perform causal reasoning calculations to determine the type, severity, and root cause of image quality problems; The clinical assessment module is used to calculate the impact of quality issues on clinical diagnosis and generate a clinical severity score. The results output module is used to generate quality control classification results and improvement suggestions.

[0012] (III) Beneficial Effects This invention provides a quality control classification method for intelligent DR images. It has the following beneficial effects: 1. This invention analyzes DR image quality based on the actual physical processes of X-ray imaging, moving beyond judging quality solely by image appearance. It accurately identifies the root cause of image quality problems. By establishing feature libraries and corresponding analysis modules for different brands and models of DR equipment, it eliminates the need to retrain models for each device, enabling stable use on various brands of DR equipment. This is suitable for scenarios where multiple brands of equipment are used in combination in primary healthcare institutions. Furthermore, by combining the actual needs of clinical diagnosis to assess image quality, it can distinguish between serious problems that affect disease diagnosis and minor problems that do not, effectively reducing unnecessary re-images and lowering the radiation dose received by patients. This analysis method can directly provide clear causes of quality problems and specific improvement suggestions, making it easier for doctors and technicians to understand and accept the AI's quality control results. Detailed Implementation

[0013] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Example 1: The core of the intelligent DR image quality control classification method of this invention is to construct a causal reasoning framework based on the physical laws of X-ray imaging, thereby achieving a technological leap from "image feature correlation judgment" to "causal reasoning of the imaging process". The specific implementation steps are as follows: S1. Based on the fundamental physical principle of X-ray-matter interaction, a causal model of a directed acyclic graph (DAG) structure containing four levels of nodes is established. The nodes at each level are defined as follows: Equipment parameter node (input layer): contains two types of variables: discrete variables and continuous variables. Discrete variables include equipment brand, equipment model, and detector type; continuous variables include tube voltage (kV), tube current (mAs), exposure time (ms), and source-image distance (SID).

[0015] Imaging process nodes (intermediate physical layer): used to characterize the complete physical process of X-rays from generation to reception by the detector, including four continuous variables: incident X-ray intensity, scattered radiation ratio, detector response value, and quantum noise level. The relationship between these variables strictly follows the physical laws of X-ray imaging.

[0016] Image quality node (intermediate feature layer): used to characterize the quality features of the final generated DR image, including four continuous variables: overall grayscale mean, contrast, spatial resolution, and noise variance, as well as three discrete variables: artifact type, artifact location, and artifact area ratio.

[0017] Clinical impact node (output layer): used to characterize the actual impact of image quality on clinical diagnosis, including continuous variables of key anatomical structure visibility and discrete variables of clinical severity.

[0018] A device feature library was constructed, collecting inherent imaging characteristic parameters of different brands and models of DR devices, including detector response curves, noise models, and contrast transfer functions. For each type of device, a corresponding causal model submodule was trained. The causal connections between nodes were determined using a structure learning algorithm, and the strength coefficients of each causal connection were determined using a parameter learning algorithm.

[0019] S2. Obtain the standard DICOM file of the DR image to be evaluated, and extract equipment parameters and imaging parameters such as equipment model, tube voltage, tube current, exposure time, and source-image distance from the DICOM header information; extract the DR image pixel data from the DICOM file and convert it into a grayscale image.

[0020] The extracted DR images are subjected to standardized preprocessing, including: adjusting the grayscale using standard window widths and levels corresponding to the examination areas; linearly mapping pixel values ​​to a unified range; removing high-frequency noise from the images using Gaussian filtering; and normalizing the image size to a fixed size.

[0021] Multi-dimensional image features are extracted from the preprocessed DR images, including: statistical features (gray-level mean, gray-level variance, gray-level histogram quantiles); texture features (contrast, correlation, energy, and entropy of the gray-level co-occurrence matrix); and edge features (edge ​​intensity and orientation distribution).

[0022] S3. Based on the device model extracted from the DICOM header information, call the corresponding causal model submodule from the pre-built device feature library and load the inherent imaging characteristic parameters of the device.

[0023] The extracted device parameters are input into the device parameter node of the causal model, and the extracted image features are input into the image quality node of the causal model. A causal inference algorithm is used to calculate the causal effects between nodes. Intervention analysis is used to determine the root cause of image quality anomalies, and the type, location, and severity of the quality problem are output.

[0024] The calculation of the severity of quality problems takes into account three factors: the intensity of causal effect, the proportion of artifact area, and the location of artifact. Among them, the location weight of critical anatomical areas is higher than that of non-critical areas.

[0025] S4. Using a pre-trained anatomical structure segmentation model, key anatomical structures corresponding to the examination area are segmented from the DR images. The image quality index of each key anatomical structure region is calculated and compared with the corresponding index of the normal image of the same device to obtain the display score of each anatomical structure.

[0026] Based on a clinical knowledge base constructed by radiology experts, this knowledge base includes the correspondence between common diseases and key anatomical structures, as well as the influence weight of different visualization levels on disease diagnosis. According to the visualization scores and corresponding weights of each anatomical structure, the degree of influence of quality issues on the diagnosis of each common disease is calculated, and the maximum value of the influence on the diagnosis of all diseases is taken as the final clinical severity score.

[0027] The clinical severity score is divided into four levels: Level 0 (no impact), Level 1 (minor impact), Level 2 (moderate impact), and Level 3 (severe impact).

[0028] S5. Generate quality control classification results based on clinical severity scores, where Level 0 and Level 1 are considered qualified images, and Level 2 and Level 3 are considered unqualified images.

[0029] Based on the root causes derived from causal reasoning, quality problems are categorized into three types: equipment problems, operational problems, or patient problems. A natural language explanation report is generated, describing the specific manifestations of the quality problems, their causes, and their impact on clinical diagnosis, and providing specific, actionable improvement suggestions for addressing the root causes.

[0030] Example 2: This invention also provides an intelligent DR image quality control system for implementing the above method. The system adopts a modular design, and the functions of each module are as follows: Data acquisition module: Supports the DICOM 3.0 standard protocol, and can automatically acquire DICOM files of DR images from the hospital PACS system, and parse and extract DR image data and DICOM header information.

[0031] Causal model storage module: Used to store pre-built DR imaging physical chain causal models, device feature libraries and clinical knowledge bases, supporting model version management and incremental updates.

[0032] Causal Reasoning Module: The core processing module, used to perform causal reasoning calculations to determine the type, severity, and root cause of image quality problems.

[0033] Clinical assessment module: used to perform anatomical segmentation and clinical severity calculation, and generate clinically relevant assessment results.

[0034] Results output module: Used to generate standardized quality control reports, supporting integration with hospital HIS, RIS and PACS systems, automatically displaying quality control results and improvement suggestions on the doctor's workstation.

[0035] The system supports both in-hospital and cloud deployment modes. In in-hospital deployment mode, all data processing is completed on the hospital's internal servers, ensuring medical data security. In cloud deployment mode, data encryption and access control technologies are used, allowing multiple hospitals to use the system simultaneously, reducing deployment and maintenance costs.

[0036] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A quality control classification method for intelligent DR images, characterized in that, Includes the following steps: S1. Construct a causal model of the physical chain of DR imaging. The causal model is a directed acyclic graph structure that includes device parameter nodes, imaging process nodes, image quality nodes, and clinical impact nodes. S2. Obtain the DICOM file of the DR image to be evaluated, and extract the DR image data and the corresponding DICOM header information from it; S3. Extract device parameters and imaging parameters from the DICOM header information and input them into the device parameter node of the causal model; extract image features from the DR image data and input them into the image quality node of the causal model; S4. Calculate the causal effects between nodes using a causal reasoning algorithm to determine the type, severity, and root cause of image quality problems; S5. Based on the causal reasoning results, calculate the degree of impact of quality problems on clinical diagnosis and generate a clinical severity score; S6. Generate quality control classification results based on clinical severity scores, and generate improvement suggestions based on root causes.

2. The quality control classification method for intelligent DR images according to claim 1, characterized in that: The steps for constructing the causal model of the physical chain of DR imaging include: (1) Collect imaging data of DR equipment of different brands and models and build equipment feature library; (2) For each type of device, train a corresponding causal model submodule, wherein the submodule contains the inherent imaging characteristic parameters of the device; (3) During the inference phase, the device model is automatically identified based on the DICOM header information, and the corresponding causal model submodule is called for calculation.

3. The quality control classification method for intelligent DR images according to claim 1, characterized in that: The device feature library contains inherent imaging characteristic parameters including detector response curves, noise models, and contrast transfer functions.

4. The quality control classification method for intelligent DR images according to claim 1, characterized in that: The steps for calculating the impact of quality issues on clinical diagnosis include: (1) Calculate the degree of impact of image quality issues on the display of key anatomical structures; (2) Calculate the degree of impact of abnormalities in key anatomical structures on the diagnosis of common diseases; (3) The clinical severity score is obtained by combining the two levels of influence mentioned above.

5. The quality control classification method for intelligent DR images according to claim 1, characterized in that: The clinical severity score is divided into four levels: Level 0: No impact; does not affect the diagnosis of any disease. Level 1: Minor impact, does not affect the diagnosis of the main disease; Level 2: Moderate impact, may affect the diagnosis of some diseases, reshoot recommended; Level 3: Severe impact, diagnosis impossible, re-exposure required.

6. The quality control classification method for intelligent DR images according to claim 1, characterized in that: The steps for generating improvement suggestions include: (1) Based on the causal reasoning results, the root causes of quality problems are classified as equipment problems, operational problems, or patient problems; (2) Generate a natural language interpretation report to describe the specific manifestations of the quality problem, its causes, and its impact on diagnosis; (3) Provide specific suggestions for parameter adjustment or operation improvement based on the root cause.

7. A quality control system for intelligent DR images, characterized in that, include: The data acquisition module is used to acquire DICOM files of DR images and extract DR image data and DICOM header information; The causal model storage module is used to store pre-built causal models of the physical chain of DR imaging and a library of device features; The causal reasoning module is used to perform causal reasoning calculations to determine the type, severity, and root cause of image quality problems; The clinical assessment module is used to calculate the impact of quality issues on clinical diagnosis and generate a clinical severity score. The results output module is used to generate quality control classification results and improvement suggestions.