Medical image quality control report generation method and system and storage medium
By constructing a multimodal medical image database and a large visual language model, the system automatically diagnoses and generates quality control reports in a unified format, solving the problem of non-recognition of medical image examination results and achieving standardized credentials and resource optimization.
Patent Information
- Application Number
- CN202410967641.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-18
- Publication Date
- 2024-11-08
AI Technical Summary
The lack of mutual recognition of domestic medical imaging examination results forces patients to undergo repeated scans, resulting in a waste of medical resources and health consequences. There is also a lack of unified and rapid model methods for real-time quality control processing.
A multimodal medical image database was constructed, and a large visual language model was used to automatically diagnose and generate quality control reports, including integrity scans, metal foreign bodies and artifacts, respiratory motion artifacts and dose noise checks. The OpenCLIP library and ConvNeXt/Transformer architecture were used to align and train image and text encoders to form a unified format of quality control reports.
It has enabled standardized credentials for medical images, solved the problem of non-recognition of examination results between different hospitals, optimized the radiology scanning process, helped patients seek medical treatment in a timely manner, and saved medical resources.
Smart Images

Figure CN118918079A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of medical image analysis, and in particular relates to a method, system and storage medium for generating a medical image quality control report. Background Art
[0002] At present, there is a serious problem of non-mutual recognition of medical imaging examination results in China. Patients who go to multiple hospitals often need to repeatedly scan CT, MRI and other medical images, which not only causes a serious waste of medical resources, but also has a negative impact on the physical and mental health of patients, and even affects the progress of disease diagnosis and treatment. Designing new methods and new ideas for medical image homogeneity and mutual recognition of examination results is an imperative path.
[0003] Currently, there is a great lack of relevant research work. Preliminary research work (such as the Chinese invention patent application: CN116982999A Multimodal CT Image Homogenization Auxiliary Method and System) has provided a preliminary solution for medical image quality control, but it does not involve the mutual recognition of quality control results, and only targets CT image processing. Different inspection modules use different algorithms for processing, which brings challenges to the real-time quality control processing of medical images. There is a lack of a unified and fast model method to generate quality control reports from images in real time. Summary of the invention
[0004] In view of the problem that the existing technology lacks solutions for promoting mutual recognition of medical imaging examination results, the present invention provides a medical imaging quality control report generation method, system and storage medium to assist radiologists in achieving full-process, real-time medical imaging quality control and generation of imaging quality control reports and promotion of mutual recognition of examination results.
[0005] In order to achieve the above technical purpose, the technical solution adopted by the present invention is:
[0006] The method for generating a medical image quality control report includes the following specific steps:
[0007] S1, build a multimodal medical imaging database, and have professional radiologists perform image quality inspections to form a text description of the quality inspection results; image quality inspections include: integrity scan inspection, metal foreign body and artifact inspection, respiratory motion artifact inspection, and dose noise inspection;
[0008] S2, based on paired medical images and quality inspection text descriptions, build and train a large visual language model for medical image quality control;
[0009] S3, obtains the target medical image, uses the visual language large model to automatically diagnose and generate a text description of the quality inspection results, forms a quality control text report in a unified format and exports it.
[0010] Preferably, the multimodal medical image database includes CT and MRI images of various parts of the body, and the professional radiologist has more than 10 years of work experience.
[0011] Preferably, the integrity scan check is to check whether the corresponding organs in the medical image are fully scanned; the metal foreign body and artifact check is to check whether there are metal foreign bodies or artifacts in the medical image, and record the CT layer number where the foreign bodies or artifacts are located; the respiratory motion artifact check is to check whether there are respiratory and motion artifacts in the medical image, and record the CT layer number where they are located; the dose noise check is to check the clarity and signal-to-noise ratio of the medical image; all inspection results are recorded to form a text description of the quality inspection results.
[0012] Preferably, the visual language large model uses the OpenCLIP library to implement the visual language model algorithm, uses CLIP weights to initialize model parameters, inputs medical images through an image encoder, and inputs text descriptions of quality inspection results through a text encoder, aligns and trains the vector embeddings of the two to obtain the final visual language large model.
[0013] Furthermore, the image encoder adopts the ConvNeXt network architecture, the text encoder adopts the Transformer architecture, and uses CLIP BPE text tagging technology for text processing.
[0014] Furthermore, the alignment and training of vector embeddings uses cosine similarity to optimize the loss function to learn the mapping relationship between medical images and text descriptions.
[0015] The present invention also provides a medical image quality control report generation system, which is used to execute the above-mentioned medical image quality control report generation method, comprising:
[0016] A medical image input module, used to read target medical image data;
[0017] A medical image quality control module, wherein the medical image quality control module is embedded with a trained visual language large model of medical image quality control, and is used for automatically diagnosing and generating a text description of the quality inspection results based on the input target medical image data;
[0018] The quality control report output module is embedded with a fixed quality control report template. It extracts keywords through text and is used to form a quality control report in a unified format and export it from the text description generated by the visual language large model.
[0019] The present invention also provides a computer-readable storage medium, on which is stored at least one computer program, and when the computer program is executed by a processor, the steps of the above-mentioned method for generating a medical image quality control report are implemented.
[0020] The beneficial effects of the present invention are:
[0021] The medical image quality control report generation method, system and storage medium of the present invention are aimed at the quality control of multimodal medical images such as CT and MRI. By constructing a large visual language model for medical image quality control, it can automatically perform integrity scanning inspection, metal foreign body and artifact inspection, respiratory motion artifact inspection and dose noise inspection on medical images, and automatically form an image quality control report in a unified format, providing homogeneous credentials for medical images, which can help solve the problem of non-mutual recognition of examination results between different hospitals, greatly optimize the scanning process of the radiology department, facilitate patients to see a doctor in time, and save huge financial and material resources for the country and society.
[0022] The present invention provides a new method and new idea for achieving domestic medical image homogeneity and mutual recognition of examination results, which can meet the actual needs of society and clinical application and has good application prospects in most parts of the country.
[0023] Obviously, according to the above contents of the present invention, in accordance with common technical knowledge and customary means in the art, without departing from the above basic technical ideas of the present invention, other various forms of modification, replacement or change may be made. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 Schematic diagram of a method for generating a medical image quality control report according to Embodiment 1 of the present invention.
[0025] Figure 2 Schematic diagram of a medical image quality control report generation system according to Embodiment 2 of the present invention.
[0026] Figure 3 A schematic diagram of an image quality control report generated for an experimental example of the present invention. DETAILED DESCRIPTION
[0027] The above contents of the present invention are further described in detail below through specific implementation methods in the form of embodiments. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following examples. All technologies realized based on the above contents of the present invention belong to the scope of the present invention.
[0028] It should be noted that the algorithms of data collection, transmission, storage and processing steps not specifically described in the embodiments, as well as the hardware structure, circuit connection, etc. not specifically described can all be implemented through the contents disclosed in the prior art.
[0029] Example 1 Method for Generating Medical Image Quality Control Report
[0030] The flow chart of the method for generating a medical image quality control report in this embodiment is as follows: Figure 1 As shown, the specific steps include:
[0031] S1, build a large-scale multimodal medical imaging database, and have professional radiologists perform image quality inspections, including: integrity scan inspection, metal foreign body and artifact inspection, respiratory motion artifact inspection, dose noise inspection; and form a text description of the quality inspection results.
[0032] The large-scale multimodal medical imaging database includes CT and MRI images of all parts of the body, with a data volume of tens of thousands. The professional radiologist has more than 10 years of work experience and performs quality inspection and annotation of images layer by layer.
[0033] The integrity scan check is to check whether the corresponding organs are fully scanned; the metal foreign body and artifact check is to check whether there are metal foreign bodies or artifacts in the image, and record the CT layer number where the foreign bodies or artifacts are located; the respiratory motion artifact check is to check whether there are respiratory and motion artifacts in the image, and record the CT layer number where they are located; the dose noise check is to check the clarity and signal-to-noise ratio of the image; all inspection results are recorded to form a text description of the quality inspection results.
[0034] S2, based on paired medical images and quality inspection text descriptions, build and train a large visual language model for medical image quality control.
[0035] The visual language model uses the OpenCLIP library to implement the visual language model algorithm, uses CLIP weights to initialize model parameters, inputs medical images through an image encoder, and inputs text descriptions of quality inspection results through a text encoder, aligns and trains the vector embeddings of the two to obtain the final visual language model.
[0036] Preferably, the image encoder adopts the ConvNeXt network architecture, the text encoder adopts the Transformer architecture, and uses the CLIP BPE text tagging technology for text processing. The alignment and training learning of vector embedding uses cosine similarity to optimize the loss function to learn the mapping relationship between medical images and text descriptions.
[0037] S3, obtains the target medical image, uses the visual language large model to automatically diagnose and generate a text description of the quality inspection results, forms a quality control report in a unified format and exports it.
[0038] Example 2 Medical Image Quality Control Report Generation System
[0039] The medical image quality control report generation system disclosed in this embodiment is used to execute the above-mentioned medical image quality control report generation method, such as Figure 2 As shown, specifically including:
[0040] Medical image input module, used to read medical image data of any modality in real time;
[0041] A medical image quality control module, wherein the medical image quality control module is embedded with a trained visual language large model of medical image quality control, and is used for automatically diagnosing and generating a text description of the quality inspection results based on the input target medical image data;
[0042] The quality control report output module is embedded with a fixed quality control report template. It extracts keywords through text and is used to form a quality control report in a unified format from the text description generated by the visual language large model and export it.
[0043] Embodiment 3 A computer readable storage medium
[0044] This embodiment provides a computer-readable storage medium, on which at least one computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for generating a medical image quality control report are implemented.
[0045] Experimental Example 1
[0046] This experimental example is a case of generating a medical image quality control report by executing the above-mentioned medical image quality control report generation method through the above-mentioned medical image quality control report generation system.
[0047] This experiment takes chest CT as an example. The processing results of image quality control using the above-mentioned medical image quality control report generation method and medical image quality control report generation system are as follows:
[0048] S1, inputting the target chest CT image into the medical image quality control module embedded with the visual language large model of medical image quality control through the medical image input module;
[0049] S2, the medical image quality control module automatically implements integrity scanning inspection, metal foreign body and artifact inspection, respiratory motion artifact inspection and dose noise inspection through the visual language large model, and automatically generates a text description of the inspection results;
[0050] S3, the quality control report output module automatically generates an image quality control report based on text descriptions and fixed report templates, forms a PDF report file and exports it; the image quality control report results output by this experimental example are as follows Figure 3 shown.
[0051] In this experimental example, the final image quality control report generated generally concluded that: the lung scan was complete; suspected metal artifacts were detected; suspected respiratory motion artifacts were detected; and the dose noise check failed. The image quality was unqualified and the image results could not be mutually recognized.
[0052] Of course, the present invention may have many other embodiments. Without departing from the spirit and essence of the present invention, technicians familiar with the field may make various corresponding changes and deformations based on the present invention, but these corresponding changes and deformations should all fall within the scope of protection of the claims attached to the present invention.
Claims
1. A method for generating a medical image quality control report, characterized in that: The specific steps include: S1, build a multimodal medical imaging database, and have professional radiologists perform image quality inspections to form a text description of the quality inspection results; image quality inspections include: integrity scan inspection, metal foreign body and artifact inspection, respiratory motion artifact inspection, and dose noise inspection; S2, based on paired medical images and quality inspection text descriptions, build and train a large visual language model for medical image quality control; S3, obtains the target medical image, uses the visual language large model to automatically diagnose and generate a text description of the quality inspection results, forms a quality control text report in a unified format and exports it.
2. The method for generating a medical image quality control report according to claim 1, characterized in that: The multimodal medical image database includes CT and MRI images of various parts of the body, and the professional radiologist has more than 10 years of work experience.
3. The method for generating a medical image quality control report according to claim 1, characterized in that: The integrity scan check is to check whether the corresponding organs in the medical image are fully scanned; the metal foreign body and artifact check is to check whether there are metal foreign bodies or artifacts in the medical image, and record the CT layer number where the foreign bodies or artifacts are located; the respiratory motion artifact check is to check whether there are respiratory and motion artifacts in the medical image, and record the CT layer number where they are located; the dose noise check is to check the clarity and signal-to-noise ratio of the medical image; all inspection results are recorded to form a text description of the quality inspection results.
4. The method for generating a medical image quality control report according to claim 1, characterized in that: The visual language model uses the OpenCLIP library to implement the visual language model algorithm, uses CLIP weights to initialize model parameters, inputs medical images through an image encoder, and inputs text descriptions of quality inspection results through a text encoder, aligns and trains the vector embeddings of the two to obtain the final visual language model.
5. The method for generating a medical image quality control report according to claim 4, characterized in that: The image encoder adopts the ConvNeXt network architecture, the text encoder adopts the Transformer architecture, and uses the CLIP BPE text tagging technology for text processing.
6. The method for generating a medical image quality control report according to claim 4, characterized in that: The alignment and training of vector embeddings uses cosine similarity to optimize the loss function to learn the mapping relationship between medical images and text descriptions.
7. A medical image quality control report generation system, used to execute the medical image quality control report generation method according to any one of claims 1 to 6, characterized in that: include: A medical image input module, used to read target medical image data; A medical image quality control module, wherein the medical image quality control module is embedded with a trained visual language large model of medical image quality control, and is used for automatically diagnosing and generating a text description of the quality inspection results based on the input target medical image data; The quality control report output module is embedded with a fixed quality control report template. It extracts keywords through text and is used to form a quality control report in a unified format and export it from the text description generated by the visual language large model.
8. A computer-readable storage medium, characterized in that: The storage medium stores at least one computer program, and when the computer program is executed by the processor, the steps of the method for generating a medical image quality control report according to any one of claims 1 to 6 are implemented.
Citation Information
Cited By
Image report quality control method and device based on large model, and storage medium
CN121439071A
Automatic segmentation quality control method and related equipment
CN121505397A
Automatic segmentation quality control method and related devices
CN121505397B
