Head and neck tumor multi-modal image database construction method and computer storage medium
By collecting and integrating CT and MRI data of head and neck tumors from multiple centers, the problems of multimodality and incomplete information in existing databases have been solved, and a high-quality multimodal imaging database has been constructed to support AI model training and clinical applications, and to realize data resource sharing and scientific research cooperation.
Patent Information
- Application Number
- CN202511220018.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-28
- Publication Date
- 2026-01-09
AI Technical Summary
Existing head and neck tumor imaging databases suffer from limited sample size, are mostly limited to single-modality images, lack multimodal data, have incomplete clinical information, and have inconsistent annotation quality, which limits the application of artificial intelligence in the diagnosis and treatment of head and neck tumors.
By collecting CT and MRI data from head and neck tumor patients at multiple centers, integrating multi-scale multimodal information, performing data quality control and de-identification processing, unifying the resolution of resampled images, accurately delineating ROIs, generating high-quality labeled data, and integrating complete clinical information, a multimodal imaging database is formed.
A high-quality, multimodal, and clinically comprehensive image database has been constructed, providing reliable resources for the training and validation of artificial intelligence models, supporting scientific research collaboration and clinical translation in head and neck tumors, and filling the gaps in datasets.
Smart Images
Figure CN121306441A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of multimodal image database construction technology, and more specifically, to a method for constructing a multimodal image database for head and neck tumors and a computer storage medium. Background Technology
[0002] Head and neck tumors encompass areas such as the eyes, ears, nose, throat, oral and maxillofacial region, and neck. They are the seventh most common type of cancer worldwide, characterized by complex structures, diverse pathological types, and frequent overlap in imaging findings across different histological types. Furthermore, approximately 70-80% of patients with malignant tumors are diagnosed at an advanced stage, facing a high risk of recurrence and metastasis, with an overall survival rate of less than 50%. Therefore, precise diagnosis and treatment of head and neck tumors are of paramount importance.
[0003] In recent years, artificial intelligence (AI), especially radiomics and deep learning technologies, has made significant progress in head and neck tumor segmentation, diagnosis, efficacy evaluation, and prognostic prediction, offering possibilities for assisting clinical decision-making and improving patient outcomes. However, the training and validation of these models heavily rely on large-scale, high-quality labeled datasets. Currently, publicly available high-quality labeled datasets for head and neck tumors are extremely scarce, severely restricting the application and development of AI in this field. Furthermore, existing publicly available head and neck tumor image databases generally suffer from the following shortcomings: limited sample size; mostly limited to single-modality images, lacking multimodal image data; clinical information (such as treatment plans and recurrence status) is often incomplete; annotation quality varies; and data is highly concentrated on squamous cell carcinoma, with extremely scarce data on other pathological types. Therefore, constructing a large-scale, multimodal image database with complete clinical information and accurate annotation is crucial for the clinical translation and practical application of AI-based precision diagnosis and treatment of head and neck tumors.
[0004] The preceding description is intended to provide general background information and does not necessarily constitute prior art. Summary of the Invention
[0005] The purpose of this invention is to provide a method for constructing a multimodal image database for head and neck tumors and a computer storage medium. This method addresses the current need for AI-assisted accurate diagnosis of head and neck tumors and fills the gap in artificial intelligence databases for head and neck tumor images both domestically and internationally. At the same time, it provides high-quality, detailed, multi-labeled manually annotated data and detailed clinical information.
[0006] This invention provides a method for constructing a multimodal image database of head and neck tumors, the method comprising:
[0007] Relying on Picture Archiving and Communication System (PACS) and Hospital Information System (HIS), multi-center collection of CT and MRI data of head and neck tumor patients throughout the entire preoperative and postoperative period is conducted. Simultaneously, cross-scale multimodal information such as clinical, pathological, and genetic data is integrated, and each data is strictly verified to ensure data reliability and usability, so as to build a high-quality database that deeply matches CT, MRI and clinical data.
[0008] By de-identifying the images, all privacy information on patients' CT / MRI images is completely hidden, achieving reliable protection of sensitive privacy data. At the same time, a separate folder is created for each patient's multimodal image data and named with a unique number (ID) to facilitate data management and traceability.
[0009] CT and MRI images from different sources and at different resolutions are uniformly resampled to the standard voxel spacing to ensure spatial consistency of multimodal CT and MRI data;
[0010] Based on the location and type of tumor, all CT and MRI images are annotated, and the tumor ROI and metastatic lymph nodes are accurately delineated to generate annotation masks that can be used for training and validation.
[0011] All patients' CT and MRI images are matched one-to-one with their corresponding labels using a number, ensuring data traceability and accuracy.
[0012] Based on two main tags—tumor location and tumor type—CT and MRI images are classified and organized, and the latest data is updated in real time to ensure the timeliness and completeness of the data.
[0013] For unstructured data such as clinical, pathological, and genetic data, we complete data cleaning, transformation, and quality control to form a structured data format for subsequent analysis and processing.
[0014] The present invention also provides a computer storage medium including a computer program, which, when executed, performs the above-described method for constructing a multimodal image database of head and neck tumors.
[0015] The advantages of this invention are:
[0016] This invention provides a method for constructing a multimodal image database for head and neck tumors. It aggregates multi-center, multimodal CT and MRI images, supplemented by high-resolution, multi-label, and precisely drawn manually. Simultaneously, it integrates complete clinical data such as patient demographics, pathology, and genetics to form a high-quality image-clinical paired resource. This database provides a reliable foundation for AI model training and validation, and also establishes a shared public platform for research collaboration and clinical translation in head and neck tumors, effectively filling the long-standing gap in publicly available datasets in the field of artificial intelligence for head and neck tumor CT and MRI. Attached Figure Description
[0017] Figure 1 This is a flowchart provided for an embodiment of the present invention.
[0018] Figure 2 This is a schematic diagram of global image labeling and ROI delineation provided in an embodiment of the present invention. Detailed Implementation
[0019] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.
[0020] The terms "first," "second," "third," "fourth," etc., used in the specification and claims of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.
[0021] Example 1
[0022] Figure 1 This is a flowchart provided for an embodiment of the present invention. Please refer to... Figure 1 The method for constructing a multimodal image database for head and neck tumors provided in this embodiment of the invention includes the following steps:
[0023] S1: Collect CT and MRI data of head and neck tumors, as well as related clinical, pathological, and genetic multi-scale multimodal data through the Picture Archiving and Communication System (PACS) and Hospital Information System (HIS), and perform data quality control;
[0024] Specifically, step S1 includes:
[0025] S11: Multicenter, retrospective and prospective collection of CT and MRI images of patients with pathologically confirmed or clinically diagnosed head and neck tumors before and after surgery, and removal of images with metal or motion artifacts;
[0026] The CT and MRI imaging data include CT and MRI images before and after treatment, scanning machine type, scanning parameters, contrast agent type, etc.
[0027] S12: Query patient medical records and collect general information, medical history, laboratory test results, pathological data, genetic and follow-up information of all patients;
[0028] Specifically, general information includes gender, age, smoking history, past medical history, family history, etc.; medical records include admission and discharge records, surgical records, daily medical records, radiotherapy and chemotherapy regimens, targeted or immunotherapy records, etc.; laboratory tests include blood biochemistry, blood glucose, HPV typing, EBV-DNA, tumor markers, etc.; pathological data include pathological typing, TNM staging and immunohistochemical indicators; genetic information includes whole exon / targeted sequencing results, etc.; and follow-up information includes efficacy evaluation, adverse reaction records, survival status and recurrence and metastasis time points, etc.
[0029] S2: Perform data anonymization and data desensitization processing on the acquired CT and MRI images;
[0030] Specifically, step S2 includes:
[0031] S21: Replace the content involving patient privacy information in the header files of the collected CT and MRI raw data with random characters, and only retain the multimodal CT and MRI data and disease diagnosis information;
[0032] S22: Transform the multimodal CT and MRI data according to the desensitization rules, create an independent folder for the multimodal image data of each patient, and name it using a number (ID);
[0033] It should be noted that the naming rules for multimodal CT and MRI images are as follows: CT and MRI images include CT and MRI images. Create two independent folders under the folder named with the ID, and name them CT and MRI respectively. Store the CT and MRI images in the folders. If there are multiple CT / MRI images, name them according to the examination date in the folder named with CT / MRI, such as "2023-01-01".
[0034] Figure 2 This is a schematic diagram of global image labeling and ROI delineation provided in an embodiment of the present invention. Further reference... Figure 1 , Figure 2 Step S3 of the present invention: All CT and MRI images are standardized, including data resampling and data annotation;
[0035] Specifically, step S3 includes:
[0036] S31: Resample all CT and MRI images collected from multiple centers at a fixed isomorphic resolution. Use the SimpleITK toolkit to resample all samples to the same size (1*1*1) and save them uniformly as nii.gz format data to the folder corresponding to each patient.
[0037] S32: Based on the location and type of the tumor, all resampled CT and MRI images are labeled and saved in .xml format.
[0038] It should be noted that, based on the location of the tumor, all resampled CT and MRI images are globally labeled, such as eye, ear, sinus, nasopharynx, larynx, etc.; the global labels are stored in the corresponding folder in .xml format.
[0039] S33: Head and neck tumor ROI and metastatic lymph nodes are manually delineated using ITK-SNAP software based on CT / MRI images, exported and saved in nii.gz format, and stored in the corresponding CT / MRI folder.
[0040] It should be noted that for cases with a clear pathological diagnosis, the annotations are based on the pathological type, such as squamous cell carcinoma, adenoid cystic carcinoma, etc. For prospective datasets, the annotations are based on clinical diagnoses, and are updated during subsequent follow-up based on the case information obtained at the earliest opportunity.
[0041] S4: Multimodal CT and MRI images and their corresponding annotations are categorized, organized, and updated in real time;
[0042] Specifically, step S4 includes:
[0043] S41: Systematically classify and organize all CT and MRI images according to the global labels marked in step S3;
[0044] S42: Detailed subclassification based on tumor pathology type, and establishment of a real-time update mechanism; each folder in the subclass should include the patient's CT / MRI images, corresponding tumor ROI markings, and metastatic lymph node markings.
[0045] S5: Cleaning, transforming, and quality-controlling unstructured data.
[0046] Specifically, step S5 includes:
[0047] S51: Organize each patient's general information, laboratory tests, CT and MRI parameters and diagnoses, pathological data, genetic information and treatment plans, and store them in the database in tabular form; and hide all patients' private information and correspond them to CT and MRI data by number.
[0048] For example, the ID0001 folder is used to show the composition of the database: original CT / MRI images (DICOM format), resampled images (nii.gz format); four categories of labels (.xml format) for tumor location, pathological classification, recurrence and lymph node metastasis; and tumor ROI segmentation labels (nii.gz format).
[0049] Example 2
[0050] This embodiment provides a computer storage medium including a computer program, which, when executed, performs the above-described method for constructing a multimodal image database of head and neck tumors.
[0051] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A method for constructing a multimodal image database for head and neck tumors, characterized in that, The construction method includes the following steps: S1: Collect CT and MRI data of head and neck tumors, as well as related clinical, pathological, and genetic multi-scale multimodal data, through image archiving and communication systems and hospital information systems, and perform data quality control; S2: Perform data anonymization and data desensitization processing on the acquired CT and MRI images; S3: All CT and MRI images are standardized, including data resampling and data annotation; S4: Multimodal CT and MRI images and their corresponding annotations are categorized, organized, and updated in real time; S5: Cleaning, transforming, and quality-controlling unstructured data.
2. The method for constructing a multimodal image database for head and neck tumors according to claim 1, characterized in that, Step S1 includes: S11: Multicenter, retrospective and prospective collection of CT and MRI images of patients with pathologically confirmed or clinically diagnosed head and neck tumors before and after surgery, and removal of images with metal or motion artifacts; S12: Query patient medical records and collect general information, medical history, laboratory tests, pathological data, genetic information and follow-up information of all patients.
3. The method for constructing a multimodal image database for head and neck tumors according to claim 2, characterized in that, The general information mentioned in step S12 includes gender, age, smoking history, past medical history, and family history. The medical records include admission and discharge records, surgical records, daily medical records, radiotherapy and chemotherapy regimens, and targeted or immunotherapy records. The laboratory tests include blood biochemistry, blood glucose, HPV typing, EBV-DNA, and tumor markers. The pathological data includes pathological typing, TNM staging, and immunohistochemical indicators. The genetic information includes whole exon / targeted sequencing results. The follow-up information includes efficacy evaluation, adverse reaction records, survival status, and recurrence and metastasis time points.
4. The method for constructing a multimodal image database for head and neck tumors according to claim 1, characterized in that, Step S2 includes: S21: Replace the content involving patient privacy information in the header files of the collected CT and MRI raw data with random characters, and only retain the multimodal CT and MRI data and disease diagnosis information; S22: Transform multimodal CT and MRI data according to desensitization rules to achieve reliable protection of sensitive privacy data. Create an independent folder for each patient's multimodal image data and name it using a number.
5. The method for constructing a multimodal image database for head and neck tumors according to claim 1, characterized in that, Step S3 includes: S31: Resample all CT and MRI images collected from multiple centers at a fixed isomorphic resolution, resample all samples to the same size, and save them uniformly as nii.gz format data; S32: Based on the location and type of the tumor, all resampled CT and MRI images are labeled and saved in .xml format. S33: Head and neck tumor ROI and metastatic lymph nodes are manually delineated using ITK-SNAP software based on CT / MRI images, exported and saved in nii.gz format, and stored in the corresponding CT / MRI folder.
6. The method for constructing a multimodal image database for head and neck tumors according to claim 1, characterized in that, Step S4 includes: S41: Systematically classify and organize all CT and MRI images according to the global labels marked in step S3; S42: Detailed subclassification based on tumor pathology type, and establishment of a real-time update mechanism; each folder in the subclass should include the patient's CT / MRI images, corresponding tumor ROI markings, and metastatic lymph node markings.
7. The method for constructing a multimodal image database for head and neck tumors according to claim 1, characterized in that, Step S5 includes: S51: Organize each patient's general information, laboratory tests, CT and MRI parameters and diagnoses, pathological data, genetic information and treatment plans, and store them in the database in tabular form; and hide all patients' private information and correspond them to CT and MRI data by number.
8. A computer storage medium, characterized in that, The method includes a computer program, which, when executed, performs the head and neck tumor multimodal image database construction method according to any one of claims 1 to 7.