A stroke data screening collection management system supporting large model training

Abstract

The application provides a stroke data screening collection management system supporting large model training, relates to the technical field of medical data management, and solves the problems of the lack of whole-process management of current stroke screening data and the inability to provide compliant high-quality data sets for medical large model training. In the system, the screening collection module is based on a standardized stroke screening evaluation scale, disassembles an audio and video collection subtask, executes an audio and video collection process, and stores data in a local terminal database; the local data query module performs local query playback and packs and exports data in an offline state, and uploads data to a cloud server in an online state; the online data query module queries and plays back screening records from the cloud and manages cloud data; the data import module receives local and cloud data, verifies them, and then archives them; and the large model training data output interface module converts and packs the archived data and corresponding results into a format, and outputs them as a medical large model training data set.

Description

Technical Field

[0001] This invention relates to the field of medical data management technology, specifically to a stroke data screening and collection management system that supports large model training. Background Technology

[0002] Stroke, an acute cerebrovascular disease characterized by high incidence, high disability rate, and high mortality, requires early and accurate screening to reduce its harm and improve treatment outcomes. In recent years, audio-visual analysis technology based on patients' facial, limb, and speech characteristics has become a core development direction for non-invasive and rapid stroke screening. In this process, the standardization of screening audio-visual acquisition, the controllability of data quality, multi-scenario acquisition capabilities, and the efficiency of data conversion into artificial intelligence training capabilities are the core factors determining screening effectiveness and subsequent technological iterations. In particular, the training of large-scale medical models places extremely high demands on data compliance, high quality, and handling of heterogeneity.

[0003] Currently, the mainstream implementation model for stroke screening in China still relies on paper-based questionnaires and on-site recording. Medical staff typically use standardized screening assessment scales to check and fill in the patient's various vital signs assessment results on-site. This traditional model lacks dedicated information and systematic data management system support throughout the process, and can only complete static result recording. This directly leads to a severe lack of data dimensions, greatly limiting the accuracy of screening and resulting in low overall management efficiency. At the same time, the paper-based recording method severely lacks data traceability and compliance, and the potential value of massive amounts of clinical data cannot be effectively released, making it difficult to meet the needs of modern medical data's refined management.

[0004] To compensate for the inability of paper records to capture the dynamic process of vital signs, a few medical institutions have attempted to record patients' vital sign responses during screening using civilian devices such as smartphones, tablets, or ordinary cameras. However, these institutions lack a dedicated audio and video data management system for stroke screening. The recorded videos are often stored sporadically on the personal devices of medical staff, failing to be incorporated into a standardized and regulated end-to-end management process. This model lacks both unified collection standards and supporting mechanisms for data storage, retrieval, verification, archiving, and security control, thus failing to establish a complete information chain from standardized collection to systematic management and compliant application.

[0005] Furthermore, existing audio and video capture methods heavily rely on civilian terminals and lack standardized capture management and real-time quality control mechanisms. This results in frequent issues such as insufficient resolution, image jitter, and incomplete images of key vital signs, making it difficult to meet overall data quality standards and ultimately unsuitable for accurate clinical diagnosis and remote review. This low-quality, poorly managed, and fragmented data cannot be transformed into compliant, high-quality datasets, leading to severe data heterogeneity and high annotation costs during the training of large-scale medical models. Therefore, the industry urgently needs a system that can enhance dedicated standardized audio and video capture, adapt to different network environments and multi-scenario applications, and achieve efficient end-to-end management to improve the efficiency and accuracy of on-site screening and provide reliable core data support for the training and iteration of large-scale medical models. Summary of the Invention

[0006] The purpose of this invention is to address the problems in existing stroke screening methods, such as the lack of unified standards and real-time quality control in data collection, and the absence of end-to-end management, which leads to insufficient clinical diagnostic support and the inability to provide compliant, high-quality datasets for training large-scale medical models. This invention proposes a stroke data screening and collection management system that supports large-scale model training. It achieves integrated management of patient basic information, standardized video screening and collection processes, and the entire lifecycle of screening data. It supports both offline standalone operation and networked data synchronization management, and provides compliant, high-quality datasets for training large-scale medical models, including intelligent stroke recognition models.

[0007] The present invention employs the following technical solutions to achieve its objective: A stroke data screening, collection, and management system that supports large-scale model training includes the following functional modules: The screening and acquisition module is configured to be based on a standardized stroke screening and assessment scale. It breaks down the assessment items into standardized audio and video acquisition sub-tasks and dynamically adjusts the execution steps of the acquisition sub-tasks according to the patient's real-time feedback. The screening and acquisition module is also used to execute the audio and video acquisition process and store the acquired patient information and screening audio and video data in the local terminal database. The local data query module is configured to query screening records in the local terminal database and play back the corresponding screening audio and video data; the local data query module is also used to package and export screening data in offline mode, and to upload locally stored screening data to the cloud server in online mode; The online data query module is configured to query screening records from the cloud server and play back the corresponding screening audio and video data in online mode; the online data query module is also used to manage screening data stored in the cloud. The data import module is configured to receive standardized data packets from a local terminal database and / or a cloud server, and to perform data validity verification on the standardized data packets; the data import module is also used to archive the verified standardized data packets into the database. The large model training data output interface module is configured to acquire standardized and legally verified screening audio and video data and corresponding evaluation results; the large model training data output interface module is also used to convert and package the screening audio and video data and evaluation results into a medical large model training dataset.

[0008] Preferably, the system also includes a login module, which is configured to provide system login verification and support for manual switching between online and local modes; the login module is also configured to implement an institutional data isolation mechanism and hierarchical permission control to isolate stored data from different institutions; the login module is also configured to display the patient directory and system function operation interface within the corresponding permission scope according to the permission level of the logged-in user.

[0009] Preferably, the screening and acquisition module is further configured to establish basic patient information by creating a new patient file before executing the audio and video acquisition process, and to automatically create a file after reading the patient's identity information by connecting an external ID card reader; the screening and acquisition module is further configured to provide compatibility support for different models or types of cameras, handheld terminals and industrial acquisition terminals, and to receive audio and video acquisition signals transmitted from different hardware terminals; the screening and acquisition module is further configured to display the patient list within the permissions before the acquisition starts, and to provide an interface for selecting or creating a patient list.

[0010] Furthermore, the screening and data collection module is configured to break down facial nerve function assessment, limb motor function assessment, speech function assessment, coordination assessment, and sensory function assessment into corresponding data collection sub-tasks based on a standardized stroke screening and assessment scale. The screening and data collection module is also configured to preset unified data collection specifications and assessment standards for each data collection sub-task. The screening and data collection module is also configured to adjust the execution steps and prompts of the data collection tasks based on the patient's real-time complaints, physical signs, and physical tolerance, and to skip unnecessary data collection items without relevant symptoms.

[0011] Furthermore, the screening and acquisition module is also configured to provide real-time preview and verification of the acquisition screen during the audio and video acquisition process when performing acquisition sub-tasks, and to provide support for pausing and resuming acquisition from breakpoints during the acquisition process; the screening and acquisition module is also configured to bind and associate the assessment results of the standardized stroke screening assessment scale with the corresponding acquired screening audio and video data one by one after the acquisition is completed; the screening and acquisition module is also configured to store the bound assessment results and screening audio and video data as a whole in the local terminal database.

[0012] Specifically, the local data query module is also configured to encapsulate patient information and screening audio and video data into standardized data packets when packaging and exporting screening data in offline mode, and to provide support for batch export of standardized data packets; The online data query module is also configured to provide retrieval and online playback of screening records and screening audio and video data when managing screening data stored in the cloud in online mode; the online data query module is also configured to provide support for batch deletion of screening data stored in the cloud.

[0013] Specifically, the data import module is also configured to provide a selection interface for standardized data packets and receive the selected standardized data packets when receiving them; the data import module is also configured to verify the data integrity, format specifications, and security encryption identifiers of the standardized data packets when performing data legality verification; the data import module is also configured to batch import the patient information and screening audio and video data in the standardized data packets into the centralized archive database after the verification is passed.

[0014] Preferably, the large model training data output interface module is also configured to desensitize the screening audio and video data to remove patient privacy information when outputting the medical large model training dataset, and then align and annotate the desensitized audio and video data with the evaluation results according to the data format required for medical large model training. The system also includes a full-process encryption unit, which is configured to encrypt patient information, screening audio and video data, and standardized data packets throughout their entire lifecycle, from local storage and cloud transmission to archiving.

[0015] This invention also provides a method for screening, collecting, and managing stroke data based on the aforementioned system, the method comprising the following steps: S1. Based on the standardized screening and assessment scale for stroke, the assessment items are broken down into standardized audio and video acquisition sub-tasks, and the execution steps of the acquisition sub-tasks are dynamically adjusted according to the patient's real-time feedback; the audio and video acquisition process is executed, and the acquired patient information and screening audio and video data are stored in the local terminal database. S2. Query screening records in the local terminal database and play back the corresponding screening audio and video data. In offline mode, package and export the screening data. In online mode, upload the locally stored screening data to the cloud server. In online mode, query screening records from the cloud server and play back the corresponding screening audio and video data, and perform management operations on the screening data stored in the cloud. S3. Receive standardized data packets from the local terminal database and / or cloud server and perform data integrity and format standardization verification on them; parse the patient information and screening audio and video data in the verified standardized data packets, and batch import them into the centralized archive database for archiving management; S4. Obtain the standardized and legally verified screening audio and video data and the corresponding evaluation results; convert and package the screening audio and video data and evaluation results into a medical large model training dataset.

[0016] Furthermore, in step S1, the assessment items are broken down into standardized audio and video acquisition sub-tasks, specifically including: breaking down facial nerve function assessment, limb motor function assessment, speech function assessment, coordination motor assessment and sensory function assessment into one-to-one corresponding acquisition sub-tasks, and presetting unified acquisition specifications and assessment standards for each acquisition sub-task. Based on the patient's real-time feedback, the execution steps of the collection sub-tasks are dynamically adjusted. Specifically, this includes adjusting the execution steps and prompts of the collection tasks based on the patient's real-time complaints, physical signs, and physical tolerance, and skipping unnecessary collection items without relevant symptoms. During the audio and video acquisition process, real-time preview and verification of the acquired footage are provided, along with support for pausing and resuming acquisition from breakpoints. After acquisition is completed, the evaluation results are bound and stored one by one with the corresponding acquired screening audio and video data.

[0017] In summary, due to the adoption of this technical solution, the beneficial effects of this invention are as follows: This invention uses the NIHSS Stroke Standardized Assessment Scale as the industry gold standard. By decomposing the scale's assessment dimensions and mapping tasks, it establishes a unified standard for stroke screening and assessment. Combined with a dynamic task scheduling mechanism based on real-time patient feedback, it achieves a unified approach between standardized screening procedures and individualized treatment needs. This addresses the problems of non-standard procedures, poor assessment consistency, and missed positive signs in traditional screening models, significantly improving the accuracy and specificity of early stroke screening.

[0018] This invention, with a local-first architecture at its core, combines functions such as patient identity verification and record building, standardized audio and video acquisition, result review, and data archiving to construct a fully digital management and control chain that completely replaces the traditional paper-based record-keeping model. As a result, this invention effectively eliminates manual data entry errors, significantly reduces the labor and management costs of screening, and significantly improves the efficiency and standardization of stroke screening.

[0019] This invention achieves network environment adaptability through a local / online dual-mode switching architecture. In a network-free environment, the entire screening process can be completed offline, while in a network-connected environment, it supports cloud data synchronization and multi-terminal collaborative management. It is fully adaptable to screening applications in both in-hospital and out-of-hospital scenarios, providing core technical support for the large-scale implementation of a broad-coverage early screening system for stroke. This invention also adopts hierarchical access control for institutions to prevent data leakage and unauthorized access risks, and the entire process complies with relevant laws and regulations on medical data management and the security management standards of medical institutions.

[0020] This invention achieves a strong correlation between screening audio and video data and patient information and clinical assessment results through a standardized data collection process. It can directly generate a standardized dataset that conforms to the training specifications of artificial intelligence models, supporting the training and iteration of intelligent recognition models for multi-dimensional signs of stroke. This invention realizes the transformation of clinical screening data into technology research and development, as well as the assetization of clinical data, providing a high-quality data foundation for the continuous optimization of non-invasive intelligent screening technology for stroke. Attached Figure Description

[0021] The present invention is described in detail with reference to the following figures, which include six figures as follows: Figure 1 This is a schematic diagram of the overall module architecture of the stroke data screening and collection management system of the present invention; Figure 2 This is a schematic diagram of the workflow of the screening and acquisition module in the system of the present invention; Figure 3 This is a schematic diagram of the workflow of the local data query module in the system of the present invention; Figure 4 This is a schematic diagram of the workflow of the online data query module in the system of the present invention; Figure 5 This is a schematic diagram of the workflow of the data import module in the system of the present invention; Figure 6 This is a schematic diagram illustrating the overall process of the stroke data screening, collection, and management method of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0024] Example 1 A stroke data screening, collection, and management system that supports large-scale model training. Figure 1 The main modules of the system are shown and can be viewed simultaneously; the key modules of the system can be summarized as follows: The screening and acquisition module is configured to be based on a standardized stroke screening and assessment scale. It breaks down the assessment items into standardized audio and video acquisition sub-tasks and dynamically adjusts the execution steps of the acquisition sub-tasks according to the patient's real-time feedback. The screening and acquisition module is also used to execute the audio and video acquisition process and store the acquired patient information and screening audio and video data in the local terminal database. The local data query module is configured to query screening records in the local terminal database and play back the corresponding screening audio and video data; the local data query module is also used to package and export screening data in offline mode, and to upload locally stored screening data to the cloud server in online mode; The online data query module is configured to query screening records from the cloud server and play back the corresponding screening audio and video data in online mode; the online data query module is also used to manage screening data stored in the cloud. The data import module is configured to receive standardized data packets from a local terminal database and / or a cloud server, and to perform data validity verification on the standardized data packets; the data import module is also used to archive the verified standardized data packets into the database.

[0025] In this embodiment, the key advantage of the system is that it supports large-scale model training, especially medical large-scale models that include intelligent stroke recognition models. Therefore, the system also preferably includes a large-scale model training data output interface module, which is configured to acquire standardized and legally verified screening audio and video data and corresponding evaluation results. The large-scale model training data output interface module is also used to convert and package the screening audio and video data and evaluation results into a medical large-scale model training dataset.

[0026] Therefore, the core functional feature of this embodiment is the collaborative operation of the data import module and the large-scale model training data output interface module. The data import module, through legality verification and centralized archiving mechanisms, ensures the standardization and integrity of stroke screening data collected from multiple terminals during the aggregation process, establishing a high-quality data foundation for subsequent data applications. The large-scale model training data output interface module connects the data path between clinical screening operations and artificial intelligence research and development, transforming standardized and verified audio-visual data and evaluation results into a standard dataset that meets the training requirements of large-scale medical models. The application of these two modules in the system not only facilitates the full lifecycle management of screening data but also solves the problems of data heterogeneity and high annotation costs faced in large-scale medical model training, significantly improving the compliant reuse value and intelligent transformation efficiency of clinical data.

[0027] Example 2 Based on Example 1, this example describes the additional login module included in the system, such as... Figure 1 As shown, the login module is the starting point for all system functions. The login module is configured to provide system login verification and support for manually switching between online and local modes.

[0028] After starting the system, users can manually select local or online mode based on their current network status, choose their affiliated institution, and then enter their account and password for login verification. The login module is also configured to implement an institutional data isolation mechanism and hierarchical permission control to isolate stored data from different institutions and display the patient directory and system function operation interface within the corresponding permission scope according to the logged-in user's permission level.

[0029] When a user chooses local login mode, the system only accesses the local terminal database and does not establish any data connection with the cloud server. In this mode, the system can independently complete the entire screening process, including patient registration, audio and video capture, data storage, and local querying. This operating mechanism enables the system to operate normally in environments with no or weak network connection, ensuring that data collection in scenarios such as primary care screening or mobile clinics is not limited by network conditions.

[0030] When a user chooses to log in in online mode, the system establishes a stable data connection with the cloud server. While retaining all existing functions of the local mode, the online mode adds extended features such as local data synchronization to the cloud, remote access to cloud data, and data import. This mode is primarily designed for intranet-based hospital scenarios, leveraging the cloud server's data aggregation and distribution capabilities to meet the needs of multi-terminal collaborative management and cross-departmental data sharing.

[0031] Regarding access control and data isolation, the login module incorporates a tiered access control mechanism that assigns corresponding operational permissions based on preset user roles. Users with different permission levels can only view and operate the patient directory and system function interface within their authorized scope after logging into the system, thus preventing unauthorized access. Simultaneously, the institutional data isolation mechanism ensures that the stored data of different medical institutions are independent of each other. Data from each institution is effectively isolated at both the underlying storage and upper-level display levels, guaranteeing the security and compliance of medical data from the system architecture level.

[0032] Example 3 Based on any of the above embodiments, this embodiment describes the screening and data collection module in the system. Figure 2 The flowchart of the screening and data collection module is shown and can be viewed simultaneously.

[0033] The screening and data collection module is the core business module of the system in this embodiment, mainly responsible for standardized audio and video collection and customized screening tasks for stroke. Before data collection begins, this module is configured to display the registered patient list of the target institution within the current operating permissions, facilitating quick location of target patients by operators. The system supports incomplete matching queries based on name, ID number, and hospital number. When the query keyword is empty, it defaults to querying the list of all registered patients under the current institution. Simultaneously, the module provides an interface for selecting or creating patient lists to meet the patient management needs in different scenarios.

[0034] When creating a new patient file, the system requires the entry of mandatory fields such as name, gender, hospital visited, and date of onset. To improve file creation efficiency, the screening and data collection module supports external ID card readers. After successfully reading the patient's identity information, it automatically fills in the resident ID number, name, gender, and other information, and stores the patient information in the local terminal database after triggering the save operation. Furthermore, this module is compatible with different models or types of cameras, handheld terminals, and industrial data collection terminals, and can receive audio and video acquisition signals transmitted from different hardware terminals. When no patient is selected, the "Start Collection" button is unavailable. After selecting a patient and triggering the "Start Collection" operation, the system initializes the camera hardware used for screening and then enters the data collection interface.

[0035] After entering the data collection interface, the screening module customizes screening tasks based on the standardized stroke screening and assessment scale. The system breaks down the scale's full-dimensional assessment items into standardized collection sub-tasks, specifically covering screening items such as consciousness and language level, dysarthria, command and facial paralysis, eye movement, upper limb muscle strength, lower limb movement, and coordination. Each collection sub-task has pre-set unified collection specifications and assessment standards. During the data collection process, the system dynamically adjusts the execution steps and prompts based on the patient's real-time complaints, physical signs, and physical tolerance, automatically skipping unnecessary collection items without relevant symptoms, thus achieving personalized screening process customization.

[0036] During the audio and video acquisition process, the screening acquisition module provides real-time preview and verification of the acquired footage, as well as support for pausing and resuming acquisition from breakpoints. The video acquisition interface is divided into a patient information preview area and a video preview area. The patient information preview area displays detailed information about the patient being acquired and the current acquisition step, while the video preview area allows real-time viewing of the camera's captured content. Once the start recording operation is triggered, high-quality and controllable acquisition of the entire patient screening process can begin, while also supporting flexible pausing and termination of acquisition to ensure that the acquisition process can adapt to unexpected situations and the patient's actual condition.

[0037] Once the data collection subtasks are completed and the overall data collection is finished, the screening and data collection module will bind and associate the assessment results of the standardized stroke screening assessment scale with the corresponding collected screening audio and video data one by one throughout the process. This binding mechanism ensures the correspondence between the scale assessment results and the dynamic vital signs video, providing a complete basis for subsequent review and data analysis. Finally, the module will automatically store the bound assessment results and screening audio and video data as a whole in the local terminal database, completing the localized and secure management of the screening data.

[0038] Example 4 Based on any of the above embodiments, this embodiment will describe the local data query module and the online data query module in the system respectively; Figure 3 The flowchart of the local data query module is shown. Figure 4 The workflow diagram of the online data query module is shown and can be viewed simultaneously.

[0039] In this embodiment, the local data query module is configured to manage local screening data throughout its entire lifecycle. Regarding query records, this module displays all screening records of the target institution stored in the local terminal database. Each record is fully associated with the patient's basic information, screening audio / video data, collection time, and stroke standardized screening assessment scale score data. Query keywords support incomplete matching queries based on ID card number, name, and hospital number. When the query keyword is empty, the system defaults to querying all screening records under the target institution.

[0040] The local data query module supports continuous frame playback of audio and video for target screening records. This module also supports batch standardized export of screening data and audio / video, encapsulating patient information and screening audio / video data into standardized data packets and providing batch export support for these standardized data packets. Furthermore, when online mode is active, this module can encrypt locally stored screening data and upload it to the cloud server.

[0041] In this embodiment, the online data query module is configured for collaborative management and control of cloud-based data. This module displays in real-time all screening lists uploaded to the cloud by the target institution. Each record is fully associated with the patient's basic information, screening audio and video data, collection time, and stroke standardized screening assessment scale score data. Query keywords support incomplete matching based on ID card number, name, and hospital number. When the query keyword is empty, the system defaults to querying all screening records uploaded to the cloud by the target institution.

[0042] The online data query module supports continuous frame playback of audio and video for targeted screening records. Simultaneously, this module provides support for batch deletion of screening data stored in the cloud, allowing for multi-condition filtering of invalid, redundant, and expired data in the cloud, and subsequent batch deletion.

[0043] Example 5 Based on any of the above embodiments, this embodiment describes the data import module in the system; Figure 5 The flowchart of the data import module is shown and can be viewed simultaneously.

[0044] In this embodiment, the data import module is primarily responsible for the centralized collection of data from multiple terminals and sites. When receiving standardized data packets, this module provides a selection interface for standardized data packets, supporting the import of data packet files in standard compressed formats. These data packet files mainly originate from files exported in batches by the local data query module. During the data packet reception process, the system automatically identifies the format of the data packets and provides immediate prompts for non-standard format data packets, thereby avoiding invalid import operations and ensuring that the received data source conforms to the system's specifications.

[0045] Upon receiving the selected standardized data packet, the data import module performs a data validity check. This check comprehensively verifies the data integrity, format specifications, and security encryption identifiers of the standardized data packet, specifically including format integrity verification and data correlation verification mechanisms. In the format integrity verification stage, the system checks whether the file structure and data fields within the data packet are complete, confirming whether any core information is missing. In the data correlation verification stage, the system verifies the correlation and authenticity between patient information, screening data, and audio / video files within the data packet, ensuring that all data corresponds correctly and has not been tampered with.

[0046] Once the standardized data packets pass the aforementioned data validity verification, the data import module will perform a batch data entry and archiving operation. The system will automatically parse the verified data packets and batch synchronize the patient files, screening records, and screening audio and video data within the data packets to the centralized archiving database. This process achieves unified data entry and centralized management of multi-source screening data, completes the final data archiving operation, and establishes a reliable data foundation for subsequent data retrieval, review, and output of training data for large-scale medical models.

[0047] Example 6 Based on any of the above embodiments, this embodiment introduces and describes the large model training data output interface module in the system, and also introduces the preferred full-process encryption unit included in the system.

[0048] In this embodiment, the large-scale model training data output interface module serves as a node connecting clinical screening operations with artificial intelligence research and development. Its main configuration involves acquiring standardized and legally validated screening audio and video data, along with corresponding evaluation results. This module extracts high-quality multimodal clinical data by calling the underlying database interface, converts and packages the data, and ultimately outputs a standard dataset that meets the training requirements of large-scale medical models.

[0049] In this embodiment, the medical big model refers to a large-scale deep learning network with multimodal understanding and generation capabilities, which can assist in diagnosis and disease prediction by learning from massive amounts of medical data. The dataset output by this module is the key foundation for driving the iterative optimization of such models.

[0050] During the output of the medical large-scale model training dataset, the large-scale model training data output interface module is first configured to perform strict anonymization processing on the screening audio and video data and patient information to completely remove patient privacy information. This anonymization mechanism ensures that the output data retains key pathological features for vital sign analysis while meeting the relevant laws and regulations on medical data compliance and personal information protection.

[0051] After the anonymization process is completed, the large-scale model training data output interface module will precisely align and annotate the anonymized audio and video data with the evaluation results according to the data format required for training the medical large-scale model. The module will convert the scores from the standardized stroke screening and assessment scale into structured labels that the model can parse, and align the video timeline with the patient's specific action nodes using timestamps. For example, it will bind the start and end timestamps of the limb lifting test with the corresponding muscle strength scores, generating an annotation file containing time series, video image paths, and multi-dimensional evaluation labels. This embodiment, through this multimodal data alignment and standardization conversion, greatly reduces the data cleaning and manual annotation costs in the early stages of medical large-scale model training, effectively solving the data heterogeneity problem.

[0052] As a preferred embodiment, to ensure the absolute security of medical data at every stage of its flow, the system also includes a full-process encryption unit. This unit is configured to encrypt patient information, screening audio / video data, and standardized data packets throughout their entire lifecycle, from local storage and cloud transmission to archiving. During local storage, this unit employs a high-strength symmetric encryption algorithm to encrypt sensitive fields in the database and audio / video files in the local file system, ensuring that even if physical devices are lost or stolen, the underlying data cannot be illegally parsed.

[0053] During the cloud transmission phase, the end-to-end encryption unit establishes an encrypted channel based on a secure transport layer protocol to encrypt the screening data uploaded to the cloud server and the standardized data packets sent out in real time. This effectively prevents network sniffing and man-in-the-middle attacks, ensuring the confidentiality and integrity of data transmission in a public network environment. For the archiving phase, the end-to-end encryption unit, in conjunction with a key management system, provides multiple layers of encryption protection for static data in the centralized archive database and implements key rotation and access control policies. Through this encryption mechanism that spans the entire data lifecycle, the system constructs a medical data security defense line from the underlying architecture, complying with the security standards and audit requirements for the informatization construction of medical institutions.

[0054] Example 7 Based on any of the above embodiments, this embodiment provides a stroke data screening and collection management method that supports large model training, based on a stroke data screening and collection management system. The hardware basis of the method is the system in any of the above embodiments. Figure 6 The overall process of the method is briefly described below and can be viewed concurrently; the key steps of the method can be summarized as follows: S1. Based on the standardized screening and assessment scale for stroke, the assessment items are broken down into standardized audio and video acquisition sub-tasks, and the execution steps of the acquisition sub-tasks are dynamically adjusted according to the patient's real-time feedback; the audio and video acquisition process is executed, and the acquired patient information and screening audio and video data are stored in the local terminal database. S2. Query screening records in the local terminal database and play back the corresponding screening audio and video data. In offline mode, package and export the screening data. In online mode, upload the locally stored screening data to the cloud server. In online mode, query screening records from the cloud server and play back the corresponding screening audio and video data, and perform management operations on the screening data stored in the cloud. S3. Receive standardized data packets from the local terminal database and / or cloud server and perform data integrity and format standardization verification on them; parse the patient information and screening audio and video data in the verified standardized data packets, and batch import them into the centralized archive database for archiving management; S4. Obtain the standardized and legally verified screening audio and video data and the corresponding evaluation results; convert and package the screening audio and video data and evaluation results into a medical large model training dataset.

[0055] This embodiment will describe in detail the entire process and preferred methods based on the order of the above steps.

[0056] In this embodiment, the method begins with system startup and login verification. After starting the system, the user manually selects either local or online mode to log in via the login module, depending on the current network environment. The system implements an institutional data isolation mechanism and hierarchical permission control to ensure that the stored data of different institutions are isolated from each other. Based on the logged-in user's permission level, the system displays the patient directory and system function operation interface within the corresponding permission scope, thereby providing a secure and compliant basic operating environment for subsequent screening and data collection.

[0057] After logging into the system, operators can use the screening and data collection module to view the patient directory within their permissions to search for target patients, or create patient profiles to establish basic patient information. When creating a new profile, the system supports connecting an external ID card reader to automatically read the patient's identity information and complete the profile creation. The audio and video data collection process then begins. Based on a standardized stroke screening and assessment scale, the system breaks down comprehensive assessment items, including facial nerve function assessment, limb motor function assessment, speech function assessment, ataxia assessment, and sensory function assessment, into corresponding standardized audio and video data collection sub-tasks. Each sub-task has pre-set unified collection specifications and assessment standards. During this process, the system dynamically adjusts the execution steps and prompts of the data collection tasks based on the patient's real-time complaints, physical signs, and physical tolerance, automatically skipping unnecessary data collection items without relevant symptoms, thus achieving personalized screening process customization.

[0058] During the audio and video acquisition process, the screening acquisition module provides real-time preview and verification of the acquired footage, allowing operators to view the camera content in real time to ensure acquisition quality. Simultaneously, the system offers flexible pause and resume capabilities to handle unexpected situations. Once the acquisition sub-tasks are completed and the overall acquisition is finished, the system will individually map and associate the assessment results of the standardized stroke screening assessment scale with the corresponding acquired screening audio and video data. The bound assessment results and screening audio and video data will then be automatically stored as a whole in the local terminal database, completing the localized and secure management of the screening data.

[0059] After data collection, operators can manage the local screening data throughout its entire lifecycle using the local data query module. Regarding record queries, the system displays all screening records from the target institution stored in the local database, supporting incomplete matching queries based on ID card number, name, and hospital number. For target screening records, the system supports continuous frame playback of audio and video, facilitating secondary verification of screening results and review of vital signs by medical staff. Furthermore, the local data query module supports encapsulating patient information and screening audio / video data into standardized data packets and provides batch export support for these standardized data packets. When online mode is active, this module can also encrypt locally stored screening data and upload it to the cloud server, achieving synchronization between local data and the cloud management system.

[0060] In online mode, operators can collaboratively manage cloud-based data through the online data query module. This module displays the full screening list uploaded to the cloud by the target institution in real time. Each record is fully associated with the patient's basic information, screening audio and video data, collection time, and stroke standardized screening assessment scale score data. Operators can search for cloud screening records using multiple conditions and perform continuous frame playback of audio and video for target records to complete remote verification and vital sign review. Simultaneously, the online data query module provides batch deletion support for screening data stored in the cloud, allowing for multi-condition filtering and batch deletion of invalid, redundant, and expired data in the cloud, thereby optimizing cloud storage resources.

[0061] To achieve centralized data collection from multiple terminals and sites, operators can import data using the data import module. This module provides a standardized data package selection interface, supports importing data package files in standard compressed formats, and automatically identifies the data package format during reception, providing immediate prompts for non-standard formats. Upon receiving the selected standardized data package, the system performs rigorous data validity verification, comprehensively checking the data integrity, format specifications, and security encryption identifiers of the data package. Specifically, this includes verifying the integrity of the file structure and data fields, as well as the correlation and authenticity between patient information, screening data, and audio / video files. After successful verification, the data import module batch synchronizes the patient files, screening records, and screening audio / video data within the data package to the centralized archiving database, completing the final data archiving operation.

[0062] After standardized data collection and legality verification, the large-scale model training data output interface module acquires the screening audio and video data and corresponding evaluation results. This module first anonymizes the screening audio and video data. Then, following the data format required for training the medical large-scale model, it precisely aligns and annotates the anonymized audio and video data with the evaluation results, converting the scale scoring results into structured labels that the model can parse, and aligning the video timeline with specific action nodes using timestamps. Finally, the module converts and packages the transformed data, outputting a high-quality medical large-scale model training dataset, providing core data support for the research and iteration of medical artificial intelligence.

Claims

1. A stroke data screening, collection, and management system supporting large model training, characterized in that, The system includes the following functional modules: The screening and acquisition module is configured to be based on a standardized stroke screening and assessment scale. It breaks down the assessment items into standardized audio and video acquisition sub-tasks and dynamically adjusts the execution steps of the acquisition sub-tasks according to the patient's real-time feedback. The screening and acquisition module is also used to execute the audio and video acquisition process and store the acquired patient information and screening audio and video data in the local terminal database. The local data query module is configured to query screening records in the local terminal database and play back the corresponding screening audio and video data; the local data query module is also used to package and export screening data in offline mode, and to upload locally stored screening data to the cloud server in online mode; The online data query module is configured to query screening records from the cloud server and play back the corresponding screening audio and video data in online mode; the online data query module is also used to manage screening data stored in the cloud. The data import module is configured to receive standardized data packets from a local terminal database and / or a cloud server, and to perform data validity verification on the standardized data packets; the data import module is also used to archive the verified standardized data packets into the database. The large model training data output interface module is configured to acquire standardized and legally verified screening audio and video data and corresponding evaluation results; the large model training data output interface module is also used to convert and package the screening audio and video data and evaluation results into a medical large model training dataset.

2. The stroke data screening and collection management system according to claim 1, characterized in that: The system also includes a login module, which is configured to provide system login verification and support manual switching between online and local modes. The login module is also configured to implement an institutional data isolation mechanism and hierarchical permission control to isolate stored data from different institutions. The login module is also configured to display the patient directory and system function operation interface within the corresponding permission scope according to the permission level of the logged-in user.

3. The stroke data screening and collection management system according to claim 1, characterized in that: The screening and acquisition module is also configured to establish basic patient information by creating a new patient file before executing the audio and video acquisition process, and to automatically create a file after reading the patient's identity information by connecting an external ID card reader; the screening and acquisition module is also configured to provide compatibility support for different models or types of cameras, handheld terminals and industrial acquisition terminals, and to receive audio and video acquisition signals transmitted from different hardware terminals; the screening and acquisition module is also configured to display the patient list within the permissions before the acquisition starts, and provide an interface for selecting or creating a patient list.

4. The stroke data screening and collection management system according to claim 1, characterized in that: The screening and data collection module is also configured to be based on a standardized stroke screening and assessment scale, which breaks down facial nerve function assessment, limb motor function assessment, speech function assessment, coordination assessment, and sensory function assessment into one-to-one corresponding data collection sub-tasks. The screening and collection module is also configured to preset unified collection specifications and evaluation standards for each collection sub-task; the screening and collection module is also configured to adjust the execution steps and prompts of the collection task based on the patient's real-time complaints, physical signs and physical tolerance, and skip unnecessary collection items without relevant symptoms.

5. The stroke data screening and collection management system according to claim 4, characterized in that: The screening and acquisition module is also configured to provide real-time preview and verification of the acquisition screen during the audio and video acquisition process when performing acquisition sub-tasks, and to provide support for pausing and resuming acquisition from breakpoints during the acquisition process. The screening and data collection module is also configured to bind and associate the assessment results of the standardized stroke screening assessment scale with the corresponding collected screening audio and video data one by one after the data collection is completed; the screening and data collection module is also configured to store the bound assessment results and screening audio and video data as a whole in the local terminal database.

6. The stroke data screening and collection management system according to claim 1, characterized in that: The local data query module is also configured to encapsulate patient information and screening audio and video data into standardized data packets when packaging and exporting screening data in offline mode, and to provide support for batch export of standardized data packets; The online data query module is also configured to provide retrieval and online playback of screening records and screening audio and video data when managing screening data stored in the cloud in online mode; the online data query module is also configured to provide support for batch deletion of screening data stored in the cloud.

7. The stroke data screening and collection management system according to claim 1, characterized in that: The data import module is also configured to provide a selection interface for standardized data packets and receive the selected standardized data packets when receiving them; the data import module is also configured to verify the data integrity, format standardization, and security encryption identifier of the standardized data packets when performing data legality verification; the data import module is also configured to batch import the patient information and screening audio and video data in the standardized data packets into the centralized archive database after the verification is passed.

8. The stroke data screening and collection management system according to claim 1, characterized in that: The large model training data output interface module is also configured to desensitize the screening audio and video data and patient information to remove patient privacy information when outputting the medical large model training dataset. Then, according to the data format required for medical large model training, the desensitized audio and video data and evaluation results are aligned and labeled. The system also includes a full-process encryption unit, which is configured to encrypt patient information, screening audio and video data, and standardized data packets throughout their entire lifecycle, from local storage and cloud transmission to archiving.

9. A method for screening, collecting, and managing stroke data according to the system of claim 1, characterized in that, The method includes the following steps: S1. Based on the standardized screening and assessment scale for stroke, the assessment items are broken down into standardized audio and video acquisition sub-tasks, and the execution steps of the acquisition sub-tasks are dynamically adjusted according to the patient's real-time feedback; the audio and video acquisition process is executed, and the acquired patient information and screening audio and video data are stored in the local terminal database. S2. Query screening records in the local terminal database and play back the corresponding screening audio and video data. In offline mode, export the screening data in a package. In online mode, upload the locally stored screening data to the cloud server. In online mode, it can query screening records from the cloud server and play back the corresponding screening audio and video data, and manage the screening data stored in the cloud. S3. Receive standardized data packets from the local terminal database and / or cloud server and perform data integrity and format compliance verification on them; The patient information and screening audio and video data in the standardized data packets that have passed the verification are parsed and then batch-entered into a centralized archive database for archiving management. S4. Obtain the standardized and legally verified screening audio and video data and the corresponding evaluation results; convert and package the screening audio and video data and evaluation results into a medical large model training dataset.

10. The stroke data screening, collection, and management method according to claim 9, characterized in that, In step S1, the assessment items are broken down into standardized audio and video acquisition sub-tasks, specifically including: facial nerve function assessment, limb motor function assessment, speech function assessment, coordination motor assessment and sensory function assessment are broken down into corresponding acquisition sub-tasks, and a unified acquisition specification and assessment standard are preset for each acquisition sub-task. Based on the patient's real-time feedback, the execution steps of the collection sub-tasks are dynamically adjusted. Specifically, this includes adjusting the execution steps and prompts of the collection tasks based on the patient's real-time complaints, physical signs, and physical tolerance, and skipping unnecessary collection items without relevant symptoms. During the audio and video acquisition process, real-time preview and verification of the acquired footage are provided, along with support for pausing and resuming acquisition from breakpoints. After acquisition is completed, the evaluation results are bound and stored one by one with the corresponding acquired screening audio and video data.

Images