Automatic follow-up visit system for stroke patients

Abstract

The embodiment of the invention provides an automatic follow-up visit system for a stroke patient, and relates to the technical field of medical follow-up visit systems.The automatic follow-up visit system comprises a data receiving module, a large language model module, a data output module and a scoring module, the question and answer data are answers of the stroke patient according to greetings or question data; the big language model module is used for analyzing the answer data by utilizing a Baichuan-M2 medical big language model based on the cerebral apoplexy health risk assessment and an improved Rankin scale to generate question data; the data output module is used for outputting problem data to the stroke patient; the scoring module is used for generating a stroke score of the stroke patient based on the question data and the answer data, and the stroke score is used for evaluating the rehabilitation condition of the stroke patient. The human cost can be greatly reduced, and the recording and follow-up visit efficiency is improved.

Description

Technical Field

[0001] This disclosure relates to the field of medical follow-up system technology, and more particularly to an automated follow-up system for stroke patients. Background Technology

[0002] Stroke, commonly known as "apoplexy," is characterized by "four highs": high incidence, high disability rate, high mortality rate, and high recurrence rate. It has a rapid onset, progresses quickly, and has serious consequences. Follow-up of stroke patients' rehabilitation after discharge is a crucial part of medical services. It not only allows for real-time monitoring of the recovery progress of motor function, language ability, and daily activities, and timely screening for complications such as infection, deep vein thrombosis, and early signs of stroke recurrence, but also enables dynamic adjustments to home rehabilitation training plans based on the patient's actual recovery, answers questions from patients and their families regarding medication and diet, alleviates rehabilitation anxiety, and improves treatment adherence.

[0003] Traditional follow-up methods mainly rely on manual telephone calls or offline questionnaires. Follow-up personnel typically call patients one by one to ask questions, record the answers, and then enter the results into the hospital database. Manual follow-up methods are labor-intensive, have a heavy workload for follow-up personnel, and are prone to omissions and errors during the follow-up process, resulting in low recording efficiency. Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, this disclosure provides an automated follow-up system for stroke patients.

[0005] This disclosure provides an automated follow-up system for stroke patients. The automated follow-up system includes a data receiving module, a large language model module, a data output module, and a scoring module. The data receiving module receives response data from stroke patients, which consists of the patient's answers to greetings or questions. The large language model module analyzes the response data using the Baichuan-M2 medical large language model based on a stroke health risk assessment scale to generate question data. The data output module outputs the question data to the stroke patients. The scoring module generates a stroke score for the stroke patients based on the question and response data, and the stroke score is used to assess the patients' rehabilitation status.

[0006] This application utilizes a data receiving module to accurately receive patient feedback, ensuring stable acquisition of patient responses and providing complete and effective raw input for subsequent analysis, avoiding data omissions or transmission errors. Leveraging the Baichuan-M2 medical large-scale language model and stroke assessment scale, it can precisely break down rehabilitation assessment points based on patient responses, generating targeted and medically logical question data without requiring manually pre-set fixed question-and-answer scripts, adapting to the individualized rehabilitation situations of different patients. The data output module can quickly output the question data generated by the large-scale language model to patients, ensuring a smooth question-and-answer process, reducing patient waiting time, and improving the follow-up experience. The scoring module automatically generates stroke scores based on the question-and-answer data, avoiding subjective biases in manual scoring and achieving standardized assessment of rehabilitation status. This application's automated follow-up system for stroke patients eliminates the need for dedicated follow-up personnel to participate in question-and-answer, recording, and scoring throughout the process, significantly reducing labor costs and workload. Utilizing the large-scale language model avoids human errors and improves recording and follow-up efficiency. Attached Figure Description

[0007] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0008] Figure 1 A schematic diagram of an automated follow-up system for stroke patients provided in an embodiment of this disclosure; Figure 2 A schematic diagram of another automated follow-up system for stroke patients provided in an embodiment of this disclosure; Figure 3 A flowchart illustrating an automated follow-up method for stroke patients provided in this embodiment of the present disclosure; Figure 4 This is a schematic diagram of the structure of a computing device provided in an embodiment of the present disclosure. Detailed Implementation

[0009] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0010] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0011] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0012] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0013] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0014] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0015] Traditional follow-up methods for stroke patients after discharge mainly rely on manual telephone calls or offline questionnaires. Follow-up personnel typically call each patient individually, ask questions, record answers, and then enter the results into the hospital database. First, manual follow-up is labor-intensive, with a huge workload for staff. Each staff member can only complete an average of ~15 effective calls per hour. This repetitive dialing and recording work is not only inefficient but also prone to omissions and errors. Second, different follow-up personnel may use inconsistent questioning methods and recording standards, leading to omissions or errors in manual recording. The collected data is unstructured, making subsequent analysis difficult, and lacks unified data standards and verification mechanisms. Furthermore, since many patients are elderly and heavily reliant on telephone follow-up, manual follow-up is limited by manpower and time constraints, making it difficult to conduct large-scale, high-frequency follow-up.

[0016] To address the aforementioned problems, this disclosure provides an automated follow-up system for stroke patients. This system includes: a data receiving module for receiving response data from stroke patients, where the response data consists of the patient's answers to greetings or questions; a large language model module for analyzing the response data using the Baichuan-M2 medical large language model based on a stroke health risk assessment scale to generate question data; a data output module for outputting the question data to the stroke patient; and a scoring module for generating a stroke score for the patient based on the question and response data. This application, by constructing an automated follow-up system for stroke patients, automatically conducts medical follow-ups with discharged stroke patients to inquire about their recovery progress. This automated follow-up system eliminates the need for dedicated follow-up personnel to participate in the entire process of questioning, recording, and scoring, significantly reducing labor costs and workload. Utilizing a large language model, it avoids human error and improves recording and follow-up efficiency.

[0017] The method will be described below with reference to specific embodiments.

[0018] Figure 1 This diagram illustrates an automated follow-up system for stroke patients, provided as an embodiment of this disclosure. The system can be implemented using software and / or hardware, and is generally integrated into a computing device. Figure 1 As shown, the automated follow-up system includes a data receiving module, a large language model module, a data output module, and a scoring module.

[0019] The data receiving module is used to receive the response data from stroke patients.

[0020] Specifically, the automated follow-up system of the computer equipment automatically sends follow-up requests and questions to stroke patients and receives their responses. The question-and-answer data consists of the stroke patient's answers based on the greetings or questions.

[0021] In one possible implementation, when the automatic follow-up system initiates a follow-up request for the first time, it automatically sends a greeting to the stroke patient, who then replies according to the greeting. The data receiving module receives the stroke patient's response data.

[0022] In another possible implementation, as the automated follow-up system continues to ask questions based on the stroke patient's answers, the stroke patient responds based on the question data, and the data receiving module receives the stroke patient's answer data.

[0023] The large language model module is used to analyze the response data based on the stroke assessment scale and generate question data using the Baichuan-M2 medical large language model.

[0024] For example, the stroke health risk assessment scale is the Modified Rankin Scale (MRS). The mRS is a clinical tool used to measure the neurological function recovery status of stroke patients, as well as their activities of daily living and degree of disability. Its scoring ranges from level zero to level six, a total of seven levels, each representing a different degree of neurological function recovery and disability status.

[0025] Among them, the Baichuan-M2 Medical Language Model is a medical augmented reasoning model that can be used for doctor-patient communication.

[0026] Specifically, the computer device stores a stroke assessment scale for the rehabilitation assessment of stroke patients. The large language model module analyzes the stroke assessment scale to extract the target information required for the assessment of stroke patients. It then uses the Baichuan-M2 medical large language model to combine the target information with the question-and-answer data of stroke patients to analyze the information that still needs to be obtained and generate corresponding question data to request stroke patients to give corresponding answers.

[0027] The data output module is used to output problem data to stroke patients.

[0028] Specifically, when the large language model module generates question data, the computer device asks questions to the stroke patient based on the question data.

[0029] Furthermore, stroke patients respond again based on the question data, continuously cycling through the large language model module, data output module, and data receiving module to generate question data and receive response data. The cycle ends when the response data covers the target information required by the stroke assessment scale.

[0030] The scoring module is used to generate stroke scores for stroke patients based on question and answer data.

[0031] Specifically, the scoring module generates a stroke score for stroke patients based on key information in the question and answer data, following the assessment methods of the stroke assessment scale. The stroke score can be used to assess the rehabilitation status of stroke patients.

[0032] The data receiving module accurately receives patient feedback, stably acquiring patient response data to provide complete and effective raw input for subsequent analysis, avoiding data omissions or transmission errors. Relying on the Baichuan-M2 medical big language model and stroke assessment scale, it can accurately break down rehabilitation assessment points based on patient responses, generating targeted and medically logical question data without requiring manually pre-set fixed question-and-answer scripts, adapting to the personalized rehabilitation status of different patients. The data output module can quickly output the question data generated by the big language model to the patient, ensuring a smooth question-and-answer process, reducing patient waiting time, and improving the follow-up experience. The scoring module automatically generates a stroke score based on the question-and-answer data, avoiding subjective bias in manual scoring and achieving standardized assessment of rehabilitation status. This application's automated follow-up system for stroke patients does not rely on dedicated follow-up personnel to participate in question-and-answer, recording, and scoring throughout the process, significantly reducing labor costs and workload. Utilizing the big language model, it avoids human errors and improves recording and follow-up efficiency.

[0033] Figure 2 This diagram illustrates another automated follow-up system for stroke patients provided in an embodiment of this disclosure. The system can be implemented using software and / or hardware, and is generally integrated into a computing device, a private server cluster, or a cloud server. Figure 2 As shown, the automated follow-up system in this embodiment is deployed using a microservice architecture, with loosely coupled functional modules for easy expansion and maintenance. The main services include a call service subsystem and an artificial intelligence service subsystem. The call service subsystem includes a real-time communication module, a control module, and a scoring module. The scoring module includes a database service. The artificial intelligence service subsystem includes a data receiving module, a large language model module, and a data output module, and is a containerized service.

[0034] Each functional service module communicates via API (Application Programming Interface), employing high-speed networks and RPC (Remote Procedure Call) calls to ensure real-time performance. The specific deployment topology is as follows: telephone lines connect to the call server, which sends audio streams to the ASR (Automatic Speech Recognition) service of the data receiving module via Socket or HTTP interface and obtains audio output from the TTS (Text to Speech) service of the data output module. Simultaneously, it interacts with the dialogue management service to exchange text data. The dialogue management service coordinates the transmission of results from the data receiving module to the large language model, obtains responses, and then submits them to the synthesis process of the data output module. The database service stores follow-up plans, patient information, and follow-up results data, with a front-end management interface for administrators to view and operate. The system supports horizontal scaling, allowing the deployment of multiple instances of ASR and TTS services to handle concurrent calls, using load balancing to schedule calls to idle instances. The following provides a detailed description of the aforementioned functional modules.

[0035] The real-time communication module is used to transmit question data and answer data.

[0036] Specifically, the real-time communication module ensures the transmission and processing of audio data during the automatic follow-up call, guarantees real-time interaction between the automatic follow-up system and the stroke patient on the telephone line, and maintains the stability and real-time nature of the call.

[0037] In one possible implementation, the real-time communication module encodes, decodes, and cancels the audio of the automated follow-up system and the stroke patient, and interfaces with the data receiving module and the data output module.

[0038] The control module is used to read the stroke patient database, access the hospital's telephone network via the SIP protocol, and initiate telephone follow-ups with stroke patients in the database.

[0039] IP (Session Initiation Protocol) is a multimedia communication protocol developed by the IETF (Internet Engineering Task Force). It is a text-based application layer control protocol used to create, modify, and release sessions of one or more participants.

[0040] Specifically, the control module connects to the hospital's existing telephone network or call center platform via the standard SIP protocol to automatically make follow-up calls to stroke patients. It also utilizes open-source VoIP (Voice over IP) software to interact with and manage calls, enabling automatic dialing and call connection. The steps by which the control module automatically initiates the telephone follow-up call include: Step a1: Read the stroke patient database.

[0041] The stroke patient database contains information on stroke patients who have been treated and discharged from this hospital, such as patient lists, treatment times, discharge times, and patient phone numbers.

[0042] The control module accesses the stroke patient database and retrieves the list of patients requiring follow-up and their phone numbers.

[0043] Step a2: Access the hospital's telephone network via the SIP protocol and dial the patient's number.

[0044] The control module automatically dials the phone numbers of stroke patients in the stroke patient database via an external SIP connection. Once the call is connected, step a3 is executed.

[0045] The use of the SIP protocol ensures compatibility with various telephone lines, including PSTN (Public Switched Telephone Network) analog lines and IMS (IP Multimedia Subsystem) digital lines.

[0046] Step a3: Once the call is connected, establish a communication connection between the automatic follow-up system and the stroke patient, and process the call media stream through the real-time communication module.

[0047] Step a4: Send a greeting to the stroke patient.

[0048] After a stroke patient connects to the phone, the control module automatically generates a greeting, and the automatic follow-up conversation officially begins. The greeting is synthesized based on the current follow-up scenario. For example, the greeting voice is that of a middle-aged female nurse.

[0049] In one possible implementation, the control module also includes a redial strategy. When a call fails, a redial is initiated using the method described above after a preset time period.

[0050] The data receiving module includes a voice receiving submodule and a voice conversion submodule. The voice receiving submodule receives the voice responses from stroke patients, while the voice conversion submodule recognizes the voice responses in real time and converts them into text responses.

[0051] The Baichuan-M2 medical language model cannot directly understand and recognize speech information. Therefore, a data receiving module is used to convert speech into text information that the Baichuan-M2 medical language model can understand and recognize.

[0052] Specifically, the data receiving module integrates with the real-time communication module through a standard interface, acquiring real-time audio from the real-time communication module. Furthermore, the data receiving module employs a local speech recognition engine to optimize the current call scenario. This local speech recognition engine boasts high accuracy and low latency, enhancing robustness in noisy environments.

[0053] In one possible implementation, the local speech recognition engine is trained to recognize multiple language types to adapt to different patient populations. For example, language types include Mandarin, common dialects, and common foreign languages.

[0054] In one possible implementation, the data receiving module employs churn detection technology to convert the patient's voice response into a text response word by word in real time. In this approach, the system can transcribe the text word by word in real time without the patient needing to finish speaking, significantly reducing patient waiting time. The transcribed text is then immediately processed by the Baichuan-M2 medical large-scale language model, enabling efficient interaction.

[0055] The large language model module is used to analyze the response data based on the stroke assessment scale and generate question data using the Baichuan-M2 medical large language model.

[0056] The Baichuan-M2 medical language model was selected as the language model for the automated follow-up system. First, a large amount of dialogue data related to stroke follow-up was collected and a dataset was generated. The Baichuan-M2 medical language model was then trained, and the trained Baichuan-M2 medical language model was used to generate question data.

[0057] During the model training phase, the large language model module includes a data collection submodule, a dataset construction submodule, and a large language model training submodule.

[0058] The data collection submodule collects dialogue data from stroke follow-up.

[0059] Specifically, data from manual telephone follow-ups and simulated question-and-answer sessions were obtained as corpus for stroke follow-up dialogues. The dataset construction submodule generates multiple sets of question and response data pairs based on the stroke follow-up dialogue corpus to construct the training dataset. Further, the dataset construction submodule includes a data filtering unit and a dataset construction unit. The data filtering unit filters the stroke follow-up dialogue corpus based on the mRS scale to obtain stroke follow-up dialogue corpus containing the content required for mRS scale assessment; the dataset construction unit constructs the training dataset based on the filtered stroke follow-up dialogue corpus.

[0060] Specifically, the dialogue data is organized according to the key content of the mRS scale to obtain a question-answer pair dataset, which is used as the training dataset for the model.

[0061] For example, the compiled dataset of conversations between specialists and patients is as follows: Specialist: "Hello, I'm a stroke rehabilitation follow-up specialist. I'm making a phone call to check on your progress. How is your recovery going? Do you still require any ongoing care?"

[0062] Patient: "No, I don't need it."

[0063] Specialist: "Okay, do you need assistance from others when you are meeting your physical needs or walking?"

[0064] Patient: "Oh, this? No need."

[0065] Commissioner: "In cases of unusual daily life activities, such as physical assistance, verbal guidance, or supervision from others, is it still necessary to have assistance from others?"

[0066] Patient: "I can eat and sleep on my own now, I'm recovering quite well."

[0067] Specialist: "How does it feel compared to before? You can still do everything, just a little slower?"

[0068] Patient: "It's not slow either; I can walk and jump now."

[0069] Commissioner: "The recovery is going very well, and you can do everything by yourself, right?" Patient: "Yes." Specialist: "That's great! We wish you a smooth recovery and hope you maintain your healthy lifestyle. We'll contact you again during our next follow-up visit. Goodbye!" The large language model training submodule trains the Baichuan-M2 medical large language model based on the training dataset. Further, the large language model training submodule includes a construction unit, a prefix input unit, and a training unit.

[0070] The building blocks are used to construct the supervised fine-tuning loss function.

[0071] In one possible implementation, the building units employ supervised autoregressive fine-tuning (SFT) to construct the loss function, which is:

[0072] Specifically, the label smoothing function is set in the supervised fine-tuning loss function using the first setting subunit. ,in, , The label smoothing function smooths the labeled data in the training dataset, preventing the model from being overconfident in its predictions and improving generalization ability. A masking cue is added to the supervised fine-tuning loss function using the second setting subunit. ,in, Masking cues are used to control the calculation range of the supervised fine-tuning loss. If the token terminology belongs to medical follow-up dialogue or structured output, This is included in the loss calculation; if the token term belongs to system prompts, fill-in content, or unlearnable segments, That is, the weights of each term are not included in the loss. The third setting subunit is used to configure the weights of each term in the supervised fine-tuning loss function. ,in, For example, assigning higher weights to key tokens such as medical terms, warning words, and JSON key names enhances the professionalism and format stability of the output data.

[0073] The prefix input unit is used to write prompts, which are then used as the input prefix for the Baichuan-M2 medical big language model.

[0074] The prompts include follow-up identity information, dialogue objectives, and key information for data collection. Follow-up identity information describes the follow-up role, such as a stroke rehabilitation follow-up specialist; dialogue objectives aim to achieve the mRS scoring target; and key information for data collection outlines the content required for the mRS scale assessment.

[0075] Specifically, a template concatenation training method is adopted, in which system prompts S, user utterances Ui, and follow-up staff utterances Ai, containing follow-up identity information, dialogue target information, and key information to be collected, are injected as model input prefixes into the Baichuan-M2 medical big language model. The code block of the prefix input unit is represented as follows: <bos> <sys> S< / sys> <usr> U1< / usr> <ast> A1< / ast> ... <usr> Uk< / usr> <ast> Ak< / ast> <eos> This application utilizes prefix input units to ensure that the model-generated responses are both medically logical and humanistically relevant.

[0076] The training unit is used to train the Baichuan-M2 medical language model based on the training dataset. By minimizing the supervised fine-tuning loss function, the trained Baichuan-M2 medical language model is generated.

[0077] Specifically, during training, a fixed decoding order is adopted, and a combination of autoregression and teacher forcing is used to predict the current token based on the true preceding tokens. At the same time, a regular penalty is added to the EOS (sequence terminator) to prevent premature termination. Furthermore, the model parameters are iteratively optimized. Minimize the loss function to gradually adapt the model to the medical follow-up dialogue task.

[0078] The Baichuan-M2 medical language model, finely tuned using this application, can understand patients' statements regarding their functional status and ask and respond questions according to a pre-defined process. When patients stray from the topic or provide incomplete information, the model can also clarify or guide the conversation based on the context, ensuring a smooth question-and-answer process.

[0079] The data output module includes a text conversion submodule and a speech output submodule. The text conversion submodule determines the target timbre based on the follow-up scenario and uses speech synthesis technology to convert text questions into speech questions with the target timbre; the speech output submodule plays the speech questions to stroke patients.

[0080] During voice calls, the text generated by the Baichuan-M2 medical language model cannot be directly sent to stroke patients. Therefore, the data output module is used to convert the text into speech information that patients can understand.

[0081] Specifically, the data output module uses a natural-sounding Chinese speech synthesis engine to convert text questions into speech questions and then play them.

[0082] In one possible implementation, a suitable target timbre is determined based on the current follow-up scenario, such as the timbre of a middle-aged female nurse. The target timbre is then synthesized, which can improve the patient's auditory comfort and sense of trust.

[0083] In one possible implementation, the synthesis engine of the data output module is set to have low latency to ensure that text questions can be quickly converted into speech questions, thus avoiding excessive silences during the call.

[0084] Furthermore, the data output module is integrated with the real-time communication module through a standard interface, and the voice information generated by the data output module is injected into the downlink audio of the call through the real-time communication module.

[0085] Meanwhile, the data receiving module will continue to be used to receive response data from stroke patients.

[0086] The scoring module includes a data recording unit, an extraction unit, and a scoring unit. The data recording unit stores question and answer data and generates a dialogue log based on the timestamps of the question and answer data; the extraction unit extracts key semantics from the dialogue log based on the content required for the stroke assessment scale; the scoring unit is used to combine multiple key semantics to generate a stroke score for stroke patients.

[0087] Specifically, the data recording unit in the scoring module records voice questions and answers, saving the entire call audio as a recording file in a pop format, and saving text questions and answers as text files, forming a structured dialogue log in timestamp format. The dialogue log contains the text content of each round of questions and the patient's answers. After the call ends, the extraction unit extracts key semantics from the patient's answers, such as their ability to care for themselves and the degree of limitation in their mobility, and infers the mRS score range based on matching results, thus obtaining the overall mRS score.

[0088] Furthermore, the stroke score results and the corresponding key judgment criteria, i.e. the corresponding key semantics, are stored in the stroke patient database and associated with the information of stroke patients.

[0089] Furthermore, the automated follow-up system also includes a visual system interface, through which medical staff can query and analyze the mRS scores and call records of stroke patients during each follow-up visit, and generate follow-up reports.

[0090] In one possible implementation, encryption and access control units are set up in the data transmission of the real-time transmission module and the data storage of the scoring module to ensure information security.

[0091] In another possible implementation, a backup module is set up. When the current module fails, the backup module is automatically switched to achieve fault transfer and ensure that the follow-up service of the automatic follow-up system is continuous and uninterrupted.

[0092] This implementation method has the following beneficial effects: 1. Reduced Labor Costs: The automated follow-up system uses artificial intelligence to automatically make calls and interact intelligently, replacing the intensive work of one-on-one manual calls and significantly reducing the manpower required for follow-up. It can support batch parallel follow-ups, with a daily follow-up volume far exceeding that of manual methods, and can work continuously, greatly improving follow-up efficiency and significantly reducing the burden on medical staff.

[0093] 2. Standardized Follow-up Process: A large language model generates dialogue strategies according to a unified standard, ensuring consistent follow-up logic for all patients and avoiding process deviations caused by personnel differences in manual follow-up. Automated speech recognition and recording can avoid omissions or subjective biases in manual recording, ensuring the consistency and completeness of follow-up data.

[0094] 3. Automated Scoring and Accurate Recording: First, the automated follow-up system automatically completes the mRS score after the call, eliminating the need for manual recalculation, reducing human error, and improving the accuracy and objectivity of the score. Second, call recordings and written Q&A logs are automatically archived and stored, ensuring a complete record of the follow-up process. This facilitates statistical analysis, quality control, and supervision by medical institutions, and also provides data support for clinical research.

[0095] 4. High scalability: The system architecture and dialogue model have good versatility and can be extended to follow-up scenarios for other chronic diseases such as diabetes and hypertension. It is also suitable for postoperative rehabilitation follow-up, health management, and other fields. The underlying layer adopts a combination of trainable large language models and modular design, which has high customizability and portability. It can flexibly adapt to the follow-up and diagnosis needs of different departments and diseases, and has wide application value.

[0096] Correspondingly, this application proposes an automated follow-up method for stroke patients, applied to the automated follow-up system of this application. Please refer to [link to application]. Figure 3 , Figure 3 This is a flowchart illustrating an automated follow-up method for stroke patients provided in an embodiment of the present disclosure, as shown below. Figure 3 As shown, the automated follow-up method includes: S301, Receive response data from stroke patients.

[0097] S302. Based on the stroke assessment scale, the Baichuan-M2 medical big language model is used to analyze the response data and generate question data.

[0098] S303, Output problem data to stroke patients.

[0099] S304. Generate stroke scores for stroke patients based on text questions and text answers.

[0100] The specific content of this method is the same as the implementation content of each module of the aforementioned automated follow-up system, and will not be repeated here. To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the automatic follow-up method for stroke patients in the above embodiments.

[0101] Figure 4 This is a schematic diagram of the structure of a computing device provided in an embodiment of the present disclosure.

[0102] The following is a detailed reference. Figure 4 The diagram illustrates a structural schematic suitable for implementing the computing device 400 in the embodiments of this disclosure. The computing device 400 in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The computing device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0103] like Figure 4 As shown, the computing device 400 may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the computing device 400. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0104] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows computing device 400 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 A computing device 400 with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0105] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 409, or installed from memory 408, or installed from ROM 402. When the computer program is executed by processor 401, it performs the functions defined in the automated follow-up method for stroke patients according to embodiments of this disclosure.

[0106] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0107] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol), and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks ("LANs"), wide area networks ("WANs"), the Internet (e.g., the Internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0108] The aforementioned computer-readable medium may be included in the aforementioned computing device; or it may exist independently and not assembled into the computing device.

[0109] The aforementioned computer-readable medium carries one or more programs, which, when executed by the computing device, cause the computing device to perform the aforementioned automated follow-up method for stroke patients.

[0110] The computing device can be programmed with computer program code for performing the operations of this disclosure in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0111] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0112] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0113] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0114] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0115] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0116] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0117] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.< / eos> < / bos>

Claims

1. An automatic follow-up system for stroke patients, characterized by, The automatic follow-up system comprises a data receiving module, a large language model module, a data output module and a scoring module, wherein, The data receiving module is configured to receive answer data of a stroke patient, the answer data being an answer of the stroke patient to a greeting or a question data; The large language model module is configured to analyze the answer data based on a stroke health risk assessment scale and using a Baichuan-M2 medical large language model to generate a question data; The data output module is configured to output the question data to the stroke patient; The scoring module is configured to generate a stroke score of the stroke patient based on the question data and the answer data, the stroke score being used to evaluate the rehabilitation of the stroke patient.

2. The automatic follow-up system for stroke patients according to claim 1, characterized in that, The large language model module comprises: A data collection sub-module configured to collect stroke follow-up dialogue corpus, the stroke follow-up dialogue corpus comprising artificial telephone follow-up data and simulated question and answer data; A data set construction sub-module configured to generate a plurality of data pairs of question data and answer data based on the stroke follow-up dialogue corpus to construct a training data set; A large language model training sub-module configured to train the Baichuan-M2 medical large language model based on the training data set.

3. The system for automatic follow-up of stroke patients according to claim 2, characterized in that, The stroke health risk assessment scale is a modified Rankin scale, and the data set construction sub-module comprises: A data screening unit configured to screen the stroke follow-up dialogue corpus based on the modified Rankin scale to obtain stroke follow-up dialogue corpus containing the required content of the modified Rankin scale; A data set construction unit configured to construct a training data set based on the screened stroke follow-up dialogue corpus. 4.The automatic follow-up system for stroke patients according to claim 2, characterized in that, The large language model training sub-module comprises: A construction unit configured to construct a supervised fine-tuning loss function; A training unit configured to train the Baichuan-M2 medical large language model based on the training data set, minimize the supervised fine-tuning loss function, and generate a trained Baichuan-M2 medical large language model.

5. The system for automatic follow-up of stroke patients according to claim 4, characterized in that, The construction unit comprises: A first setting sub-unit configured to set a label smoothing function in the supervised fine-tuning loss function, the label smoothing function being used to smooth the label data in the training data set; A second setting sub-unit configured to set a masking prompt in the supervised fine-tuning loss function, the masking prompt being used to control the calculation range of the supervised fine-tuning loss; A third setting sub-unit configured to configure the weight of each word element in the supervised fine-tuning loss function. 6.The automatic follow-up system for stroke patients according to claim 4, characterized in that, The language model training sub-module further comprises: A prefix input unit configured to write prompt information, the prompt information comprising follow-up identity information, dialogue target information and collection key information; The prompt information is used as an input prefix of the Baichuan-M2 medical large language model. 7.The automatic follow-up system for stroke patients according to claim 1, wherein The scoring module comprises: A data recording unit configured to store the question data and the answer data and generate a dialogue log based on the time stamps of the question data and the answer data; An extraction unit is configured to extract key semantics in the dialogue log based on required content of the stroke assessment scale; A scoring unit is configured to generate a stroke score of the stroke patient in combination with the key semantics. 8.The automatic follow-up system of stroke patients according to claim 1, characterized in that, The question data is in text format, and the data receiving module comprises: A voice receiving sub-module configured to receive voice answers of the stroke patient; A voice conversion sub-module configured to recognize the voice answers in real time and convert the voice answers into text answers. 9.The automatic follow-up system for stroke patients according to claim 1, wherein, The data output module comprises: A text conversion sub-module configured to determine a target voice tone based on a follow-up scenario, and convert the text questions into voice questions in the target voice tone by using a voice synthesis technology; A voice output sub-module configured to play the voice questions to the stroke patient. 10.The automatic follow-up system for stroke patients according to claim 1, wherein, The automatic follow-up system further comprises: A control module configured to read a stroke patient database, access a hospital telephone network through a SIP protocol, and initiate a telephone follow-up to the stroke patient in the stroke patient database; A real-time communication module configured to transmit the question data and the answer data.

Images