Clinical informed agreement intelligent recommendation device based on BERT model

Abstract

The invention relates to a clinical informed agreement intelligent recommendation device based on a BERT model, and belongs to the field of digital medical treatment. Comprising a data acquisition and processing module used for acquiring and preprocessing historical medical record information, medical knowledge and clinical informed agreement samples of a patient; the model construction module is used for obtaining a trained clinical informed agreement generation model; the knowledge base construction module is used for extracting key information and terms through the trained clinical informed agreement generation model and constructing a clinical informed agreement template library and a term list library; the clinical informed agreement generation module is used for inputting data of a to-be-tested patient into a trained clinical informed agreement generation model and calling a clinical informed agreement template and a term list to generate a personalized clinical informed agreement; and the doctor-patient interaction platform is used for providing a chat window. According to the device, a personalized clinical informed agreement book is automatically generated for a patient, so that the patient can more accurately understand the content in the clinical informed agreement book while the working efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of digital medical technology, and in particular to a clinical informed consent intelligent recommendation device based on a BERT model. Background Art

[0002] In clinical medicine, the clinical informed consent form plays a key role in conveying treatment details and risks to patients to ensure that they fully understand the treatment content. However, existing clinical informed consent forms usually rely on doctors to write, which is time-consuming and inefficient. On the other hand, they are not customized according to the actual situation of the patient, resulting in the patient's inability to fully and accurately understand the content, thus affecting the patient's ability to make the right treatment choice.

[0003] Therefore, an intelligent recommendation device is needed to achieve personalized customization of clinical informed consent forms, thereby improving doctors' work efficiency and helping patients better understand treatment risks and improve patient satisfaction. Summary of the invention

[0004] In view of the above analysis, the present invention aims to provide an intelligent recommendation device for clinical informed consent based on the BERT model to solve the problem that the current clinical informed consent form relies on doctors to write it on the one hand, and is not personalized according to the actual situation of the patient on the other hand.

[0005] The present invention provides a clinical informed consent intelligent recommendation device based on a BERT model, the device comprising the following modules:

[0006] The data collection and processing module is used to collect the patient's historical medical information, medical knowledge and clinical informed consent form samples and pre-process them, and send the pre-processed data to the model building module;

[0007] A model building module, used to build a clinical informed consent form generation model based on the BERT model, and pre-train the clinical informed consent form generation model based on the preprocessed data to obtain a trained clinical informed consent form generation model;

[0008] The knowledge base construction module is used to extract key information and terms from the preprocessed data through the trained clinical informed consent form generation model and label them to build a knowledge base, build a clinical informed consent form template library based on the key information in the knowledge base, and build a term list library based on the terms in the knowledge base;

[0009] A clinical informed consent form generation module is used to input the pre-processed historical medical record information of the patient to be tested into the trained clinical informed consent form generation model and call the corresponding clinical informed consent form template and term list to generate a personalized clinical informed consent form for the patient to be tested;

[0010] The doctor-patient interaction platform is used to display the clinical informed consent form and provide an interaction window between doctors and patients.

[0011] Furthermore, preprocessing the collected data includes:

[0012] Analyze and clean the collected data;

[0013] The cleaned data is segmented using the WordPiece algorithm;

[0014] Encode the segmented data;

[0015] Each word is encoded according to the segmented data, and the encoding includes a Token corresponding to the word, a sentence information encoding corresponding to the word, and an absolute position information encoding of the word in the sentence.

[0016] Furthermore, the clinical informed consent form generation model includes an encoder layer and a decoder layer, the encoder layer includes a vocabulary embedding module, a sentence embedding module, a position embedding module and a Tm module, the vocabulary embedding module is used to obtain the Token corresponding to each vocabulary, the sentence embedding module is used to obtain the sentence information encoding corresponding to the vocabulary, the position embedding module is used to obtain the absolute position information encoding of the vocabulary in the sentence, the Tm module includes a multi-head self-attention mechanism and a feedforward neural network, the multi-head self-attention mechanism allows the model to consider all other input vocabulary when processing each vocabulary, the feedforward neural network performs a nonlinear transformation on the output of the multi-head self-attention mechanism and outputs a high-dimensional vector sequence, and the decoder layer decodes the high-dimensional vector sequence output by the encoder layer to output a target vocabulary sequence.

[0017] Furthermore, the expression of the input data of the Tm module is:

[0018] a i =[E(x i ), E Ai , E i ],

[0019] Among them, a i Represents x i The encoded data after the vocabulary embedding module, sentence embedding module and position embedding module, x i represents the i-th word in the sentence, E(x i ) represents x i The corresponding token, E Ai Represents x i The corresponding sentence information encoding, E i Represents x i The corresponding absolute position information encoding.

[0020] Furthermore, the content related to the keywords of the clinical informed consent form is obtained through the tags of the key information to construct a clinical informed consent form template;

[0021] Obtaining a surgical risk level according to the type of surgery, wherein the surgical risk level includes low risk, medium risk, and high risk;

[0022] Three templates are constructed for each type of surgery according to the surgical risk level, and the surgical type and surgical risk level corresponding to the template are used as its label. When the surgical risk level is low risk, the corresponding template does not contain medical history keywords and is supplemented with surgical risks and postoperative care content based on actual conditions. When the surgical risk level is medium risk and high risk, the corresponding template contains medical history keywords and is supplemented with surgical risks and postoperative care content based on actual conditions.

[0023] Furthermore, the medical knowledge level of the patients is set to a professional level, a good level or a general level according to the degree to which they understand medical terms; alternative words of different difficulty of understanding are set for each term according to different medical knowledge levels, and surgical type labels are marked; terms with the same surgical type label constitute a term list.

[0024] Furthermore, the clinical informed consent generation module includes a risk level identification unit, a knowledge base calling unit and a model running unit. The risk level identification unit is used to obtain a corresponding surgical risk level based on the surgical type of the patient to be tested; the knowledge base calling unit is used to obtain a corresponding clinical informed consent template and terminology list according to the surgical type and surgical risk level of the patient to be tested; the model running unit is used to run a trained clinical informed consent generation model, input the pre-processed data of the patient to be tested into the trained clinical informed consent generation model, and call the corresponding informed consent template and terminology list to generate a personalized clinical informed consent for the patient to be tested.

[0025] Furthermore, when the historical treatment plan of the patient to be tested includes similar operations, the differences and similarities between this operation and similar operations are added to the corresponding clinical informed consent template for dynamic updating; when the patient to be tested has a history of allergies, the drugs and materials used and their potential allergic reaction risks are added to the corresponding clinical informed consent template for dynamic updating; when the patient to be tested has a history of chronic disease, relevant chronic disease management recommendations and the risks and complications of related chronic diseases during surgery are added to the corresponding clinical informed consent template for dynamic updating, and a personalized clinical informed consent for the patient to be tested is generated based on the updated clinical informed consent template.

[0026] Furthermore, the clinical informed consent form generation model is pre-trained based on the pre-processed data to obtain a trained clinical informed consent form generation model including:

[0027] Get the dataset from the preprocessed data and perform data augmentation;

[0028] The data-augmented dataset is used as the input of the clinical informed consent form generation model, and pre-training is performed according to the pre-training task. The trained clinical informed consent form generation model is obtained by updating the parameters in its network structure.

[0029] Furthermore, the doctor-patient interaction platform includes a front-end module, a communication module, a management module, and a back-end module. The front-end module is used to establish a user interface through a component library, the communication module is used to realize communication between the front-end module and the back-end module by establishing a full-duplex communication channel, the management module is used to manage the user's status, and the back-end module is used to realize user authentication and authority control, message processing, and data storage.

[0030] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0031] 1. The present invention constructs a clinical informed consent form template and a term list library, and automatically generates high-quality clinical informed consent forms based on the BERT model, thereby improving work efficiency and solving the problem that the current clinical informed consent forms rely on doctors to write and are inefficient.

[0032] 2. The present invention customizes a personalized clinical informed consent form for the patient based on the patient's medical history information, medical knowledge level, treatment plan and risk level, so that the patient can understand the medical terms, treatment information and treatment risks more comprehensively and accurately, and thus make the right treatment choice.

[0033] 3. The present invention constructs a clinical informed consent form template library and a term list library by collecting a large number of clinical informed consent form samples, medical literature, and medical standards, and continuously updates the clinical informed consent form template library and the term list library by supplementing new medical literature and medical standards to obtain the latest medical knowledge and terminology, thereby improving the accuracy of generating clinical informed consent forms.

[0034] 4. The present invention provides a chat function for patients and doctors through a doctor-patient interaction platform to conduct consultation and feedback related to the clinical informed consent form, thereby promoting communication between doctors and patients and improving patient satisfaction.

[0035] In the present invention, the above-mentioned technical solutions can also be combined with each other to achieve more preferred combination solutions. Other features and advantages of the present invention will be described in the subsequent description, and some advantages can become obvious from the description, or can be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the contents particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings are only for the purpose of illustrating specific embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0037] Figure 1 A block diagram of an intelligent recommendation device for clinical informed consent based on a BERT model according to an embodiment of the present invention;

[0038] Figure 2 A schematic diagram of input data for a clinical informed consent generation model according to an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the model structure for generating a clinical informed consent form according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0041] A specific embodiment of the present invention discloses an intelligent recommendation device for clinical informed consent based on a BERT model. Figure 1 As shown, the device comprises:

[0042] The data collection and processing module is used to collect the patient's historical medical information, medical knowledge and clinical informed consent form samples and pre-process them, and send the pre-processed data to the model building module;

[0043] A model building module, used to build a clinical informed consent form generation model based on the BERT model, and pre-train the clinical informed consent form generation model based on the preprocessed data to obtain a trained clinical informed consent form generation model;

[0044] The knowledge base construction module is used to extract key information and terms from the preprocessed data through the trained clinical informed consent form generation model and label them to build a knowledge base, build a clinical informed consent form template library based on the key information in the knowledge base, and build a term list library based on the terms in the knowledge base;

[0045] A clinical informed consent form generation module is used to input the pre-processed historical medical record information of the patient to be tested into the trained clinical informed consent form generation model and call the corresponding clinical informed consent form template and term list to generate a personalized clinical informed consent form for the patient to be tested;

[0046] The doctor-patient interaction platform is used to display the clinical informed consent form and provide an interaction window between doctors and patients.

[0047] Furthermore, the data acquisition and processing module includes a data acquisition unit and a data processing unit. The data acquisition unit is used to collect the patient's historical medical record information through the electronic medical record system, crawl medical-related knowledge through the Internet, collect clinical informed consent form samples through various medical institutions and send them to the data processing unit; the data processing unit is used to pre-process the received data.

[0048] Specifically, the patient's medical history information includes the patient's personal information, medical information, examination / imaging information, doctor's advice information, diagnosis information, treatment information, medical history information, hospitalization records, and follow-up information. A large amount of medical-related knowledge was crawled from the Internet, including medical terminology, medical standards, disease information, drug information, treatment information, medical literature, medical e-books and other documents. A large number of informed consent form samples were collected from various medical institutions, covering various diseases, treatment plans and medical fields to ensure the comprehensiveness and diversity of the collected data content.

[0049] Specifically, disease information includes the definition, symptoms, diagnostic criteria, and treatment methods of the disease; drug information includes the name, purpose, usage, dosage, and side effects of the drug; treatment information includes examinations (such as blood tests, imaging examinations) and their steps and precautions, and treatments (such as surgery, radiotherapy, chemotherapy) and their steps and precautions.

[0050] Furthermore, preprocessing the collected data includes:

[0051] Analyze and clean the collected data;

[0052] The cleaned data is segmented using the WordPiece algorithm;

[0053] Encode the segmented data.

[0054] Specifically, parsing refers to parsing data in different formats (such as PDF, Word, HTML, etc.) into a unified data format (such as a string). Cleaning includes removing irrelevant symbols, removing duplicate parts, standardizing, and removing irrelevant information from the parsed data; the removal of irrelevant symbols refers to removing symbols that have no effect on semantics, such as line breaks, HTML tags, page numbers, etc.; the removal of duplicate parts refers to removing semantically repeated parts to avoid redundant information in subsequent retrieval; the standardization refers to unifying terms with the same semantics, such as unifying "ASA" and "aspirin"; the removal of irrelevant information refers to removing information without substantive content, such as headers, footers, signatures, etc. The WordPiece algorithm is used to segment continuous text (such as the description of the inspection process) into independent sentences, and the sentences are segmented into independent words. Each word is encoded according to the segmented data, and the encoding includes the token corresponding to the word, the sentence information encoding corresponding to the word, and the absolute position information encoding of the word in the sentence. The encoded data is the basic unit that can be understood and processed by the clinical informed consent form generation model constructed later.

[0055] For example, the preprocessed data is as follows: Figure 2 As shown, [CLS] indicates the start position, [SEP] indicates the sentence separation, [E A ] represents the sentence information encoding of the first sentence, [E B ] represents the sentence information encoding of the second sentence. According to the position of "disease" in "disease heart valve disease", its absolute position information is encoded as [E 1 ].

[0056] Furthermore, the network structure of the clinical informed consent generation model is as follows: Figure 3 As shown, it includes an encoder layer and a decoder layer, wherein the encoder layer is used to convert the preprocessed data into a high-dimensional vector sequence and input the high-dimensional vector sequence into the decoder layer, and the decoder layer is used to generate a target vocabulary sequence from the output of the decoder.

[0057] Specifically, the encoder layer includes a vocabulary embedding module, a sentence embedding module, a position embedding module and a Tm module. The vocabulary embedding module is used to obtain the token corresponding to each vocabulary, such as Figure 3 "E[CLS]", "E[disease]", "E[SEP]" in the sentence embedding module; the sentence embedding module is used to obtain the sentence information encoding corresponding to the vocabulary, such as Figure 3 The "E" in A "; the position embedding module is used to obtain the absolute position information encoding of the word in the sentence, such as Figure 3 The "E" in 0 ”, “E 1 ”, “E N"; The Tm module includes a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism allows the model to consider all other input words when processing each word. The feedforward neural network performs a nonlinear transformation on the output of the multi-head self-attention mechanism and outputs a high-dimensional vector sequence to capture the semantic and grammatical relationships between words, thereby representing the semantic information of the text. The decoder layer decodes the high-dimensional vector sequence output by the encoder layer and outputs a target vocabulary sequence, such as Figure 3 The "T" in 0 ”, “T 1 ”, “T N ”.

[0058] Specifically, the expression of the input data of the Tm module is:

[0059]

[0060] Among them, a i Represents x i The encoded data after the vocabulary embedding module, sentence embedding module and position embedding module, x i represents the i-th word in the sentence, E(x i ) represents x i The corresponding token, E Ai Represents x i The corresponding sentence information encoding, E i Represents x i The corresponding absolute position information encoding.

[0061] It is understandable that since the encoder of the BERT model does not rely on convolution operations or loop structures, it is unable to capture the position information of a single word when processing a word sequence. Therefore, the present application encodes the absolute position information of each word during preprocessing, thereby introducing the position encoding of the word in the encoder layer of the clinical informed consent form generation model. By adding the token corresponding to the word, the sentence information encoding, and the absolute position information encoding, the clinical informed consent form generation model can capture the positional relationship of the word in the sequence, thereby being able to more accurately understand the contextual information.

[0062] Furthermore, the pre-training of the clinical informed consent form generation model based on the pre-processed data to obtain a trained clinical informed consent form generation model includes:

[0063] Get the dataset from the preprocessed data and perform data augmentation;

[0064] The data-augmented dataset is used as the input of the clinical informed consent form generation model, and pre-training is performed according to the pre-training task. The trained clinical informed consent form generation model is obtained by updating the parameters in its network structure.

[0065] Specifically, data enhancement is performed by randomly deleting words, randomly replacing words, and inserting synonyms or antonyms. The pre-training task includes masking training (MLM, Mask Language Model) and next sentence prediction training (NSP, Next Sentence Prediction). The masking training refers to randomly masking certain input words, and then predicting the masked words based on the context to represent the current context, thereby improving the model's ability to understand the context; the next sentence prediction training refers to judging whether there is a logical association between two sentences, thereby improving the model's ability to understand the context.

[0066] It should be noted that during the training process, taking into account the particularity of medical texts, such as the diversity of terms and abbreviations, differences in text length and other factors, this application adopts a strategy of multiple rounds of training and dynamic masking to improve the generalization ability and robustness of the model.

[0067] It is understandable that pre-training only fine-tunes the model parameters through training tasks, which reduces the training time. After pre-training, the trained model can be used to extract information and perform semantic analysis on new pre-processed data.

[0068] Furthermore, the key information includes the type of surgery (name of surgery), potential risks and complications related to the surgery, patient information, treatment purpose, expected treatment effect, postoperative care instructions and precautions, etc. The terms include medical terms (such as cardiac arrest, cardiac output, atherosclerosis), surgery-related terms (such as incision, suture, depth of anesthesia), etc.

[0069] Furthermore, newly added medical knowledge-related documents are added to the knowledge base in the form of RAG to update it, so as to supplement the latest medical knowledge, thereby improving the basis for improving the accuracy of generating clinical informed consent forms.

[0070] Furthermore, the content related to the clinical informed consent keywords is obtained through the tags of key information to construct a clinical informed consent template.

[0071] Specifically, the keywords of the clinical informed consent form include the patient's name, gender, age, hospitalization number, department, bed number, type of surgery, purpose of surgery, preoperative diagnosis, surgical risk, postoperative care, and medical history. The surgical risk level is obtained according to the type of surgery, and the surgical risk level includes low risk, medium risk, and high risk. A clinical informed consent form template is constructed based on the above keywords and their related content, wherein three templates are constructed for each type of surgery according to the surgical risk level, and the surgical type and surgical risk level corresponding to the template are used as its label. When the surgical risk level is low risk, the corresponding template does not contain the medical history keyword and supplements the surgical risk and postoperative care content according to the actual situation. When the surgical risk level is medium risk and high risk, the corresponding template contains the medical history keyword and supplements the surgical risk and postoperative care content according to the actual situation.

[0072] It should be noted that all clinical informed consent form templates constitute a clinical informed consent form template library, and the corresponding templates can be called through the surgery type and surgery risk level labels.

[0073] Specifically, the medical knowledge level of the patient is set to professional level, good level or general level according to the degree of understanding of medical terms. For each term, different replacement words with different understanding difficulties are set according to different medical knowledge levels, and the surgical type label is marked. Terms with the same surgical type label constitute a term list, and all term lists constitute a term list library.

[0074] It should be noted that when setting replacement terms, it is necessary to ensure that the replacement terms are semantically consistent with the original terms and meet the needs of patients with different levels of medical knowledge.

[0075] For example, the term list is as follows:

[0076]

[0077] It can be understood that the present invention constructs a clinical informed consent form template library and a term list library by collecting a large number of clinical informed consent form samples, medical literature, and medical standards, and continuously updates the clinical informed consent form template library and the term list library by supplementing new medical literature and medical standards to obtain the latest medical knowledge and terminology, thereby improving the accuracy of generating clinical informed consent forms.

[0078] Furthermore, the clinical informed consent generation module includes a risk level identification unit, a knowledge base calling unit and a model running unit. The risk level identification unit is used to obtain a corresponding surgical risk level based on the surgical type of the patient to be tested; the knowledge base calling unit is used to obtain a corresponding clinical informed consent template and terminology list according to the surgical type and surgical risk level of the patient to be tested; the model running unit is used to run a trained clinical informed consent generation model, input the pre-processed data of the patient to be tested into the trained clinical informed consent generation model, and call the corresponding informed consent template and terminology list to generate a personalized clinical informed consent for the patient to be tested.

[0079] Specifically, the data collection and processing module collects the historical medical record information of the patient to be tested and performs preprocessing to obtain the preprocessed data of the patient to be tested, and the preprocessed data of the patient to be tested includes his name, age, gender, surgery type, medical history information, drug use information, diagnosis information, disease type, treatment information, and medical knowledge level. The risk level identification unit queries the corresponding level in the surgery classification catalog (national standard) according to the surgery type of the patient to be tested. The levels include level one to level four, and level one or level two is set to low risk, level three is set to medium risk, and level four is set to high risk. The knowledge base calling unit retrieves the clinical informed consent form template with corresponding labels in the clinical informed consent form template library according to the surgery type and surgery risk level of the patient to be tested, and retrieves the term list with corresponding labels in the term list library according to the surgery type of the patient to be tested. The trained clinical informed consent generation model identifies relevant terms, disease type, medical history information and treatment plan based on the input name, age, gender, surgery type, medical history information, medication information, diagnosis information and treatment information of the patient to be tested, outputs the content corresponding to each keyword in the clinical informed consent template, searches for each term in the term list in the output content, and if found, replaces it with the term corresponding to the medical knowledge level column, thereby generating a personalized clinical informed consent form for the patient to be tested.

[0080] Specifically, when the historical treatment plan of the patient to be tested includes similar operations, the differences and similarities between the current operation and similar operations are added to the corresponding clinical informed consent template for dynamic update; when the patient to be tested has a history of allergies, the drugs and materials used and their potential allergic reaction risks are added to the corresponding clinical informed consent template for dynamic update; when the patient to be tested has a history of chronic disease, the relevant chronic disease management recommendations and the risks and complications of the relevant chronic disease during surgery are added to the corresponding clinical informed consent template for dynamic update, and a personalized clinical informed consent for the patient to be tested is generated based on the updated clinical informed consent template.

[0081] It should be noted that the similar surgeries refer to surgeries that are classified into one category according to the ICD10 surgical codes.

[0082] Exemplarily, it is assumed that the medical history information of the patient to be tested is hypertension and heart valve disease, the type of surgery is heart bypass surgery, and the patient's medical knowledge level is general. According to heart bypass surgery, the surgical risk type is high risk. According to heart bypass surgery and high risk, a clinical informed consent template with heart bypass surgery and high risk labels is retrieved in the clinical informed consent template library, and a term list with heart bypass surgery labels is retrieved in the term list library according to heart bypass surgery. The trained clinical informed consent generation model generates a clinical informed consent for heart bypass surgery based on the information of the patient to be tested, wherein, since the patient has a history of hypertension and heart valve disease, the content corresponding to the keywords of preoperative and postoperative precautions will detail the antihypertensive drug information that the patient needs to take according to the "drug use information", and the content corresponding to the surgical risk keyword will describe the risks of hypertension in surgery and related complications in detail; and, each term in the term list is searched in the output content, and if it is found, it is replaced with the term corresponding to the term in the general level column to make it easier for the patient to be tested to understand.

[0083] It should be noted that clinical informed consent forms usually include surgical informed consent forms, special examination informed consent forms and special treatment informed consent forms. There are many types of surgical informed consent forms and the contents they contain are more complex. Therefore, the technical solution of this application is explained using the surgical informed consent form as an example. Other types of clinical informed consent forms have corresponding templates and labels, which are generated by calling the corresponding templates through the labels using the above method.

[0084] It can be understood that the present invention automatically generates high-quality clinical informed consent forms based on the BERT model by constructing a clinical informed consent form template and a term list library, thereby improving work efficiency and solving the problem that the current clinical informed consent forms rely on doctors to write and are inefficient. The present invention customizes a personalized clinical informed consent form for the patient based on the patient's medical history information, medical knowledge level, treatment plan and risk level, so that the patient can understand the medical terms, treatment information and treatment risks more comprehensively and accurately, and thus make the right treatment choice.

[0085] Furthermore, the doctor-patient interaction platform includes a front-end module, a communication module, a management module, and a back-end module. The front-end module is used to establish a user interface through a component library, the communication module is used to realize communication between the front-end module and the back-end module by establishing a full-duplex communication channel, the management module is used to manage the user's status, and the back-end module is used to realize user authentication and authority control, message processing, and data storage.

[0086] It should be noted that the user interface includes a chat main interface, a message input box, a user list, and a message list, and the user interface can adapt to screens of different sizes and different types of devices. The user status includes the user's connection status (including the user's login and logout) and online status. The message processing includes parsing the message content. The doctor-patient interaction platform provides a chat function so that patients and doctors can communicate at any time and place, so that patients and doctors can consult and provide feedback on the clinical informed consent form, thereby promptly answering patients' doubts about the clinical informed consent form.

[0087] Exemplarily, the front-end module uses the Vue3 framework to develop the components required for the user interface through the Composition API. The communication module uses WebSocket to achieve communication between the front-end module and the back-end module by establishing a full-duplex communication connection between them. The management module uses the Vuex state management library to manage user login, message sending and receiving, online, logout and other states. The back-end module uses the Express framework to build a Node.js server, integrates Socket.IO to monitor user connections and message sending and receiving events, and realizes message processing.

[0088] It can be understood that the present invention provides a chat function for patients and doctors through a doctor-patient interaction platform to conduct consultation and feedback related to clinical informed consent, thereby promoting communication between doctors and patients and improving patient satisfaction.

[0089] Compared with the prior art, the beneficial effects of the clinical informed consent intelligent recommendation device based on the BERT model provided by the present invention are as follows:

[0090] 1. The present invention constructs a clinical informed consent form template and a term list library, and automatically generates high-quality clinical informed consent forms based on the BERT model, thereby improving work efficiency and solving the problem that the current clinical informed consent forms rely on doctors to write and are inefficient.

[0091] 2. The present invention customizes a personalized clinical informed consent form for the patient based on the patient's medical history information, medical knowledge level, treatment plan and risk level, so that the patient can understand the medical terms, treatment information and treatment risks more comprehensively and accurately, and thus make the right treatment choice.

[0092] 3. The present invention constructs a clinical informed consent form template library and a term list library by collecting a large number of clinical informed consent form samples, medical literature, and medical standards, and continuously updates the clinical informed consent form template library and the term list library by supplementing new medical literature and medical standards to obtain the latest medical knowledge and terminology, thereby improving the accuracy of generating clinical informed consent forms.

[0093] 4. The present invention provides a chat function for patients and doctors through a doctor-patient interaction platform to conduct consultation and feedback related to the clinical informed consent form, thereby promoting communication between doctors and patients and improving patient satisfaction.

[0094] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0095] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A clinical informed consent intelligent recommendation device based on the BERT model, characterized in that: The device comprises: The data collection and processing module is used to collect the patient's historical medical information, medical knowledge and clinical informed consent form samples and pre-process them, and send the pre-processed data to the model building module; A model building module, used to build a clinical informed consent form generation model based on the BERT model, and pre-train the clinical informed consent form generation model based on the preprocessed data to obtain a trained clinical informed consent form generation model; The knowledge base construction module is used to extract key information and terms from the preprocessed data through the trained clinical informed consent form generation model and label them to build a knowledge base, build a clinical informed consent form template library based on the key information in the knowledge base, and build a term list library based on the terms in the knowledge base; A clinical informed consent form generation module is used to input the pre-processed historical medical record information of the patient to be tested into the trained clinical informed consent form generation model and call the corresponding clinical informed consent form template and term list to generate a personalized clinical informed consent form for the patient to be tested; The doctor-patient interaction platform is used to display the clinical informed consent form and provide an interaction window between doctors and patients.

2. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 1, characterized in that: The preprocessing of the collected data includes: Analyze and clean the collected data; The cleaned data is segmented using the WordPiece algorithm; Encode the segmented data; Each word is encoded according to the segmented data, and the encoding includes a Token corresponding to the word, a sentence information encoding corresponding to the word, and an absolute position information encoding of the word in the sentence.

3. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 1, characterized in that: The clinical informed consent form generation model includes an encoder layer and a decoder layer. The encoder layer includes a vocabulary embedding module, a sentence embedding module, a position embedding module and a Tm module. The vocabulary embedding module is used to obtain the token corresponding to each vocabulary, the sentence embedding module is used to obtain the sentence information encoding corresponding to the vocabulary, and the position embedding module is used to obtain the absolute position information encoding of the vocabulary in the sentence. The Tm module includes a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism allows the model to consider all other input vocabulary when processing each vocabulary. The feedforward neural network performs a nonlinear transformation on the output of the multi-head self-attention mechanism and outputs a high-dimensional vector sequence. The decoder layer decodes the high-dimensional vector sequence output by the encoder layer and outputs a target vocabulary sequence.

4. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 2, characterized in that: The expression of the input data of the Tm module is: Among them, a i Represents x i The encoded data after the vocabulary embedding module, sentence embedding module and position embedding module, x i represents the i-th word in the sentence, E(x i ) represents x i The corresponding token, E Ai Represents x i The corresponding sentence information encoding, E i Represents x i The corresponding absolute position information encoding.

5. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 1, characterized in that: Through the labels of key information, the content related to the keywords of clinical informed consent form is obtained to build a clinical informed consent form template; Obtaining a surgical risk level according to the type of surgery, wherein the surgical risk level includes low risk, medium risk, and high risk; Three templates are constructed for each type of surgery according to the surgical risk level, and the surgical type and surgical risk level corresponding to the template are used as its label. When the surgical risk level is low risk, the corresponding template does not contain medical history keywords and is supplemented with surgical risks and postoperative care content based on actual conditions. When the surgical risk level is medium risk and high risk, the corresponding template contains medical history keywords and is supplemented with surgical risks and postoperative care content based on actual conditions.

6. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 5, characterized in that: The medical knowledge level of the patients is set to professional level, good level or general level according to the degree of their understanding of medical terms; for each term, alternative words with different levels of difficulty of understanding are set according to different medical knowledge levels, and the surgical type labels are marked; terms with the same surgical type label constitute a term list.

7. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 1, characterized in that: The clinical informed consent form generation module includes a risk level identification unit, a knowledge base calling unit and a model running unit, wherein the risk level identification unit is used to obtain a corresponding surgical risk level based on the surgical type of the patient to be tested; The knowledge base calling unit is used to obtain the corresponding clinical informed consent form template and term list according to the surgery type and surgery risk level of the patient to be tested; The model running unit is used to run the trained clinical informed consent form generation model, input the pre-processed data of the patient to be tested into the trained clinical informed consent form generation model and call the corresponding informed consent form template and term list to generate a personalized clinical informed consent form for the patient to be tested.

8. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 7, characterized in that: When the patient's historical treatment plan includes similar surgeries, the differences and similarities between this surgery and similar surgeries will be added to the corresponding clinical informed consent template for dynamic update; when the patient has a history of allergies, the drugs and materials used and their potential allergic reaction risks will be added to the corresponding clinical informed consent template for dynamic update; when the patient has a history of chronic disease, the relevant chronic disease management recommendations and the risks and complications of the relevant chronic disease during surgery will be added to the corresponding clinical informed consent template for dynamic update, and a personalized clinical informed consent for the patient will be generated based on the updated clinical informed consent template.

9. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 1, characterized in that: The clinical informed consent form generation model is pre-trained based on the pre-processed data to obtain a trained clinical informed consent form generation model including: Get the dataset from the preprocessed data and perform data augmentation; The data-augmented dataset is used as the input of the clinical informed consent form generation model, and pre-training is performed according to the pre-training task. The trained clinical informed consent form generation model is obtained by updating the parameters in its network structure.

10. The clinical informed consent intelligent recommendation device based on the BERT model according to claim 1, characterized in that: The doctor-patient interaction platform includes a front-end module, a communication module, a management module, and a back-end module. The front-end module is used to establish a user interface through a component library, the communication module is used to achieve communication between the front-end module and the back-end module by establishing a full-duplex communication channel, the management module is used to manage the user's status, and the back-end module is used to achieve user authentication and authority control, message processing, and data storage.

Images