Artificial intelligence-based medical information publishing method and system

Abstract

The application provides a medical information publishing method and system based on artificial intelligence. The method can input the Nth semantic information and the Nth determined plurality of information types into a medical information conversion model to determine the Nth medical information for medical staff and the Nth medical information for patients corresponding to the original medical information; perform quality assessment on the Nth medical information for medical staff according to the professional degree, the seriousness of the emotional tendency, and the accuracy of the Nth medical information for medical staff to obtain a first quality assessment value; perform quality assessment on the Nth medical information for patients according to the understandability, the friendliness of the emotional tendency, and the accuracy of the Nth medical information for patients to obtain a second quality assessment value; and determine a third quality assessment value according to the matching degree of the medical staff version sub-information of each information type and the patient version sub-information of the corresponding information type, so that the medical information published by the hospital has a high adaptation degree for medical staff and patients.

Description

Technical Field

[0001] This application relates to the field of medical information publishing technology, and in particular to a medical information publishing method and system based on artificial intelligence. Background Technology

[0002] In some scenarios, hospitals need to simultaneously release medical information to healthcare staff and patients. This medical information can include, but is not limited to, information on changes to the medical process, public health and safety information, information on large-scale hospital events, or information on emergencies.

[0003] Currently, hospital administration typically compiles and distributes medical information to healthcare workers and patients. However, due to the different roles doctors and patients play in the hospital and their varying levels of professional knowledge, the medical information released by hospitals is often poorly suited to both healthcare workers and patients. This results in a less than ideal reading experience for both groups, and they struggle to accurately grasp the meaning of the information. Summary of the Invention

[0004] This application provides an artificial intelligence-based medical information publishing method and system to address the problems in the existing technology where, due to the different roles doctors and patients play in hospitals and the different professional knowledge they possess, the medical information published by hospitals has low adaptability for medical staff and patients, resulting in a poor reading experience for both and difficulty in accurately grasping the meaning of the medical information.

[0005] Firstly, this application provides an artificial intelligence-based method for publishing medical information, applied to a hospital integrated server. The method provided in this application includes:

[0006] Step 1: Receive raw medical information to be published from the hospital's administrative terminal;

[0007] Step 2: For the Nth time, semantic analysis is performed on the original medical information based on the pre-trained semantic analysis model to obtain the Nth semantic information of the original medical information. The semantic analysis model is trained by inputting multiple first training samples into the first neural network. Each first training sample includes historical original medical information labeled with historical semantic information, and N is an integer greater than or equal to 2.

[0008] Step 3: Input the Nth semantic information into the pre-trained information type determination model, and determine the multiple information types corresponding to the semantic information for the Nth time. The information type determination model is trained by inputting multiple second training samples into the second neural network. Each second training sample includes the semantic information of the historical original medical information and multiple historical sub-information types labeled by experts.

[0009] Step 4: Input the Nth semantic information and the multiple information types determined in the Nth time into the pre-trained medical information conversion model, so that the medical information conversion model can determine the Nth medical care version and the Nth patient version medical information corresponding to the original medical information based on the Nth semantic information. The Nth medical care version medical information includes multiple medical care version sub-information, each of which corresponds to an information type. The Nth patient version medical information includes multiple patient version sub-information, each of which corresponds to an information type. The multiple information types of the Nth patient version medical information correspond one-to-one with the multiple information types of the medical care version medical information. The medical information conversion model is trained by inputting multiple third training samples into a third neural network. Each third training sample includes the semantic information of the historical original medical information pair and its corresponding multiple historical information types, the corresponding historical medical care version medical information, and the corresponding historical patient version medical information.

[0010] Step 5: Analyze the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, and analyze the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information;

[0011] Step 6: Based on the professionalism, emotional seriousness, and accuracy of the Nth version of medical information, conduct a quality assessment of the Nth version of medical information to obtain the first quality assessment value of the Nth version of medical information.

[0012] Step 7: Based on the comprehensibility, emotional friendliness, and accuracy of the Nth patient version of the medical information, conduct a quality assessment of the Nth patient version of the medical information to obtain a second quality assessment value for the Nth patient version of the medical information.

[0013] Step 8: Determine the matching degree between the medical staff version sub-information of each information type in the Nth medical staff version and the corresponding patient version sub-information of the Nth patient version;

[0014] Step 9: Based on the matching degree between the medical staff version sub-information of each information type and the patient version sub-information of the corresponding information type, determine the third quality assessment value of the Nth patient version medical information and the Nth medical staff version medical information;

[0015] Step 10: Based on the first quality assessment value, the second quality assessment value, and the third quality assessment value, obtain the Nth comprehensive quality assessment value;

[0016] Step 11: If the Nth comprehensive quality assessment value is less than the set quality assessment value, use the first reinforcement learning model to update the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model based on the Nth comprehensive quality assessment value; and increment the value of N by 1, repeating steps 2-10 until the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

[0017] Step 12: Publish the Nth medical information (medical care version) to the medical care terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, and publish the Nth patient information (patient version) to the patient terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

[0018] In some implementations, the analysis includes assessing the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, as well as the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information, including:

[0019] The professionalism of the Nth medical and nursing version of medical information is determined by the first proportion of the number of professional terms in the Nth medical and nursing version of medical information to the total number of terms in the Nth medical and nursing version of medical information. The first proportion is positively correlated with the professionalism.

[0020] The seriousness of the Nth medical information is determined by the number of words with a serious emotional tendency in the Nth version of medical information and the second proportion of the total number of words in the Nth version of medical information. The second proportion is positively correlated with the seriousness.

[0021] Determine the third percentage of the number of defective statements in the Nth version of medical information relative to the total number of statements in the Nth version of medical information. The third percentage is negatively correlated with accuracy. Defects are defined as at least one of the following: ambiguous statements, incomplete statements, and semantic inconsistencies in statements.

[0022] In some implementations, the quality of the Nth healthcare version of medical information is assessed based on its professionalism, the seriousness of its emotional tone, and its accuracy, resulting in a first quality assessment value for the Nth healthcare version of medical information, including:

[0023] Based on the professionalism of the Nth medical information, find the corresponding professionalism score from the preset scoring relationship table;

[0024] Based on the seriousness of the emotional tendency of the Nth medical information, find the emotional tendency score corresponding to the seriousness from the preset scoring relationship table;

[0025] Based on the accuracy of the Nth medical information, find the accuracy score corresponding to the accuracy from the preset scoring relationship table;

[0026] The first quality assessment value of the Nth version of medical information is obtained by weighting the professionalism score, accuracy score, and sentiment score.

[0027] In some implementations, the quality of the Nth patient's medical information is assessed based on its comprehensibility, emotional friendliness, and accuracy, resulting in a second quality assessment value for the Nth patient's medical information, including:

[0028] The comprehensibility of the Nth patient's medical information is determined by the fourth percentage of the number of technical terms in the Nth patient's medical information compared to the total number of terms in the Nth patient's medical information. The fourth percentage is negatively correlated with the comprehensibility.

[0029] The friendliness level of the Nth patient's medical information is determined by the number of words with a friendly sentiment in the Nth patient's medical information and the fifth percentage of the total number of words in the Nth patient's medical information. The fifth percentage is positively correlated with the friendliness level.

[0030] Determine the sixth percentage of the number of defective statements in the Nth patient version of medical information relative to the total number of statements in the Nth patient version of medical information. The sixth percentage is negatively correlated with accuracy. Defects are defined as at least one of the following: ambiguous statements, incomplete statements, and semantic inconsistencies in statements.

[0031] In some implementations, determining the matching degree between the medical staff version sub-information of each information type in the Nth medical staff version and the corresponding patient version sub-information of the Nth patient version of the medical information includes:

[0032] For each information type, determine the semantic similarity between the medical staff version information and the patient version information under the information type, and define the semantic similarity as the matching degree between the medical staff version information and the patient version information under the information type.

[0033] In some implementations, a third quality assessment value is determined for the Nth patient version of medical information and the Nth medical staff version of medical information based on the matching degree between the medical staff sub-information of each information type and the corresponding patient version sub-information.

[0034] Based on the matching degree between the medical staff sub-information of each information type and the patient sub-information of the corresponding information type, the matching degree score corresponding to each information type is retrieved from the preset scoring relationship table;

[0035] The matching scores for each information type are weighted and averaged to obtain the third quality assessment value for the Nth patient version of medical information and the Nth nurse version of medical information.

[0036] In some implementations, the Nth comprehensive quality assessment value is obtained based on the first quality assessment value, the second quality assessment value, and the third quality assessment value, including:

[0037] The weighted average of the first, second, and third quality assessment values ​​is used to obtain the Nth comprehensive quality assessment value.

[0038] In some implementations, publishing the Nth healthcare version of medical information to the healthcare terminal when the Nth overall quality assessment value is greater than or equal to a set quality assessment value, and publishing the Nth patient version of medical information to the patient terminal when the Nth overall quality assessment value is greater than or equal to a set quality assessment value, includes:

[0039] If the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, determine whether the Nth comprehensive quality assessment value and the pre-recorded N-1th comprehensive quality assessment value are greater than the preset difference threshold.

[0040] If so, the second reinforcement learning model is used to update the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model based on the Nth comprehensive quality assessment value.

[0041] Increment the value of N by 1, and repeat steps 2-10 until the Nth comprehensive quality assessment value and the (N-1)th comprehensive quality assessment value are less than or equal to the preset difference threshold.

[0042] The Nth version of medical information, where the Nth overall quality assessment value and the (N-1)th overall quality assessment value are greater than a preset difference threshold, is published to the medical staff terminal. The Nth version of medical information, where the Nth overall quality assessment value and the (N-1)th overall quality assessment value are greater than a preset difference threshold, is published to the patient terminal.

[0043] In some implementations, the network parameters of the semantic analysis model, the information type determination model, and the medical information conversion model are the states of the first reinforcement learning model, updating the network parameters of the semantic analysis model, the information type determination model, and the medical information conversion model is the action of the first reinforcement learning model, and the Nth comprehensive quality evaluation value is the reward of the reinforcement learning model.

[0044] Secondly, this application also provides an artificial intelligence-based medical information publishing system, applied to a hospital's integrated server. The system provided by this application includes:

[0045] The data receiving unit is used to receive raw medical information to be published from the hospital's administrative terminal.

[0046] The semantic analysis unit is used to perform semantic analysis on the original medical information for the Nth time based on the pre-trained semantic analysis model to obtain the Nth semantic information of the original medical information. The semantic analysis model is trained by inputting multiple first training samples into the first neural network. Each first training sample includes historical original medical information labeled with historical semantic information, and N is an integer greater than or equal to 2.

[0047] The information type determination unit is used to input the Nth semantic information into the pre-trained information type determination model and determine multiple information types corresponding to the semantic information for the Nth time. The information type determination model is trained by inputting multiple second training samples into the second neural network. Each second training sample includes the semantic information of historical original medical information and multiple historical sub-information types labeled by experts.

[0048] The medical information conversion unit is used to input the Nth semantic information and the multiple information types determined in the Nth time into the pre-trained medical information conversion model, so that the medical information conversion model can determine the Nth medical care version and the Nth patient version medical information corresponding to the original medical information based on the Nth semantic information. The Nth medical care version medical information includes multiple medical care version sub-information, each of which corresponds to an information type. The Nth patient version medical information includes multiple patient version sub-information, each of which corresponds to an information type. The multiple information types of the Nth patient version medical information correspond one-to-one with the multiple information types of the medical care version medical information. The medical information conversion model is trained by inputting multiple third training samples into a third neural network. Each third training sample includes the semantic information of the historical original medical information pair and its corresponding multiple historical information types, the corresponding historical medical care version medical information, and the corresponding historical patient version medical information.

[0049] The medical information analysis unit is used to analyze the professionalism, seriousness, and accuracy of the Nth version of medical information, as well as the comprehensibility, friendliness, and accuracy of the Nth version of medical information.

[0050] The first medical information evaluation unit is used to evaluate the quality of the Nth medical information based on its professionalism, the seriousness of its emotional tone, and its accuracy, and to obtain the first quality evaluation value of the Nth medical information.

[0051] The second medical information evaluation unit is used to evaluate the quality of the Nth patient version of medical information based on its comprehensibility, emotional friendliness, and accuracy, and to obtain the second quality evaluation value of the Nth patient version of medical information.

[0052] The matching degree determination unit is used to determine the matching degree between the medical care version sub-information of each information type in the Nth medical care version and the corresponding information type patient version sub-information in the Nth patient version of medical information.

[0053] The third medical information evaluation unit is used to determine the third quality evaluation value of the Nth patient version medical information and the Nth medical care version medical information based on the matching degree between the medical care version sub-information of each information type and the corresponding patient version sub-information of the information type.

[0054] The comprehensive quality assessment unit is used to obtain the Nth comprehensive quality assessment value based on the first quality assessment value, the second quality assessment value, and the third quality assessment value.

[0055] The reinforcement learning unit is used to update the network parameters of the semantic analysis model, the information type determination model, and the medical information conversion model based on the Nth comprehensive quality assessment value when the Nth comprehensive quality assessment value is less than the set quality assessment value. The unit also increments the value of N by 1 until the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

[0056] The medical information publishing unit is used to publish the Nth medical care version of medical information to the medical care terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, and to publish the Nth patient version of medical information to the patient terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

[0057] This application provides a method and system for publishing medical information based on artificial intelligence. The method inputs the Nth semantic information into a pre-trained information type determination model, which determines multiple information types corresponding to the Nth semantic information. Since the information type determination model is trained by inputting multiple second training samples into a second neural network, each second training sample includes the semantic information of historical original medical information and multiple historical sub-information types annotated by experts. Therefore, the reliability of the multiple information types corresponding to the Nth semantic information is high, and it more accurately represents the Nth semantic information.

[0058] The Nth semantic information and the multiple information types determined in the Nth iteration are then input into a pre-trained medical information conversion model. This allows the model to determine the Nth healthcare provider version and the Nth patient version of the original medical information based on the Nth semantic information. The Nth healthcare provider version includes multiple healthcare provider sub-information items, each corresponding to an information type. Similarly, the Nth patient version includes multiple patient sub-information items, each corresponding to an information type. The multiple information types in the Nth patient version correspond one-to-one with the multiple information types in the healthcare provider version. This ensures a high degree of semantic matching between the Nth healthcare provider version and the Nth patient version.

[0059] Furthermore, since the medical information conversion model is trained by inputting multiple third training samples into a third neural network, each third training sample includes the semantic information of the historical original medical information pair and its corresponding multiple historical information types, as well as the corresponding historical medical information for medical staff and the corresponding historical medical information for patients. This ensures high reliability of the obtained Nth medical information for medical staff and Nth medical information for patients.

[0060] Next, we analyze the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, and the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information. Based on the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, we conduct a quality assessment to obtain a first quality assessment value for the Nth healthcare version of medical information. Then, based on the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information, we conduct a quality assessment to obtain a second quality assessment value for the Nth patient version of medical information. Thus, the first and second quality assessment values ​​obtained are highly accurate.

[0061] Next, the matching degree between the medical staff version sub-information of each information type in the Nth medical staff version and the corresponding patient version sub-information of the Nth patient version can be determined. Based on the matching degree between the medical staff version sub-information of each information type and the corresponding patient version sub-information, a third quality assessment value for the Nth patient version and the Nth medical staff version can be determined. Understandably, the aforementioned matching degree can accurately characterize the quality of the Nth patient version and the Nth medical staff version, thus the obtained third quality assessment value has high reliability.

[0062] Since the reliability of the first, second, and third quality assessment values ​​is relatively high, the reliability of obtaining the Nth comprehensive quality assessment value based on the first, second, and third quality assessment values ​​is also high.

[0063] If the Nth overall quality assessment value is less than the set quality assessment value, it indicates that the quality of the obtained Nth patient version medical information and Nth nurse version medical information is not high enough. Therefore, the first reinforcement learning model is used to update the network parameters of the semantic analysis model, the information type determination model, and the medical information conversion model based on the Nth overall quality assessment value. The value of N is incremented by 1, and steps 2-11 are repeated until the Nth overall quality assessment value is greater than or equal to the set quality assessment value, which can make the quality of the obtained Nth patient version medical information and Nth nurse version medical information sufficiently high.

[0064] Finally, the Nth version of medical information (medical staff version) is published to the medical staff terminal when the Nth overall quality assessment value is greater than or equal to the set quality assessment value, and the Nth version of medical information (patient version) is published to the patient terminal when the Nth overall quality assessment value is greater than or equal to the set quality assessment value. In this way, the hospital administration can obtain very high-quality patient and medical staff versions of medical information simply by editing a single original medical record, which is efficient and saves on labor costs.

[0065] Furthermore, it can increase patients' acceptance and understanding of the published patient-version medical information, resulting in a better reading experience. Similarly, it can increase medical staff's acceptance and understanding of the published doctor-version medical information, resulting in a better reading experience. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 A schematic diagram illustrating the interaction between the hospital integrated server provided in this application embodiment and the medical staff terminal, the patient terminal, and the hospital administrative terminal, respectively.

[0068] Figure 2 A flowchart illustrating an AI-based medical information publishing method provided in this application embodiment. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments made by those skilled in the art under the guidance of these embodiments are within the scope of protection of this application.

[0070] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0071] The technical solution of the present invention and how the technical solution of the present invention solves the above-mentioned technical problems will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0072] This application provides an artificial intelligence-based medical information publishing method, applied to a hospital integrated server. For example... Figure 1 As shown, the hospital integrated server provided in this embodiment can communicate with medical staff terminals, patient terminals, and hospital administrative terminals respectively. Figure 2 As shown, the method provided in this application embodiment includes:

[0073] S201: Receive raw medical information to be published from the hospital's administrative terminal.

[0074] S202: The Nth time, based on the pre-trained semantic analysis model, semantic analysis is performed on the original medical information to obtain the Nth semantic information of the original medical information.

[0075] The semantic analysis model is trained by inputting multiple first training samples into the first neural network. Each first training sample includes historical original medical information labeled with historical semantic information, where N is an integer greater than or equal to 2.

[0076] S203: Input the Nth semantic information into the pre-trained information type determination model, and determine the multiple information types corresponding to the semantic information for the Nth time.

[0077] The information type determination model is trained by inputting multiple second training samples into a second neural network. Each second training sample includes semantic information of historical original medical information and multiple historical sub-information types annotated by experts. These multiple information types may include time type, core content type, response plan type, and prompt type, etc., without limitation.

[0078] S204: Input the Nth semantic information and the multiple information types determined in the Nth time into the pre-trained medical information conversion model so that the medical information conversion model can determine the Nth medical care version and the Nth patient version of the original medical information based on the Nth semantic information.

[0079] The Nth medical care version includes multiple medical care version sub-information, each corresponding to an information type. The Nth patient version includes multiple patient version sub-information, each corresponding to an information type. The multiple information types of the Nth patient version correspond one-to-one with the multiple information types of the medical care version. The medical information conversion model is obtained by inputting multiple third training samples into a third neural network for training. Each third training sample includes the semantic information of the historical original medical information pair and its corresponding multiple historical information types, the corresponding historical medical care version, and the corresponding historical patient version.

[0080] S205: Analyze the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, and analyze the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information.

[0081] S205 can be specifically implemented as follows: The professionalism of the Nth medical information version is determined by the first ratio of the number of professional terms (such as HIS, EMR, or other medical professional terms) in the Nth medical information version to the total number of terms in the Nth medical information version, where the first ratio is positively correlated with professionalism; the seriousness of the Nth medical information version is determined by the second ratio of the number of words with a serious emotional tendency (such as strict, prohibited, must, timely, etc.) in the Nth medical information version to the total number of terms in the Nth medical information version, where the second ratio is positively correlated with seriousness; and the number of defective sentences in the Nth medical information version is determined by the third ratio of the number of sentences in the Nth medical information version to the total number of sentences in the Nth medical information version, where the third ratio is negatively correlated with accuracy, and defects include at least one of the following: ambiguous sentences, incomplete sentences, and semantic contradictions.

[0082] S206: Based on the professionalism, emotional seriousness, and accuracy of the Nth medical information, conduct a quality assessment of the Nth medical information to obtain the first quality assessment value of the Nth medical information.

[0083] S206 can be specifically implemented as follows: based on the professionalism of the Nth medical information version, find the professionalism score corresponding to the professionalism from a preset scoring relationship table; based on the seriousness of the emotional tendency of the Nth medical information version, find the emotional tendency score corresponding to the seriousness from a preset scoring relationship table; based on the accuracy of the Nth medical information version, find the accuracy score corresponding to the accuracy from a preset scoring relationship table; and perform a weighted average of the professionalism score, accuracy score, and emotional tendency score to obtain the first quality assessment value of the Nth medical information version.

[0084] S207: Based on the comprehensibility, emotional friendliness, and accuracy of the Nth patient version of the medical information, a quality assessment is conducted on the Nth patient version of the medical information to obtain a second quality assessment value for the Nth patient version of the medical information.

[0085] S207 can be specifically implemented as follows: The comprehensibility of the Nth patient's medical information is determined by the fourth ratio of the number of technical terms (such as HIS, EMR, or other medical technical terms) in the Nth patient's medical information to the total number of terms in the Nth patient's medical information, where the fourth ratio is negatively correlated with comprehensibility; the friendliness of the Nth patient's medical information is determined by the fifth ratio of the number of friendly-sounding terms (such as "please," "warmly," "thank you," "sorry," etc.) in the Nth patient's medical information to the total number of terms in the Nth patient's medical information, where the fifth ratio is positively correlated with friendliness; and the number of defective statements in the Nth patient's medical information is determined by the sixth ratio of the total number of statements in the Nth patient's medical information, where the sixth ratio is negatively correlated with accuracy, and defects include at least one of the following: ambiguous statements, incomplete statements, and semantic inconsistencies.

[0086] S208: Determine the matching degree between the medical staff version sub-information of each information type in the Nth medical staff version and the corresponding patient version sub-information of the Nth patient version.

[0087] For example, S208 can be specifically implemented as follows: for each information type, determine the semantic similarity between the medical staff version information and the patient version information under the information type, and determine the semantic similarity as the matching degree between the medical staff version information and the patient version information under the information type.

[0088] For example, if the original medical information is "Due to system upgrade, the hospital's HIS system will be suspended from 9:00 to 12:00 on August 10, 2025. During this period, functions such as outpatient registration, electronic medical records, and laboratory report inquiries will be suspended", the Nth medical staff version and the Nth patient version of medical information can be shown in Table 1 below.

[0089]

[0090] Table 1

[0091] S209: Based on the matching degree between the medical staff version sub-information of each information type and the patient version sub-information of the corresponding information type, determine the third quality assessment value of the Nth patient version medical information and the Nth medical staff version medical information.

[0092] For example, based on the matching degree between the medical staff version sub-information of each information type and the corresponding patient version sub-information of the same information type, a matching degree score corresponding to each information type is retrieved from a preset scoring relationship table. A weighted average of the matching degree scores corresponding to each information type is then calculated to obtain the third quality assessment value for the Nth patient version medical information and the Nth medical staff version medical information. For example, when there are M information types, the weight of the matching degree score corresponding to each information type is 1 / M. This results in a high accuracy of the obtained third quality assessment value.

[0093] S210: Based on the first quality assessment value, the second quality assessment value, and the third quality assessment value, obtain the Nth comprehensive quality assessment value.

[0094] In some implementations, a weighted average can be taken from the first quality assessment value, the second quality assessment value, and the third quality assessment value to obtain the Nth comprehensive quality assessment value. For example, the weight of the first quality assessment value can be 0.3, the weight of the second quality assessment value can be 0.3, and the weight of the third quality assessment value can be 0.4.

[0095] S211: If the Nth comprehensive quality assessment value is less than the set quality assessment value, the first reinforcement learning model updates the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model based on the Nth comprehensive quality assessment value; and increments the value of N by 1, repeating S202-S210 until the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

[0096] For example, the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model are the states of the first reinforcement learning model, updating the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model are the actions of the first reinforcement learning model, and the Nth comprehensive quality evaluation value is the reward of the reinforcement learning model.

[0097] S212: Publish the Nth medical information (medical care version) to the medical care terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, and publish the Nth patient information (patient version) to the patient terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

[0098] In summary, the AI-based medical information publishing method provided in this application allows the Nth semantic information to be input into a pre-trained information type determination model, thereby determining multiple information types corresponding to the Nth semantic information. Since the information type determination model is trained by inputting multiple second training samples into a second neural network, each second training sample includes the semantic information of historical original medical information and multiple historical sub-information types annotated by experts. Therefore, the reliability of the multiple information types corresponding to the Nth semantic information is high, and it more accurately represents the Nth semantic information.

[0099] The Nth semantic information and the multiple information types determined in the Nth iteration are then input into a pre-trained medical information conversion model. This allows the model to determine the Nth healthcare provider version and the Nth patient version of the original medical information based on the Nth semantic information. The Nth healthcare provider version includes multiple healthcare provider sub-information items, each corresponding to an information type. Similarly, the Nth patient version includes multiple patient sub-information items, each corresponding to an information type. The multiple information types in the Nth patient version correspond one-to-one with the multiple information types in the healthcare provider version. This ensures a high degree of semantic matching between the Nth healthcare provider version and the Nth patient version.

[0100] Furthermore, since the medical information conversion model is trained by inputting multiple third training samples into a third neural network, each third training sample includes the semantic information of the historical original medical information pair and its corresponding multiple historical information types, as well as the corresponding historical medical information for medical staff and the corresponding historical medical information for patients. This ensures high reliability of the obtained Nth medical information for medical staff and Nth medical information for patients.

[0101] Next, we analyze the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, and the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information. Based on the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, we conduct a quality assessment to obtain a first quality assessment value for the Nth healthcare version of medical information. Then, based on the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information, we conduct a quality assessment to obtain a second quality assessment value for the Nth patient version of medical information. Thus, the first and second quality assessment values ​​obtained are highly accurate.

[0102] Next, the matching degree between the medical staff version sub-information of each information type in the Nth medical staff version and the corresponding patient version sub-information of the Nth patient version can be determined. Based on the matching degree between the medical staff version sub-information of each information type and the corresponding patient version sub-information, a third quality assessment value for the Nth patient version and the Nth medical staff version can be determined. Understandably, the aforementioned matching degree can accurately characterize the quality of the Nth patient version and the Nth medical staff version, thus the obtained third quality assessment value has high reliability.

[0103] Since the reliability of the first, second, and third quality assessment values ​​is relatively high, the reliability of obtaining the Nth comprehensive quality assessment value based on the first, second, and third quality assessment values ​​is also high.

[0104] If the Nth overall quality assessment value is less than the set quality assessment value, it indicates that the quality of the obtained Nth patient version medical information and Nth nurse version medical information is not high enough. Therefore, the first reinforcement learning model is used to update the network parameters of the semantic analysis model, the information type determination model, and the medical information conversion model based on the Nth overall quality assessment value. The value of N is incremented by 1, and steps 2-11 are repeated until the Nth overall quality assessment value is greater than or equal to the set quality assessment value, which can make the quality of the obtained Nth patient version medical information and Nth nurse version medical information sufficiently high.

[0105] Finally, the Nth version of medical information (medical staff version) is published to the medical staff terminal when the Nth overall quality assessment value is greater than or equal to the set quality assessment value, and the Nth version of medical information (patient version) is published to the patient terminal when the Nth overall quality assessment value is greater than or equal to the set quality assessment value. In this way, the hospital administration can obtain very high-quality patient and medical staff versions of medical information simply by editing a single original medical record, which is efficient and saves on labor costs.

[0106] Furthermore, it can increase patients' acceptance and understanding of the published patient-version medical information, resulting in a better reading experience. Similarly, it can increase medical staff's acceptance and understanding of the published doctor-version medical information, resulting in a better reading experience.

[0107] Furthermore, the above-mentioned S212 can be specifically implemented as follows:

[0108] Step A: If the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, determine whether the Nth comprehensive quality assessment value and the pre-recorded N-1th comprehensive quality assessment value are greater than the preset difference threshold. If so, proceed to Step B.

[0109] If there is no preset difference threshold between the Nth overall quality assessment value and the pre-recorded N-1th overall quality assessment value, it indicates that there is still room for improvement in the quality of the Nth patient version of medical information and the Nth nurse version of medical information. Therefore, the subsequent step B can be performed.

[0110] Step B: Using the second reinforcement learning model, update the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model based on the Nth comprehensive quality assessment value.

[0111] In some implementations, the network parameters of the semantic analysis model, the information type determination model, and the medical information conversion model are the states of the second reinforcement learning model, updating the network parameters of the semantic analysis model, the information type determination model, and the medical information conversion model is the action of the second reinforcement learning model, and the Nth comprehensive quality evaluation value is the reward of the reinforcement learning model.

[0112] Step C: Increment the value of N by 1, and repeat S202-S210 until the Nth comprehensive quality assessment value and the (N-1)th comprehensive quality assessment value are less than or equal to the preset difference threshold.

[0113] Based on steps A-C above, the highest quality patient-level and nurse-level medical information can be obtained to the greatest extent possible.

[0114] Step D: Publish the Nth medical information (medical care version) to the medical care terminal when the Nth comprehensive quality assessment value and the (N-1)th comprehensive quality assessment value are greater than the preset difference threshold, and publish the Nth patient information (patient version) to the patient terminal when the Nth comprehensive quality assessment value and the (N-1)th comprehensive quality assessment value are greater than the preset difference threshold.

[0115] In addition, this application also provides an artificial intelligence-based medical information publishing system applied to a hospital's integrated server. It should be noted that the basic principles and technical effects of the artificial intelligence-based medical information publishing system provided in this embodiment are the same as those in the above embodiments. For the sake of brevity, any parts not mentioned in the embodiments of this invention can be referred to the corresponding content in the above embodiments. Specifically, the system provided in this application includes:

[0116] The data receiving unit is used to receive raw medical information to be published from the hospital's administrative terminal.

[0117] The semantic analysis unit is used to perform semantic analysis on the original medical information for the Nth time based on the pre-trained semantic analysis model to obtain the Nth semantic information of the original medical information. The semantic analysis model is trained by inputting multiple first training samples into the first neural network. Each first training sample includes historical original medical information labeled with historical semantic information, and N is an integer greater than or equal to 2.

[0118] The information type determination unit is used to input the Nth semantic information into the pre-trained information type determination model and determine multiple information types corresponding to the semantic information for the Nth time. The information type determination model is trained by inputting multiple second training samples into the second neural network. Each second training sample includes the semantic information of historical original medical information and multiple historical sub-information types labeled by experts.

[0119] The medical information conversion unit is used to input the Nth semantic information and the multiple information types determined in the Nth time into the pre-trained medical information conversion model, so that the medical information conversion model can determine the Nth medical care version and the Nth patient version medical information corresponding to the original medical information based on the Nth semantic information. The Nth medical care version medical information includes multiple medical care version sub-information, each of which corresponds to an information type. The Nth patient version medical information includes multiple patient version sub-information, each of which corresponds to an information type. The multiple information types of the Nth patient version medical information correspond one-to-one with the multiple information types of the medical care version medical information. The medical information conversion model is trained by inputting multiple third training samples into a third neural network. Each third training sample includes the semantic information of the historical original medical information pair and its corresponding multiple historical information types, the corresponding historical medical care version medical information, and the corresponding historical patient version medical information.

[0120] The medical information analysis unit is used to analyze the professionalism, seriousness, and accuracy of the Nth version of medical information, as well as the comprehensibility, friendliness, and accuracy of the Nth version of medical information.

[0121] The first medical information evaluation unit is used to evaluate the quality of the Nth medical information based on its professionalism, the seriousness of its emotional tone, and its accuracy, and to obtain the first quality evaluation value of the Nth medical information.

[0122] The second medical information evaluation unit is used to evaluate the quality of the Nth patient version of medical information based on its comprehensibility, emotional friendliness, and accuracy, and to obtain the second quality evaluation value of the Nth patient version of medical information.

[0123] The matching degree determination unit is used to determine the matching degree between the medical care version sub-information of each information type in the Nth medical care version and the corresponding information type patient version sub-information in the Nth patient version of medical information.

[0124] The third medical information evaluation unit is used to determine the third quality evaluation value of the Nth patient version medical information and the Nth medical care version medical information based on the matching degree between the medical care version sub-information of each information type and the corresponding patient version sub-information of the information type.

[0125] The comprehensive quality assessment unit is used to obtain the Nth comprehensive quality assessment value based on the first quality assessment value, the second quality assessment value, and the third quality assessment value.

[0126] The reinforcement learning unit is used to update the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model based on the Nth comprehensive quality assessment value when the Nth comprehensive quality assessment value is less than the set quality assessment value. The unit also increments the value of N by 1 and repeats steps 2-11 until the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

[0127] The medical information publishing unit is used to publish the Nth medical care version of medical information to the medical care terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, and to publish the Nth patient version of medical information to the patient terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for publishing medical information based on artificial intelligence, characterized in that, Applied to a hospital integrated server, the method includes: Step 1: Receive raw medical information to be published from the hospital's administrative terminal; Step 2: For the Nth time, semantic analysis is performed on the original medical information based on the pre-trained semantic analysis model to obtain the Nth semantic information of the original medical information. The semantic analysis model is trained by inputting multiple first training samples into the first neural network. Each first training sample includes historical original medical information labeled with historical semantic information, and N is an integer greater than or equal to 2. Step 3: Input the Nth semantic information into the pre-trained information type determination model, and determine the multiple information types corresponding to the semantic information for the Nth time. The information type determination model is obtained by inputting multiple second training samples into the second neural network for training. Each second training sample includes the semantic information of historical original medical information and multiple historical sub-information types labeled by experts. Step 4: Input the Nth semantic information and the multiple information types determined in the Nth time into the pre-trained medical information conversion model, so that the medical information conversion model can determine the Nth medical care version and the Nth patient version medical information corresponding to the original medical information based on the Nth semantic information. The Nth medical care version medical information includes multiple medical care version sub-information, each of which corresponds to one of the information types. The Nth patient version medical information includes multiple patient version sub-information, each of which corresponds to one of the information types. The multiple information types of the Nth patient version medical information correspond one-to-one with the multiple information types of the medical care version medical information. The medical information conversion model is trained by inputting multiple third training samples into a third neural network. Each third training sample includes the semantic information of the historical original medical information pair and its corresponding multiple historical information types, the corresponding historical medical care version medical information, and the corresponding historical patient version medical information. Step 5: Analyze the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, and analyze the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information; Step 6: Based on the professionalism, emotional seriousness, and accuracy of the Nth medical information, conduct a quality assessment of the Nth medical information to obtain a first quality assessment value for the Nth medical information. Step 7: Based on the comprehensibility, emotional friendliness, and accuracy of the Nth patient version of the medical information, perform a quality assessment of the Nth patient version of the medical information to obtain a second quality assessment value of the Nth patient version of the medical information; Step 8: Determine the matching degree between the medical staff version sub-information of each information type in the Nth medical staff version and the corresponding patient version sub-information of the Nth patient version; Step 9: Based on the matching degree between the medical staff version sub-information of each information type and the patient version sub-information of the corresponding information type, determine the third quality assessment value of the Nth patient version medical information and the Nth medical staff version medical information; Step 10: Obtain the Nth comprehensive quality assessment value based on the first quality assessment value, the second quality assessment value, and the third quality assessment value; Step 11: If the Nth comprehensive quality assessment value is less than the set quality assessment value, the first reinforcement learning model updates the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model based on the Nth comprehensive quality assessment value; and increments the value of N by 1, repeating steps 2-10 until the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value. Step 12: Publish the Nth medical information (medical care version) to the medical care terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, and publish the Nth patient information (patient version) to the patient terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

2. The method according to claim 1, characterized in that, The analysis of the professionalism, seriousness, and accuracy of the Nth healthcare version of medical information, and the analysis of the comprehensibility, friendliness, and accuracy of the Nth patient version of medical information, include: The professionalism of the Nth medical care version of medical information is determined by the first ratio of the number of professional terms in the Nth medical care version of medical information to the total number of terms in the Nth medical care version of medical information, wherein the first ratio is positively correlated with the professionalism. The seriousness of the Nth medical information is determined by the second ratio of the number of words with a serious emotional tendency in the Nth medical information to the total number of words in the Nth medical information, wherein the second ratio is positively correlated with the seriousness. Determine a third ratio of the number of defective statements in the Nth medical care version to the total number of statements in the Nth medical care version, wherein the third ratio is negatively correlated with the accuracy, and the defect is at least one of the following: ambiguous statements, incomplete statements, and semantic inconsistencies in statements.

3. The method according to claim 2, characterized in that, The quality assessment of the Nth medical information version is performed based on its professionalism, emotional seriousness, and accuracy to obtain a first quality assessment value for the Nth medical information version, including: Based on the professionalism of the Nth medical information, the professionalism score corresponding to the professionalism is found from the preset scoring relationship table; Based on the seriousness of the emotional tendency of the Nth medical information, the emotional tendency score corresponding to the seriousness is found from the preset scoring relationship table; Based on the accuracy of the Nth medical information, the accuracy score corresponding to the accuracy is found from the preset scoring relationship table; The first quality assessment value of the Nth medical information version is obtained by weighting and averaging the professionalism score, the accuracy score, and the sentiment score.

4. The method according to claim 1, characterized in that, The quality assessment of the Nth patient's medical information is performed based on its comprehensibility, emotional friendliness, and accuracy, resulting in a second quality assessment value for the Nth patient's medical information. This includes: The comprehensibility of the Nth patient version of medical information is determined by the fourth ratio of the number of technical terms in the Nth patient version of medical information to the total number of terms in the Nth patient version of medical information, wherein the fourth ratio is negatively correlated with the comprehensibility. The friendliness level of the Nth patient's medical information is determined by the fifth ratio of the number of words with a friendly sentiment in the Nth patient's medical information to the total number of words in the Nth patient's medical information, wherein the fifth ratio is positively correlated with the friendliness level. Determine the sixth ratio of the number of defective statements in the Nth patient version of medical information to the total number of statements in the Nth patient version of medical information, wherein the sixth ratio is negatively correlated with the accuracy, and the defect is at least one of the following: ambiguous statements, incomplete statements, and semantic inconsistencies in statements.

5. The method according to claim 1, characterized in that, Determining the matching degree between the medical staff version sub-information of each information type in the Nth medical staff version and the corresponding patient version sub-information of the Nth patient version, including: For each information type, determine the semantic similarity between the medical staff version information and the patient version information under that information type, and define the semantic similarity as the matching degree between the medical staff version information and the patient version information under that information type.

6. The method according to claim 5, characterized in that, The third quality assessment value of the Nth patient version medical information and the Nth medical staff version medical information is determined based on the matching degree between the medical staff version sub-information of each information type and the corresponding patient version sub-information of the information type. Based on the matching degree between the medical staff sub-information of each information type and the patient sub-information of the corresponding information type, the matching degree score corresponding to each information type is retrieved from the preset scoring relationship table; The matching scores corresponding to each of the aforementioned information types are weighted and averaged to obtain the third quality assessment value of the Nth patient version of medical information and the Nth nurse version of medical information.

7. The method according to claim 1, characterized in that, The step of obtaining the Nth comprehensive quality assessment value based on the first quality assessment value, the second quality assessment value, and the third quality assessment value includes: The first quality assessment value, the second quality assessment value, and the third quality assessment value are weighted and averaged to obtain the Nth comprehensive quality assessment value.

8. The method according to any one of claims 1-7, characterized in that, The step of publishing the Nth healthcare version of medical information to the healthcare terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, and publishing the Nth patient version of medical information to the patient terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, includes: If the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, determine whether the Nth comprehensive quality assessment value and the pre-recorded N-1th comprehensive quality assessment value are greater than a preset difference threshold. If so, the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model are updated using the second reinforcement learning model based on the Nth comprehensive quality assessment value. Increment the value of N by 1, and repeat steps 2-10 until the Nth comprehensive quality assessment value and the (N-1)th comprehensive quality assessment value are less than or equal to the preset difference threshold. The Nth medical care version of medical information, where the Nth comprehensive quality assessment value and the (N-1)th comprehensive quality assessment value are both greater than a preset difference threshold, is published to the medical care terminal. The Nth patient version of medical information, where the Nth comprehensive quality assessment value and the (N-1)th comprehensive quality assessment value are both greater than a preset difference threshold, is published to the patient terminal.

9. The method according to claim 1, characterized in that, The network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model are the state of the first reinforcement learning model. Updating the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model is the action of the first reinforcement learning model. The Nth comprehensive quality evaluation value is the reward of the reinforcement learning model.

10. A medical information publishing system based on artificial intelligence, characterized in that, The system, applied to a hospital integrated server, includes: The data receiving unit is used to receive raw medical information to be published from the hospital's administrative terminal. The semantic analysis unit is used to perform semantic analysis on the original medical information for the Nth time according to the pre-trained semantic analysis model to obtain the Nth semantic information of the original medical information. The semantic analysis model is trained by inputting multiple first training samples into the first neural network. Each first training sample includes historical original medical information labeled with historical semantic information, and N is an integer greater than or equal to 2. An information type determination unit is used to input the Nth semantic information into a pre-trained information type determination model and determine multiple information types corresponding to the semantic information for the Nth time. The information type determination model is trained by inputting multiple second training samples into a second neural network. Each second training sample includes the semantic information of historical original medical information and multiple historical sub-information types labeled by experts. A medical information conversion unit is used to input the Nth semantic information and the multiple information types determined in the Nth time into a pre-trained medical information conversion model, so that the medical information conversion model determines the Nth medical care version and the Nth patient version medical information corresponding to the original medical information based on the Nth semantic information. The Nth medical care version includes multiple medical care version sub-information, each corresponding to one of the information types. The Nth patient version medical information includes multiple patient version sub-information, each corresponding to one of the information types. The multiple information types of the Nth patient version medical information correspond one-to-one with the multiple information types of the medical care version medical information. The medical information conversion model is trained by inputting multiple third training samples into a third neural network. Each third training sample includes the semantic information of a historical original medical information pair and its corresponding multiple historical information types, the corresponding historical medical care version medical information, and the corresponding historical patient version medical information. The medical information analysis unit is used to analyze the professionalism, seriousness, and accuracy of the Nth version of medical information, as well as the comprehensibility, friendliness, and accuracy of the Nth version of medical information. The first medical information evaluation unit is used to evaluate the quality of the Nth medical information based on its professionalism, the seriousness of its emotional tone, and its accuracy, and to obtain a first quality evaluation value for the Nth medical information. The second medical information evaluation unit is used to evaluate the quality of the Nth patient version of medical information based on its comprehensibility, emotional friendliness, and accuracy, and to obtain a second quality evaluation value for the Nth patient version of medical information. A matching degree determination unit is used to determine the matching degree between the medical care version sub-information of each information type in the Nth medical care version and the corresponding information type patient version sub-information in the Nth patient version; The third medical information evaluation unit is used to determine the third quality evaluation value of the Nth patient version medical information and the Nth medical staff version medical information based on the matching degree between the medical staff version sub-information of each information type and the patient version sub-information of the corresponding information type. A comprehensive quality assessment unit is used to obtain the Nth comprehensive quality assessment value based on the first quality assessment value, the second quality assessment value, and the third quality assessment value. The reinforcement learning unit is used to update the network parameters of the semantic analysis model, the network parameters of the information type determination model, and the network parameters of the medical information conversion model based on the Nth comprehensive quality assessment value when the Nth comprehensive quality assessment value is less than the set quality assessment value; and to increment the value of N by 1 until the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value. The medical information publishing unit is used to publish the Nth medical care version of medical information to the medical care terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value, and to publish the Nth patient version of medical information to the patient terminal when the Nth comprehensive quality assessment value is greater than or equal to the set quality assessment value.

Images