CN116453674A

CN116453674A - Intelligent medical system

Info

Publication number: CN116453674A
Application number: CN202310465786.3A
Authority: CN
Inventors: 李四雄
Original assignee: Guangzhou Southwing Information Technology Co ltd
Current assignee: Guangzhou Southwing Information Technology Co ltd
Priority date: 2023-04-27
Filing date: 2023-04-27
Publication date: 2023-07-18

Abstract

The invention discloses an intelligent medical system, which comprises a background server, a database and a client, wherein the database and the client are communicated with the background server; the database is used for storing data; the client comprises a user end and a doctor end, and the background server comprises: the system comprises a data processing module, a symptom identification module, a laboratory sheet auxiliary disease diagnosis module, a medical image auxiliary disease diagnosis module and a disease diagnosis module, wherein the disease diagnosis module is used for inputting the patient information and at least one of symptom information, an abnormal index result and medical image characteristics into a convolutional neural network model and outputting a patient disease result. The intelligent medical system can more accurately and intelligently analyze the diseases suffered by the patient, and can realize the functions of intelligent disease diagnosis, intelligent doctor matching, intelligent question answering and the like in terms of functions.

Description

Intelligent medical system

Technical Field

The invention belongs to the technical field of medical systems, and particularly relates to an intelligent medical system.

Background

At present, the development of medical systems in China is imperfect, the treatment procedure is complex, the queuing time is long, doctors can only treat patients in fixed working time, and the patients need to be queued again after being tested in the inquiry process, so that a great amount of time is wasted. The traditional medical mode needs the user to finish operations such as registering, paying fee, taking reports and the like, so that the user experience is poor, the service quality is low, a reasonable scheme cannot be allocated to patients in a hospital, and the number of treatment channels which can be selected by people is small, so that the medical mode is a problem facing the current situation and is a problem of most concern to the public.

Along with the development of the times and the perfection of hardware facilities, in order to solve the problems that high-quality medical resources are short, the queuing time of patients going to a hospital is long, doctors which are most good at treating own illness can not be found by random registration, illness state delay is caused by finding out suitable doctors, and the like, an intelligent medical system which is convenient for the patients is developed, and the essence of the intelligent medical system is that the technology of the Internet of things is combined with a terminal to reduce unnecessary flows under the online condition.

However, with the continuous development of technology, the current intelligent medical system has a great room for innovation and improvement.

Disclosure of Invention

The invention aims to solve the technical problems and provides an intelligent medical system.

In order to solve the problems, the invention is realized according to the following technical scheme:

the invention provides an intelligent medical system, which comprises a background server, a database and a client, wherein the database and the client are communicated with the background server; the database is used for storing data; the client comprises a user end and a doctor end, wherein the user end is used for providing a man-machine interaction access interface for a patient, and the doctor end is used for providing a man-machine interaction access interface for a doctor;

wherein, the backend server includes:

the data processing module is used for acquiring internet medical information data, identifying medical words from the medical information data by using a statistical method, and carrying out word embedding training on the new medical words through word2vec training to obtain word embedding vectors;

a symptom identification module for acquiring patient information and inquiry information input by the client and identifying symptom information in the patient information and inquiry information;

the laboratory sheet auxiliary disease diagnosis module is used for acquiring the laboratory sheet picture input by the client, identifying a plurality of index information in the laboratory sheet picture, comparing the index information one by one and outputting an abnormal index result;

the medical image auxiliary disease diagnosis module is used for acquiring medical images input by the client, identifying the medical images by adopting a convolution lifting network and outputting medical image characteristics;

and the disease diagnosis module is connected with the data processing module, the symptom identification module, the laboratory sheet auxiliary disease diagnosis module and the medical image auxiliary disease diagnosis module, and is used for inputting the patient information and at least one of the symptom information, the abnormal index result and the medical image characteristic into a convolutional neural network model and outputting the patient disease result.

Preferably, the background server includes:

and the doctor matching module is connected with the disease diagnosis module and is used for acquiring a patient disease result and performing doctor-patient matching on the patient disease result by adopting a collaborative filtering algorithm.

Preferably, the doctor matching module performs doctor-patient matching, including the following steps:

obtaining a disease result of a patient;

identifying historical case data of a plurality of patients with the same disease as the disease result of the patient from the pre-stored historical case data in a database; the historical case data comprises patient information, doctor information, patient evaluation and doctor satisfaction evaluation;

identifying doctor adept fields in doctor information, and carrying out doctor screening according to the correlation degree between the doctor adept fields and patient disease results;

calculating the recommendation degree of each doctor reserved after screening according to the patient evaluation, the satisfaction degree evaluation and the recommendation degree formula of the doctor;

and taking the doctor with the highest recommendation degree as the recommended doctor.

Preferably, the background server includes:

and the intelligent medical question and answer module is used for executing response operation according to the inquiry information sent by the user side.

Preferably, the intelligent medical question and answer module is configured to perform the steps of:

acquiring inquiry information sent by the user side;

determining an information tag from the query information, the information tag being used to represent a query intent of the patient, the information tag including greetings, interviews, and knowledge of medical knowledge;

responding to the information label as a greeting, and calling a corresponding preset reply statement according to inquiry information to return to the user side;

responding to the information label as a consultation, sending the information label to a disease diagnosis module according to the consultation information to output a patient disease result, inputting the patient disease result to a doctor matching module to perform doctor-patient matching, and returning the patient disease result and a recommended doctor to the user side;

and responding to the information label to know medical knowledge, identifying keywords in the query information, querying in internet medical information data according to the keywords, and returning a query result to the user side.

Preferably, the smart medical system further comprises:

the video terminal equipment is deployed in the ICU ward, each video terminal equipment is provided with a unique number, and the unique numbers are associated with sickbeds of the ICU ward in a one-to-one correspondence manner;

the visit information configuration module is used for enabling the doctor end to write in patient information of the ICU ward and associating the patient information with the video terminal equipment;

the ICU visit module is used for receiving a patient visit request of the user side, wherein the patient visit request carries patient information, a sickbed number of an ICU ward and a verification code; and establishing connection between the user side and corresponding video terminal equipment according to the verification results of the patient information, the sickbed number and the verification code.

Compared with the prior art, the invention has the beneficial effects that:

wherein, the backend server includes: the data processing module is used for acquiring internet medical information data, identifying medical words from the medical information data by using a statistical method, and carrying out word embedding training on the new medical words through word2vec training to obtain word embedding vectors; a symptom identification module for acquiring patient information and inquiry information input by the client and identifying symptom information in the patient information and inquiry information; the laboratory sheet auxiliary disease diagnosis module is used for acquiring the laboratory sheet picture input by the client, identifying a plurality of index information in the laboratory sheet picture, comparing the index information one by one and outputting an abnormal index result; the medical image auxiliary disease diagnosis module is used for acquiring medical images input by the client, identifying the medical images by adopting a convolution lifting network and outputting medical image characteristics; and the disease diagnosis module is connected with the data processing module, the symptom identification module, the laboratory sheet auxiliary disease diagnosis module and the medical image auxiliary disease diagnosis module, and is used for inputting the patient information and at least one of the symptom information, the abnormal index result and the medical image characteristic into a convolutional neural network model and outputting the patient disease result.

The intelligent medical system can more accurately and intelligently analyze diseases suffered by patients under the joint cooperation of the data processing module, the symptom identification module, the laboratory sheet auxiliary disease diagnosis module, the medical image auxiliary disease diagnosis module and the disease diagnosis confirming module, and can realize functions of intelligent disease diagnosis, intelligent doctor matching, intelligent question answering and the like in terms of functions, realize high-accuracy doctor-patient matching, and enable the patients to diagnose and treat own illness state by using the mobile phone at home, so that on one hand, the flow of people in hospitals is reduced, and on the other hand, the possibility of cross infection of infectious diseases is also reduced.

Drawings

The invention is described in further detail below with reference to the attached drawing figures, wherein:

FIG. 1 is a system diagram of a smart medical system of the present invention;

fig. 2 is a block diagram of a convolutional neural network of the present invention.

Description of the embodiments

The preferred embodiments of the present invention will be described below with reference to the accompanying drawings, it being understood that the preferred embodiments described herein are for illustration and explanation of the present invention only, and are not intended to limit the present invention.

Examples

As shown in fig. 1, the intelligent medical system of the present invention includes a background server, a database in communication with the background server, and a client; the database is used for storing data; the intelligent medical system has the management function of resources, and the system is divided into a user end and a doctor end by using users as a whole, wherein the client end comprises a user end and a doctor end, the user end is used for providing a man-machine interaction access interface for a patient, and the doctor end is used for providing a man-machine interaction access interface for a doctor.

The client can be deployed on a mobile terminal, such as a mobile phone, a tablet computer and the like.

In one implementation, the client has the following functional modules:

the doctor selects the module: the system receives a patient's condition description and examination materials (laboratory sheets, medical image materials, etc.), and then derives the condition of the patient through analysis of the system and performs doctor's assignment. The system recommends several doctors with highest matching degree to the patient, and the patient can select one or more doctors from the doctors recommended by the system through the doctor selection unit to diagnose the diseases of the patients.

The health reminding module is used for: the patient can synchronize the health information (data such as step number, heart rate, exercise time) in the mobile phone to the system in real time, the system carries out health reminding on the patient through background analysis on the data, and if the exercise time of the patient is overlong, the heart rate is too fast or too slow, the system can issue a message to the patient to remind the patient of the exercise intensity.

And the message pushing module is used for pushing the medical common sense to the users, for example, the high-incidence disease information and the prevention method corresponding to the current season are pushed to all users using the system.

The intelligent question-answering module: the patient may have a conversation with the system assuming the patient greetings the system such as: "hello", "bye", etc., the system will reply humanizedly. The patient can also submit the disease description and examination materials to the system, and then first make a free initial disease diagnosis, and the patient can ask about knowledge related to the disease, such as: introduction to the disease, how the disease is treated, etc.

The appointment registering module is used for registering in all hospitals connected with the intelligent medical system. The system is connected to the hospital systems of all hospitals through the external interfaces, and then registration and reservation operations are carried out, so that queuing time and medical resource waste can be effectively reduced.

And the hospital navigation module is used for introducing a Goldmap, a hundred-degree map or a Tencel map through an API external link, so that a patient can directly navigate the map through the system without additionally downloading the APP of the hundred-degree map, and the operation complexity of a user is reduced.

The medicine purchasing module is used for purchasing medicine, delivering medicine, going to the gate and the like through a pharmacy connected with the intelligent medical system.

The diagnosis feedback module is used for analyzing and selecting the diagnosis of a plurality of doctors, determining how to recover the disease of the doctor, evaluating the diagnosis of all the doctors after recovery, and the like, so that the later patient can be helped to match with the doctor capable of rapidly solving the disease diagnosis.

Correspondingly, at least the following modules are arranged in the doctor end:

and the diagnosis result submitting module is used for submitting the diagnosis result judged by the doctor to the system and feeding the diagnosis result back to the patient through the system.

And the distribution feedback module is used for evaluating the doctor-patient matching of the time. In cases where the patient has little knowledge of the medical knowledge, the patient will have a high probability of selecting the first doctor or doctors recommended by the system. However, if the patient's disease is a relatively rare disease, all doctors assigned by the system are wrong, and in this case, the doctor first switches the patient to other doctors, and then feeds back to the system to be assigned as wrong. After the system records, similar patients are assigned to the correct doctor.

In a specific implementation, the background server adopts a cloud platform, and the cloud platform can use a Hadoop open source framework to establish a plurality of different components so as to realize a plurality of functions, such as: acquiring data, managing tasks, coordinating workflows, and the like. On the other hand, the data processing method is used in cooperation with a big data platform, and through the introduction of a data warehouse technology, analysis dimension of medical information data in a hospital is enriched, quality and usability of the data are improved, and basic data support is provided for data mining and intelligent application.

Specifically, the background server comprises a data processing module, a symptom identification module, a laboratory sheet auxiliary disease diagnosis module, a medical image auxiliary disease diagnosis module and a disease diagnosis confirming module.

In another example, the backend server includes a doctor matching module and a smart health question-answering module.

Specifically, the invention describes each module of the intelligent medical system in detail:

the data processing module is used for acquiring internet medical information data, identifying medical words from the medical information data by using a statistical method, and carrying out word embedding training on the new medical words through word2vec training to obtain word embedding vectors.

In a specific implementation, the data processing module obtains medical information data on a network by using a crawler technology, and next step of data cleaning includes: removing symbols and stop words, segmenting words by using a jieba tool, finding medical words and new vocabulary by using a statistical method, embedding vectors by word2vec training words, and constructing a data base for the operation of a subsequent module.

Specifically, the web crawler is used to crawl public medical information, such as diseases and corresponding symptoms in hundreds of degrees encyclopedia, medical questions and answers in a medical question and answer website, and the like, on the internet. The first step after obtaining the basic data is to clean the data, and when the text of Chinese is processed, the text is required to be segmented. Because the word segmentation result of the jieba word segmentation tool is not ideal in the medical field, the words which cannot be segmented by the word segmentation tool by using the statistical method are segmented again, and more accurate word segmentation results can be obtained, so that the operations such as named entity labeling and the like can be conveniently carried out at the later stage. Besides the data cleaning of the characters, the data cleaning of the pictures is an indispensable step, and most of the pictures can be input into a convolutional neural network by cleaning and denoising, so that more accurate results are obtained.

Wherein, only through the jieba word segmentation tool, all words can not be clearly segmented, and the words which are not segmented by the word segmentation tool and recorded in the dictionary are called as 'new words'. The reasons for the appearance of the new words are numerous, for example, a great number of named entities exist in the medical field, such as the names of diseases such as psoriasis and white line hernia, the names of symptoms such as inflammation red sickness and pulse floating, and the names of medicines such as amide alcohols and nitrofurans. Therefore, the invention establishes a database containing word meaning, type, word constitution and other aspects, and then carries out cosine operation on vectors to obtain new medical vocabulary, and the word segmentation flow based on a statistical method is as follows:

(1) The document is preprocessed, and two steps of processing are needed after the text is obtained.

(1) Firstly, sentence dividing is carried out on the text according to punctuation marks, and the first level is to divide the Chinese sentence into periods. "and english period", "is taken as the end of a sentence, and the text is split. The second level is to further phrase signs such as comma "," bracket "()".

(2) And secondly, setting a plurality of sliding windows, and selecting N-gram words under all windows by the candidate word strings. N-gram means that N adjacent words in a text are selected, such as I love learning. The N-gram words are: "me", "love", "learn", "love", "loving", "learn", "loving", "learning", "loving learning". It can be seen that if the value of the width N of the sliding window is increased, the calculation difficulty is greatly increased, and since the number of words with a large number of words is not large, the sliding window width N is basically selected to be 5 words or 6 words.

(2) And selecting statistics.

And after the candidate word strings are identified, extracting the candidate word strings, calculating the values of four statistics of word frequency, left and right adjacent entropy and point mutual information through a formula, designing a threshold value, and comparing the four probabilities of the words with the threshold value to obtain whether the candidate word strings can be used as new words.

The term frequency means the frequency of occurrence of all n-gram candidate word strings obtained in the previous step in all documents. Mutual Information (MI) may indicate whether there is an association between two different values a, b, and the strength of the relationship of the two values. Mutual information means that the degree of correlation between the front and rear texts can be calculated by one formula. The value of mutual information is often used as one of the statistics of new word mining.

The calculation formula of mutual information is:in this formula z represents the target word itself, a and b represent the subwords that make up the word string z, N represents the total number of candidate word strings in the document, +.>Representing the frequencies of occurrence of three candidate word strings in a document, a, b, z, respectively.

The numerator is the multiplication probability of the candidate word string, and the denominator is the probability of the individual word formation of the sub word string of the candidate word string. From this, it is inferred that the probability that a candidate word string can become a new word is greater when the relationship between two sub-word strings is smaller.

The statistics method is used for mining new words, and the complete knowledge base is not available, for example: the word "lupus erythematosus" may have the possibility of the word segmentation results of "red" and "lupus erythematosus", "erythema" and "lupus", "red wolf" and "sore", "red" plus "plaque" and "lupus", and the like. Still another concept within mutual information is point mutual information. In probability, assuming that two values a, b are not related, p (a, b) =p (a) p (b). The greater the link between the two, the less p (a, b) will be than p (a) p (b).

The calculation formula of the left and right adjacency entropy is as follows:in the formula, a represents the word on the left side of the candidate word string, ++>Representing the probability that a is clung to the left of t, b represents the right word of the candidate word string z, +.>Representing the probability that b is close to the right of z.

Statistical-based word segmentation processes are available to those skilled in the art and are not described in detail herein.

In the present invention, the Word2vec model is simply a simplified neural network, and the present invention is not so much described, and Word2vec is used for Word embedding training of the obtained data, and the obtained Word embedding vector is used in subsequent disease diagnosis.

Specifically, "word embedding" is mapping a word into a vector space, i.e., a word may be represented by a vector. Unlike normal space vectors, word embedding uses vector representations of a large number of corpus unsupervised training words, which are rich in semantic information, and can represent words as a lower-dimensional vector, which also has certain abstract semantics. This abstract semantics is manifested in that even if two words are completely different in character, similar meanings such as "bellyache" and "abdominal pain" are expressed, and the similarity of vectors of these two words in vector space is very large. The word embedding vector may also represent relationships between words.

And the symptom identification module is used for acquiring the patient information and the inquiry information input by the client and identifying symptom information in the patient information and the inquiry information.

In the module, character recognition technology and voice recognition technology are mainly adopted, and symptom information is obtained from recognition in patient information and inquiry information. In one implementation, the medical vocabulary library is established through the data processing module, medical vocabularies in the medical vocabulary library are used as keywords, and when the inquiry information input by the patient is text information, the same symptom information is identified from the patient information and the inquiry information.

Specifically, in the system, some patients can not learn words or use pinyin, but can not give the illness state description to the system, and the illness state can be described in a voice mode, so that the user experience and audience are improved. When the patient inputs speech, the speech is converted into text information.

And the laboratory sheet auxiliary disease diagnosis module is used for acquiring the laboratory sheet picture input by the client, identifying a plurality of index information in the laboratory sheet picture, comparing the index information one by one, and outputting an abnormal index result.

In the present invention, the disease analysis is assisted according to the patient's laboratory sheet and divided into two steps: the first step is to identify the characters in the picture, the second step is to judge the values of various indexes, and finally output the result to the convolutional neural network model of the disease diagnosis module to improve the accuracy of the model. Specifically, the method comprises the following steps:

(1) Character recognition: the laboratory sheet (noisy image) is processed by methods including graying, binarizing, tilt detection and correction, and the like. And then carrying out layout processing, layout analysis, layout understanding and layout reconstruction. Finally, image segmentation is carried out, and the single characters can be conveniently identified after the segmentation treatment. Tesseact is a tool for character recognition, which can be quickly realized by combining Python.

(2) The numerical value range of each index is judged to extract the index with abnormal numerical value, the abnormal index is used as a characteristic of a patient to be input into the penultimate layer of the convolutional neural network model of the disease diagnosis module, and the accuracy of the diagnosis result is improved.

And the medical image auxiliary disease diagnosis module is used for acquiring the medical image input by the client, identifying the medical image by adopting a convolution lifting network and outputting the characteristics of the medical image.

In the field, the application technology of the convolutional neural network in medical image processing is becoming popular, the most prominent capability of the convolutional neural network is to extract data features, and the convolutional structure can reduce the memory occupied by deep networks, so that the convolutional neural network has wide application in aspects of image processing, voice recognition, text classification and the like. The medical image auxiliary disease diagnosis module adopts the prior art in the industry, uses a convolutional neural network to identify the medical image, and inputs the identification result as the characteristic of a patient into a convolutional neural network model of the disease diagnosis module.

The input picture can obtain the characteristics of the picture through the convolutional neural network, and the characteristics are used as additional symptom characteristics of a patient before being input into a full-connection layer for disease diagnosis, so that the accuracy of a disease diagnosis module for disease diagnosis can be increased.

In the invention, a disease diagnosis module analyzes the disease description of a patient by using a convolutional neural network model, and based on the disease description of the patient, test materials such as a test sheet provided by the patient are added into the convolutional neural network model as the characteristics of the patient to obtain the probability of various diseases of the patient, and finally the diseases of the patient are finally determined by a TF-IDF technology.

In the present invention, unlike picture processing, the input of the convolutional neural network is not a simple pixel point but a sentence and a document expressed in a matrix at the time of processing a text. Each row of the matrix corresponds to a vector, each corresponding to a word or a word, which vectors may be represented as single hot vectors, but most cases are represented using word-embedded vectors. Only one bit in the composition of the single heat vector is 1, and the other bits are 0, so that the index position of the word in the vocabulary is represented; while the word embedding vector is relatively lower in dimension and richer in characterization. Specifically, the specific technique of the disease diagnosis module is as follows:

(1) The network structure, as shown in fig. 2, is a text classification convolutional neural network structure used by the model.

Order theIs the k-dimensional vector of the i-th word in the corresponding sentence. A sentence of length m may be expressed asThe method comprises the steps of carrying out a first treatment on the surface of the Here->Representing a series of symbols. Use->To representAnd are serially connected.

The convolution operation comprises a filter w, and a new feature is obtained by using the trained filter to operate on a sliding window of length h. Sliding windowBy the formula->Generating features。

The next step is to use a max pooling operation for each feature map, the specific operation is to takeAs a feature value corresponding to the feature map. The purpose of this is to take the most important feature, the most valued feature, for each feature map. Each filter generates a feature. The present model uses different filters W for different size windows to produce a plurality of eigenvalues. These eigenvalues constitute the penultimate layer of the graph and pass through a fully connected softmax layer whose output is the probability distribution over the individual class labels.

(2) Regularization: dropout's method is used to avoid overfitting.

(3) And (3) adding the characteristics: several additional features are important in disease judgment of patients, such as: the sex, age, whether surgery has been performed, etc. The present invention incorporates three features of gender, age and whether surgery was done in the neural network.

In the neural network of the technology, there are x filters, so that an x-dimensional vector is generated at the penultimate layer, three dimensions are added behind the generated x-dimensional vector, and the first dimension represents gender, wherein male is 1, and female is 0; the second dimension represents age, and ten digits are placed in the second dimension by rounding off digits, such as: taking 2 at 15 years old and 1 at 14 years old; the third dimension represents whether surgery was done, 1 and 0. When the patient provides a laboratory sheet or a CT (computed tomography) film in addition to the most basic disease description, the results obtained by analysis by other functional modules are added into the convolutional neural network as two features, so that the accuracy of disease diagnosis can be improved, and the results are not affected if the laboratory sheet or the CT film is not provided.

(4) Convolution output: through the softmax layer, the output of the network is a probability distribution over the individual labels.

(5) TF-IDF: since different diseases may have the same symptoms, which may lead to an outcome that does not determine which disease the patient is ultimately suffering from, TF-IDF was introduced to address this problem. TF-IDF means word frequency-inverse text frequency, which consists of two parts, word frequency and inverse text frequency. Word frequency is the frequency with which all candidate word strings appear in all documents. The inverse text frequency may show the importance of a word or a word, for example, whether it is important to see only the number of times that the word or word has appeared, and the contribution of the word or word to the sentence, for example, is considered comprehensively. Words of little practical significance such as "and" have "in substantially all articles and, although their word frequency is very high, their contribution to understanding the main content of the article is far less than the contribution of the named entity to the sentence. So if a word appears more frequently, its reverse text frequency is lower; the lower the frequency of occurrence of the opposite word, the higher its inverse text frequency value, and if a word has occurred many times in all text, its inverse text frequency should be close to zero.

Specifically, the basic formula of the IDF of the word x is as follows: IDF (x) =log (N/N (x)); where N represents the total number of text in the corpus and N (x) represents the total number of text in the corpus that contains word x. The calculation formula of TF-IDF is as follows: TF-IDF (x) =tf (x) ∗ IDF (x), where TF (x) refers to the word frequency of word x in the current text.

Specifically, the TF-IDF calculation of the present invention can be expressed by the following formula:；/>representing the number of times symptom i appears in the current disease set, and s represents the number of symptom entities possessed by all diseases.

(6) Outputting a result: the convolutional neural network outputs probabilities of symptoms described for the patient in all diseases, and as the description of the patient may not be sufficiently standard and accurate, by using the TD-IDF method, several symptoms with the greatest probability of the symptoms of the N diseases in the convolutional output contributing the greatest to the diagnosis of the disease are calculated. After calculation, the most likely disease of the patient is screened out of 5 diseases according to the symptom information of the patient, and finally the disease of the patient is determined.

In one implementation, the doctor matching module performs doctor-patient matching including the following steps:

s1: and obtaining the disease result of the patient.

S2: identifying historical case data of a plurality of patients with the same disease as the disease result of the patient from the pre-stored historical case data in a database; the historical case data comprises patient information, doctor information, patient evaluation and satisfaction evaluation of doctors.

In one example, the symptom that occurs when patient X is a "gastrointestinal cold". It is necessary to find out the history of patients in gastroenterology and patients who have the same diseases as patient X, including the doctor's skill area and the doctor's satisfaction with the last match to obtain a table with patient X (see Table 1).

Table 1 historic patient table

Patient serial number	Disease of the human body	Treating doctor	Patient evaluation	Satisfaction evaluation of doctor
					1	Gastrointestinal cold	A	5	Satisfactory satisfaction
2	Gastrointestinal cold	A	5	Satisfactory satisfaction
					3	Gastrointestinal cold	B	4	Satisfactory satisfaction
4	Gastrointestinal cold	C	3	Dissatisfaction with
					5	Gastrointestinal cold	A	5	Satisfactory satisfaction
6	Gastrointestinal cold	C	4	Dissatisfaction with
					7	Gastrointestinal cold	B	5	Satisfactory satisfaction
8	Gastrointestinal cold	A	4	Satisfactory satisfaction
					……	……	……	……	……

S3: identifying doctor adept fields in doctor information, and carrying out doctor screening according to the correlation degree between the doctor adept fields and patient disease results;

s4: and calculating the recommendation degree of each doctor reserved after screening according to the patient evaluation, the satisfaction degree evaluation of the doctor and the recommendation degree formula.

In the invention, doctor-patient matching is firstly carried out according to the doctor's good field, if there is a history patient, the evaluation (1-5 points) of the history patient on the above is added into the index considered by the doctor-patient matching, and the recommendation degree of each doctor is calculated according to whether the doctor is satisfied with the last matching result (whether the disease of the matched patient is good for the treatment field) or not, wherein the calculation formula is as follows:the method comprises the steps of carrying out a first treatment on the surface of the Wherein->For doctor's recommendation->For the scoring of patient i,for the physician to be satisfied with the dispensing.

S5: and taking the doctor with the highest recommendation degree as the recommended doctor.

In one implementation, doctor a recommends: for 4 treatments, the patient aggregate score 19, doctor is good at treating cold, patient type is gastrointestinal cold, doctor is satisfied with patient assignment, recommendation is 3.9. The recommendation degree of the doctor B is 3.8; the recommendation degree of doctor C is 2.8. It can be seen that physician a is the recommended physician most appropriate for patient a.

In the traditional technology, in actual situations, patients can have certain blindness when selecting doctors, namely, expert doctors are preferentially selected to visit without considering the self situation, and the reasonable application of medical resources is not facilitated. And the existing patient number calling system in the hospital department, although alleviating the problem of selecting doctors for patients to a certain extent, does not consider the relationship between doctor expertise and patient symptoms. Leading to the situation that doctors may possibly take a diagnosis of ill patients with poor quality, leading to misdiagnosis and referral, reducing the doctor experience of the patients and increasing the risk of contradiction between doctors and patients.

Therefore, the invention solves the technical problems by the doctor matching module, uses the collaborative filtering algorithm to recommend doctors to patients, firstly finds out the historical data of a plurality of patients with the same illness state as the existing patients, and distributes doctors to new patients by considering the treatment condition of the historical patients, the doctor's adequacy field and the satisfaction degree of the doctor to the last matching. The patient is matched with the doctor for diagnosis.

In one implementation, the intelligent medical question and answer module is configured to perform the steps of:

s10: acquiring inquiry information sent by the user side;

s20: an information tag is determined from the query information, the information tag being used to represent a query intent of the patient, the information tag including greetings, interviews, and knowledge of medical knowledge.

The realization technology can adopt the intention judgment of multi-label classification, carries out multi-label classification on the question of the patient, extracts the labels from the query of the patient, knows what the type of the question he wants to know, and finally answers the question by the matching of the labels. This is prior art in the art and is not described here too much.

S30: responding to the information label as a greeting, and calling a corresponding preset reply statement according to inquiry information to return to the user side;

s40: responding to the information label as a consultation, sending the information label to a disease diagnosis module according to the consultation information to output a patient disease result, inputting the patient disease result to a doctor matching module to perform doctor-patient matching, and returning the patient disease result and a recommended doctor to the user side;

s50: and responding to the information label to know medical knowledge, identifying keywords in the query information, querying in internet medical information data according to the keywords, and returning a query result to the user side.

Specifically, medical knowledge can be processed by using a TextRank algorithm, the main purpose of the algorithm is to generate keywords and abstracts for texts, the importance of partial sentences for medical treatment only articles is calculated by using the TextRank algorithm, and then the first few most important sentences are taken as answers to be replied to a patient according to the importance order of the sentences.

Examples

The intelligent medical system of embodiment 2 has a system composition identical to that of embodiment 1, except for aunt, and the intelligent medical system of embodiment 2 further includes:

In one implementation, the verification code may be a password set by the doctor end or a mobile phone verification code log-in.

Intensive Care Units (ICU) are special care units that focus on the technological forces and advanced instrumentation of hospitals and perform continuous, dynamic, qualitative and quantitative monitoring and treatment of emergency, critical and critically ill patients. The family members are the main social support system of ICU patients and play a vital role in rehabilitation. The ICU patient is separated from the family members after being admitted, anxiety and depressed emotion are easy to generate, psychological wound symptoms are reduced, various medical care measures are smoothly developed, and the intelligent medical system develops video visit for the ICU patient and family members, and has good effects in meeting psychological demands of the patient and family members and relieving depression and anxiety.

Other structures of an intelligent medical system according to this embodiment are described in the prior art.

The present invention is not limited to the preferred embodiments, and any modifications, equivalent variations and modifications made to the above embodiments according to the technical principles of the present invention are within the scope of the technical proposal of the present invention.

Claims

1. The intelligent medical system is characterized by comprising a background server, a database and a client, wherein the database and the client are communicated with each other; the database is used for storing data; the client comprises a user end and a doctor end, wherein the user end is used for providing a man-machine interaction access interface for a patient, and the doctor end is used for providing a man-machine interaction access interface for a doctor;

wherein, the backend server includes:

2. The smart medical system of claim 1, wherein the backend server comprises:

3. The intelligent medical system according to claim 2, wherein the doctor matching module performs doctor-patient matching comprising the steps of:

obtaining a disease result of a patient;

4. The smart medical system of claim 1, wherein the backend server comprises:

5. The smart medical system of claim 4, wherein the smart question-answering module is configured to perform the steps of:

acquiring inquiry information sent by the user side;

6. The smart medical system of claim 1, further comprising: