CN112071388A - Intelligent medicine dispensing and preparing method based on deep learning - Google Patents

Abstract

The invention discloses an intelligent dispensing method based on deep learning. The method performs data preprocessing on the patient's condition data collected by the intelligent device, and transmits the processed data to the compounding drug control server. The network model is used to accurately analyze the unknown data in the follow-up time, output the proportioning table of the amount of drugs required by different patients, and the preparation drug control server regularly receives the pre-processed patient data, and tracks and adjusts the disease in time. The proportioning table of the amount of medicines required by patients, so as to achieve differentiated and customized medicine configuration services for patients, at the same time, it can improve the audience scope of medicines for manufacturers, and realize intelligent and differentiated management of medicines.

Description

A deep learning-based intelligent pharmaceutical dispensing method

technical field

The invention relates to the field of medical intelligent dispensing and pharmacy, in particular to a method for realizing intelligent control and differentiated management of the components of the medicines in the production and configuration of medicines combined with artificial intelligence technology.

Background technique

At present, domestic pharmaceutical manufacturers usually formulate the dosage ratio table of a certain drug in advance, and extract various components of the drug according to this table, and process it on the machine. This method of pre-customizing the dosage and proportioning table by professional pharmacists is usually configured according to professional knowledge, lacking differentiated and personalized configuration. Professional pharmacists investigate the etiology, pathology and prescription of patients on a large scale on the spot. , there are many challenges in reality; if the patient's condition is tracked for a long time, it will also consume a lot of human, material and financial resources. In addition, the patient's patient's condition varies from person to person and from time to time. Direct communication and the probability of understanding the disease are rare in reality. According to the principle that demand leads the market, manufacturers should formulate drugs based on user needs, formulate drugs in a differentiated manner, and realize intelligent drug preparation and customized services.

With the rapid development of technologies such as artificial intelligence and the Internet of Things, the application of some intelligent configurations has also been widely popularized. Combined with intelligent equipment and artificial intelligence technology, "private customization" and differentiated management of drug dosage control for dispensing and pharmaceuticals can be realized, which has important market value for future intelligent, differentiated and dynamic services in medical applications.

SUMMARY OF THE INVENTION

In view of the above-mentioned domestic pharmaceutical manufacturers, they usually lack real-time tracking of patients and patients when formulating drugs, and cannot dynamically obtain user needs. , proposes a deep learning-based intelligent pharmaceutical dispensing method. The method performs data preprocessing on the patient's condition data collected by the smart device, and uses the deep learning-based intelligent dispensing and pharmaceutical algorithm to train and learn the processed data. Using precise analysis, output the proportioning table of the amount of drugs needed by different patients, so as to achieve differentiated and customized drug configuration services for patients. Intelligent differentiated management of medicines.

In order to achieve the above purpose, the technical solution adopted in the present invention is: a deep learning-based intelligent dispensing and pharmaceutical method. The method includes the following steps:

Step A1: The smart device collects the current patient's condition data, preprocesses the data, and transmits it to the compounding drug control server; the data preprocessing is to archive the condition data and organize it into structured data; the structured data According to the patient's name, gender, age, medical history, physical measurement data, etiology, treatment data, drug use, duration and status, classify and record and file;

Step A2: The compounding drug control server adopts the intelligent dispensing and pharmaceutical algorithm based on deep learning to output the proportioning table of the amount of medicines required by different patients;

Step A3: The preparation drug control server periodically collects the preprocessed condition data of the patient through the intelligent device in real time, and uses the deep learning-based intelligent dispensing and pharmaceutical algorithm proposed by the present invention to dispense the required medicine of the patient. The ratio is changed, and the changed dose ratio table is transmitted to the patient or doctor.

Further, the described deep learning-based intelligent dispensing pharmaceutical algorithm includes the following steps:

Step B1: Randomly initialize the parameters w and θ of the Actor network and the Critic network respectively;

Step B2: determine whether the current time t is less than the preset iteration period T, if so, go to step B3, otherwise go to step B10;

Step B3: Initialize the state set S, and obtain its feature vector φ(S) after data preprocessing;

Step B4: The Actor network uses φ(S) as the data input, outputs the corresponding behavior set A at the current moment, and obtains the next moment state S' and the immediate reward R of the Critical network based on the behavior set A;

Step B5: The Critic network uses φ(S) and φ(S') as data input respectively, and obtains the output total cumulative reward V(S) of the current state and the total cumulative reward V(S') of the subsequent state. Equations (1) and (2) calculate V(S) and V(S'), respectively:

V(S')=R(t)+βV[S', argmax _A' V(S', A'|w|θ)] (2)

Among them, R(t) represents the immediate reward at time t, β represents the discount factor, A′ represents the behavior at the next moment, and S _t and A _t represent the state and behavior at time t, respectively;

Step B6: The Critic network randomly selects D samples from the playback buffer pool, and calculates the error function δ according to formula (3):

δ=R(t)+αV(S′)-V(S) (3)

Among them, α represents the update rate, and its range is [0, 1];

Step B7: The Critic network takes the loss function L(w) according to the formula (4), calculates the gradient of the loss function according to the formula (5), and updates the critical network parameter w according to the formula (6), which is used to update the gradient of the critical network parameters:

Among them, k represents the statistical variable of sample count, γ represents the learning step size, represents the learning rate, and its range is [0, 1];

Step B8: Update Actor's network parameter θ according to formula (7):

为学习步长，即学习率，范围为[0，1]；π表示Actor网络在状态S下，采取行为A所依据的策略，

表示关于参数θ的梯度；in,

is the learning step size, that is, the learning rate, in the range [0, 1]; π indicates that the Actor network adopts the strategy based on the behavior A in the state S,

represents the gradient with respect to the parameter θ;

Step B9: Output the policy set π of the Actor network;

Step B10: stop the cycle;

Preferably, the compounding drug control server is an APP and a hardware device capable of running an intelligent dispensing and pharmaceutical algorithm based on deep learning. The APP is an application program that executes the application software on the hardware device, and the hardware Devices refer to hardware devices included in computers or mobile phones.

Preferably, the dosage ratio table includes at least the following two items: the name of the medicine and the required dosage.

Preferably, the playback buffer pool is used to store Actor network and Critic network samples, and the samples include (S, A, R, S').

Beneficial effects: The present invention proposes an intelligent dispensing method based on deep learning. The method performs data preprocessing on the patient's condition data collected by the smart device, and transmits the processed data to the compounding drug control server. The network model is used to accurately analyze the unknown data at the subsequent time, output the proportioning table of the amount of drugs required by different patients, and the preparation drug control server regularly receives the preprocessed patient data, and timely tracks and adjusts the disease. The proportioning table of the amount of medicines required by patients, so as to achieve differentiated and customized medicine configuration services for patients, at the same time, it can improve the audience scope of medicines for manufacturers, and realize intelligent and differentiated management of medicines.

Description of drawings

Fig. 1 is a kind of schematic flow chart of the intelligent dispensing pharmaceutical method based on deep learning;

Fig. 2 is a kind of schematic diagram of the operation principle of the algorithm in the intelligent dispensing and pharmaceutical method based on deep learning;

Fig. 3 is a schematic diagram of an algorithm flow in a deep learning-based intelligent pharmaceutical dispensing method.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the specific embodiments and the accompanying drawings. It should be understood that these embodiments are only used to explain the technical principle of the present invention, and are not intended to limit the protection scope of the present invention. Also, in the following description, descriptions of well-known structures and techniques are omitted to avoid unnecessarily obscuring the concepts of the present invention.

Basic knowledge: The deep learning-based intelligent dispensing and pharmaceutical algorithm consists of three networks: Actor, Critic network and Critic target network. Actor network is composed of a network, which also adopts the structure of deep convolutional neural network, namely input layer, hidden layer and output layer. The network receives the input data and the reward or punishment signal fed back by the system, and continuously adjusts the parameters according to the objective function through the hidden layer to train and learn the model of the network. It is a policy-based Q network. The Critic network consists of two networks, similar to the deep Q network, an original network for training the model, and a target Q network for periodically saving the network parameter model of the original network. Here we refer to the target Q network as the Critic target network.

The idea of intelligent dispensing and pharmaceutical algorithm based on deep learning is as follows:

1. Actor network adopts policy-based Q network, first observes the current state, adopts ∈-greedy selection method, and executes a behavior;

2. The Critic network gives a score based on the current state and behavior of the Actor network. This score is based on the score of the total reward accumulated from the current time t to the end of the iteration and the comprehensive evaluation of the Critic target network score to continuously adjust the scoring strategy. The total cumulative reward is given by system or environmental feedback.

3. Actor network adjusts its behavior strategy according to the score of Critic network, which is the reward.

4. The Critic network constantly adjusts its own scoring strategy according to the system or environment's real reward and the Critic target network's scoring. Actor network and Critic network undergo continuous network training, and finally Actor network outputs the optimal behavior set.

The present invention provides a deep learning-based intelligent dispensing and pharmaceutical method, and the schematic flowchart of the method is shown in Figure 1, including the following steps:

Step A1: The smart device collects the current patient's condition data, preprocesses the data, and transmits it to the compounding drug control server; the data preprocessing is to archive the condition data and organize it into structured data; the structured data According to the patient's name, gender, age, medical history, physical measurement data, etiology, treatment data, drug use, duration and status, classify and record and file;

Specific embodiment 1: The patient collects disease data by using a smart phone, a smart bracelet with medical functions, or devices such as blood pressure, blood oxygen, blood sugar, and thermometer.

Specific embodiment 2: The patient transmits the disease data to the drug preparation control server or the drug dispenser through the medical staff's examination and diagnosis sheet, the CT image taken by the machine, etc. through the remote medical equipment; the drug preparation control server adopts the deep learning-based Intelligent dispensing The pharmaceutical method is differentiated and intelligent to generate the drug dispensing table.

Specific embodiment 3: The medical professional or the patient transmits the condition data through the smart device to the drug preparation control server or the drug dispenser through voice or video.

Specific embodiment 4: Business database acquisition of medical equipment: medical records, electronic medical records and clinical data, including patient admission, diagnosis, hospitalization, treatment, examination, medication and discharge information.

Step A2: The compounding drug control server adopts the intelligent dispensing and pharmaceutical algorithm based on deep learning to output the proportioning table of the amount of medicines required by different patients;

Step A3: The preparation drug control server periodically collects the preprocessed condition data of the patient through the intelligent device in real time, and uses the deep learning-based intelligent dispensing and pharmaceutical algorithm proposed by the present invention to dispense the required medicine of the patient. The ratio is changed, and the changed dose ratio table is transmitted to the patient or doctor.

Figure 2 depicts a schematic diagram of the operating principle of the algorithm in a deep learning-based intelligent pharmaceutical method. First, the Actor network initializes the current state, adopts a certain strategy, and executes a behavior;

Specific embodiment 1, the Actor network adopts the ∈-greedy selection strategy, that is, selects a certain behavior with the probability of ∈, and selects a certain behavior from the playback buffer pool with the probability of 1-∈ to execute.

In the specific embodiment 2, the Actor network adopts a random selection strategy, that is, a certain behavior is randomly selected for execution.

In the specific embodiment 3, the Actor network adopts a priority sampling method to select a strategy, that is, a certain behavior is selected and executed in the order of the preset priority when the sample is sampled.

Further, the Actor network feeds back the current state and execution behavior to the Critic network. The Critic network initially uses a random strategy to score (ie, reward) the behavior of the Actor network. Based on this reward, the Actor network adjusts its own strategy and behavior, and Obtain the state of the next moment, send the obtained state to the Critic network, and store the current state, behavior, reward and the state of the next moment as a sample tuple in the playback buffer pool. In the following time, the Critic network learns the scoring strategy according to the real reward given by the environment and the score of the Critic target network and the samples currently accessed by itself. The Critic network and the Critic target network calculate the loss function according to formula (4), using formula ( 5) Calculate the gradient of the loss function, so as to estimate the error, continuously adjust the network parameters of the Critic network, and periodically copy the network parameters after training to the Critic target network, and the Critic target network saves the network model parameters with better training.

The present invention provides an intelligent pharmaceutical dispensing method based on deep learning. The schematic diagram of the algorithm flow of the method is shown in FIG. 3 .

Specific embodiment 1: The dosage ratio table includes: the name of a certain medicine, the type it belongs to, the properties of the medicine and the required dosage.

Specific embodiment 2: the dosage ratio table includes: the patient's name, the name of the drug, the required amount of the drug and the number of days recommended for taking it.

The described preparation drug control server is an APP and a hardware device capable of running an intelligent dispensing and pharmaceutical algorithm based on deep learning. The described APP is an application program that executes application software on the hardware device, and the hardware device refers to a computer. Included hardware devices.

Specific embodiment 1: The control server for compounding medicines adopts a computer running an APP based on a deep learning-based intelligent compounding and pharmaceutical algorithm.

Specific embodiment 2: The control server for compounding medicine adopts the mobile phone of the APP running the intelligent medicine compounding and pharmaceutical algorithm based on deep learning.

Specific embodiment 3: The preparation drug control server adopts a workstation running an APP based on an intelligent dispensing and pharmaceutical algorithm based on deep learning.

Through the above method, the present invention provides a deep learning-based intelligent dispensing and pharmaceutical method, which realizes real-time, differentiated, and intelligent formulation of drug ratios for patients, and at the same time, provides intelligent and customized services for the dispensing and pharmaceutical mode of dispensing pharmaceutical manufacturers. Transformation provides an efficient and convenient method with broad market application prospects.

The above description is only provided as an implementable technical solution of the present invention, and not as a single limitation of the technical solution itself.

Claims (5)

1. An intelligent medicine dispensing and preparing method based on deep learning comprises the following steps:

step A1: the intelligent equipment collects the disease condition data of a current patient, preprocesses the data and transmits the data to the control server for preparing the medicine; the data preprocessing is to file the disease data and arrange the disease data into structured data; the structured data is classified, recorded and filed according to the name, sex, age, medical history, physical examination data, etiology, treatment data, drug use condition, duration and current situation of the patient;

step A2: the prepared medicine control server adopts an intelligent medicine preparation algorithm based on deep learning to output a ratio table of the medicine amount required by different patients;

step A3: the prepared medicine control server regularly collects the preprocessed disease condition data of the patient in real time through intelligent equipment, changes the dosage ratio of the medicine required by the patient by adopting the intelligent medicine dispensing and preparing algorithm based on deep learning, and transmits the changed dosage ratio table to the patient or a doctor.

2. The intelligent medicine dispensing and preparing method based on deep learning of claim 1, wherein: the intelligent medicine dispensing and preparing algorithm based on deep learning comprises the following steps:

step B1: respectively initializing parameters w and theta of an Actor network and a Critic network at random;

step B2: judging whether the current time T is smaller than a preset iteration period T, if so, turning to a step B3, otherwise, turning to a step B10;

step B3: initializing a state set S, and obtaining a characteristic vector phi (S) of the state set S after data preprocessing;

step B4: the Actor network uses phi (S) as data input, outputs a corresponding behavior set A at the current moment, and obtains a state S' at the next moment and an immediate reward R of the Critic network based on the behavior set A;

step B5: the Critic network uses phi (S), phi (S ') as data inputs, respectively, to get the total accumulated award V (S) for the current state and the total accumulated award V (S ') for the subsequent state of the output, where V (S) and V (S ') are calculated according to equations (1) and (2), respectively:

V(S′)＝R(t)+βV[S′，argmax_A′V(S′，A′|w|θ)] (2)

wherein R (t) represents an instant prize at time t, β represents a discount factor, A' represents a behavior at the next time, S_tAnd A_tRespectively representing the state and the behavior at the moment t;

step B6: the criticic network randomly draws D samples from the playback buffer pool and calculates an error function according to equation (3):

＝R(t)+αV(S′)-V(S) (3)

wherein α represents an update rate, which ranges from [0, 1 ];

step B7: the criticic network is used as a loss function L (w) according to the formula (4), the gradient of the loss function is calculated according to the formula (5), and the criticic network parameter w is updated according to the formula (6) and used for updating the gradient of the criticic network parameter:

where k represents the statistical variable of the sample count, γ represents the learning step size, represents the learning rate, and ranges from [0, 1 ];

step B8: updating the network parameter θ of the Actor according to equation (7):

wherein,

for the learning step size, i.e. learning rate, the range is 0, 1](ii) a Pi represents the strategy according to which the Actor network takes action A in the state S,

representing a gradient with respect to a parameter θ;

step B9: outputting a strategy set pi of the Actor network;

step B10: the cycle is terminated.

3. The intelligent medicine dispensing and preparing method based on deep learning of claim 1, wherein: the drug preparation control server is an APP and hardware equipment capable of running an intelligent drug preparation algorithm based on deep learning, the APP is an application program and executes application software on the hardware equipment, and the hardware equipment refers to hardware equipment included in a computer or a mobile phone.

4. The intelligent medicine dispensing and preparing method based on deep learning of claim 1, wherein: the dosage proportioning table at least comprises the following two items: the name of the drug and the amount of drug required.

5. The intelligent medicine dispensing and preparing method based on deep learning of claim 2, wherein: the playback buffer pool in step B6 is used to store the Actor network and criticc network samples, which include (S, a, R, S').

Images