A prescription recommendation system

Abstract

The present invention discloses a prescription recommendation system, including a feature extraction module, a prescription prediction module, a drug extension module, a drug contraindication checking module, a dosage friendliness module and a compatibility contraindication module, wherein the prescription prediction module includes a plurality of specialist prescription models constructed according to specialist categories. The present invention realizes simultaneous training and prediction of drug types and dosages in the same task, thereby providing patients with more accurate and personalized recommended prescriptions.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a prescription recommendation system. Background Art

[0002] In clinical work, the prescription recommendation system can improve work efficiency for tertiary hospitals and achieve the goal of "sinking knowledge from higher-level hospitals" and realize the effect of knowledge inheritance for grassroots hospitals, which is of great value.

[0003] Existing prescription recommendation systems mainly use collaborative filtering technology or model classifier mode. Collaborative filtering technology is based on the statistics of medication information of patients with the same diagnosis, obtains its statistical distribution, and then recommends drugs that are "similar" to the target patient and have the highest usage rate. This method lacks sufficient generalization ability; while the model classifier mode can only train drug types, and it is ineffective when encountering numerical variables such as dosage. Summary of the invention

[0004] To solve the above problems, the present invention provides a prescription recommendation system.

[0005] The present invention adopts the following technical solutions:

[0006] A prescription recommendation system, comprising a feature extraction module, a prescription prediction module, a drug expansion module, a drug contraindication checking module, a dosage friendlier module and a compatibility contraindication module, wherein the prescription prediction module comprises a plurality of specialist prescription models respectively constructed according to specialist categories;

[0007] The workflow of the system is as follows:

[0008] S1. Input medical record information and the specialty category to which it belongs, and the feature extraction module extracts patient features from the medical record;

[0009] S2, the prescription prediction module inputs the patient characteristics into the corresponding specialist prescription model according to the specialist category to which the medical record information belongs to perform prescription prediction, and outputs the recommended drugs and their corresponding dosages and usages;

[0010] S3, the drug expansion module searches for similar drugs of the same category and having the same main ingredients as the recommended drug from the drug library;

[0011] S4, the drug contraindication checking module verifies whether there are any contraindications based on the medical record information;

[0012] S5. The dosage-friendly module converts the dosage of the main ingredients of the drug into a user-friendly administration method according to the specific specifications of the recommended drug and similar drugs;

[0013] S6. The incompatibility module further checks whether there are incompatibility between drugs in the prescription and outputs the final recommended prescription.

[0014] Furthermore, the characteristics of the medical record information include symptoms and signs, test results, medical history, patient age and patient gender.

[0015] Furthermore, the specialist prescription model is used to classify and score drugs, and it divides drugs into several main categories according to their functions. Each main category contains several sub-categories of drugs with the same functions and can be interchanged with each other, and each sub-category of drugs contains several commonly used doses.

[0016] Furthermore, the specialist prescription model includes a representation layer, an embedding layer, a multi-residual layer, an application layer and a multi-task loss function. The representation layer is used to collect drug functional characteristics, and the application layer includes a main classification scoring layer, a sub-classification scoring layer and a dosage level scoring layer.

[0017] Furthermore, the main classification scoring layer is used to weight score the main category; the sub-classification scoring layer is used to weight score the sub-category drugs; and the dosage level scoring layer is used to weight score the commonly used dosages of the sub-category drugs.

[0018] Furthermore, the multi-task loss function L is specifically:

[0019]

[0020] Among them, p main represents the probability estimate of the main classification; Represents the probability estimate of the sub-category corresponding to the i-th main category; represents the probability estimate of the dose corresponding to the jth sub-category of the i-th main category; gt main The fact label representing the main category; Indicates the fact label of the sub-category corresponding to the i-th main category; The fact label of the dosage corresponding to the jth sub-category of the i-th main category; Indicates whether the dose classification is legal.

[0021] After adopting the above technical solution, the present invention has the following advantages compared with the background technology:

[0022] 1. The prescription prediction module of the prescription recommendation system of the present invention adopts a specialist prescription model. The specialist prescription model is constructed and trained according to the specialist category, which greatly reduces the number of drugs that need to be distinguished by a single model. The model is a deep learning model with the same representation of multiple tasks. It uses a deep cross network to achieve the fusion of multimodal features. Specifically, its representation layer is patient characteristics, and its application layer divides the corresponding drugs into several major categories according to their functions for different specialties and drug functions. There is no overlap between categories, and the dosage and usage of each drug are also independently formed into a branch, that is, the same representation layer information is used to train the selection of major categories, the selection of drugs in the major categories, and the selection of drug dosage and usage at the application layer, so as to achieve simultaneous training and prediction of drug types and dosages in the same task, overcoming the disadvantage that the existing model classifier cannot be effective for numerical variables of dosage;

[0023] 2. By combining the prescription prediction module and the drug expansion module, the system fully considers the patient's medical history information and expands similar drugs from the recommended drugs, which has good generalization ability;

[0024] 3. The system also uses the medical records of excellent clinical doctors as training materials. The feature extraction module extracts a lot of information such as symptoms and signs, test examinations and medical history in the medical records, and combines demographic information such as age and gender as features to provide patients with more personalized recommended prescriptions;

[0025] 4. The dosage-friendly module combines the specifications of the drug and converts the prescription recommendation results into a dosage-friendly form, making it easier for patients to take the drug according to the prescription. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a system flow chart of the present invention;

[0027] Figure 2 It is a classification structure diagram of the medicines of the present invention;

[0028] Figure 3 It is a structural diagram of the specialist prescription model of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0030] Example

[0031] like Figure 1As shown, a prescription recommendation system includes a feature extraction module, a prescription prediction module, a drug expansion module, a drug contraindication checking module, a dosage friendlier module and a compatibility contraindication module, wherein the prescription prediction module includes a plurality of specialist prescription models constructed according to specialist categories;

[0032] The workflow of the system is as follows:

[0033] S1. Input medical record information and the specialty category to which it belongs, and the feature extraction module extracts patient features from the medical record;

[0034] S2, the prescription prediction module inputs the patient characteristics into the corresponding specialist prescription model according to the specialist category to which the medical record information belongs to perform prescription prediction, and outputs the recommended drugs and their corresponding dosages and usages;

[0035] S3, the drug expansion module searches for similar drugs of the same category and having the same main ingredients as the recommended drug from the drug library;

[0036] S4, the drug contraindication checking module verifies whether there are any contraindications based on the medical record information;

[0037] S5. The dosage-friendly module converts the dosage of the main ingredients of the drug into a user-friendly administration method according to the specific specifications of the recommended drug and similar drugs;

[0038] S6. The incompatibility module further checks whether there are incompatibility between drugs in the prescription and outputs the final recommended prescription.

[0039] The specialist prescription model is used to classify and score drugs, such as Figure 2 As shown, the drugs are divided into several main categories according to their functions, each main category contains several sub-categories of drugs with the same functions and which can be interchanged with each other, and each sub-category of drugs contains several commonly used dosages.

[0040] like Figure 3 As shown, the specialist prescription model includes a representation layer, an embedding layer, a multi-residual layer, an application layer, and a multi-task loss function. The representation layer is used to collect drug functional characteristics, and the application layer includes a main classification scoring layer, a sub-classification scoring layer, and a dosage level scoring layer. The main classification scoring layer is used to weight the main category; the sub-classification scoring layer is used to weight the sub-class drugs; and the dosage level scoring layer is used to weight the commonly used doses of the sub-class drugs.

[0041] The multi-task loss function L is specifically:

[0042]

[0043] Among them, p main represents the probability estimate of the main classification; Represents the probability estimate of the sub-category corresponding to the i-th main category; represents the probability estimate of the dose corresponding to the jth sub-category of the i-th main category; gt main The fact label representing the main category; Indicates the fact label of the sub-category corresponding to the i-th main category; The fact label of the dosage corresponding to the jth sub-category of the i-th main category; Indicates whether the dose classification is legal. Indicates that the dose classification is illegal. Indicates that the dosage classification is legal.

[0044] The following is an example of a prescription recommendation process:

[0045] 1. Enter complete medical record information;

[0046] 2. After the feature extraction module, different types of features are extracted: one-hot type, such as acute upper respiratory tract infection, 10 means the disease is diagnosed, 01 means the disease is not diagnosed; time type (recorded in days), such as coughing days 3; floating point type, such as age (in years), 3.5 means 3 and a half years old; all different types of features are connected in series into a long vector to form a "complete feature", which is the patient feature;

[0047] 3. Send the patient characteristics to the prescription prediction module to output the recommended drugs and corresponding dosage and usage, such as azithromycin (three times a day, 20g each time);

[0048] 4. The above information is sent to the drug expansion module to expand the drug clarithromycin;

[0049] 5. Check the drug contraindications module. If there are no special symptoms, you can take the drug.

[0050] 6. Dosage-friendly module, output friendly administration method, for example, if azithromycin is 10g / bag, then the output is azithromycin (three times a day, 2 bags each time);

[0051] 7. Incompatibility contraindications module: if there is no incompatibility contraindications in the prescription, the prescription can be issued.

[0052] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A prescription recommendation system, characterized in that: It includes a feature extraction module, a prescription prediction module, a drug expansion module, a drug contraindication checking module, a dosage friendliness module and a compatibility contraindication module, wherein the prescription prediction module includes several specialist prescription models constructed according to specialist categories; The specialist prescription model includes a representation layer, an embedding layer, a multi-residual layer, an application layer, and a multi-task loss function. The representation layer is used to collect patient characteristics, and the application layer includes a main classification scoring layer, a sub-classification scoring layer, and a dose level scoring layer. In the specialist prescription model, drugs are divided into several main categories according to their functions, each main category contains several sub-categories of drugs with the same functions and interchangeable with each other, and each sub-category of drugs contains several commonly used dosages; The main classification scoring layer is used to weight the main classification; the sub-classification scoring layer is used to weight the sub-class drugs; the dosage level scoring layer is used to weight the commonly used dosages of the sub-class drugs; The workflow of the system is as follows: S1. Input medical record information and the specialty category to which it belongs, and the feature extraction module extracts patient features from the medical record; S2, the prescription prediction module inputs the patient characteristics into the corresponding specialist prescription model according to the specialist category to which the medical record information belongs to perform prescription prediction, and outputs the recommended drugs and their corresponding dosages and usages; S3, the drug expansion module searches for similar drugs of the same category and having the same main ingredients as the recommended drug from the drug library; S4, the drug contraindication checking module verifies whether there are any contraindications based on the medical record information; S5, the dosage-friendly module converts the dosage of the main ingredients of the drug into a user-friendly administration method according to the specific specifications of the recommended drug and similar drugs; S6. The incompatibility module further checks whether there are incompatibility between drugs in the prescription and outputs the final recommended prescription.

2. A prescription recommendation system as claimed in claim 1, characterized in that: The patient characteristics in the medical records include symptoms and signs, test results, medical history, patient age, and patient gender.

3. A prescription recommendation system as claimed in claim 2, characterized in that: The multi-task loss function L is specifically: Among them, p main represents the probability estimate of the main classification; Represents the probability estimate of the sub-category corresponding to the i-th main category; represents the probability estimate of the dose corresponding to the jth sub-category of the i-th main category; gt main The fact label representing the main category; Indicates the fact label of the sub-category corresponding to the i-th main category; The fact label of the dosage corresponding to the jth sub-category of the i-th main category; Indicates whether the dose classification is legal.

Images