A privacy protection and image recognition-based drug delivery method and device

Abstract

The application provides a drug delivery method and device based on privacy protection and image recognition, which comprises the following steps: after a patient visits a doctor and purchases drugs, the prescribed dosage, drug name, oral disease name and personal information are input into a central processing device; the oral lesion images of the patient are collected by multiple terminals of the hospital and treatment institutions and the drug delivery device, and privacy protection is performed according to the number of cases; an oral positioning model is trained by federated learning, which is used for oral lesion partition positioning; a disease analysis model is trained, which is used for determining the drug dosage; the drug delivery device is initialized; the patient uses the drug delivery device to deliver drugs, and the oral lesion images are uploaded by collection; the drug delivery device establishes a patient drug delivery project, and performs expiration reminding, expiration locking and device reuse; the drug delivery device is a handheld spray drug delivery device that meets the requirements of human oral drug delivery, comprising a drug storage chamber, an image acquisition and analysis device, a drug delivery probe and an information interconnection device.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical intelligent devices, and particularly relates to a drug delivery method and device based on privacy protection and image recognition. BACKGROUND

[0002] When a patient is given a spray drug for an oral disease, the patient often faces the problems of inaccurate drug delivery position, unadjustable drug delivery amount according to the disease condition, and expired drug. Image recognition technology has been widely used in the medical field. By adding a recognition device to the drug delivery device, the lesion position and the severity of the disease can be recognized, so that the drug delivery can be adjusted. Due to the individual differences in the oral conditions of patients, the recognition effect of the same disease condition may be quite different. In addition, the drug delivery device can be provided with an information sharing and transmission function. Based on the oral image samples from various patients, the recognition function can be further trained through federated learning. This will involve the problem of patient privacy protection, which needs to be solved by a technical solution. At the same time, to solve the problem of expired spray drug for the oral cavity, the information transmission function of the drug delivery device can be combined with the blockchain and smart contract technology to manage the drug delivery, so as to prevent the expiration of the spray drug for the oral cavity and the abuse of the drug. SUMMARY

[0003] The present application aims to solve the above technical problems, and provides a drug delivery method based on privacy protection and image recognition.

[0004] Another object of the present application is to provide an oral drug delivery device based on privacy protection and image recognition.

[0005] The object of the present application can be achieved by adopting the following technical solutions: After a patient visits a doctor and purchases a drug, the prescribed amount, the drug name, the oral disease name, and the personal medical information of the patient are input into a central processing device; the oral lesion images of the patient are collected by multiple terminals of the hospital, the treatment institution, and the drug delivery device, and are clustered and protected by privacy according to the number of cases; an oral positioning model is trained by federated learning, which is used for oral lesion partition positioning; a disease analysis model is trained, which is used for drug delivery amount determination; the drug delivery device is initialized; the patient uses the drug delivery device to deliver the drug, and uploads the oral lesion images collected; the drug delivery device establishes a patient drug delivery project, and performs expiration reminding, expiration locking, and device reuse.

[0006] Further optionally, after the patient visits a doctor and purchases a drug, the prescribed amount, the drug name, the oral disease name, and the personal medical information of the patient are input into a central processing device, which includes:

[0007] The oral disease name is contained in the following oral disease categories: caries, oral mucosa disease, periodontal disease. The prescribed amount of medical advice refers to the recommended amount of the physician at the time of consultation. If the doctor does not give the recommended amount, it is set to the amount specified in the drug instruction manual. The personal consultation information refers to the age, gender, drug opening time, and other auxiliary information of the patient. The other auxiliary information refers to the patient's past medical history, disease inducement, and disease stage information collected from the patient's medical history by text analysis method, such as: preference for spicy food and sweets, and dental pulp root canal treatment. The central processing device is deployed in the information center of the hospital or treatment institution, which can collect, sort, store and transmit the patient's prescribed amount of medical advice, drug name, oral disease name, and personal consultation information data. Further optionally, the patient's oral lesion image is collected by the hospital and treatment institution, and the drug delivery device multi-terminal, clustered and privacy protected according to the number of cases, including: the oral lesion image can be collected by the hospital and treatment institution after the patient's consent, or collected by the drug delivery device. The oral lesion image is clustered based on the oral disease name, the personal consultation information and the other auxiliary information.

[0008] For example: according to the oral disease name, a patient has periodontal disease, and the age is 65 years old, then it is allocated to two categories with a patient of 17 years old and with periodontal disease, which is mainly due to the great difference between the oral conditions of the elderly and the young.

[0009] After clustering is stable, the number of cases in each cluster is counted, sorted from low to high, and privacy is protected according to the sorting, adding noise to the personal consultation information and the other auxiliary information, and the cluster with less amount is allocated more privacy budget to ensure that the cluster with less sample size can provide more available information during model training. The number of cases refers to the number of oral lesion images contained in each cluster. The privacy protection refers to using random noise to ensure that the result of public visible information query request will not leak individual privacy information, that is, to remove individual characteristics to protect user privacy on the premise of retaining the statistical characteristics of the database; The privacy protection technology adopted by the present application is local differential privacy protection based on Laplace mechanism, which mainly protects structured information, including: the personal consultation information and the other auxiliary information, the greater the privacy budget, the lower the protection level of the information, and the more available information.

[0010] Further optionally, the training of the oral cavity positioning model through federated learning for oral cavity lesion partition positioning comprises: the oral cavity positioning model is used for identifying the position of a lesion inside an oral cavity, and the specific training manner comprises: 1) partitioning the internal structure of the oral cavity, that is, matching the positions of various organs of the oral cavity to a plurality of oral cavity partitions; for example, for teeth, four regions of right upper, left upper, left lower and right lower can be classified according to the tooth arrangement classification used in clinics, the right upper is classified as a first region, that is, the right upper central incisor is tooth 11, the right upper second molar is tooth 17, the left upper is classified as a second region, then the left upper central incisor is tooth 21, and so on;

[0011] 2) the pre-collected oral cavity lesion image is used for labeling the oral cavity partitions and the names of oral cavity diseases, and a machine learning model is trained;

[0012] 3) the obtained oral cavity positioning model can judge whether the oral cavity image newly collected by the drug delivery device contains a lesion of an oral cavity disease, and rapidly judge the partition of the lesion in the oral cavity; 4) the oral cavity lesion images of patients treated by the same doctor are uniformly collected and trained; at the same time, in order to avoid the diagnosis difference of different doctors, the models trained at the multiple terminals of the doctors are uploaded to a federated learning server, the updated parameters are returned, and the models of the terminals are updated.

[0013] The federal learning is a horizontal federal learning, that is, all data are distributed to different machines, each machine downloads a model from a server, then trains the model using local data, and then returns parameters that need to be updated to the server; the server aggregates the returned parameters on each machine, updates the model, and feeds back the latest model to each machine. Further optionally, the training disease analysis model for determining the amount of administration includes: pre-training the disease analysis model based on image recognition technology, and the obtained model can judge the cluster of the patient's oral disease based on the newly collected oral lesion image of the administration device. Determine the prescribed amount of the different disease conditions corresponding to the cluster as the administration amount of this administration. The disease stage refers to the severity of the oral disease, including the size of the lesion and the duration. The prescribed amount of the different disease conditions refers to the amount of the drug determined by clinical trials under the disease severity and the patient's physical condition. Further optionally, the initialization of the administration device includes: the initialization refers to the central processing device entering the prescribed amount of the medical order, the drug name, the oral disease name, the prescribed amount of the different disease conditions, and the personal medical information data into the administration device; at the same time, the oral positioning model and the disease analysis model are embedded into the administration device. Further optionally, the patient uses the administration device for administration, and uploads the oral lesion image through collection, including: the patient uses the administration device for administration, and the administration device identifies the lesion position in the oral cavity through the oral positioning model and the disease analysis model, and administers based on the prescribed amount of the different disease conditions or the prescribed amount of the different disease conditions. According to the number of cases in the cluster, the central processing device can also collect the oral lesion image, that is, send a collection push to the administration device of the patient who meets the cluster condition with fewer cases to expand the sample size; the patient can upload the oral lesion image through the administration device, and the image is learned after privacy protection, including: 1) the administration device can collect the oral lesion image and the personal medical information of the patient after the patient agrees; 2) the oral lesion image is processed by image segmentation, and non-oral facial organs are processed by Gaussian blur to prevent leakage of patient privacy; 3) remove the identity information such as name, phone number and medical date when uploading the personal medical information;

[0014] An incentive mechanism is adopted for patients who voluntarily upload images after collection, that is, patients who have uploaded the oral lesion image after collection can match the original doctor through the administration device since the date of uploading, and obtain a remote pre-consultation.

[0015] Further optionally, the administration device establishes a patient administration project, and performs expiration reminding, expiration locking and device reuse, including:

[0016] Since the oral spray administration often faces the situation of drug expiration, through the administration device combined with the function of the smart contract of the block chain, the administration situation can be monitored, and the device can be reused.

[0017] The smart contract of the block chain presets the drug name, the administration device issuing time and the drug validity period on the day of issuing the administration device, and establishes the administration project.

[0018] The smart contract can establish the administration project for management for each administration device in the block chain; meanwhile, according to the drug validity period and the administration device issuing time, the time is calculated, and when the time after the administration device is issued exceeds the drug validity period, the expiration reminder is carried out through the administration device or the expiration locking of the administration device is carried out.

[0019] The drug validity period includes the drug validity period determined by the instruction manual and the use time recommended by the doctor's advice, for example: long-term use of some drugs will cause discomfort in the pharynx, so the doctor will make recommendations on the use time. For the drug validity period determined by the instruction manual, the expiration locking of the administration device is carried out, that is, the administration cannot be carried out again; for the drug validity period recommended by the doctor's advice, the expiration reminder of the administration device is carried out, but the patient can still choose to continue to use.

[0020] After the expiration reminder and the expiration locking, the patient can apply to the hospital through the administration device, if the patient needs to continue to use the drug or replace other drugs after being diagnosed by the doctor, the device cleaning and drug supplementing can be carried out in the hospital, that is, the administration device can be reused repeatedly.

[0021] When other drugs are used, the related parameters of the new drug need to be set again, and a new administration project is established.

[0022] Another object of the application can be achieved by adopting the following technical solutions:

[0023] A drug administration method and device based on privacy protection and image recognition, characterized in that the device comprises:

[0024] The drug delivery device is a handheld spray-on drug delivery device that meets the requirements for oral administration, including a drug storage chamber, an image acquisition and analysis device, a drug delivery probe, and an information interconnection device. The drug storage chamber is used for storing the drug solution. The image acquisition and analysis module can acquire images of the oral cavity from a specified distance and direction in front of the mouth, and includes the oral cavity positioning model and the disease analysis model, enabling image comparison and determination of the lesion's oral cavity partition and the drug dosage. The drug delivery probe delivers the prescribed amount of drug solution to the lesion's oral cavity partition, specifically including: 1) The drug delivery probe can rotate according to the judgment result of the oral cavity positioning model, and its top has a macro camera to acquire images and transmit them back to the image acquisition and analysis module for image comparison, until the top of the drug delivery probe faces the same direction as the lesion's oral cavity partition; 2) After the drug dosage is determined, the drug delivery probe draws the prescribed amount of drug solution and administers the drug; The information interconnection module is located on the body of the drug storage chamber and mainly has the following functions:

[0025] a. Used to establish the drug administration program, provide expiration reminders and expiration lockouts, thereby enabling patients to submit device reuse requests to the hospital;

[0026] b. Used to send recruitment pushes to patients, who then upload images of the oral lesions and their personal medical information, which are then used for model learning after privacy protection.

[0027] c. Used to record the time and number of times the oral lesion images are uploaded, and to send an appointment notification to the original attending physician when the patient needs to make an appointment for consultation.

[0028] The technical solutions provided by the embodiments of this invention can include the following beneficial effects: The method and apparatus provided by this invention can confirm the location and severity of oral lesions in patients through image recognition, and perform precise drug administration based on pre-set dosages for different conditions; it adopts a federated learning method, combined with clustering results, to selectively collect missing oral image samples from the target population; at the same time, the clustering results also affect the level of privacy protection for different samples, so as to ensure that clustering with a small number of cases can present more effective information during model training. Furthermore, the method and apparatus used in this invention not only help to expand the training samples of oral lesion images, but also facilitate the establishment of a reward mechanism for patients who voluntarily upload images, and can establish a drug administration management program to prevent the expiration and abuse of spray drugs, and promote device reuse. [Attached Image Description] Figure 1 This is a flowchart of a drug delivery method based on privacy protection and image recognition according to the present invention. Figure 2 This is a structural diagram of an oral drug delivery device based on privacy protection and image recognition according to the present invention.

Detailed Implementation Methods

[0029] The oral disease names are included in the following categories: dental caries, oral mucosal diseases, and periodontal diseases. The prescribed dosage refers to the dosage recommended by the physician at the time of consultation; if the physician does not provide a recommended dosage, the dosage specified in the drug instructions will be used. The personal medical information refers to the patient's age, gender, prescription date, and other auxiliary information. The other auxiliary information refers to information collected from the patient's medical records using text analysis methods, such as the patient's past medical history, causes of illness, and stage of illness, for example, preferences for spicy and sweet foods, and previous root canal treatment. The central processing unit, deployed in the hospital or treatment institution's information center, can collect, organize, store, and transmit patient prescribed dosages, drug names, oral disease names, and personal medical information data. S2: Collecting patient oral lesion images through multiple terminals including the hospital, treatment institution, and drug delivery device, clustering them, and protecting privacy based on the number of cases includes: The oral lesion images can be collected by the hospital or treatment institution with the patient's consent, or collected through the drug delivery device. The oral lesion images are clustered based on the name of the oral disease, the personal medical information, and other auxiliary information.

[0030] For example, based on the name of the oral disease, if a patient has periodontal disease and is 65 years old, then he / she and a 17-year-old patient with periodontal disease are assigned to two different categories. This is mainly because there are significant differences in the oral conditions of the elderly and teenagers.

[0031] After clustering stabilizes, the number of cases in each cluster is counted and sorted from low to high. Privacy protection is implemented based on this sorting by adding noise to the individual's medical information and other auxiliary information. Clusters with fewer cases are allocated a larger privacy budget to ensure that clusters with smaller sample sizes can provide more usable information during model training. The number of cases refers to the number of oral lesion images contained in each cluster. Privacy protection refers to using random noise to ensure that the results of query requests for publicly visible information do not leak individual privacy information; that is, removing individual features while preserving the statistical characteristics of the database to protect user privacy. The privacy protection technology adopted in this invention is local differential privacy protection based on the Laplace mechanism, which mainly protects structured information, including the individual's medical information and other auxiliary information. The larger the privacy budget, the lower the level of information protection and the more usable information.

[0032] S3: Training an oral cavity localization model through federated learning for oral lesion localization includes: The oral cavity localization model is used to identify the location of lesions inside the oral cavity. The specific training methods include: 1) Dividing the internal structure of the oral cavity into multiple oral cavity partitions; for example, for teeth, they can be divided into four regions according to the clinically used dental arch classification: upper right, upper left, lower left, and lower right. The upper right is divided into the first region, i.e., the upper right central incisor is tooth 11, the upper right second molar is tooth 17, the upper left is divided into the second region, i.e., the upper left central incisor is tooth 21, and so on.

[0033] 2) The pre-collected oral lesion images are labeled with the oral cavity regions and the names of the oral diseases, and then used to train a machine learning model;

[0034] 3) The obtained oral cavity localization model can determine whether the image contains oral disease lesions based on the newly acquired oral cavity image of the drug delivery device, and quickly determine the oral cavity region where the lesion is located; 4) The oral lesion images of patients who visit the same doctor are uniformly collected and trained; at the same time, in order to avoid diagnostic differences between different doctors, the model trained at multiple terminals belonging to different doctors is uploaded to the federated learning server, updated and returned, and each terminal updates the model.

[0035] The federated learning is a horizontal federated learning method, where all data is distributed to different machines. Each machine downloads the model from the server, trains the model using local data, and then returns the parameters that need updating to the server. The server aggregates the parameters returned from each machine, updates the model, and then feeds the latest model back to each machine. S4: Training a disease analysis model for dosage determination includes: pre-training a disease analysis model based on image recognition technology. The resulting model can determine the cluster of oral lesions in the patient based on newly acquired images of oral lesions from the drug delivery device. The prescribed dosage for each cluster corresponding to different disease conditions is determined as the dosage for this administration. The disease stage refers to the severity of the oral disease, including the size and duration of the lesions. The prescribed dosage for different disease conditions refers to the dosage of the drug determined through clinical trials for different disease severity levels and patient conditions, based on the prescribed dosage. S5: Initializing the drug delivery device includes: The initialization refers to the central processing unit inputting the prescribed dosage, drug name, oral disease name, prescribed dosage for different conditions, and personal medical information data into the drug delivery device; simultaneously embedding the oral cavity positioning model and the disease analysis model into the drug delivery device. S6: The patient uses the drug delivery device to administer medication, and the collected and uploaded oral lesion images include: The patient uses the drug delivery device to administer medication, and the drug delivery device identifies the location of lesions in the oral cavity through the oral cavity positioning model and the disease analysis model, and administers medication based on the prescribed dosage for different conditions. Based on the number of cases after clustering, the central processing device can also collect oral lesion images, that is, send collection pushes to the drug delivery devices of patients with fewer cases who match the clustering conditions, in order to expand the sample size; patients can upload oral lesion images through the drug delivery device, which are then used for model learning after privacy protection, including: 1) The drug delivery device can collect the patient's oral lesion images and personal medical information after obtaining the patient's consent; 2) The oral lesion images are processed by image segmentation, and non-oral facial organs are Gaussian blurred to prevent leakage of patient privacy; 3) When the personal medical information is uploaded, identity information, such as name, telephone number, and date of visit, is removed;

[0036] A reward mechanism is implemented for patients who voluntarily upload images after being recruited. That is, patients who have uploaded the oral lesion images after being recruited can obtain one remote appointment consultation by matching their original attending physician through the drug delivery device from the date of uploading.

[0037] S7: The drug delivery device establishes patient drug delivery programs, provides expiration reminders, locks out expired drugs, and allows for device reuse.

[0038] Since oral spray administration often faces the problem of expired drugs, the aforementioned drug delivery device, combined with blockchain smart contract functionality, can monitor drug delivery and enable device reuse.

[0039] The blockchain pre-sets a smart contract, which can record the drug name, the drug delivery time, and the drug expiration date on the date the drug delivery device is issued, and establish the drug delivery program.

[0040] The smart contract can establish and manage the drug delivery project on the blockchain for each drug delivery device; at the same time, it can calculate the time based on the drug's expiration date and the drug delivery device's distribution time, and when the time after the drug delivery device is distributed exceeds the drug's expiration date, it can send an expiration reminder through the drug delivery device or lock the drug delivery device upon expiration.

[0041] The drug's expiration date includes the expiration date specified in the instructions and the recommended usage duration as prescribed by the physician. For example, prolonged use of some medications can cause throat discomfort, so physicians will provide recommendations on usage time. For the expiration date specified in the instructions, the delivery device will lock the medication upon expiration, meaning further administration is no longer permitted. For the expiration date recommended by the physician, the delivery device will provide an expiration reminder, but the patient may still choose to continue using the medication.

[0042] After the expiration reminder and expiration lock, the patient can submit an application to the hospital through the drug delivery device. If the patient is diagnosed by a doctor as needing to continue using the drug or to switch to another drug, the device can be cleaned and the drug replenished at the hospital, meaning the drug delivery device can continue to be reused.

[0043] When using other medications, the relevant parameters for the new medication need to be set again, and a new dosing program needs to be established.

[0044] Figure 2 This is a structural diagram of an oral drug delivery device based on privacy protection and image recognition according to the present invention. Figure 2 As shown, the oral drug delivery device based on privacy protection and image recognition in this embodiment may specifically include:

[0045] The drug delivery device is a handheld spray-on drug delivery device that meets the requirements for oral administration, including a drug storage chamber, an image acquisition and analysis device, a drug delivery probe, and an information interconnection device. The drug storage chamber is used for storing the drug solution. The image acquisition and analysis module can acquire images of the oral cavity from a specified distance and direction in front of the mouth, and includes the oral cavity positioning model and the disease analysis model, enabling image comparison and determination of the lesion's oral cavity partition and the drug dosage. The drug delivery probe delivers the prescribed amount of drug solution to the lesion's oral cavity partition, specifically including: 1) The drug delivery probe can rotate according to the judgment result of the oral cavity positioning model, and its top has a macro camera to acquire images and transmit them back to the image acquisition and analysis module for image comparison, until the top of the drug delivery probe faces the same direction as the lesion's oral cavity partition; 2) After the drug dosage is determined, the drug delivery probe draws the prescribed amount of drug solution and administers the drug; The information interconnection module is located on the body of the drug storage chamber and mainly has the following functions:

[0046] a. Used to establish the drug administration program, provide expiration reminders and expiration lockouts, thereby enabling patients to submit device reuse requests to the hospital;

[0047] b. Used to send recruitment pushes to patients, who then upload images of the oral lesions and their personal medical information, which are then used for model learning after privacy protection.

[0048] c. Used to record the time and number of times the oral lesion images are uploaded, and to send an appointment notification to the original attending physician when the patient needs to make an appointment for consultation.

[0049] Another embodiment of the present invention provides an apparatus for running federated learning in a drug delivery method based on privacy protection and image recognition, mainly comprising: an infrastructure module, a cloud network control CNC module, and a DApps application module.

[0050] The infrastructure module includes virtual hosts, virtual storage, virtual lines, and TEE trusted hardware, which are sourced from cloud development vendors, enterprise clouds, and personal computers.

[0051] The cloud network control CNC module is used to integrate and schedule the redundant computing power of all terminal nodes running on the infrastructure module through the underlying cloud network chain technology, providing hardware computing support for Dapps application modules.

[0052] The infrastructure module controls the CNC module via the cloud network to provide hardware computing support for the DApps application module.

[0053] DApps application modules, through the infrastructure modules including open cloud platforms, enterprise clouds, and personal computers, form hardware nodes that support computing. These hardware nodes are combined to create subnets. The subnets host software containers, where users upload interoperable computing units, including code, state, privacy computing frameworks, and algorithms, including but not limited to Federated Learning (FL) algorithms.

[0054] This invention integrates infrastructure modules, interconnection modules, cloud network control CNC modules, and DApps application modules to provide federated learning services to the Internet. The method and program of this invention are developed using DevOps integrated development and operation tools. The above descriptions are merely embodiments of this invention and do not limit the patent scope of this invention. Any equivalent structural or procedural transformations made using the content of this specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this invention. The program used to implement information control in this invention can be written in one or more programming languages ​​or a combination thereof to execute computer program code for performing the operations of this invention. The programming languages ​​include object-oriented programming languages—such as Java, Python, and C++—and conventional procedural programming languages—such as C or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can connect to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or it can connect to an external computer (e.g., via the Internet using an Internet service provider). In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and other division methods may exist in actual implementation. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or in the form of hardware plus software functional units. The integrated unit implemented as a software functional unit described above can be stored in a computer-readable storage medium. This software functional unit, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of the present invention.The aforementioned storage media include: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks or optical disks, and other media that can store program code.

Claims

1. A drug delivery method based on privacy protection and image recognition, characterized in that, The method includes: After a patient visits a doctor and purchases medication, their prescribed dosage, medication name, oral disease name, and personal medical information are entered into a central processing unit. Images of the patient's oral lesions are collected from multiple terminals, including the hospital, treatment facilities, and medication delivery devices. These images are clustered, and privacy is protected based on the number of cases. This privacy protection involves using random noise to ensure that the results of queries that request publicly visible information do not leak individual privacy information; that is, individual features are removed while preserving the statistical characteristics of the database to protect user privacy. A federated learning model is used to train an oral cavity localization model for lesion localization. A disease analysis model is trained for dosage determination. The initialization of the medication delivery system is then performed. The medication delivery device includes a central processing unit that inputs the prescribed dosage, drug name, oral disease name, prescribed dosage for different conditions, and personal medical information into the delivery device; simultaneously, it embeds the oral cavity positioning model and the disease analysis model into the delivery device; the patient uses the delivery device to administer medication, and images of the oral lesions are collected and uploaded; the delivery device establishes patient medication programs, provides expiration reminders, locks the device upon expiration, and allows for device reuse; the delivery device is a handheld spray delivery device that meets the requirements for oral medication administration, including a drug storage compartment, an image acquisition and analysis device, a delivery probe, and an information interconnection device.

2. The method according to claim 1, wherein, After a patient visits a doctor and purchases medication, their prescribed dosage, medication name, name of oral disease, and personal medical information are entered into a central processing unit, including: The oral disease names are included in the following oral disease categories: dental caries, oral mucosal diseases, and periodontal diseases; The dosage prescribed by the doctor refers to the dosage recommended by the doctor at the time of the consultation. If the doctor does not give a recommended dosage, the dosage prescribed in the drug instructions shall be used. The personal medical information refers to the patient's age, gender, medication prescription time, and other auxiliary information; The other auxiliary information refers to information on the patient's past medical history, causes of illness, and stages of illness collected from the patient's medical records through text analysis methods. The central processing unit is deployed in the information center of the hospital or treatment institution to collect, organize, store, and transmit patient medical records, including prescribed dosages, drug names, oral disease names, and personal medical information.

3. The method according to claim 1, wherein, The process involves collecting images of patients' oral lesions through multiple terminals, including hospitals, treatment facilities, and drug delivery devices, clustering these images, and protecting privacy based on the number of cases. The oral lesion images are collected by the hospital and treatment institution with the patient's consent, or by the drug delivery device; the oral lesion images are clustered based on the name of the oral disease, the individual's medical information, and other auxiliary information; After the clustering stabilizes, the number of cases in each cluster is counted and sorted from low to high. Privacy protection is implemented based on the sorting by adding noise to the individual medical information and other auxiliary information. The cluster with fewer cases is allocated more privacy budget to ensure that the cluster with fewer cases provides more usable information during model training. The number of cases refers to the number of oral lesion images contained in each cluster; The privacy protection technology employed is local differential privacy protection based on the Laplace mechanism, which mainly protects structured information, including: the personal medical information and other auxiliary information. The larger the privacy budget, the lower the level of information protection and the more information is available.

4. The method according to claim 1, wherein, The method of training an oral cavity localization model through federated learning for localizing oral lesions includes: The oral cavity localization model is used to identify the location of lesions inside the oral cavity. Specific training methods include: 1) Divide the internal structure of the oral cavity into zones, that is, match the location of various organs in the oral cavity to multiple oral zonal zones; 2) The pre-collected oral lesion images are labeled with the oral cavity regions and the names of the oral diseases, and then used to train a machine learning model; 3) The obtained oral cavity localization model determines whether the image contains oral disease lesions based on the newly acquired oral cavity image of the drug delivery device, and quickly determines the oral cavity region where the lesion is located; 4) For patients who visit the same doctor, the images of their oral lesions are collected and used for training in a unified manner; at the same time, in order to avoid diagnostic differences among different doctors, the models trained on multiple terminals belonging to different doctors are uploaded to the federated learning server, updated and returned to the parameters, and each terminal updates the model.

5. The method according to claim 1, wherein, The trained disease analysis model, used for dosage determination, includes: The disease analysis model is pre-trained based on image recognition technology, and the resulting model determines the clustering of the patient's oral diseases based on the newly acquired oral lesion images from the drug delivery device. Determine the prescribed dosage for different conditions corresponding to the cluster, and use it as the dosage for this administration. Stage of disease refers to the severity of oral disease, including the size of the lesions and the duration of the disease. The prescribed dosage for different conditions refers to the dosage of the drug determined through clinical trials based on the dosage prescribed in the doctor's order, for different degrees of severity of the disease and the patient's physical condition.

6. The method according to claim 1, wherein, The patient administers medication using the drug delivery device, and images of the oral lesions are collected and uploaded, including: The patient uses the drug delivery device to administer medication. The drug delivery device identifies the location of lesions in the oral cavity through the oral cavity positioning model and the disease analysis model, and administers medication based on the prescribed dosage for different conditions or the prescribed dosage for different conditions. Based on the number of cases after clustering, the central processing device also collects oral lesion images, that is, it sends collection pushes to the drug delivery devices of patients with fewer cases who match the clustering conditions, in order to expand the sample size; the patient uploads oral lesion images through the drug delivery device, which are then used for model learning after privacy protection, including: 1) The drug delivery device, with the patient's consent, collects the patient's oral lesion images and personal medical information; 2) The oral lesion images undergo image segmentation processing, and non-oral facial organs are Gaussian blurred to prevent leakage of patient privacy; 3) When the personal medical information is uploaded, the identity recognition information is removed; A reward mechanism is implemented for patients who voluntarily upload images after being recruited. That is, patients who have uploaded the oral lesion images after being recruited will be matched with their original attending physician through the drug delivery device and obtain one remote appointment consultation from the date of upload.

7. The method according to claim 1, wherein, The drug delivery device establishes patient drug delivery schedules, provides expiration reminders, locks the device upon expiration, and allows for device reuse, including: Since oral spray administration often faces the problem of drug expiration, the aforementioned drug delivery device, combined with blockchain smart contract functionality, can monitor drug delivery and enable device reuse. The blockchain pre-sets a smart contract, which records the drug name, the drug delivery time, and the drug expiration date on the date the drug delivery device is issued, and establishes the drug delivery program. The smart contract manages the drug delivery project for each drug delivery device by establishing a drug delivery project on the blockchain. It also calculates the time based on the drug's expiration date and the drug delivery device's distribution time. If the time after the drug delivery device is distributed exceeds the drug's expiration date, it sends an expiration reminder through the drug delivery device or locks the drug delivery device upon expiration. The drug's effective period includes the effective period specified in the instructions and the recommended usage duration prescribed by the doctor. For the effective period specified in the instructions, the drug delivery device will lock the drug upon expiration, meaning that it can no longer be administered. For the effective period recommended by the doctor, the drug delivery device will provide an expiration reminder, but the patient may still choose to continue using the drug. After the expiration reminder and the expiration lock, the patient submits an application to the hospital through the drug delivery device. If the patient is diagnosed by a doctor as needing to continue using the drug or to change to another drug, the device is cleaned and the drug is replenished at the hospital, meaning the drug delivery device can continue to be used repeatedly. When using other medications, the relevant parameters for the new medication need to be set again, and a new dosing program needs to be established.

8. A drug delivery device based on privacy protection and image recognition, performing the drug delivery method based on privacy protection and image recognition as described in claim 1, wherein, The drug delivery device is a handheld spray drug delivery device that meets the requirements for oral drug delivery, including a drug storage compartment, an image acquisition and analysis device, a drug delivery probe, and an information interconnection device, comprising: The medicine storage compartment is used for preserving the medicine solution; The image acquisition and analysis device acquires images of the inside of the oral cavity at a specified distance and direction in front of the oral cavity, and includes the oral cavity positioning model and the disease analysis model, and performs image comparison, lesion zoning in the oral cavity, and determination of the drug dosage; The drug delivery probe delivers the prescribed amount of medication to the oral cavity region of the lesion, specifically including: 1) The drug delivery probe rotates according to the oral cavity positioning model judgment result. The top has a macro camera to acquire images and transmit them back to the image acquisition and analysis module for image comparison until the top of the drug delivery probe is oriented in the same direction as the lesion oral cavity partition. 2) After the dosage is determined, the drug delivery probe draws up the prescribed amount of drug solution and administers the drug. The information interconnection device, located on the body of the medicine storage compartment bottle, mainly has the following functions: a. Used to establish the drug delivery program, provide the expiration reminder and expiration lock, thereby enabling the patient to submit a device reuse application to the hospital; b. Used to send recruitment pushes to patients, who then upload images of the oral lesions and their personal medical information, which are then used for model learning after privacy protection. c. Used to record the time and number of times the oral lesion images are uploaded, and to send an appointment notification to the original attending physician when the patient needs to make an appointment for consultation.

Images