An intelligent registration method and system for a smart patient service platform

Abstract

The present invention discloses an intelligent registration method and system for a smart patient service platform. Through a face change network, based on the current registered face image and medical records, the face changes under the symptom change state are detected to obtain multiple improved face images and multiple worsened face images. Based on the medical records, multiple improved face images correspond to multiple worsened face images, and identity matching is performed to obtain matching medical records. Through the face change network, the face image at the medical record detection time point predicted by the current registered face image is used to find the face images predicted at multiple medical record detection time points, and then match them with the faces in the database. If all matches are the same patient, it means that the patient corresponding to the current registered face image and the patient corresponding to the medical record patient face image stored in the database are successfully matched. In this way, the medical record file corresponding to the patient stored in the database can be obtained to register.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an intelligent registration method and system for a smart patient service platform. Background Art

[0002] Currently, hospitals are able to store patient numbers, identification information, and symptoms during testing. Repeated re-examinations can also reveal changes in symptoms. The challenge is how to use this information to generate medical records so that the patient's face can be accurately matched with the medical record during registration, thereby obtaining the patient's identity. Predicting changes in a patient's symptoms based on the patient's current face at registration and facial images recorded at multiple medical examination points is difficult. Furthermore, a user's medical record is crucial information, so setting permissions to ensure greater privacy is also a challenge. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent registration method and system for a smart patient service platform to solve the above-mentioned problems existing in the prior art.

[0004] In a first aspect, an embodiment of the present invention provides an intelligent registration method for a smart patient service platform, comprising:

[0005] Obtaining a patient's medical record and a facial image of the patient in the medical record; the patient's medical record includes a patient number, patient identification information, patient symptoms, and symptom change status; the symptom change status includes a symptom severity value and a corresponding medical record detection time point; the patient's facial image in the medical record represents an image of the face recorded at multiple medical record detection time points;

[0006] Establishing an association between the facial image of the patient in the medical record and the patient's medical record to generate a medical record file;

[0007] Obtaining a current registered face image; the current registered face image represents a face image of the patient entered when registering at the current time point; the current time point is later than the medical record detection time point in the medical record file;

[0008] Using a face change network, based on the current registered face image and medical records, facial changes under changing disease states are detected to obtain multiple improved face images and multiple worsening face images; one improved face image and one worsening face image are corresponding to one medical record detection time point;

[0009] Based on the medical records, the multiple improved facial images correspond to the multiple deteriorated facial images, performing identity matching to obtain matching medical records;

[0010] Smart registration is carried out based on matching medical records.

[0011] Optionally, the facial change network detects facial changes under changing symptom conditions based on the current registered facial image and medical records, and obtains multiple improved facial images and multiple deteriorated facial images, including:

[0012] The face change network includes a symptom improvement network and a symptom deterioration network; the symptom improvement network is used to predict the face of the current registered face image corresponding to the medical record detection time point when the symptom of the preset patient improves; the symptom deterioration network is used to predict the face of the current registered face image corresponding to the medical record detection time point when the symptom of the preset patient deteriorates;

[0013] The disease improvement network includes a first identity neural network, a first improved face feature extraction network, a first improved face reconstruction convolutional network and a second improved face reconstruction convolutional network;

[0014] The disease deterioration network includes a second identity neural network, a first deteriorated face feature extraction network, a first deteriorated reconstructed face convolutional network, and a second deteriorated reconstructed face convolutional network;

[0015] Using a symptom improvement network, based on the medical records and the currently registered facial image, the facial changes of the patient under the condition that the symptom is improved are detected to obtain a plurality of improved faces;

[0016] Through the disease deterioration network, based on the medical records and the current registered face image, the facial changes of the patient under the condition of deterioration of the disease are detected to obtain multiple deteriorated faces.

[0017] Optionally, the symptom improvement network detects facial changes under the condition that the patient's symptom is improved based on the medical records and the currently registered facial image, and obtains multiple improved faces, including:

[0018] Based on multiple medical record detection time points and the current time point, time segmentation is performed to obtain a set of patient detection time points;

[0019] Based on the patient detection time point set, the patient's symptoms and the current registered facial image, a plurality of first predicted facial images are obtained through a first improved face detection network;

[0020] Sort the plurality of first predicted face images according to time points from early to late to obtain a first predicted three-dimensional face image;

[0021] Based on the patient detection time point set, patient symptoms, symptom severity value and the first predicted three-dimensional facial image, multiple improved facial images are obtained through the second improved face detection network; one improved facial image corresponds to one medical record detection time point.

[0022] Optionally, the method of obtaining a plurality of first predicted facial images based on the patient detection time point set, the patient's symptoms, and the current registered facial image through a first improved face detection network includes:

[0023] Inputting the patient identity information and patient symptoms in the medical record file into a first identity neural network, extracting characteristics of the impact of the symptoms on patients with different patient identity information, and obtaining identity information characteristics;

[0024] The number of neurons in the output layer of the first identity neural network is n;

[0025] Inputting the current registered face image into a first improved face feature extraction network to extract features of the patient's face at the current time point to obtain a first current face feature map;

[0026] Superimposing the patient's symptoms and identity information features on the first current facial feature map to obtain a second current facial feature map;

[0027] Inputting the second current facial feature map into the first improved reconstructed facial convolutional network, predicting a facial image at a time point adjacent to and earlier than the current time point in the patient detection time point set, to obtain a first predicted facial image;

[0028] The s time points in the patient detection time point set correspond to s-1 first predicted face images.

[0029] Optionally, the step of obtaining a plurality of improved facial images by a second improved face detection network based on the patient detection time point set, the patient's symptoms, the symptom severity value, and the first predicted three-dimensional facial image includes:

[0030] Extracting two adjacent time points corresponding to values ​​of 1 from the patient detection time point set to obtain a first time point and a second time point; the first time point is earlier than the second time point;

[0031] Extracting the facial image between the first time point and the second time point and corresponding to the first time point from the first predicted three-dimensional facial image to obtain a first three-dimensional image to be tested;

[0032] superimposing the patient's symptoms and the symptom severity value with the first three-dimensional image to be measured to obtain a second three-dimensional image to be measured;

[0033] Obtain a second three-dimensional convolution kernel; the second three-dimensional convolution kernel is 2*2*z;

[0034] Here, z is smaller than the number of time points corresponding to the second three-dimensional image to be measured.

[0035] Convolving the second three-dimensional convolution kernel with a step size of 1 on the second three-dimensional image to be tested, to obtain a second face change feature map;

[0036] The second facial change feature map is input into a second improved reconstructed face convolutional network, and the face of the patient at the second time point is predicted based on the change relationship of the patient's face when the condition improves, to obtain an improved face image.

[0037] Optionally, multiple improved facial images are obtained corresponding to multiple symptom detection time points.

[0038] Optionally, the training process of the disease improvement network includes:

[0039] Obtaining the training patient's identity information, training patient's symptoms, training symptom severity value and corresponding training medical record detection time point;

[0040] Acquire multiple historical time points and corresponding multiple historical patient facial images; the historical time points represent time points at which a patient's symptoms improved and recorded their facial features at fixed time intervals; the historical patient facial images represent facial images of a patient recorded at a historical time point;

[0041] Randomly mark s time points from multiple historical time points as annotated medical record detection time points;

[0042] The latest historical time point among multiple historical time points is used as the current training time point; the historical face image corresponding to the training time point is used as the current registration face image for training;

[0043] Using multiple historical time points earlier than the training time point as a first annotated time point set; using historical facial images corresponding to the historical time points in the first annotated time point set as annotated patient facial images;

[0044] By training historical time points, training patient identity information, training patient symptoms, and training current registered face images, the loss is calculated by annotating patient face images, and the first identity neural network, the first improved face detection network, and the first improved reconstruction face convolutional network are trained;

[0045] Through the trained first identity neural network, the first improved face detection network and the first improved reconstructed face convolutional network, according to the marked medical record detection time point, the training disease severity value, the training patient disease, the loss is calculated by marking the patient's face image, and the second three-dimensional convolution kernel and the second improved reconstructed face convolutional network are trained.

[0046] Optionally, the time segmentation is performed based on multiple medical record detection time points and the current time point to obtain a patient detection time point set, including:

[0047] Get fixed split length;

[0048] Find a medical record detection time point earlier than other medical record detection time points among the multiple medical record detection time points as the front-end medical record detection time point;

[0049] Divide the time difference between the current time point and the front-end medical record detection time point by the fixed segmentation length to obtain the number of segments;

[0050] Construct a set with the number of elements equal to the number of partitions and all element values ​​0 as the first set;

[0051] Construct a key-value pair from two time points separated by a fixed length in the time period between the current time point and the front-end medical record detection time point, and obtain multiple time key-value pairs;

[0052] Input the time key-value pairs into the first set in order from early to late to obtain the first time point set;

[0053] If there is a medical record detection time point in the time period of the time key-value pair corresponding to the first time point set, the value of the corresponding subscript in the first set is set to 1 to obtain the patient detection time point set.

[0054] Optionally, performing identity matching based on the medical record files, the plurality of improved facial images, and the plurality of deteriorated facial images to obtain matching medical records includes:

[0055] Obtaining a currently authorized facial image; the currently authorized facial image represents a facial image of a person with authorization to obtain the patient's medical records;

[0056] Matching the current authorized person's face image with the annotated authorized person's face image in the database to obtain a second matching value;

[0057] If the second matching value is greater than the matching threshold, the database is opened to view the patient's facial image in the medical record;

[0058] Matching the improved face image with all face images stored in the database to obtain a better matching user; multiple improved face images correspond to multiple better matching users;

[0059] Match the degraded face image with all face images stored in the database to obtain a degraded matching user; multiple degraded face images correspond to multiple degraded matching users;

[0060] If multiple improved matching users are the same or multiple deteriorated matching users are the same, the matching user is obtained;

[0061] The patient medical records corresponding to the matching user are used as matching medical records.

[0062] In a second aspect, an embodiment of the present invention provides an intelligent registration system for a smart patient service platform, including:

[0063] An acquisition module is configured to acquire a patient's medical record and a facial image of the patient in the medical record; the patient's medical record includes a patient number, patient identification information, patient symptoms, and symptom change status; the symptom change status includes a symptom severity value and a corresponding medical record detection time point; the facial image of the patient in the medical record represents an image of the face recorded at multiple medical record detection time points; and obtain a current registration facial image; the current registration facial image represents the facial image of the patient recorded at the current time point when registering; the current time point is later than the medical record detection time point in the medical record file;

[0064] A medical record file generation module is used to establish an association relationship between the facial image of the patient in the medical record and the patient's medical record to generate a medical record file;

[0065] The module for detecting improved and worsened faces is configured to detect facial changes under changing symptom conditions based on the current registered facial image and medical records through a facial change network, thereby obtaining multiple improved and worsened facial images; one improved and one worsened facial image are associated with each medical record detection time point;

[0066] A medical record matching module, configured to perform identity matching based on the medical record files, the plurality of improved facial images, and the plurality of deteriorated facial images to obtain matching medical records;

[0067] The registration module is used for intelligent registration based on matching medical records.

[0068] Compared with the prior art, the embodiments of the present invention achieve the following beneficial effects:

[0069] In this application, a patient's number, identity information, and symptoms are associated with the patient's facial image and the time of each reexamination in the medical record, automatically generating a medical record. The medical records are stored in a database. When a patient registers, they can simply scan their face and retrieve the patient's information without presenting their ID or medical card. This allows for quick registration or access to the patient's medical records, assisting doctors in diagnosing the patient based on their medical history. Accessing a patient's medical records requires authorized personnel. After authorization is granted, the database stores the patient's previous facial images. However, as symptoms worsen or improve with treatment, the facial image at the current time differs from the facial image in the medical record. Therefore, a facial change network is used to predict the facial image at the time of each medical record examination using the current registration facial image. Multiple facial images predicted at each medical record examination time are then matched against faces in the database. If all matches are identical, the patient corresponding to the current registration facial image successfully matches the patient corresponding to the facial image in the medical record stored in the database. This allows access to the medical records corresponding to the patient stored in the database. Registration is done based on the patient number, patient identity information, patient symptoms, and symptom status in the medical record, achieving the technical effect of increasing the accuracy and privacy of smart registration. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a flow chart of an intelligent registration method for a smart patient service platform provided by an embodiment of the present invention.

[0071] Figure 2 This is a structural diagram of a symptom improvement network in an intelligent registration method of an intelligent patient service platform provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The present invention will be described in detail below with reference to the accompanying drawings.

[0073] Example 1

[0074] like Figure 1 As shown, an embodiment of the present invention provides an intelligent registration method for a smart patient service platform, the method comprising:

[0075] S101: Obtain the patient's medical record and the patient's facial image in the medical record; the patient's medical record includes the patient number, age, patient's symptoms, and symptom change status; the symptom change status includes the symptom severity value and the corresponding detection time point; the patient's facial image in the medical record represents the face entered when recording the patient's medical record.

[0076] In this embodiment, the disease is assigned a value from 0 to 1 based on the degree to which it can be detected to cause destructive damage to the body. For example, the early stage of the disease is assigned a value of 0.25, the middle stage is assigned a value of 0.5, and the late stage is assigned a value of 0.75.

[0077] The symptom severity value represents the severity of the symptom. For example, if a user is diagnosed with early-stage gastric cancer on January 1, 2024, then the patient's symptom is gastric cancer, the symptom severity value is 0.25, and the detection time is January 1, 2024.

[0078] In this embodiment, the patient identity information includes gender, age, height and weight.

[0079] S102: Establish an association between the facial image of the patient in the medical record and the patient's medical record to generate a medical record file.

[0080] In this embodiment, the patient's facial image is used as the key in a key-value pair, and the patient's medical history is used as the value in the key-value pair. The patient's facial image is associated with the patient's medical history, and the patient's facial image, medical history, and associated relationship constitute a medical record file and are stored in a database.

[0081] Among them, one patient number corresponds to one medical record file.

[0082] S103: Obtain a current registered face image; the current registered face image represents the face image of the patient entered when registering at the current time point; the current time point is later than the medical record detection time point in the medical record file.

[0083] S104: Using a face change network, based on the current registered face image and medical records, detect face changes under symptom change conditions, and obtain multiple improved face images and multiple worsened face images; one medical record detection time point corresponds to one improved face image and one worsened face image.

[0084] S105: Based on the medical record file, the multiple improved facial images and the multiple deteriorated facial images, identity matching is performed to obtain matching medical records.

[0085] S106: Perform intelligent registration based on matching medical records.

[0086] Among them, the corresponding data is entered when registering according to the patient number, patient identity information, patient symptoms, and symptom change status in the medical record file.

[0087] Optionally, the facial change network detects facial changes under changing symptom conditions based on the current registered facial image and medical records, and obtains multiple improved facial images and multiple deteriorated facial images, including:

[0088] The face change network includes a symptom improvement network and a symptom deterioration network; the symptom improvement network is used to predict the face of the current registered face image corresponding to the medical record detection time point when the symptom of the preset patient improves; the symptom deterioration network is used to predict the face of the current registered face image corresponding to the medical record detection time point when the symptom of the preset patient deteriorates.

[0089] The disease improvement network includes a first identity neural network, a first improved face feature extraction network, a first improved face reconstruction convolutional network and a second improved face reconstruction convolutional network.

[0090] The structural diagram of the disease improvement network is as follows: Figure 2 shown.

[0091] The disease deterioration network includes a second identity neural network, a first deteriorated face feature extraction network, a first deteriorated reconstructed face convolutional network and a second deteriorated reconstructed face convolutional network.

[0092] The structure of the symptom worsening network is the same as that of the symptom improvement network. The structure of the second identity neural network is the same as that of the first identity neural network, but the parameters after training are different. The structure of the first deteriorating facial feature extraction network is the same as that of the first improving facial feature extraction network, but the parameters after training are different. The structure of the first improving reconstructed facial convolutional network is the same as that of the first deteriorating reconstructed facial convolutional network, but the parameters after training are different. The structure of the second improving reconstructed facial convolutional network is the same as that of the second deteriorating reconstructed facial convolutional network, but the parameters after training are different.

[0093] Through the symptom improvement network, based on the medical records and the current registered face images, the facial changes of the patient under the condition that the patient's symptoms are improved are detected, and multiple improved faces are obtained.

[0094] The improvement of the symptoms is determined by the severity of the symptoms. The earlier the medical record is detected, the greater the severity of the symptoms, indicating that the symptoms have improved.

[0095] The improved face refers to a face image predicted from the current registered face image at a medical record detection time point earlier than the current time point, under the condition that the patient's condition has improved.

[0096] Through the disease deterioration network, based on the medical records and the current registered face image, the facial changes of the patient under the condition of deterioration of the disease are detected to obtain multiple deteriorated faces.

[0097] The deteriorated face refers to a face image predicted from the current registered face image at a medical record detection time point earlier than the current time point under the condition that the patient's condition is deteriorating.

[0098] The training method used for the symptom worsening network is the same as that used for the symptom improvement network. The training data used for the symptom worsening network are facial images of patients whose symptoms worsened, taken at multiple time points.

[0099] The worsening of the disease is determined by the disease severity value. The earlier the medical record detection time point, the smaller the corresponding disease severity value, indicating that the disease has worsened.

[0100] Optionally, the symptom improvement network detects facial changes under the condition that the patient's symptom is improved based on the medical records and the currently registered facial image, and obtains multiple improved faces, including:

[0101] Based on multiple medical record detection time points and the current time point, time segmentation is performed to obtain a set of patient detection time points;

[0102] Based on the patient detection time point set, the patient's symptoms and the current registered facial image, a plurality of first predicted facial images are obtained through a first improved face detection network;

[0103] Sort the plurality of first predicted face images according to time points from early to late to obtain a first predicted three-dimensional face image;

[0104] Based on the patient detection time point set, patient symptoms, symptom severity value and the first predicted three-dimensional facial image, multiple improved facial images are obtained through the second improved face detection network; one improved facial image corresponds to one medical record detection time point.

[0105] Optionally, the method of obtaining a plurality of first predicted facial images based on the patient detection time point set, the patient's symptoms, and the current registered facial image through a first improved face detection network includes:

[0106] The patient identity information and patient symptoms in the medical record file are input into the first identity neural network, and the characteristics of the impact of the symptoms on patients with different patient identity information are extracted to obtain identity information characteristics.

[0107] Wherein, the first identity neural network is a fully connected neural network (Fully Connected Neural Network).

[0108] The identity information feature represents the impact of the disease on facial changes at multiple disease detection time points based on the patient's own physical condition.

[0109] The number of neurons in the output layer of the first identity neural network is n.

[0110] The current registered face image is input into a first improved face feature extraction network to extract the features of the patient's face at the current time point to obtain a first current face feature map.

[0111] Among them, the first improved facial feature extraction network is a convolutional neural network (CNN).

[0112] Among them, the number of channels of the convolution kernel of the last layer of the first improved facial feature extraction network is m.

[0113] Among them, the first current facial feature map represents the facial features of the current registered facial image.

[0114] The patient's symptoms and identity information features are superimposed on the first current facial feature map to obtain a second current facial feature map.

[0115] The number of rows in the second current facial feature map is equal to the number of rows in the first current facial feature map; the number of columns in the second current facial feature map is equal to the number of columns in the first current facial feature map; the number of channels in the second current facial feature map is equal to 1+n+m; the channel corresponding to 1 represents the patient's condition, the channel corresponding to n represents the identity information feature, and the channel corresponding to m represents the first current facial feature map. The values ​​of any row and column corresponding to the channels corresponding to the patient's condition and the identity information feature in the second current facial feature map are all the same.

[0116] The second current facial feature map includes features of the patient's condition, patient identity information, and the current registered facial image. The impact of the patient's condition on facial changes is extracted from the identity information features. The patient's condition is also separately incorporated into the construction of the second current facial feature map, increasing the importance of the patient's condition's correlation with facial changes.

[0117] The second current facial feature map is input into the first improved reconstructed facial convolutional network to predict the facial image of a time point adjacent to and earlier than the current time point in the patient detection time point set to obtain a first predicted facial image.

[0118] The first improved facial reconstruction convolutional network includes multiple 4*4*(n+1+m) three-dimensional convolution kernels. Each time, the first improved facial reconstruction convolutional network adds three rows above and below the boundary of the feature map and three columns to the left and right, using the corresponding three-dimensional convolution kernels for convolution with a stride of 1.

[0119] Wherein, n and m are natural numbers greater than 0.

[0120] The number of rows of the first predicted face image is equal to the number of rows of the current registered face image, the number of columns of the first predicted face image is equal to the number of columns of the current registered face image, and the number of channels of the first predicted face image is 3, indicating three RGB channels.

[0121] The s time points in the patient detection time point set correspond to s-1 first predicted face images.

[0122] Wherein, s is a natural number greater than 0.

[0123] The method includes performing facial reconstruction based on the acquired first predicted facial image through a first improved facial feature extraction network and a first improved reconstructed facial convolutional network to obtain a new first predicted facial image; wherein the time point corresponding to the new first predicted facial image is adjacent to the time point corresponding to the first predicted facial image in the set of patient detection time points; and the time point corresponding to the new first predicted facial image is earlier than the time point corresponding to the first predicted facial image in the set of patient detection time points.

[0124] Optionally, the step of obtaining a plurality of improved facial images by a second improved face detection network based on the patient detection time point set, the patient's symptoms, the symptom severity value, and the first predicted three-dimensional facial image includes:

[0125] Two adjacent time points corresponding to values ​​of 1 in the patient detection time point set are extracted to obtain a first time point and a second time point; the first time point is earlier than the second time point.

[0126] The first time point refers to a medical record examination time point stored in the medical record file, and the second time point refers to a medical record examination time point stored in the medical record file for a reexamination after the first time point.

[0127] The facial image between the first time point and the second time point and corresponding to the first time point is extracted from the first predicted three-dimensional facial image to obtain a first three-dimensional image to be tested.

[0128] The first three-dimensional image to be tested does not include a facial image in the first predicted three-dimensional facial image corresponding to the second time point. The facial image corresponding to the second time point needs to be predicted.

[0129] The patient's symptoms and the symptom severity value are superimposed on the first three-dimensional image to be measured to obtain a second three-dimensional image to be measured.

[0130] The number of channels of the second three-dimensional image to be tested is equal to 1+1+the number of channels of the first three-dimensional image to be tested, one "1" represents the patient's symptom, and one "1" represents the severity of the symptom.

[0131] Obtain a second three-dimensional convolution kernel; the second three-dimensional convolution kernel is 2*2*z.

[0132] Here, z is smaller than the number of time points included in the first three-dimensional image to be measured.

[0133] Wherein, z is the minimum value of two adjacent symptom detection time points minus 1, and z is a natural number greater than 0. In this embodiment, z is 3.

[0134] Convolving the second three-dimensional convolution kernel with a step size of 1 on the second three-dimensional face image to be tested, to obtain a second face change feature map;

[0135] The second facial change feature map is input into a second improved reconstructed face convolutional network, and the face of the patient at the second time point is predicted based on the change relationship of the patient's face when the condition improves, to obtain an improved face image.

[0136] The second improved face reconstruction convolutional network includes multiple 4*4 two-dimensional convolution kernels. The second improved face reconstruction convolutional network adds 3 rows above and below the boundary of the feature map and 3 columns to the left and right each time, and uses the corresponding 2D convolution kernel with a stride of 1 for convolution.

[0137] Optionally, multiple improved facial images are obtained corresponding to multiple symptom detection time points.

[0138] Optionally, the training process of the disease improvement network includes:

[0139] Obtain the training patient's identity information, training patient's symptoms, training symptom severity value and corresponding training medical record detection time point.

[0140] Among them, one patient corresponds to one training patient identity information, one training patient symptom, one training symptom severity value, and one training medical record detection time point.

[0141] Acquire multiple historical time points and corresponding multiple historical patient facial images; the historical time points represent time points at which a patient's symptoms improved and recorded their facial features at fixed time intervals; the historical patient facial images represent facial images of a patient recorded at a historical time point;

[0142] Randomly mark s time points from multiple historical time points as annotated medical record detection time points;

[0143] The latest historical time point among multiple historical time points is used as the current training time point; the historical face image corresponding to the training time point is used as the current registration face image for training;

[0144] Using multiple historical time points earlier than the training time point as a first annotated time point set; using historical facial images corresponding to the historical time points in the first annotated time point set as annotated patient facial images;

[0145] By training time points, training patient identity information, training patient symptoms and labeling patient face images, the first identity neural network, the first improved face detection network and the first improved face reconstruction convolutional network are trained;

[0146] Through the trained first identity neural network, first improved face detection network and first improved reconstruction face convolutional network, the first identity neural network, first improved face detection network and first improved reconstruction face convolutional network are trained according to the training time point, training patient identity information, training patient symptoms and labeled patient face images.

[0147] Optionally, the time segmentation is performed based on multiple medical record detection time points and the current time point to obtain a patient detection time point set, including:

[0148] Gets the fixed split length.

[0149] In this embodiment, the fixed segmentation length is 3 days.

[0150] Find a medical record detection time point among the multiple medical record detection time points that is earlier than other medical record detection time points as the front-end medical record detection time point.

[0151] For example, if a test is conducted on May 1, 2024, the current time point is 20240501. For example, if the time points for multiple medical record tests are 20240501, 20240601, 20240613, and 20240618, the front-end medical record test time point is 20240501.

[0152] The time difference between the current time point and the medical record detection time point is divided by the fixed segmentation length to obtain the number of segments.

[0153] The number of divisions is rounded down when dividing. The specific calculation method is to subtract the month and day separately. For example, 06-05=01. Based on the month of May in 2024, the corresponding day is 31. 18-01=17, 17+31=48. 48 / 3=16. Therefore, the number of divisions is 16.

[0154] Construct a set whose number of elements is the number of partitions and whose element values ​​are all 0 as the first set.

[0155] Among them, the first set is {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0}.

[0156] Construct key-value pairs for two time points separated by a fixed segmentation length within the time period between the current time point and the front-end medical record detection time point, obtaining multiple time key-value pairs.

[0157] Among them, in this embodiment, in 20240501 - 20240618, <2024050>1, 20240503>, <2024050>4, 20240505>…<2024061>5, 20240618>.

[0158] Input the time key-value pairs into the first set in order from earliest to latest, obtaining the first time point set. <000032⑤>

[0159] Among them, the first time point set is {<2024050>1, 20240503>, <2024050>4, 20240505>…<2024061>5, 20240618>}.

[0160] If there is a medical record detection time point in the time period of the time key-value pair corresponding to the first time point set, set the value of the corresponding subscript in the first set to 1, obtaining the patient detection time point set.

[0161] Among them, {1,0,0,0,0,0,0,0,0,0,1,②,0,0,1,1}.

[0162] Optionally, the identity matching based on the medical record file, multiple improved face images corresponding to multiple deteriorated face images to obtain a matching medical record includes:

[0163] Obtain the current authorized face image; the current authorized face image represents the face image with permission to预想获取患者的病历档案 (preferably: obtain the medical record file of the patient).

[0164] Among them, in this embodiment, the current authorized face image is the face image of a doctor.

[0165] Match the current authorized face image with the labeled authorized face images in the database to obtain a second matching value.

[0166] Among them, the Euclidean distance is used for matching.

[0167] If the second matching value is greater than the matching threshold, open the permission to view the face image of the patient with the medical record in the database.

[0168] Among them, the matching threshold is 0.9. When the face image of the doctor is not matched, the face image of the patient with the medical record in the database is encrypted and cannot be directly viewed.

[0169] It should be noted that there seems to be an unclear expression "预想获取患者的病历档案" in the original text, and I have made a tentative translation "obtain the medical record file of the patient". If there are more specific requirements or corrections for this part, it can be adjusted accordingly.The improved face image is matched with all face images stored in the database to obtain an improved matching user; multiple improved face images correspond to multiple improved matching users.

[0170] Among them, all facial images stored in the database include multiple facial images of patients in medical records stored at multiple medical record detection time points for multiple patients.

[0171] Match the degraded face image with all face images stored in the database to obtain a degraded matching user; multiple degraded face images correspond to multiple degraded matching users;

[0172] If multiple improved matching users are the same or multiple deteriorated matching users are the same, a matching user is obtained.

[0173] The matching users are associated using patient numbers.

[0174] The patient medical records corresponding to the matching user are used as matching medical records.

[0175] Example 2

[0176] Based on the above-mentioned intelligent registration method of a smart patient service platform, an embodiment of the present invention further provides an intelligent registration system of a smart patient service platform, the system comprising:

[0177] An acquisition module is configured to acquire a patient's medical record and a facial image of the patient in the medical record; the patient's medical record includes a patient number, patient identification information, patient symptoms, and symptom change status; the symptom change status includes a symptom severity value and a corresponding medical record detection time point; the facial image of the patient in the medical record represents an image of the face recorded at multiple medical record detection time points; and obtain a current registration facial image; the current registration facial image represents the facial image of the patient recorded at the current time point when registering; the current time point is later than the medical record detection time point in the medical record file;

[0178] A medical record file generation module is used to establish an association relationship between the facial image of the patient in the medical record and the patient's medical record to generate a medical record file;

[0179] The module for detecting improved and worsened faces is configured to detect facial changes under changing symptom conditions based on the current registered facial image and medical records through a facial change network, thereby obtaining multiple improved and worsened facial images; one improved and one worsened facial image are associated with each medical record detection time point;

[0180] The medical record matching module is used to perform identity matching based on the medical record files, multiple improved facial images corresponding to multiple deteriorated facial images, and obtain matching medical records.

[0181] The registration module is used for intelligent registration based on matching medical records.

Claims

1. An intelligent registration method for a smart patient service platform, characterized in that: include: Obtaining patient medical records and facial images of patients in the medical records; the patient medical records include patient number, patient identity information, patient symptoms, and symptom change status; The symptom change state includes the symptom degree value and the corresponding medical record detection time point; the medical record patient face image represents the image of the face recorded at multiple medical record detection time points; Establishing an association between the facial image of the patient in the medical record and the patient's medical record to generate a medical record file; Obtaining a current registered face image; the current registered face image represents a face image of the patient entered when registering at the current time point; the current time point is later than the medical record detection time point in the medical record file; Through the face change network, based on the current registered face image and medical record files, the face changes under the condition change state are detected to obtain multiple improved face images and multiple deteriorated face images; one medical record detection time point corresponds to one improved face image and one deteriorated face image; Based on the medical record file, the plurality of improved face images and the plurality of deteriorated face images, identity matching is performed to obtain matching medical records; Make smart registration based on matching medical records; The face change network detects face changes under the condition change state based on the current registered face image and medical records, and obtains multiple improved face images and multiple deteriorated face images, including: The face change network includes a symptom improvement network and a symptom deterioration network; the symptom improvement network is used to predict the face of the current registered face image corresponding to the medical record detection time point when the symptom of the preset patient improves; the symptom deterioration network is used to predict the face of the current registered face image corresponding to the medical record detection time point when the symptom of the preset patient deteriorates; The disease improvement network includes a first identity neural network, a first improved face feature extraction network, a first improved face reconstruction convolutional network and a second improved face reconstruction convolutional network; The disease deterioration network includes a second identity neural network, a first deteriorated face feature extraction network, a first deteriorated reconstructed face convolution network, and a second deteriorated reconstructed face convolution network; Through the symptom improvement network, based on the medical records and the current registered face images, the facial changes of the patient under the condition that the symptom is improved are detected to obtain a plurality of improved face images; Using a disease deterioration network, based on the medical records and the current registered facial image, detecting facial changes of the patient under the condition that the disease is deteriorating, and obtaining a plurality of deteriorated facial images; The performing of identity matching based on the medical record file, the plurality of improved face images and the plurality of deteriorated face images to obtain matching medical records includes: Obtaining a currently authorized facial image; the currently authorized facial image indicates a facial image of a person with authority for obtaining the patient's medical records; Matching the current authorized person's face image with the annotated authorized person's face image in the database to obtain a second matching value; If the second matching value is greater than the matching threshold, the database is opened to the user for viewing the facial image of the patient in the medical record; Matching the improved face image with all face images stored in the database to obtain an improved matching user; multiple improved face images correspond to multiple improved matching users; Matching the degraded face image with all face images stored in the database to obtain a degraded matching user; multiple degraded face images correspond to multiple degraded matching users; If multiple improved matching users are the same or multiple deteriorated matching users are the same, a matching user is obtained; The patient medical records corresponding to the matching users are used as matching medical records.

2. The intelligent registration method of the intelligent patient service platform according to claim 1, characterized in that: The symptom improvement network detects facial changes under the condition that the patient's symptom is improved based on the medical record and the current registered facial image, and obtains a plurality of improved facial images, including: Based on multiple medical record detection time points and the current time point, time segmentation is performed to obtain a set of patient detection time points; Based on the patient detection time point set, the patient's symptoms and the current registered face image, a plurality of first predicted face images are obtained through a first improved face detection network; Sort the plurality of first predicted face images according to time points from early to late to obtain a first predicted three-dimensional face image; Based on the patient detection time point set, patient symptoms, symptom severity value and the first predicted three-dimensional face image, multiple improved face images are obtained through a second improved face detection network; one improved face image corresponds to one medical record detection time point.

3. The intelligent registration method of the intelligent patient service platform according to claim 2, characterized in that: The method of obtaining a plurality of first predicted face images based on the patient detection time point set, the patient's symptoms and the current registered face image through a first improved face detection network includes: Inputting the patient identity information and patient symptoms in the medical record file into a first identity neural network, extracting characteristics of the effects of the symptoms on patients with different patient identity information, and obtaining identity information characteristics; The number of neurons in the output layer of the first identity neural network is n; Inputting the current registered face image into a first improved face feature extraction network, extracting the features of the patient's face at the current time point, and obtaining a first current face feature map; Superimposing the patient's symptoms and identity information features on the first current face feature map to obtain a second current face feature map; Inputting the second current face feature map into the first improved reconstructed face convolutional network, predicting a face image at a time point adjacent to the current time point and earlier than the current time point in the patient detection time point set, and obtaining a first predicted face image; The s time points in the patient detection time point set correspond to s-1 first predicted face images.

4. The intelligent registration method of the intelligent patient service platform according to claim 2, characterized in that: The method of obtaining a plurality of improved face images based on the patient detection time point set, the patient's symptoms, the symptom severity value and the first predicted three-dimensional face image through a second improved face detection network includes: Extracting two adjacent time points corresponding to values ​​of 1 in the patient detection time point set to obtain a first time point and a second time point; the first time point is earlier than the second time point; Extracting the facial image between the first time point and the second time point and corresponding to the first time point in the first predicted three-dimensional facial image to obtain a first three-dimensional image to be tested; Superimposing the patient's symptoms, the symptom severity value and the first three-dimensional image to be measured to obtain a second three-dimensional image to be measured; Obtain a second three-dimensional convolution kernel; the second three-dimensional convolution kernel is 2*2*z; Wherein, z is less than the number of time points corresponding to the second three-dimensional image to be measured; Convolving the second three-dimensional convolution kernel with a step size of 1 on the second three-dimensional image to be tested in sequence to obtain a second face change feature map; The second facial change feature map is input into the second improved reconstructed facial convolutional network, and the face of the patient at the second time point is predicted based on the change relationship of the patient's face when the symptom improves, so as to obtain an improved facial image.

5. The intelligent registration method of the intelligent patient service platform according to claim 4, characterized in that: Multiple improved face images corresponding to multiple disease detection time points are obtained.

6. The intelligent registration method of the intelligent patient service platform according to claim 2, characterized in that: The training process of the disease improvement network includes: Obtaining training patient identity information, training patient symptoms, training symptom severity values, and corresponding training medical record detection time points; Acquire multiple historical time points and corresponding multiple historical patient face images; the historical time points represent time points at which the face of a patient whose symptoms have improved is recorded at fixed time intervals; the historical patient face images represent images of the face of a patient recorded at a historical time point; Randomly mark s time points among multiple historical time points as annotated medical record detection time points; The latest historical time point among the multiple historical time points is used as the current training time point; the historical face image corresponding to the training time point is used as the current registered face image for training; Using multiple historical time points earlier than the training time point as a first annotated time point set; using historical facial images corresponding to the historical time points in the first annotated time point set as annotated patient facial images; By training historical time points, training patient identity information, training patient symptoms, and training current registered face images, the loss is obtained by annotating patient face images, and the first identity neural network, the first improved face detection network, and the first improved reconstruction face convolutional network are trained; Through the trained first identity neural network, the first improved face detection network and the first improved face reconstruction convolutional network, according to the marked medical record detection time point, the training disease severity value, the training patient disease, the loss is calculated by marking the patient's face image, and the second three-dimensional convolution kernel and the second improved face reconstruction convolutional network are trained.

7. The intelligent registration method of the intelligent patient service platform according to claim 2, characterized in that: The method of performing time segmentation based on multiple medical record detection time points and the current time point to obtain a patient detection time point set includes: Get the fixed segmentation length; Find a medical record detection time point among the multiple medical record detection time points that is earlier than other medical record detection time points as the front-end medical record detection time point; Divide the time difference between the current time point and the front-end medical record detection time point by the fixed segmentation length to obtain the number of segments; Construct a set whose number of elements is the number of partitions and whose element values ​​are all 0 as the first set; Construct a key-value pair from two time points separated by a fixed split length in the time period between the current time point and the front-end medical record detection time point, and obtain multiple time key-value pairs; Input the time key-value pairs from early to late into the first set to obtain a first time point set; If there is a medical record detection time point in the time period of the time key-value pair corresponding to the first time point set, the value of the corresponding subscript in the first set is set to 1 to obtain a patient detection time point set.

8. An intelligent registration system for a smart patient service platform, characterized in that: include: An acquisition module is used to acquire a patient's medical record and a facial image of the patient in the medical record; the patient's medical record includes a patient number, patient identity information, patient symptoms, and symptom change status; the symptom change status includes a symptom severity value and a corresponding medical record detection time point; the medical record patient's facial image represents an image of a face entered at multiple medical record detection time points; obtain a current registration facial image; the current registration facial image represents a facial image of the patient entered when registering at the current time point; the current time point is later than the medical record detection time point in the medical record file; A medical record file generation module is used to establish an association relationship between the face image of the patient in the medical record and the patient's medical record to generate a medical record file; The improved face and deteriorated face detection module is used to detect the face changes under the symptom change state based on the current registered face image and the medical record through the face change network, and obtain multiple improved face images and multiple deteriorated face images; one medical record detection time point corresponds to one improved face image and one deteriorated face image; A medical record matching module, used for performing identity matching based on the medical record file, a plurality of improved face images and a plurality of deteriorated face images to obtain a matching medical record; The registration module is used to perform intelligent registration based on matching medical records; The face change network detects face changes under the condition change state based on the current registered face image and medical records, and obtains multiple improved face images and multiple deteriorated face images, including: The face change network includes a symptom improvement network and a symptom deterioration network; the symptom improvement network is used to predict the face of the current registered face image corresponding to the medical record detection time point when the symptom of the preset patient improves; the symptom deterioration network is used to predict the face of the current registered face image corresponding to the medical record detection time point when the symptom of the preset patient deteriorates; The disease improvement network includes a first identity neural network, a first improved face feature extraction network, a first improved face reconstruction convolutional network and a second improved face reconstruction convolutional network; The disease deterioration network includes a second identity neural network, a first deteriorated face feature extraction network, a first deteriorated reconstructed face convolution network, and a second deteriorated reconstructed face convolution network; Through the symptom improvement network, based on the medical records and the current registered face images, the facial changes of the patient under the condition that the symptom is improved are detected to obtain a plurality of improved face images; Using a disease deterioration network, based on the medical records and the current registered facial image, detecting facial changes of the patient under the condition that the disease is deteriorating, and obtaining a plurality of deteriorated facial images; The performing of identity matching based on the medical record file, the plurality of improved face images and the plurality of deteriorated face images to obtain matching medical records includes: Obtaining a currently authorized facial image; the currently authorized facial image indicates a facial image of a person with authority for obtaining the patient's medical records; Matching the current authorized person's face image with the annotated authorized person's face image in the database to obtain a second matching value; If the second matching value is greater than the matching threshold, the database is opened to the user for viewing the facial image of the patient in the medical record; Matching the improved face image with all face images stored in the database to obtain an improved matching user; multiple improved face images correspond to multiple improved matching users; Matching the degraded face image with all face images stored in the database to obtain a degraded matching user; multiple degraded face images correspond to multiple degraded matching users; If multiple improved matching users are the same or multiple deteriorated matching users are the same, a matching user is obtained; The patient medical records corresponding to the matching users are used as matching medical records.

Images