TWI679653B - Distributed monitoring system and method - Google Patents

Abstract

A decentralized monitoring system and method. Decentralized monitoring systems include: cloud servers, client systems, medical systems, and third-party verification agencies. The cloud server receives multiple pieces of medical data after calculation and encryption from the medical system, and receives monitoring data from the client system. After the cloud server receives the monitoring data from the client system, it sends the identification key corresponding to the monitoring data to the client system. The third-party verification organization downloads monitoring data and medical data from the cloud server, and determines whether the monitoring data is abnormal based on the medical data to generate a judgment result, and then transmits the judgment result to the cloud server. After receiving the identification key from the client system, the cloud server sends the judgment result corresponding to the identification key to the client system.

Description

Decentralized monitoring system and method

The invention relates to an artificial intelligence monitoring system and method, and more particularly to a decentralized monitoring system and method.

With the rapid aging of the overall population structure, the number of long-term care needs has also increased. The change in the state of old age actually involves many complex elements, which may be general chronic diseases, may be physical and mental function decline such as disability, dementia, etc., or may be affected by family social resources. As a result, the pressure on the care of individuals and families is increasing, which in turn leads to social and economic problems. In addition, not only is the elderly's health problem worthy of attention, modern people have a lot of work pressure, but it is also easy to ignite a red light without knowing their health. Therefore, the power of technology needs to be used to help monitor the user's health.

The invention provides a decentralized monitoring system and method, uploading encrypted medical data to a cloud server, and providing a third-party verification agency to further determine whether the monitoring data of the client system is abnormal, thereby ensuring the privacy of medical data. .

The decentralized monitoring system of the present invention includes a cloud server, a client system, a medical system, and a third-party verification agency. The client system is coupled to the cloud server, which is used to upload user monitoring data to the cloud server. The medical system is coupled to a cloud server, which calculates and encrypts multiple pathological data of multiple patients based on a neural network to obtain multiple medical data, and uploads the medical data to the cloud server. The calculated and encrypted medical data does not include the patient's identity data, but includes multiple thresholds and multiple weights. The third-party verification organization is coupled to the cloud server, and downloads the monitoring data and the medical data after calculation and encryption from the cloud server, and determines whether the monitoring data is abnormal based on the medical data after calculation and encryption, and generates a judgment result. Then send the judgment result to the cloud server. After the cloud server receives the monitoring data from the client device, the cloud server sends an identification key corresponding to the monitoring data to the client system. After the cloud server receives the identification key from the client system, the cloud server sends the judgment result corresponding to the identification key to the client system.

In an embodiment of the present invention, the above-mentioned medical system regards each patient ’s respective gender, degree of illness, and age as multiple input nodes of the neural network-like, and uses the diagnosis results of each patient as output nodes of the neural-like network. The threshold value and the weight are obtained from multiple hidden nodes of the neural network.

In an embodiment of the present invention, the client system includes at least one sensor for detecting physiological information, and the physiological information is used as monitoring data.

In an embodiment of the present invention, the at least one sensor is further used for detecting facial features, using the facial features to identify the user, and using the facial features to obtain emotional information, and combining the emotional information with the physiological information As monitoring information.

In an embodiment of the present invention, the at least one sensor is further configured to detect a sound feature, use the sound feature to identify a user, and use the sound feature to identify a cough message, and use the cough message with a physiological message for monitoring. data.

In an embodiment of the present invention, the at least one sensor is further configured to detect environmental information.

In an embodiment of the present invention, the at least one sensor is disposed in a wearable device or a home appliance.

In an embodiment of the present invention, the third-party verification agency uses deep learning to determine whether the monitoring data is abnormal.

The decentralized monitoring method of the present invention includes: receiving, in a cloud server, a plurality of medical data after calculation and encryption from a medical system, and receiving monitoring data from a client system, wherein the medical system is based on a neural network. The medical data is obtained after calculation and encryption of multiple pathological data of each patient. The medical data after calculation and encryption does not include the identity information of the patient but includes multiple thresholds and multiple weights; from the client system on the cloud server After receiving the monitoring data, the identification key corresponding to the monitoring data is sent to the client system through the cloud server; the third-party verification agency downloads the monitoring data from the cloud server and the medical data after calculation and encryption, and is based on The encrypted medical data determines whether the monitoring data is abnormal and generates a judgment result, and then transmits the judgment result to the cloud server; after the cloud server receives the identification key from the client system, it transmits the corresponding identification key through the cloud server. The judgment result is sent to the client system.

Based on the above, the medical data stored in the cloud server is calculated data, not raw data. Therefore, even if a third-party verification organization obtains medical data from the cloud server, it cannot know the privacy behind these medical data. Information to ensure medical privacy issues. In addition, only the client system that receives the identification key can obtain the final judgment result, so the privacy of the user can also be ensured.

In order to make the above features and advantages of the present invention more comprehensible, embodiments are hereinafter described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram of a decentralized monitoring system according to an embodiment of the present invention. Referring to FIG. 1, the distributed monitoring system 100 includes a medical system 110, a cloud server 120, a client system 130, and a third-party verification agency 140. The medical system 110, the client system 130, and the third-party verification agency 140 are coupled to the cloud server 120, respectively.

In this embodiment, the medical system 110 is, for example, a server used in a clinic, a hospital, a medical center, or the like, and is used to store medical data of multiple patients. The medical system 110 calculates and encrypts multiple pathological data of multiple patients based on a neural network to obtain multiple medical data, and then uploads the calculated and encrypted medical data to the cloud server 120. The encrypted medical information will only contain the scattered data, and will not include the patient's identity information. For example, the medical system 110 trains the pathological data, and the learned data is not the original data.

Therefore, the medical data uploaded to the cloud server 120 has been processed without including the patient's personal information, so the privacy of the patient can be fully ensured. In other words, the medical data is not directly uploaded to the patient's pathological data, but it is calculated by converting these pathological data into digital values (for example, a value string composed of 0 and 1), so as to remove Patient's personal message. In addition, these medical record data can be further integrated to classify medical data according to illness, gender, age, and so on. Medical information includes characteristics of various conditions.

For example, the medical system 110 uses the gender, the degree of illness (such as depression with different severity), and the age of multiple patients as the multiple input nodes of the neural-like network, and uses the diagnosis results of each patient as the neural-like The output nodes of the network obtain multiple thresholds and weights from multiple hidden nodes of the neural network, and use these thresholds and weights as medical information. For example, it includes facial features, facial features, and frequency and times of speech recorded at different times of the day. Through, for example, neural network-like deep calculations, these recorded features are used as input nodes in the input layer. , Add appropriate hidden layers and nodes, take the diagnosis result of the medical system 110 as the result value of the output layer, and then calculate and train to obtain better input nodes connected to the hidden nodes in the hidden layer and Thresholds and weights between hidden nodes and the output layer results. The medical system 110 links the facial features of facial features, severity, and gender corresponding to the above three dimensions (age, illness level, and gender) to the hidden nodes in the hidden layer and hidden layers through calculation and training. The threshold and weight between the hidden node and the result of the output layer are regarded as a calculated result. The threshold and weight between the appropriate hidden node in the hidden layer and the input layer and the output layer are encrypted and uploaded to the cloud server. 120.

The foregoing is just an example. The medical system can follow the example given above for different pathological characteristics, different data, and different patient distributions. After encryption, it can be calculated and trained before uploading to the cloud server 120, and the corresponding results obtained. It can be called a feature function. The comparison of whether there is any disease can be obtained by comparing the characteristics of the user. In this way, the medical system 110 has been calculated first, and the patient's private data has been removed. Even if the data is leaked during the upload to the cloud server 120, the obtained value cannot be restored to the pathological data of the specific patient. .

The cloud server 120 adopts, for example, a decentralized encryption technology. The cloud server 120 includes multiple hosts, and the data uploaded to the cloud server 120 will be stored on each host at the same time. Moreover, these uploaded materials cannot be tampered with. As time passes, the data in the cloud server 120 will be continuously copied. These data will be scattered among different hosts, and its mechanism is similar to blockchain technology. For example, there are three hosts A to C, and each of the hosts A to C stores data n, data n + 1, and data n + 2. During the verification, if the data of host A is found to be different from the data of hosts B and C, it is determined that the data of host A has been tampered with.

After the data is uploaded to the cloud server 120, it will be encrypted first. Therefore, when the hacker invades the cloud server 120, the hacker cannot obtain the data because the hacker does not know the encryption key. In addition, since the data uploaded to the cloud server 120 is a set of numerical strings after calculation, even if the hacker obtains the encryption key, he can only obtain a set of numerical strings, which cannot know the meaning of the numerical strings. .

The client system 130 includes at least one sensor. Sensors are used to detect monitoring information such as physiological information, facial features, sound characteristics, and environmental information. Below are examples.

FIG. 2 is a block diagram of a client system according to an embodiment of the present invention. Referring to FIG. 2, the client system 130 includes a physiological information acquisition module 210, an environmental information acquisition module 220, an identification module 230, and a medical communication module 240.

Here, the physiological information acquisition module 210, the environmental information acquisition module 220, and the identity recognition module 230 may be implemented using sensors. The medical communication module 240 is for the cloud server 120 to send and receive data, and is, for example, a communication device such as a network card.

The physiological information acquisition module 210 is, for example, a physiological information sensor, and is configured to detect a user's heartbeat, blood pressure, mood, body temperature, and sleep time and other physiological information. The physiological information acquisition module 210 may be disposed in a wearable device such as a watch, a bracelet, a neckband, a helmet, or may be disposed in a non-wearable device such as a home appliance.

The environmental information acquisition module 220 is, for example, a temperature sensor, a humidity sensor, and the like, and is used to detect indoor temperature and humidity. In addition, the environmental information acquisition module 220 can also use the Internet to obtain outdoor environmental information, such as today's temperature, humidity, and air pollution index. It can further determine whether the indoor and outdoor environmental information will affect the physiological information of the user.

The identity recognition module 230 is, for example, an image capture device, a radio device, and the like, and is configured to detect at least one of facial features and sound features of the user. The user may be identified using at least one of facial features and sound features. With the user in different positions, through the analysis and learning of features, the user's identity can be accurately identified.

For example, facial features are scanned by image capture devices, and machine learning and big data records are used to distinguish specific users. For example, an image of a face is captured 5 times per second, and calculation is performed after obtaining facial features. For example, a neural network is used to take parameters such as heartbeat, blood pressure, and facial features as input nodes, and the hidden layer uses 1.3-3 times the input nodes to accumulate records of all facial features into big data. The data of big data is repeatedly put into the neural-like deep learning module until the accuracy reaches a threshold (99.9% accurate), and then the user corresponding to the facial features is recorded.

In addition, you can use the radio to record sound (without pre-classifying different users), and process the audio into digital files to appropriately enhance the high-frequency areas of the digital files. After that, the digital file of the enhanced high-frequency area is divided into multiple frames, each frame can be 20-50ms, and the overlapping area of each frame can be 1/5 ~ 1/2. Next, the endpoints of each frame are processed to facilitate connection with the preceding and following frames. Then, Fourier transform is performed to obtain a frequency domain signal, and then a Mel filter is used to obtain a Mel-Frequency Cepstral Coefficient (MFCC) to obtain a sound characteristic.

After obtaining the sound features, the neural network-like sound features are used as the input nodes, and the hidden layer uses 1.3-3 times the input nodes to accumulate all the records of the sound features into big data, and repeatedly put the data of the big data into the class In the neural deep learning module, until the accuracy rate reaches a threshold (99.9% accurate), the users corresponding to the sound characteristics are connected, so as to distinguish the specific user's voice when multiple voices are obtained.

In addition, facial features can be used to obtain emotional messages, and voice features can be used to identify cough messages. For example, a large number of facial features with various emotions are obtained in advance, thereby training feature models corresponding to various emotions. Furthermore, a cough model is obtained through data collection and training. In this way, the models obtained after the training can be used to determine whether the subsequently detected facial features or sound features include specific emotional information or cough information. The actions can be performed in the third-party verification agency 140 or in the client system 130. That is, in one embodiment, the client system 130 determines whether the facial features or sound features include specific emotional information or cough information, and then sends the emotional information or cough information to the cloud server 120 as monitoring data. In another embodiment, the client system 130 may only collect information (face features, sound features), and the third-party verification agency 140 determines whether the facial features or sound features include specific emotional information or cough information. .

The medical communication module 240 uses distributed encryption technology to upload data to the cloud server 120 and download data from the cloud server 120.

In addition, the monitoring data uploaded by the client system 130 will not be transmitted to the medical system 110. Only when the user goes to the medical institution for diagnosis and treatment, the medical records related to the diagnosis and treatment will be added to the medical system 110. Therefore, the privacy of the user of the client system 130 can be ensured.

FIG. 3 is a flowchart of a decentralized monitoring method according to an embodiment of the present invention. Referring to FIG. 1 and FIG. 3, in steps S301 and S303, the cloud server 120 receives a plurality of medical data after calculation and encryption from the medical system 110, and receives monitoring data from the client system 130. Here, the execution order of steps S301 and S303 is not limited. Monitoring data includes physiological, emotional, cough, and environmental messages. After the cloud server 120 receives the monitoring data from the client system 130, in step S302, the cloud server 120 sends an identification key corresponding to the monitoring data to the client system 130.

In step S304, the third-party verification agency 140 downloads monitoring data and medical data from the cloud server 120. In addition, the third-party verification agency 140 determines whether the monitoring data is abnormal based on the medical data and generates a judgment result. After that, in step S305, the third-party verification agency 140 transmits the determination result to the cloud server 120.

In step S306, the client system 130 sends the identification key to the cloud server 120, and accordingly requests the cloud server 120 to return a judgment result corresponding to the monitoring data. After the cloud server 120 receives the identification key from the client system 130, in step S307, the cloud server 120 transmits the determination result corresponding to the identification key to the client system 130.

The cloud server 120 may simultaneously receive monitoring data transmitted by different client systems, and transmit different identification keys to each client system accordingly. Different client systems can obtain corresponding judgment results by identifying the key.

In addition, the client system 130 can also upload monitoring data of different users to the cloud server 120 at the same time. For example, the client system 130 includes a plurality of wearable devices, and these wearable devices are respectively mounted on different users. The cloud server 120 generates a plurality of different identification keys for monitoring data sent by different wearable devices, so that different wearable devices can obtain corresponding judgment results by using the identification keys.

The wearable device and the non-wearable device in the client system 130 can assist each other to achieve the effect of removing noise from the data. FIG. 4 is a schematic diagram of a measurement position according to an embodiment of the present invention. In FIG. 4, the user wears a wearable device 403, and in this embodiment, non-wearable devices 401 and 402 are respectively provided. For example, if the user wears an obstacle between the wearable device 403 and the non-wearable device 401, the position of the wearable device 403 can be calculated by using the non-wearable device 402. Preferably, the non-wearable devices 401 and 402 can be used to obtain the location of the wearable device 403 more accurately.

The following illustrates the practical application of the foregoing embodiment.

When the user gets up and brushes his teeth in the morning, the client system 130 can obtain the facial features of the user through the image capturing device provided on the mirror, and obtain the emotional information from the facial features. The emotional information is then encrypted and uploaded to Cloud server 120. After the third-party verification agency 140 downloads the emotional information from the cloud server 120, it uses deep learning of big data to determine whether the emotional information matches the characteristics of depression. Then, the judgment result is transmitted to the client system 130 through the cloud server 120.

In addition, assuming that the user is in the living room, the client system 130 can obtain the user's voice characteristics through a radio device disposed in the living room, and transmit the voice characteristics to the third-party verification agency 140 through the cloud server 120. If the user's voice characteristics show unconventional high-frequency signals and there are breathless features, after comparison by the third-party verification agency 140, a warning message can be sent to the client system 130 through the cloud server 120.

For another example, the client system 130 can use a wearable device to detect the user's physiological information (heartbeat, blood pressure, etc.), use a radio device to detect the user's voice information, and then, use the cloud server to pass the physiological information and voice information The device 120 is transmitted to the third-party verification authority 140. After the comparison by the third-party verification agency 140, it was found that the resting heart rate of the user was abnormally high and he did not speak all day, which represents that the user may have symptoms of severe depression, so it was issued through the cloud server 120 Warning message to client system 130.

Here, based on the privacy of the user, the cloud server 120 will not actively notify the medical institution. However, in other embodiments, an active help-seeking mechanism may also be set in the cloud server 120. When the user is in a life-threatening situation, the cloud-server 120 may actively send a help-seeking message to the medical institution, for example, actively call for help Wait.

To sum up, the above-mentioned embodiment uses decentralized encryption technology to achieve confidentiality. In addition, the above embodiments can be applied to remote home care, reducing personnel care, and not only applicable to nursing homes, hospitals, etc., but also to ordinary families. In the above embodiment, through big data and deep learning, various special problems that are specific to specific people can be obtained with accurate results. In addition, by using the above-mentioned embodiment, it is possible to reduce the human labor timing to obtain the physiological information of the user. In addition, the use of the identification key can ensure that the results of your judgment will only be transmitted back to yourself, which can fully ensure privacy issues.

Although the present invention has been disclosed as above with the examples, it is not intended to limit the present invention. Any person with ordinary knowledge in the technical field can make some modifications and retouching without departing from the spirit and scope of the present invention. The protection scope of the present invention shall be determined by the scope of the attached patent application.

100‧‧‧ decentralized monitoring system

110‧‧‧ Medical System

120‧‧‧ Cloud Server

130‧‧‧client system

140‧‧‧Third party verification agency

210‧‧‧ physiological information acquisition module

220‧‧‧Environmental information acquisition module

230‧‧‧Identification Module

240‧‧‧Medical Communication Module

S301 ～ S307‧‧‧ Decentralized monitoring method steps

401, 402‧‧‧ non-wearable devices

403‧‧‧ Wearable

FIG. 1 is a block diagram of a decentralized monitoring system according to an embodiment of the present invention.
FIG. 2 is a block diagram of a client system according to an embodiment of the present invention.
FIG. 3 is a flowchart of a decentralized monitoring method according to an embodiment of the present invention.
FIG. 4 is a schematic diagram of a measurement position according to an embodiment of the present invention.

Claims (16)

A distributed monitoring system includes:
A cloud server;
A client system coupled to the cloud server and uploading monitoring data of a user to the cloud server;
A medical system, coupled to the cloud server, calculates and encrypts multiple pathological data of multiple patients based on a type of neural network to obtain multiple medical data, and uploads the medical data to the cloud server, where The medical data after calculation and encryption do not include the identity data of the patients but include multiple thresholds and multiple weights;
A third-party verification agency is coupled to the cloud server, downloads the monitoring data and the medical data after calculation and encryption from the cloud server, and judges the medical data based on the medical data after calculation and encryption. Monitoring data for abnormality to generate a judgment result, and then transmitting the judgment result to the cloud server;
Wherein, after the cloud server receives the monitoring data from the client device, the cloud server sends an identification key corresponding to the monitoring data to the client system;
After the cloud server receives the identification key from the client system, the cloud server sends the judgment result corresponding to the identification key to the client system. The decentralized monitoring system according to item 1 of the scope of patent application, wherein the medical system uses each of the patients' gender, a degree of illness, and a year as a plurality of input nodes of the neural network, and each The diagnosis results of the patients are used as output nodes of the neural network, and the thresholds and weights are obtained from multiple hidden nodes of the neural network. The decentralized monitoring system as described in the first patent application scope, wherein the client system includes:
At least one sensor detects a physiological message and uses the physiological message as the monitoring data. The decentralized monitoring system according to item 3 of the scope of patent application, wherein the at least one sensor is further used to detect a facial feature, use the facial feature to identify the user, and use the facial feature to Obtain an emotional message, and use the emotional message with the physiological message as the monitoring data. The decentralized monitoring system according to item 3 of the scope of patent application, wherein the at least one sensor is further configured to detect a sound feature, use the sound feature to identify the user, and use the sound feature to identify a cough Message, and use the cough message with the physiological message as the monitoring data. The decentralized monitoring system according to item 3 of the scope of patent application, wherein the at least one sensor is further used to detect an environmental message. The decentralized monitoring system according to item 3 of the scope of patent application, wherein the at least one sensor is disposed in a wearable device or a household appliance. The decentralized monitoring system according to item 1 of the scope of patent application, wherein the third-party verification agency uses deep learning to determine whether the monitoring data is abnormal. A decentralized monitoring method includes:
In a cloud server, a plurality of medical data that is calculated and encrypted is received from a medical system, and a monitoring data is received from a client system. The medical system is based on a type of neural network to send multiple records of multiple patients The medical information is obtained after pathological data calculation and encryption. The medical data after calculation and encryption does not include the identity information of the patients but includes multiple thresholds and multiple weights;
After the cloud server receives the monitoring data from the client system, sending an identification key corresponding to the monitoring data to the client system through the cloud server;
A third-party verification agency downloads the monitoring data and the medical data after calculation and encryption from the cloud server, and determines whether the monitoring data is abnormal based on the medical data after calculation and encryption, and generates a judgment. As a result, the judgment result is then transmitted to the cloud server; and after the cloud server receives the identification key from the client system, the judgment result corresponding to the identification key is transmitted to the client through the cloud server. End system. The decentralized monitoring method according to item 9 of the scope of patent application, wherein the medical system regards each of the patients as a gender, a degree of illness, and a year as a plurality of input nodes of the neural network, and each The diagnosis results of the patients are used as output nodes of the neural network, and the thresholds and weights are obtained from multiple hidden nodes of the neural network. The decentralized monitoring method described in item 9 of the patent application scope further includes:
In the client system, a physiological message is detected through at least one sensor, and the physiological message is used as the monitoring data. The decentralized monitoring method described in item 11 of the patent application scope further includes:
In the client system, a facial feature is detected by the at least one sensor, the facial feature is used to identify the user, and the facial feature is used to obtain an emotional message, and the emotional message is matched with The physiological information is used as the monitoring data. The decentralized monitoring method described in item 11 of the patent application scope further includes:
In the client system, at least one sensor is used to detect a sound feature, use the sound feature to identify the user, and use the sound feature to identify a cough message, and match the cough message with the physiological message As the monitoring information. The decentralized monitoring method described in item 11 of the patent application scope further includes:
In the client system, an environmental message is detected through at least one sensor. The method of decentralized monitoring according to item 11 of the scope of patent application, wherein the at least one sensor is disposed in a wearable device or a household appliance. The decentralized monitoring method according to item 9 of the scope of patent application, wherein the third-party verification agency uses a deep learning to determine whether the monitoring data is abnormal.