Health early warning system

Abstract

The present application relates to a system for issuing an alert related to a health condition of a subject. More specifically, the system includes a processor having an artificial intelligence module configured to determine a probability that a subject is diseased and to issue an alarm when the probability exceeds a preset value.

Description

TECHNICAL FIELD

[0001] The present application relates to a system for issuing an alert related to the health condition of a subject. More specifically, the system comprises a processor having an artificial intelligence module configured to determine the probability of the subject having a disease and to issue an alert when the probability exceeds a preset value. BACKGROUND

[0002] The healthcare industry focuses on early detection of health problems and personalized treatment.

[0003] Early detection of health problems helps to detect diseases early or, in the case of patients already having a disease, to detect the worsening of the disease in time. Personalized treatment programs can make the treatment achieve the best effect for a specific patient. Early detection and personalized treatment are crucial for healthcare providers to intervene in time, prevent complications and even save lives.

[0004] There are several factors that hinder or prevent the achievement of these goals, such as long lists of patients waiting to be seen at healthcare providers, human errors, difficulties in allocating resources (healthcare providers and equipment), and limited access to healthcare for patients living in remote areas or with limited mobility.

[0005] In addition to the above factors, the strict confidentiality of healthcare data and the technical problems of data security and privacy arising from the integration of such data with existing medical devices and electronic health record (EHR) systems cannot be ignored. Incorrect integration can lead to fragmentation of healthcare data, that is, healthcare and patient personal information is scattered in different systems, and it can be difficult for healthcare providers to access this information. In this way, healthcare providers cannot have a comprehensive understanding of the patient's medical history and health condition.

[0006] Therefore, there is a need for a system that can detect health problems early, which needs to be able to solve technical problems related to access to healthcare and data integration with existing medical devices and electronic health record systems, thereby improving resource allocation and at the same time reducing human errors. SUMMARY

[0007] The present application provides a solution to the above problem, namely a system for issuing an alert related to the health condition of a subject according to claim 1. In the dependent claims, preferred embodiments of the application are defined.

[0008] The present application provides, in a first inventive aspect, a system for issuing an alert related to the health condition of a subject, the system comprising a processor configured to:

[0009] - receiving at least one health parameter of a subject, wherein the one or more health parameters are measured by at least one sensor, and

[0010] - receiving at least one personal parameter of a subject,

[0011] wherein the processor comprises an artificial intelligence (AI) module trained to generate a plurality of disease models based on:

[0012] at least one health parameter and / or at least one personal parameter of a subject suffering from a disease of at least one of the disease models, and

[0013] at least one health parameter and / or at least one personal parameter of a subject not suffering from the disease;

[0014] wherein each disease model is based on a neural network comprising:

[0015] o an input layer consisting of N nodes, where N is the number of health parameters and personal parameters used to train the disease model;

[0016] o three hidden layers, each containing an activation function and a plurality of nodes configured to receive a plurality of inputs;

[0017] o an output layer with only one node representing the probability of the subject suffering from the disease;

[0018] wherein the output of each node of each hidden layer of the neural network is the result of applying its respective activation function to the weighted sum of its respective inputs;

[0019] wherein the artificial intelligence module is configured to perform the following steps:

[0020] a) using the at least one health parameter and / or the at least one personal parameter as input for one or more disease models;

[0021] b) determining the probability of the subject suffering from the disease; and

[0022] c) issuing an alert if the probability exceeds a determined probability value.

[0023] In this document, a "subject" refers to a person who can suffer from a certain disease, or a patient who already suffers from a certain disease.

[0024] In this document, a "health parameter" refers to a parameter that provides information about the health status of a subject, which can include physiological data and / or behavioral data; for example, the heart rate of the subject, the sleep pattern of the subject or the oxygen level in the blood of the subject.

[0025] In one embodiment, the at least one health parameter comprises at least one vital sign of the subject. In the present context, a "vital sign" refers to a parameter reflecting the basic functioning of the subject's body, such as heart rate, respiratory rate, body temperature and blood pressure.

[0026] In one particular embodiment, the at least one health parameter is one or more of the following parameters:

[0027] - the type of chest pain experienced by the subject,

[0028] - the blood pressure level at rest,

[0029] - serum cholesterol,

[0030] - the maximum heart rate experienced by the subject,

[0031] - exercise-induced angina pectoris,

[0032] - ST segment depression induced by exercise compared to the resting state,

[0033] - the subject having diabetes,

[0034] - the subject having depression,

[0035] - the subject carrying the APOE ε3 allele,

[0036] - the oxygen content in the blood,

[0037] - the heart rate,

[0038] - the time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or the time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device,

[0039] - the alcohol content in the blood,

[0040] - the body temperature,

[0041] - the weight,

[0042] - the sleep pattern,

[0043] - the walking / step motion pattern,

[0044] - the clinical dementia rating for indicating the severity of dementia symptoms,

[0045] - the mini-mental state examination for indicating the level of cognitive function,

[0046] - the normalized total brain volume,

[0047] - the estimated total intracranial volume,

[0048] - the atlas scaling factor,

[0049] - heart rate variability,

[0050] - skin conductance activity,

[0051] - respiratory rate and breathing pattern,

[0052] - blood pressure levels,

[0053] - activity level,

[0054] - social interaction patterns,

[0055] - history of drug abuse,

[0056] - self-reported anxiety and / or depressive symptoms,

[0057] - mental health diagnosis or treatment history,

[0058] - body mass index,

[0059] - number of weekly binge eating episodes,

[0060] - number of weekly purging episodes,

[0061] - body dissatisfaction score, to indicate a self-reported level of body dissatisfaction,

[0062] - eating disorder questionnaire score, to indicate a score of eating disorder symptoms according to a standard scale,

[0063] - depression score, to indicate a score of a depression assessment scale,

[0064] - anxiety score, to indicate a score of an anxiety assessment scale, and

[0065] - eating disorder diagnosis.

[0066] At least one health parameter is measured by at least one sensor. In one embodiment, the sensor is part of the system. In one embodiment, the sensor is part of a wearable device.

[0067] In one embodiment, one or more health parameters are symptom descriptions provided by the subject, for example, a type of chest pain experienced by the subject, self-reported anxiety symptoms, or a body dissatisfaction score.

[0068] In one embodiment, one or more health parameters include behavioral data, for example, social interaction patterns of the subject.

[0069] In one embodiment, one or more health parameters reflect lifestyle factors, for example, a history of drug abuse by the subject.

[0070] In one embodiment, the one or more health parameters are a previously diagnosed disease of the subject, for example, the subject has diabetes or depression.

[0071] In one embodiment, the one or more health parameters are obtained from a standard questionnaire answered by the subject, for example, a score of the Eating Disorder Examination-Questionnaire.

[0072] In one embodiment, the one or more health parameters are based on medical imaging studies, for example, normalized whole brain volume or atlas scaling factor.

[0073] In one embodiment, the processor of the system receives the health parameters from the one or more sensors in real-time or periodically.

[0074] In one embodiment, the processor of the system receives the one or more health parameters from a medical database and / or the subject and / or a caregiver and / or a medical service provider.

[0075] In this document, a "personal parameter" refers to a parameter that provides personal data information of the subject; for example, the subject's age or gender.

[0076] In one embodiment, the processor of the system receives at least one personal parameter from a medical database and / or the subject and / or a caregiver and / or a medical service provider.

[0077] The processor of the system is configured to receive at least one health parameter, one or more of which is measured by the at least one sensor, and at least one personal parameter, for example, from the subject's clinical history, an electronic health record (EHR) system, the subject's self-report, or the subject's personal profile previously registered in the system.

[0078] In one embodiment, the transmission of the health and / or personal parameters is through a mobile application or a cloud-based platform.

[0079] The processor comprises a trained artificial intelligence module for generating a disease model. The training takes into account N health parameters and / or personal parameters of real subjects, wherein the real subjects include subjects having at least one disease represented by the disease model, and subjects not having the disease. Such data are taken, for example, from a medical database. In one embodiment, N is greater than 1.

[0080] Each disease model is a neural network having an input layer with N nodes, three hidden layers, and an output layer.

[0081] Each hidden layer comprises an activation function and a plurality of nodes for receiving a plurality of inputs.

[0082] In the present context, an "activation function" refers to a mathematical function that defines the output of a node as a function of the input to the node. Specifically, the output of each node in a neural network is the result of applying its respective activation function to the weighted sum of its respective inputs.

[0083] The first hidden layer receives as input the output of the input layer; the second hidden layer receives as input the output of the first hidden layer; the third hidden layer receives as input the output of the second hidden layer.

[0084] The output layer receives as input the output of the third hidden layer. The output layer comprises a node that provides a probability of the subject developing a disease predicted by the disease model.

[0085] After the artificial intelligence module has generated a disease model, it determines a probability of the subject developing the disease based on at least one health parameter and / or at least one personal parameter received by the processor as input to the disease model. If the determined probability exceeds a predetermined value, the artificial intelligence module issues an alert.

[0086] In one embodiment, for at least one disease model, the health parameters and / or personal parameters used by the artificial intelligence module to determine the probability of the subject developing the disease are at least a subset of the health parameters and / or personal parameters used by the artificial intelligence module to generate the disease model. Preferably, all health parameters and / or personal parameters used by the artificial intelligence module to determine the probability are all health parameters and / or personal parameters used by the artificial intelligence module to generate the disease model.

[0087] In one embodiment, the processor is configured to receive at least one health parameter of the subject measured by a sensor, and the artificial intelligence module is configured to compare the value of the at least one health parameter to a predefined range of values. Furthermore, the artificial intelligence module is configured to issue an alert if the measured value is outside the predefined range of values.

[0088] In one embodiment, the medical service provider defines the range of values according to a standard, and the system is configured to store the range of values, e.g. in an internal database. In one embodiment, the medical service provider modifies the range of values according to the disease and / or symptoms and / or health condition and / or personal parameters of the subject.

[0089] In one embodiment, the processor is configured to receive at least one health parameter of the subject measured by a sensor over time, and the artificial intelligence module is configured to compare the current value of the at least one health parameter to a previously measured value of the at least one health parameter. Furthermore, the artificial intelligence module is configured to issue an alert if the current measured value has changed abruptly compared to the previous value.

[0090] In one embodiment, the system alerts the medical service provider and / or the caregiver and / or the subject's family members and / or any other contact and / or the subject himself / herself about the subject's health condition.

[0091] In this document, "medical service provider" refers to any member of a medical service team that can take care of a subject, such as a doctor, a nurse or a psychologist.

[0092] The first innovation of this system is the use of artificial intelligence neural networks and predictive modeling techniques to predict future health outcomes, thus enabling the medical service provider to predict the worsening of chronic diseases and prevent adverse events.

[0093] Thanks to the system of the present application, the medical service provider can remotely monitor the health condition of the subject, track health parameters and personal parameters to track the occurrence and development of diseases, and intervene when necessary without having to see the patient in person.

[0094] The system provided by the present application has more advantages as follows.

[0095] - With the help of artificial intelligence prediction, the medical service provider can make wise decisions and avoid human errors, thus predicting any disease that the patient may suffer from.

[0096] - Health parameters can be continuously monitored in various ways, such as using sensors on wearable devices. In this way, the health parameters obtained can be provided to the system in real time, thus helping to make more informed decisions and provide personalized care.

[0097] - Continuous monitoring can detect subtle changes in the subject's condition early, which may indicate the onset of disease. This early detection helps to intervene in time, prevent complications, reduce hospitalization, improve patient outcomes, and possibly save lives.

[0098] - Remote monitoring is particularly important for subjects living in remote areas or with limited mobility, who have limited access to medical resources, especially those with chronic diseases or those recovering from surgery. Subjects can be monitored comfortably at home, thus reducing hospitalization and the burden on medical institutions.

[0099] - Continuous monitoring also helps prioritize care based on the severity of the patient's condition and needs, thus optimizing resource allocation. This relieves the pressure on medical service providers and resources and ensures that patients receive timely care.

[0100] - Automation of subject monitoring can also reduce the risk of transcription errors and misinterpretation of data (human errors).

[0101] In one embodiment, the system is configured to transmit and receive data with at least one existing medical device and / or at least one electronic health record (EHR) system. This addresses the issue of data fragmentation in the medical service field, where patient information is scattered across different systems and can be difficult for medical service providers to access. This enables the system to provide a comprehensive view of a patient’s medical history and health condition.

[0102] In one embodiment, to address the technical problem of protecting patient data, the system implements subject data encryption and / or access control and / or compliance with medical service privacy regulations (e.g. HIPAA).

[0103] In one embodiment, the system includes a dashboard that enables medical service providers to continuously access real-time health parameters and / or personal parameters of a subject and / or artificial intelligence predictions.

[0104] In this way, medical service providers can detect abnormalities in the subject’s indicators and take timely action if necessary. For example, if the heart rate of a subject diagnosed with heart disease suddenly increases or decreases sharply, medical personnel can immediately notify the emergency department.

[0105] In one embodiment, the system is configured to transmit and receive data with clinical decision support tools used by medical service providers. The advantage of the system is that it can provide real-time guidance and decision support for medical service providers.

[0106] In one embodiment, the system is configured to automatically input information from new subjects with one or more diseases and the latest medical literature into the artificial intelligence module to train the artificial intelligence module. The advantage is that the disease model is improved, and the accuracy and effectiveness of the prediction are improved over time.

[0107] In one embodiment, the artificial intelligence module is further trained to generate a plurality of disease severity models, the training being performed using at least one health parameter and / or at least one personal parameter related to subjects with different levels of disease severity;

[0108] wherein each disease severity model is constructed based on a neural network comprising:

[0109] o an input layer consisting of M nodes, where M is the number of health parameters and personal parameters used to train the disease severity model;

[0110] o three hidden layers, each containing an activation function and a plurality of nodes configured to receive a plurality of inputs;

[0111] o an output layer having only one node, representing the severity of the disease of the subject;

[0112] wherein the output of each node of each hidden layer of the neural network is the result of applying its respective activation function to the weighted sum of its respective inputs;

[0113] wherein the artificial intelligence module is further configured to perform the following steps:

[0114] 1. using at least one health parameter and / or at least one personal parameter as input to one or more disease severity models;

[0115] 2. determining a disease severity index of the disease of the subject; and

[0116] 3. issuing an alert if the disease severity index exceeds a determined severity value.

[0117] In the present embodiment, the artificial intelligence module is trained to generate disease severity models. The training takes into account M health parameters and / or personal parameters of real subjects who have one or more diseases, each of which is represented by a disease severity model, and each of which has a different severity. Such data are for example taken from a medical database. In one embodiment, M is greater than 1.

[0118] In one embodiment, the disease severity index is a numerical value representing the severity of the disease.

[0119] Each disease severity model is a neural network comprising an input layer having M nodes, three hidden layers, and an output layer.

[0120] Each hidden layer comprises an activation function and a plurality of nodes for receiving a plurality of inputs.

[0121] The first hidden layer receives as input the output of the input layer; the second hidden layer receives as input the output of the first hidden layer; and the third hidden layer receives as input the output of the second hidden layer.

[0122] The output layer receives as input the output of the third hidden layer. The output layer comprises one node which provides the disease severity index of the disease of the subject.

[0123] After the artificial intelligence module has generated a disease severity model, it determines the disease severity index of the subject based on at least one health parameter and / or at least one personal parameter received by the processor as input to the disease severity model. If the determined disease severity index exceeds a predetermined severity value, the artificial intelligence module issues an alert.

[0124] In one embodiment, for at least one disease severity model, the artificial intelligence module is configured to determine that the health parameters and / or personal parameters of the subject’s disease severity are at least a part of the health parameters and / or personal parameters used by the artificial intelligence module to generate the disease severity model. Preferably, the artificial intelligence module is configured to determine that all the health parameters and / or personal parameters of the severity are all the health parameters and / or personal parameters used by the artificial intelligence module to generate the disease severity model.

[0125] One advantage of the system is that it can continuously assess the disease severity of the subject and alert when the disease severity exceeds a pre-set threshold. For example, if the condition of a heart disease patient reaches a certain severity, the artificial intelligence module will alert an emergency call as this severity can be indicative of a heart attack.

[0126] Furthermore, by comparing the severity assessed by the system over time, the medical service provider can detect a worsening of the subject’s condition and can optionally take some measures; for example, modifying the medication regimen of the subject to adapt to the worsening condition.

[0127] The system is also useful in intensive care units and during surgical procedures. The continuous monitoring of the health parameters of the subject provides real-time data of vital signs and alerts when the condition reaches a severity so that medical personnel can immediately take intervention measures.

[0128] In one particular embodiment, the processor is configured to select the health parameters and / or personal parameters used to train the artificial intelligence module to generate each disease model, wherein the selection is performed using a plurality of decision trees, each decision tree comprising a plurality of layers, and each layer comprising at least one parent node, at least one left child node and at least one right child node, and wherein the processor is configured to perform the following steps:

[0129] - using the health parameters and / or personal parameters as inputs of the plurality of decision trees;

[0130] determining an importance value of each health parameter and each personal parameter in each layer of the decision tree, the importance value being defined as a reduction of node impurity and weighted by a probability of reaching at least one parent node of the layer; wherein

[0131] o the probability of reaching the parent node of the layer is defined as the number of health parameters and personal parameters reaching the parent node divided by the total number of health parameters and personal parameters used to train the artificial intelligence module;

[0132] o the reduction of node impurity is determined as follows:

[0133]

[0134] there are:

[0135] N left and N right are the total number of health parameters and personal parameters in the at least one left child node and the at least one right child node, respectively, of the layer, and

[0136] I p , I left and I right are the impurities of the at least one parent node, the at least one left child node and the at least one right child node, respectively; the impurities are calculated by an impurity measure, preferably a Gini impurity measure or an entropy measure;

[0137] determining a final importance value for each health parameter and each personal parameter, the final importance value being the importance value of the health parameter or personal parameter calculated in the last layer of the decision tree; and

[0138] selecting health parameters and / or personal parameters for training the artificial intelligence module, the health parameters and / or personal parameters being parameters whose final importance value is higher than a predefined threshold.

[0139] As mentioned previously, the artificial intelligence module is trained to generate a disease model. This training uses one or more health and / or personal parameters of subjects suffering from at least one disease and of subjects not suffering from the disease. However, not all health and / or personal parameters are suitable for all diseases, and it is recommended to select appropriate parameters to obtain the best disease prediction results.

[0140] In the present embodiment, the selection of parameters is done by means of a decision tree for each disease. In the present text, a “decision tree” will be understood as: a decision support hierarchical model that uses a tree model to represent decisions and their possible consequences.

[0141] With the decision tree, it is possible to calculate a final importance value for each health parameter and each personal parameter. The higher the final importance value, the higher the relevance of the parameter with the particular disease.

[0142] The selected parameters are the parameters that reach a predefined final importance threshold. In one embodiment, a corresponding threshold is selected for each disease.

[0143] In one embodiment, the artificial intelligence module is trained to generate a plurality of disease severity models, and the processor is configured to select health parameters and / or personal parameters for training the artificial intelligence module to generate each disease severity model, the selection being made using a plurality of decision trees as described above.

[0144] In one particular embodiment, the processor is configured to process the health parameters and / or the personal parameters used to train the artificial intelligence module before the disease model generation training process, the processor being configured to process the health parameters and / or the personal parameters by performing the following steps for each disease:

[0145] - creating a table in which each column represents a health parameter or a personal parameter and each row represents a different subject suffering from the disease;

[0146] - assigning a predetermined numerical format to the values in the table;

[0147] - replacing the values in the table that are not numerical with the average value of the values in the column;

[0148] - identifying outliers in each column, preferably using the z-score method;

[0149] - deleting the rows containing outliers;

[0150] or

[0151] - creating a table in which each row represents a health parameter or a personal parameter and each column represents a different subject suffering from the disease;

[0152] - assigning a predetermined numerical format to the values in the table;

[0153] - replacing the values in the table that are not numerical with the average value of the values in the row;

[0154] - identifying outliers in each row, preferably using the z-score method;

[0155] - deleting the columns containing outliers.

[0156] This embodiment describes a process for preparing health parameters and / or personal parameters before training an artificial intelligence module for generating a disease model. The rows and columns of the table generated by this process can be interchanged.

[0157] First, a table is created from the information of a plurality of subjects suffering from a certain disease and the information is classified by health parameters and / or personal parameters.

[0158] Then, a format conversion is performed in which all the values in the table are assigned a predetermined numerical format. For example, an integer or a floating point number.

[0159] After the format conversion, some of the values in the table can be non-numerical values (NaN). In this case, the value is replaced with the average value of the values in the column or in the row.

[0160] Finally, outliers are detected and the data of the subject containing the outlier is deleted from the table (the column or row in which the subject is located). In this context, an "outlier" refers to an extreme value in the dataset that deviates significantly from the overall pattern of values.

[0161] In one embodiment, the identification of outliers is achieved by the method of standard scores.

[0162] The data cleaning and formatting process enables the artificial intelligence module to be trained with high quality data, thereby making subsequent predictions more accurate.

[0163] In one embodiment, the artificial intelligence module is trained to generate a plurality of disease severity models, the processor is configured to process the health parameters and / or personal parameters used to train the artificial intelligence module prior to the disease severity model generation training process, the processor is configured to perform the processing of the health parameters and / or personal parameters for each disease severity model and perform the aforementioned steps.

[0164] In one particular embodiment, the system comprises at least one sensor. In one embodiment, the system comprises a plurality of sensors. In one embodiment, one or more of the at least one sensor is configured to measure at least one health parameter of the subject.

[0165] In one specific embodiment, the system comprises at least one sensor comprising a three-axis gyroscope and a three-axis accelerometer, and the artificial intelligence module is further configured to:

[0166] detect whether the subject has fainted or fallen by analyzing the angular velocity measured by the three-axis gyroscope and the acceleration measured by the three-axis accelerometer;

[0167] if the subject has fainted or fallen, an alert is issued.

[0168] According to the present embodiment, the system disclosed herein can also detect whether the subject has fainted or fallen by analyzing information from the three-axis gyroscope and the three-axis accelerometer. In this case, the system will issue an alert, for example, to notify a medical service provider, a caregiver or a family member of the subject's fall or fainting.

[0169] The artificial intelligence module of the system detects falls and fainting by integrating information from the three-axis gyroscope and the three-axis accelerometer; that is, the artificial intelligence module takes into account the measurements of the six axes of the sensors collected at multiple points in time. The artificial intelligence module estimates the spatial posture of the patient from these measurements, and if the spatial posture estimated by the artificial intelligence module mutates within a certain period of time (for example, less than 4 seconds), the artificial intelligence module detects that the subject has fainted or fallen. Subsequently, the artificial intelligence module will issue an alert.

[0170] This feature is particularly important for elderly patients and patients with heart disease, as their fall or fainting situation needs immediate care.

[0171] In one embodiment, the artificial intelligence module considers measurements of the six axes of the sensor every 10 milliseconds.

[0172] In one specific embodiment, the system comprises at least one sensor for tracking the position of the subject, and the artificial intelligence module is further configured to issue an alert when the subject is outside a predetermined area.

[0173] In this embodiment, the system disclosed in the present application can also continuously track the position of the subject. With the position information, if the patient leaves the pre-set area (for example, his residence or ward), the artificial intelligence module can issue an alert. In this case, the system will issue an alert to inform the medical service provider, caregiver, contact or family member of the location of the subject.

[0174] In one embodiment, GPS and / or Bluetooth and / or beacon and / or geofencing technology is used to track the position of the subject.

[0175] This feature is particularly important for people with schizophrenia, dementia or Alzheimer's disease, as they are prone to disorientation.

[0176] In one specific embodiment, the artificial intelligence module is further configured to:

[0177] recommend a personalized treatment plan according to the probability of the subject having a disease; and / or

[0178] associate the subject with at least one specific medical service provider according to the probability of the subject having a certain disease.

[0179] In this embodiment, the system provides more functions.

[0180] In one aspect, the artificial intelligence module of the system can recommend a specific treatment plan for the subject according to the probability of having a disease.

[0181] In one embodiment, the system contains a database in which a plurality of treatment plans for specific diseases are stored. In this way, when the artificial intelligence module issues an alert that the probability of the subject having a certain disease exceeds a predetermined probability value, the artificial intelligence module will select a treatment plan for the disease from the system database.

[0182] The advantage of the system is that it can develop a personalized treatment plan for each subject, making the treatment as effective as possible and reducing human error.

[0183] In one embodiment, the treatment comprises a medication regimen, a physical therapy regimen, a psychological therapy regimen, a rehabilitation regimen, a surgical intervention, and / or a lifestyle change.

[0184] In one embodiment, the artificial intelligence module is further configured to provide explanations and justifications for the recommended treatment regimen, highlighting its underlying principles, evidence-based guidelines, and factors influencing the decision-making process. This is beneficial as it can improve the transparency and credibility of the artificial intelligence recommendations.

[0185] In one embodiment, the system is configured to receive reports (e.g., from medical service providers) indicating the subject's response to at least one previous prescribed treatment (e.g., a previous medication regimen). These reports can also include symptoms experienced by the subject when following the prescribed treatment regimen. The artificial intelligence module is further configured to select a specific treatment regimen for the subject based on these reports.

[0186] For example, if a subject who is taking a previously prescribed medication experiences symptoms of nausea and vomiting, and a medical service provider has recorded these symptoms in a report, the artificial intelligence module will take these symptoms into account and recommend a treatment regimen that does not include the previously prescribed medication. In this way, the system can avoid adverse reactions to the medication regimen in the patient.

[0187] In one embodiment, the system is configured to receive reports (e.g., from medical service providers) indicating the subject's response to the recommended treatment regimen (e.g., a medication regimen). The reports can also include symptoms experienced by the subject when following the recommended regimen. The artificial intelligence module is further configured to improve the subject's treatment regimen based on these reports.

[0188] For example, if a patient is receiving a recommended treatment plan that includes a medication regimen, and this medication regimen causes the patient to experience symptoms such as nausea and vomiting, and a medical service provider has recorded these symptoms in a report, the artificial intelligence module will take these symptoms into account and improve the treatment plan so that it no longer includes the previously recommended medication regimen. In this way, the system can avoid adverse reactions to the medication regimen in the subject.

[0189] In one embodiment, the system is configured to transmit and receive data with clinical decision support tools used by medical service providers. The advantage of this embodiment is that the system can provide real-time guidance and decision support to medical service providers, enabling them to make informed decisions based on personalized treatment recommendations.

[0190] In one embodiment, the artificial intelligence module of the system is further configured to generate a collaborative treatment regimen that is followed by both the medical service provider and the caregiver. In one embodiment, the collaborative treatment regimen takes into account the patient’s response to a previous prescribed treatment regimen. The advantage of such a collaborative treatment regimen is that it enables the caregiver and the medical service provider to provide patient-centered care.

[0191] In one embodiment, the system is configured to automatically input information from new subjects having one or more diseases, treatment outcomes, and medical literature updates to the artificial intelligence module to train the artificial intelligence module to improve and enhance the accuracy and effectiveness of treatment recommendations over time.

[0192] In one embodiment, the system is configured to provide the caregiver and the family member with access to an online support group so that they can keep in touch with the medical service provider.

[0193] In one embodiment, the system is configured to provide the subject’s family member, caregiver, and / or contact with relevant information about the subject’s condition and to receive updates.

[0194] In addition, the artificial intelligence module of the system can also match the subject to at least one specific medical service provider based on the probability of the subject having a disease. For example, if the subject has a mental illness, the artificial intelligence module can match the subject to a psychologist and a psychiatrist whose specialty is in handling such mental illness.

[0195] In one embodiment, the artificial intelligence module selects a suitable medical service provider based on the subject’s disease and schedules an appointment between the subject and the medical service provider, thereby associating the specific medical service provider with the subject. In one embodiment, the system is configured to send a notification containing the date and time of the appointment to the subject, the subject’s family member, and / or the subject’s contact. In one embodiment, the system is configured to send the notification via a phone call, a video conference, a text message, a mobile application message, or a web application message.

[0196] The advantage of the artificial intelligence module is that it is able to automatically match the subject to a medical service provider (e.g., a specialist, a nurse, a caregiver, a therapist, a counselor, etc.), thereby increasing the likelihood of establishing a solid therapeutic alliance.

[0197] In one embodiment, additional information such as the subject’s medical history (including laboratory test results, imaging studies, and / or genetic markers), the subject’s family history, and / or the subject’s lifestyle factors are taken into account when creating a customized treatment regimen and / or medical service provider association.

[0198] In one specific embodiment, the artificial intelligence module is further configured to:

[0199] - receive at least one textual and / or voice input from the subject;

[0200] - estimate the probability of the subject exhibiting a specific emotional trend according to the following formula:

[0201]

[0202] wherein:

[0203] B is the textual and / or voice input of the subject;

[0204] A is the emotional trend;

[0205] P(A) is the prior probability of the subject exhibiting the emotional trend A;

[0206] P(B) is the prior probability of finding at least one predefined word in the textual and / or voice input of the subject;

[0207] P(B|A) is the conditional probability of the textual and / or voice input of the subject containing at least one predefined word when the subject is exhibiting the emotional trend A;

[0208] P(A|B) is the conditional probability of the subject exhibiting the emotional trend A when the textual and / or voice input of the subject contains at least one predefined word; and

[0209] - compare the estimated probability of the subject exhibiting the specific emotional trend with a pre-set range of probabilities;

[0210] - if the estimated probability is within the pre-set range, then a health crisis is detected;

[0211] - an alert is issued when a health crisis is detected.

[0212] According to embodiments of the present application, the artificial intelligence module of the system is configured to analyze the textual and / or voice input of the subject to obtain the probability of the subject exhibiting a specific emotional trend (e.g., whether the subject is experiencing anxiety, sadness, depression, happiness, anger, stress, or pain) and to detect whether a health crisis is imminent (e.g., whether the patient is experiencing or has a suicidal tendency). If the estimated probability of the subject exhibiting the specific emotional trend is within a pre-defined range of probabilities, then a health crisis is detected; if a health crisis is detected, the artificial intelligence module will issue an alert, e.g., requesting the intervention of an emergency service.

[0213] In one embodiment, the probability range is defined by the medical service provider according to a standard, the system is configured to store the probability range, for example, in an internal database. In one embodiment, the probability range is modified by the medical service provider according to the subject's disease and / or symptoms and / or health status and / or personal parameters.

[0214] When the system issues an alarm, emergency service personnel and / or medical service providers can actively intervene, contact the person concerned, provide support, advice or adjust the treatment plan as needed. This not only enables timely intervention, but also reduces the number of emergency visits.

[0215] In one embodiment, the at least one predefined word is at least one keyword, phrase and / or language pattern.

[0216] In one embodiment, the artificial intelligence module is configured to execute a natural language processing (NLP) algorithm to identify keywords, phrases and / or language patterns in the text and / or voice input, wherein these keywords, phrases and / or language patterns are indicative of different emotional states, such as happiness, sadness, anxiety, depression, anger or stress. With this information, P(A) is estimated as the relative frequency of occurrence of the event "at least one predefined word is found in at least one text and / or voice input" received from the subject.

[0217] In one embodiment, the artificial intelligence module is trained to estimate the probability that the subject follows a specific emotional trend, wherein the training data set contains a plurality of text and / or voice inputs. In one embodiment, P(B) is determined as the relative frequency of occurrence of the event "at least one predefined word is found in the text and / or voice input", wherein the text and / or voice input considered is contained in the training data set. In one embodiment, P(B|A) is determined as the relative frequency of occurrence of the event "at least one predefined word is found in the text and / or voice input", wherein the text and / or voice input considered is contained in a subset of the training data set, which corresponds to subjects following a specific emotional trend A.

[0218] In one embodiment, the artificial intelligence module is configured to execute at least one natural language processing (NLP) algorithm to determine whether at least one word is present in the text and / or voice input in the training data set or in the text and / or voice input received from the subject.

[0219] In one embodiment, the text and / or voice input can be collected from the subject’s social media posts, online activity, recorded conversations with family or friends, and / or telemedicine recorded sessions with a healthcare provider. The text and / or voice input can also be collected directly from the subject’s reports, that is, the subject can provide information to the system about their emotional state, symptoms, stressors, emotional changes, feelings, and / or experiences.

[0220] In one embodiment, the system comprises an artificial intelligence-driven chatbot or virtual mental health assistant that can reach out to the subject when a health crisis is detected. The artificial intelligence-driven chatbot or virtual mental health assistant is configured to guide the subject to appropriate support resources, such as coping strategies, crisis hotlines, or to interface the subject to a mental health professional. For example, when the artificial intelligence module detects that the subject is likely to have a panic attack, the artificial intelligence-driven chatbot or virtual mental health assistant can provide the subject with techniques or methods to alleviate anxiety while looking for a healthcare provider that can reach out to the subject.

[0221] In one specific embodiment, the artificial intelligence module is configured to:

[0222] - receive the subject’s medical history;

[0223] - estimate the probability that the subject follows a particular emotional trend according to the following formula:

[0224]

[0225] where:

[0226] C is the subject’s medical history;

[0227] A is the emotional trend;

[0228] P(A) is the prior probability that the subject follows the emotional trend A;

[0229] P(C) is the prior probability of finding at least one predefined datum in the subject’s medical history;

[0230] P(C|A) is the conditional probability of finding at least one predefined datum in the subject’s medical history when the subject follows the emotional trend A;

[0231] P(A|C) is the conditional probability that the subject follows the emotional trend A when at least one predefined datum is found in the subject’s medical history; and

[0232] - compare the estimated probability that the subject follows a particular emotional trend with a pre-established range of probabilities;

[0233] - if the estimated probability is within a pre-defined range, then it is determined that a health crisis is detected;

[0234] - an alarm is issued when a health crisis is detected.

[0235] According to the present embodiment, the artificial intelligence module of the system disclosed herein also analyzes the medical history of the subject to obtain the probability of the subject presenting a specific emotional trend (e.g. whether the subject has recently experienced a painful episode) and to detect whether a health crisis is imminent (e.g. whether the patient is or has a tendency to commit suicide). If the estimated probability of the subject presenting a specific emotional trend is within a pre-defined probability range, then it is determined that a health crisis is detected; if a health crisis is detected, the artificial intelligence module issues an alarm, e.g. requesting the intervention of the emergency services.

[0236] In one embodiment, the probability range is defined by the medical service provider according to the criteria and the system is configured to store the value of the probability range, e.g. in an internal database. In one embodiment, the probability range is modified by the medical service provider according to the disease and / or symptoms and / or health condition and / or personal parameters of the subject.

[0237] When the system issues an alarm, the medical service provider can proactively contact the patient, provide support, counseling and / or adjust the treatment regimen as needed. This not only enables timely intervention, but also reduces the number of emergency visits.

[0238] In one embodiment, the medical history comprises personal parameters of the subject, and / or health parameters measured over time by the subject, and / or standardized cognitive and behavioral reports generated by caregivers, family members and / or the medical service provider, and / or disease diagnosis reports made by the medical service provider, and / or information provided by the subject to the system over time, including at least one of the following: their emotional state, symptoms, stressors, mood, feelings and experiences.

[0239] In one embodiment, the artificial intelligence module is configured to execute a natural language processing (NLP) algorithm to identify data (keywords, phrases and / or language patterns) in the medical history of the subject, wherein these data are indicative of different emotional states, e.g. happiness, sadness, anxiety, depression, anger or stress. With this information, P(A) is estimated as the relative frequency of the event "at least one pre-defined data is found in the medical history of the subject" to occur.

[0240] In one embodiment, the artificial intelligence module is trained to estimate the probability of the subject following a specific emotional trend, wherein the training dataset contains the medical history of a plurality of subjects. In one embodiment, P(C) is determined as the relative frequency of the event of "finding at least one predefined datum in the medical history of the subject", wherein the medical history considered is contained in the training dataset. In one embodiment, P(C|A) is determined as the relative frequency of the event of "finding at least one predefined datum in the medical history of the subject", wherein the medical history considered is contained in a subset of the training dataset, which subset corresponds to subjects following a specific emotional trend A.

[0241] In one embodiment, the artificial intelligence module is configured to execute at least one natural language processing (NLP) algorithm to determine whether at least one datum is present in the training dataset and / or in the medical history data received from the subject.

[0242] In this context, the "cognitive, behavioral and disease diagnosis report" will be understood as a report generated by a caregiver, a family member and / or a medical service provider summarizing the health status, progress, test results, medical imaging studies, genetic tests, medication and any significant changes in the health status over time of the subject.

[0243] In one embodiment, the system contains an artificial intelligence-driven chatbot or virtual mental health assistant that can contact the subject when a health crisis is detected. These artificial intelligence-driven chatbots or virtual mental health assistants aim to guide the subject to appropriate resources, such as coping strategies, crisis hotlines or contacting mental health professionals. For example, if the artificial intelligence module detects that the subject is about to have a panic attack, the artificial intelligence-driven chatbot or virtual mental health assistant can provide the subject with anti-anxiety techniques while looking for a medical service provider who can contact the subject.

[0244] In one specific embodiment, the artificial intelligence module is further configured to:

[0245] - receive at least one voice input from the subject;

[0246] - estimate the probability of the subject following a specific emotional trend according to the following formula:

[0247]

[0248] wherein:

[0249] D is the voice input of the subject;

[0250] A is the emotional trend;

[0251] P(A) is the prior probability of the subject following the emotional trend A;

[0252] P(D) is the prior probability of finding at least one predefined voice feature in the subject’s voice input;

[0253] P(D|A) is the conditional probability of finding at least one predefined voice feature in the subject’s voice input when the subject is following the emotional trend A;

[0254] P(A|D) is the conditional probability of the subject following the emotional trend A when the subject’s voice input contains at least one predefined voice feature; and

[0255] comparing the estimated probability of the subject following the specific emotional trend to a pre-set range of probabilities;

[0256] - if the estimated probability is within the pre-set range, then determining that a health crisis is detected;

[0257] - issuing an alert when a health crisis is detected.

[0258] In one embodiment, the voice features can be one or more of the following: pitch features, timbre features, tone features, and voice features. The voice features can include specific values of one or more of the pitch, timbre, voice, and / or tone features, as well as patterns of these features.

[0259] In one embodiment, the artificial intelligence module is configured to execute voice recognition and voice processing algorithms to identify voice features (e.g., changes in voice patterns, pitch, or tone) in the subject’s voice input, where the voice features are indicative of different emotional states, such as happiness, sadness, anxiety, depression, anger, or stress. With this information, P(A) is estimated as the relative frequency of the event of finding at least one predefined voice feature in the subject’s voice input.

[0260] In one embodiment, the artificial intelligence module is trained to estimate the probability of the subject following a specific emotional trend, where the training data set contains voice inputs of a plurality of subjects. In one embodiment, P(D) is determined as the relative frequency of the event of finding at least one predefined voice feature in the subject’s voice input, where the considered voice input is contained in the training data set. In one embodiment, P(D|A) is determined as the relative frequency of the event of finding at least one predefined voice feature in the subject’s voice input, where the considered voice input is contained in a subset of the training data set, the subset corresponding to subjects following the specific emotional trend A.

[0261] In one specific embodiment, the system comprises at least one wearable device having at least one sensor configured to:

[0262] - monitoring vital signs of the subject; and / or

[0263] - monitoring activity level of the subject; and / or

[0264] - monitoring sleep pattern of the subject; and / or

[0265] - monitoring heart rate of the subject; and / or

[0266] - monitoring location of the subject; and / or

[0267] - measuring blood pressure of the subject; and / or

[0268] - measuring blood oxygen saturation of the subject; and / or

[0269] - measuring body temperature of the subject; and / or

[0270] - measuring blood glucose level of the subject.

[0271] The system of the present embodiment comprises one or more wearable devices configured to measure and / or monitor one or more health parameters that can vary instantaneously or continuously over time. In the present context, a wearable device is understood to be a device that is directly attached to the skin of a human body or clothing, which contains at least one sensor for collecting biological signals. Fitness trackers (smartwatches, smartbands), smartglasses or smartT-shirts are examples of wearable devices. Sensors contained in wearable devices include biological sensors, electrocardiogram monitors, blood pressure meters and blood glucose meters, among others.

[0272] In the present context, Sp02 refers to blood oxygen saturation.

[0273] In the present embodiment, the sensors of the wearable devices measure or monitor one or more physiological and behavioral parameters that belong to a specific type of health parameter. In other embodiments, the wearable devices can also measure or monitor other health parameters, as long as they meet the requirements of the present application.

[0274] The health parameters provided by the wearable devices enable the artificial intelligence module to make informed decisions about the health status of the subject in real time, thereby enabling the healthcare provider to take timely action when needed.

[0275] In one embodiment, the wearable device for monitoring the location of the subject is based on one of the following technologies: GPS, beacon, Bluetooth or geofencing.

[0276] In one embodiment, the wearable device for monitoring the heart rate of the subject is configured to track the heart rate trend of the subject, the heart rate of the subject at rest and / or the heart rate variation of the subject when performing different activities.

[0277] In one embodiment, the wearable device that monitors the activity level of the subject is configured to track the number of steps taken by the subject over a period of time, the distance travelled by the subject over a period of time and / or the number of calories consumed by the subject over a period of time.

[0278] In one embodiment, the wearable device that monitors the sleep pattern of the subject is configured to track the length of sleep of the subject, the quality of sleep of the subject and / or the number of interruptions during the sleep of the subject.

[0279] In one embodiment, the user registers with the system and the wearable device is configured to pair with the user’s account using the user’s device (e.g. smartphone, smartwatch or tablet).

[0280] In one embodiment, the system is configured to check whether the wearable device is correctly worn on the subject to obtain accurate data. If not, the subject is asked to re-wear the wearable device.

[0281] In one embodiment, the wearable device is configured to transmit the health parameters it measures to the processor of the system in real time or periodically. In one embodiment, the transmission is made through a mobile application or a cloud-based platform.

[0282] In one embodiment, the system is configured to take security measures during transmission and storage to protect the data safety of the health parameter measurements transmitted by the wearable device.

[0283] In one embodiment, the alerts issued by the system include:

[0284] - a medical report to be sent to a medical service provider; and / or

[0285] - an emergency call to emergency medical services; and / or

[0286] - a notification to a medical service provider and / or emergency medical services and / or the subject and / or a contact.

[0287] As mentioned previously, the system of the present application can issue alerts related to the health of the subject; for example, if the system detects that the subject is falling ill, if the system detects that the subject is experiencing a mental health crisis (e.g. suicidal thoughts or a panic attack), if the system detects that the subject has fallen or fainted, or if the system detects that the subject has left a predetermined area.

[0288] The severity of the alerts varies greatly. In one embodiment, the system issues different types of alerts depending on the urgency.

[0289] In one embodiment, when an alarm is urgent (e.g., suicidal ideation or syncope is detected), the system is configured to automatically place an emergency service call. When an alarm is not urgent, the system is configured to call a contact (e.g., a caregiver or family member) and inform the contact of the subject’s health condition (e.g., when a disease is detected to a certain severity). For both types of alarms, the system can also be configured to automatically generate reports related to the patient’s health condition and send these reports to the subject’s healthcare provider.

[0290] The benefit of this is that by prioritizing alarms according to severity and urgency, medical resources can be more efficiently allocated, resulting in faster response times and a higher level of care for the subject.

[0291] In this document, “emergency call” refers to placing an emergency service call. In this document, “notification” refers to a phone call, a mobile push notification, or a message (including a cell phone text message, an email, a mobile app notification, or a web page notification). In this document, “contact” refers to any family member, caregiver, healthcare provider, or any other person designated by the subject to be contacted upon issuance of a particular alarm.

[0292] In one embodiment, when an alarm is issued, the system is configured to automatically initiate an emergency call to emergency services. If there is no response to the initial emergency call, the system is also configured to send a notification to a contact or to initiate a new emergency call and send a notification to a contact.

[0293] In one embodiment, the system includes means for tracking responses to alarms and subsequent actions.

[0294] In one embodiment, the system includes communication means configured to allow a caregiver, family member, or healthcare provider to communicate with the subject at any time, especially after an alarm is issued, through text messages, voice conference, or video conference. Examples of communication means include a cell phone, a tablet, or a personal computer.

[0295] In one embodiment, the communication means are further configured to allow communication between caregivers, family members, and / or healthcare providers to share updates and concerns.

[0296] In one embodiment, the system includes a customization feature configured to allow a healthcare provider to change alarm settings according to the needs, conditions, and preferences of individual subjects.

[0297] In one embodiment, the plurality of disease models includes a heart disease model generated by the artificial intelligence module using one or more of the following parameters of a subject with heart disease:

[0298] - age,

[0299] - the type of chest pain experienced by the subject,

[0300] - the blood pressure level at rest,

[0301] - serum cholesterol,

[0302] - the maximum heart rate experienced by the subject,

[0303] - exercise-induced angina, and

[0304] - ST segment depression induced by exercise compared to the resting state;

[0305] wherein:

[0306] o the activation function of each node in the first hidden layer is defined by the following equation:

[0307] f1(x) = max(0, x)

[0308] where x represents the weighted sum of the inputs to each node in the first hidden layer;

[0309] o the activation function of each node in the second hidden layer is defined by the following equation:

[0310] f2(x) = x • tanh(ln(1 + e x ))

[0311] where x represents the weighted sum of the inputs to each node in the second hidden layer; and

[0312] o the activation function of the third hidden layer is defined by the following equation:

[0313]

[0314] where x i represents the weighted sum of the inputs to the i-th node in the third hidden layer, and K is the total number of nodes in the third hidden layer.

[0315] In one embodiment, the artificial intelligence module generates the heart disease model using a plurality of the above-mentioned parameters (i.e. a plurality of parameters selected from the parameters of age, the type of chest pain experienced by the subject, the blood pressure level at rest, serum cholesterol, the maximum heart rate experienced by the subject, exercise-induced angina, and ST segment depression induced by exercise compared to the resting state). Preferably, all of the above-mentioned parameters (i.e. age, the type of chest pain experienced by the subject, the blood pressure level at rest, serum cholesterol, the maximum heart rate experienced by the subject, exercise-induced angina, and ST segment depression induced by exercise compared to the resting state) are used to generate the heart disease model.

[0316] Cardiac disease refers to a range of diseases affecting the heart, including vascular diseases (such as coronary artery disease), arrhythmias, congenital heart defects, myocardial diseases, and heart valve diseases.

[0317] The cardiac disease model is generated by an artificial intelligence module and is built based on a neural network. When the neural network receives input parameters at the input layer, it begins forward propagation.

[0318] In the three hidden layers of the neural network, a pre-activation and activation process takes place. The pre-activation function is a calculation of the weighted sum of the inputs to the hidden layer. The activation function acts on this weighted sum by introducing a bias, causing the data flow of the neural network to exhibit nonlinear characteristics.

[0319] The first activation function used in the first hidden layer is a simple mathematical function applied to the output of the input layer. Its purpose is to introduce nonlinearity, enabling the network to learn complex patterns and relationships in the data. If the input value is positive, the function outputs the value; if the input value is negative, it outputs zero.

[0320] The second activation function used in the second hidden layer is a complex mathematical function applied to the output of the first hidden layer. Its purpose is to introduce nonlinearity, enabling the network to learn complex patterns and relationships in the data.

[0321] The third activation function used in the third hidden layer is a complex mathematical function applied to the output of the second hidden layer. Its purpose is to introduce nonlinearity, enabling the network to learn complex patterns and relationships in the data.

[0322] The output layer has only one node, representing the probability of the subject having a cardiac disease, and its input is the output of the third hidden layer.

[0323] In one embodiment, the artificial intelligence module, after 100 rounds of training, achieved an accuracy of over 90% in subsequent artificial intelligence predictions.

[0324] In this document, "ST segment depression" refers to the phenomenon of the ST segment in the electrocardiogram (ECG) of a subject appearing depressed or flat relative to the baseline of the electrocardiogram. This parameter indicates myocardial ischemia, i.e., insufficient blood supply to the myocardium of the subject. Myocardial ischemia is often associated with coronary artery disease (CAD) or other cardiac diseases.

[0325] In one embodiment, the measurement units of the parameters are:

[0326] - age is calculated in years, and / or

[0327] - blood pressure levels in the resting state are measured in millimeters of mercury (mm / Hg), and / or

[0328] - the serum cholesterol is expressed in milligrams per deciliter (mg / dl).

[0329] In one embodiment, the artificial intelligence module uses one or more of the aforementioned parameters (i.e. age, type of chest pain of the subject, blood pressure level at rest, serum cholesterol, maximum heart rate of the subject, exercise-induced angina pectoris, and exercise-induced ST segment depression compared to the resting state) as inputs of a heart disease model to determine the probability of the subject having a heart disease.

[0330] In one embodiment, the artificial intelligence module uses a plurality of the aforementioned parameters (i.e. a plurality of parameters selected from age, type of chest pain of the subject, blood pressure level at rest, serum cholesterol, maximum heart rate of the subject, exercise-induced angina pectoris, and exercise-induced ST segment depression compared to the resting state) in the input layer. Preferably, the artificial intelligence module uses all the aforementioned parameters (i.e. age, type of chest pain of the subject, blood pressure level at rest, serum cholesterol, maximum heart rate of the subject, exercise-induced angina pectoris, and exercise-induced ST segment depression compared to the resting state).

[0331] In one embodiment, the plurality of disease severity models comprises a heart disease severity model generated by the artificial intelligence module using the same parameters as the ones used by the artificial intelligence module to generate the heart disease model. In addition, the activation functions of the three hidden layers of the heart disease severity model are the same as the activation functions of the heart disease model.

[0332] In one embodiment, the artificial intelligence module uses a plurality of the aforementioned parameters (i.e. a plurality of parameters selected from age, type of chest pain of the subject, blood pressure level at rest, serum cholesterol, maximum heart rate of the subject, exercise-induced angina pectoris, and exercise-induced ST segment depression compared to the resting state) in the input layer of the heart disease severity model. Preferably, the artificial intelligence module uses all the aforementioned parameters (i.e. age, type of chest pain of the subject, blood pressure level at rest, serum cholesterol, maximum heart rate of the subject, exercise-induced angina pectoris, and exercise-induced ST segment depression compared to the resting state).

[0333] In one embodiment, the plurality of disease models comprises a dementia model generated by the artificial intelligence module using one or more of the following parameters of a subject having dementia:

[0334] - the subject has diabetes,

[0335] - the subject has depression,

[0336] - the subject carries the APOE ε3 allele,

[0337] - the oxygen content in the blood,

[0338] - heart rate,

[0339] - age,

[0340] - the time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination being performed, or the time elapsed between the magnetic resonance imaging examination being performed and the subject wearing the wearable device,

[0341] - alcohol content in the blood,

[0342] - body temperature, and

[0343] - weight;

[0344] wherein:

[0345] o the activation function of each node in the first hidden layer is defined by the following equation:

[0346] f1(x) = max(0, x)

[0347] where x represents the weighted sum of the inputs of each node in the first hidden layer;

[0348] o the activation function of each node in the second hidden layer is defined by the following equation:

[0349] f2(x) = x · tanh(ln(1 + e x ))

[0350] where x represents the weighted sum of the inputs of each node in the second hidden layer; and

[0351] o the activation function of the third hidden layer is defined by the following equation:

[0352] f3(x) = ln(1 + e x )

[0353] where x represents the weighted sum of the inputs of each node in the third hidden layer.

[0354] In one embodiment, the dementia model is generated by an artificial intelligence module that uses a plurality of the aforementioned parameters (i.e. subjects with diabetes, subjects with depression, subjects carrying the APOE e3 allele, blood oxygen level, heart rate, age, time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device, alcohol content in the blood, body temperature and weight) as inputs for the dementia model. Preferably, all the aforementioned parameters (i.e. subjects with diabetes, subjects with depression, subjects carrying the APOE e3 allele, blood oxygen level, heart rate, age, time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device, alcohol content in the blood, body temperature and weight) are used to generate the dementia model.

[0355] Dementia is not a disease in itself, but a set of symptoms caused by brain damage from different diseases (e.g. Alzheimer's disease). Dementia is a progressive disease that affects cognitive and functional abilities. These symptoms vary depending on the location of the brain damage and can include:

[0356] Memory loss; and / or

[0357] Difficulty concentrating; and / or

[0358] Difficulty completing familiar daily tasks, such as finding the right change when shopping; and / or

[0359] Difficulty following a conversation or finding the right words; and / or

[0360] Confusion about time and place; and / or

[0361] Emotional changes.

[0362] The dementia model is generated by an artificial intelligence module and is built based on a neural network. When the neural network receives the input parameters at the input layer, it begins forward propagation.

[0363] In the three hidden layers of the neural network, the pre-activation and activation processes are performed. The pre-activation function is a calculation of the weighted sum of the inputs of the hidden layer. The activation function acts on this weighted sum by introducing a bias, making the data flow of the neural network exhibit nonlinear characteristics.

[0364] The first activation function used in the first hidden layer is a simple mathematical function applied to the output of the input layer. Its purpose is to introduce nonlinearity, enabling the network to learn complex patterns and relationships in the data. If the input value is positive, the function outputs that value; if the input value is negative, it outputs zero.

[0365] The second activation function used by the second hidden layer is a complex mathematical function applied to the output of the first hidden layer. The purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data.

[0366] The third activation function used by the third hidden layer is a complex mathematical function applied to the output of the second hidden layer. The purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data.

[0367] The output layer has only one node, representing the probability of the subject having dementia, whose input is the output of the third hidden layer.

[0368] In one embodiment, the artificial intelligence module, after 100 rounds of training, achieved over 95% accuracy in subsequent artificial intelligence predictions.

[0369] In this document, "APOE ε3 allele" refers to a specific variant of the apolipoprotein ε (APOE) gene, which plays a crucial role in lipid metabolism and cholesterol transport in the body. The presence or absence of this gene in a subject can be determined through genetic testing.

[0370] In one embodiment, the units of measurement of the parameters are:

[0371] - the unit of measurement of heart rate is the number of heartbeats per minute (bpm), and / or

[0372] - age is calculated in years, and / or

[0373] - alcohol concentration in the blood is expressed in grams per deciliter (g / dl) or milligrams per deciliter (mg / dl), and / or

[0374] - the unit of measurement of body temperature is degrees Celsius or Fahrenheit, and / or

[0375] - the unit of weight is kilograms (kg) or pounds (lbs).

[0376] In one embodiment, the artificial intelligence module uses one or more of the aforementioned parameters (i.e., subjects with diabetes, subjects with depression, subjects carrying the APOE ε3 allele, blood oxygen level, heart rate, age, the time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or the time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device, alcohol content in the blood, body temperature, and weight) as inputs to the dementia model to determine the probability of the subject having dementia.

[0377] In one embodiment, the artificial intelligence module uses in the input layer of the dementia severity model a plurality of the aforementioned parameters (i.e. a plurality of parameters selected from the following: subjects with diabetes, subjects with depression, subjects carrying the APOE ε3 allele, blood oxygen level, heart rate, age, time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device, alcohol content in blood, body temperature and weight). Preferably, the artificial intelligence module uses all the aforementioned parameters (i.e. subjects with diabetes, subjects with depression, subjects carrying the APOE ε3 allele, blood oxygen level, heart rate, age, time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device, alcohol content in blood, body temperature and weight).

[0378] In one embodiment, the artificial intelligence module is further trained to generate a plurality of disease severity models, including a dementia severity model, which is generated by the artificial intelligence module using the same parameters as for generating the dementia model. Furthermore, the activation functions of the three hidden layers of the dementia severity model are the same as those of the dementia model.

[0379] In one embodiment, the artificial intelligence module uses in the input layer of the dementia severity model a plurality of the aforementioned parameters (i.e. a plurality of parameters selected from the following: subjects with diabetes, subjects with depression, subjects carrying the APOE ε3 allele, blood oxygen level, heart rate, age, time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device, alcohol content in blood, body temperature and weight). Preferably, the artificial intelligence module uses all the aforementioned parameters (i.e. subjects with diabetes, subjects with depression, subjects carrying the APOE ε3 allele, blood oxygen level, heart rate, age, time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device, alcohol content in blood, body temperature and weight).

[0380] In one embodiment, the plurality of disease models includes a schizophrenia disease model generated by the artificial intelligence module using one or more of the following parameters of subjects with schizophrenia:

[0381] - sleep pattern,

[0382] - walking / step movement pattern, and

[0383] - heart rate;

[0384] wherein:

[0385] o the activation function of each node in the first hidden layer is defined by the following equation:

[0386] f1(x) = max(0, x)

[0387] where x represents the weighted sum of the inputs to each node in the first hidden layer;

[0388] o the activation function of each node in the second hidden layer is defined by the following equation:

[0389] f2(x) = max(0, x)

[0390] where x represents the weighted sum of the inputs to each node in the second hidden layer; and

[0391] o the activation function of the third hidden layer is defined by the following equation:

[0392]

[0393] where x i represents the weighted sum of the inputs to the i-th node in the third hidden layer, and K is the total number of nodes in the third hidden layer.

[0394] In one embodiment, the schizophrenia disease model is generated by the artificial intelligence module using a plurality of the aforementioned parameters (i.e. a plurality of parameters selected from sleep patterns, walking / step movement patterns, and heart rate). Preferably, all of the aforementioned parameters (i.e. sleep patterns, walking / step movement patterns, and heart rate) are used to generate the schizophrenia disease model.

[0395] Schizophrenia is a mental illness that affects the way a person thinks, feels, and behaves. The disease can develop slowly. It is difficult to recognize because early symptoms usually appear during adolescence.

[0396] The symptoms of schizophrenia are usually divided into the following types:

[0397] Positive symptoms: any change in behavior or thought, such as hallucinations or delusions;

[0398] Negative symptoms: the subject appears to be isolated from the world around them, is not interested in daily social interactions, and often appears cold and numb.

[0399] The schizophrenia disease model is generated by the artificial intelligence module and is constructed based on a neural network. When the neural network receives input parameters at the input layer, it begins forward propagation.

[0400] In the three hidden layers of the neural network, a pre-activation and activation process will take place. The pre-activation function is a calculation of the weighted sum of the inputs to the hidden layer. The activation function acts on this weighted sum by introducing a bias, making the data flow of the neural network exhibit non-linear characteristics.

[0401] The first activation function used in the first hidden layer is a simple mathematical function applied to the output of the input layer. The purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data. If the input value is positive, the function outputs the value; if the input value is negative, it outputs zero.

[0402] The second activation function used in the second hidden layer is a simple mathematical function applied to the output of the first hidden layer. The purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data. If the input value is positive, the function outputs the value; if the input value is negative, it outputs zero.

[0403] The third activation function used in the third hidden layer is a complex mathematical function applied to the output of the second hidden layer. The purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data.

[0404] The output layer has only one node, representing the probability of the subject having schizophrenia, whose input is the output of the third hidden layer.

[0405] In one embodiment, the artificial intelligence module, after 100 rounds of training, achieved an accuracy of over 90% in subsequent artificial intelligence predictions.

[0406] In one embodiment, the measurement unit of heart rate is the number of heartbeats per minute (bpm).

[0407] In one embodiment, the artificial intelligence module uses one or more of the aforementioned parameters (i.e., sleep pattern, walking / step movement pattern, and heart rate) as inputs to the schizophrenia model to determine the probability of the subject having schizophrenia.

[0408] In one embodiment, the artificial intelligence module uses multiple aforementioned parameters (i.e., sleep pattern, walking / step movement pattern, and heart rate) in the input layer. Preferably, the artificial intelligence module uses all the aforementioned parameters (i.e., sleep pattern, walking / step movement pattern, and heart rate).

[0409] In one embodiment, the artificial intelligence module is further trained to generate a plurality of disease severity models, including a schizophrenia disease severity model, which is generated by the artificial intelligence module using the same parameters as the schizophrenia disease model. Furthermore, the activation functions of the three hidden layers of the schizophrenia disease severity model are the same as the activation functions of the schizophrenia disease model.

[0410] In one embodiment, the artificial intelligence module uses a plurality of the aforementioned parameters (i.e. sleep pattern, walking / step motion pattern and heart rate) in the input layer of the schizophrenia disease severity model. Preferably, the artificial intelligence module uses all of the aforementioned parameters (i.e. sleep pattern, walking / step motion pattern and heart rate).

[0411] In one embodiment, the plurality of disease models includes an Alzheimer’s disease model generated by the artificial intelligence module using one or more of the following parameters of a subject suffering from Alzheimer’s disease:

[0412] - a clinical dementia rating for indicating the severity of dementia symptoms,

[0413] - a mini-mental state examination for indicating the level of cognitive function,

[0414] - a normalized total brain volume,

[0415] - an estimated total intracranial volume,

[0416] - a template scaling factor, and

[0417] - an age,

[0418] wherein:

[0419] o the activation function of each node in the first hidden layer is defined by the following equation:

[0420] f1(x) = max(0, x)

[0421] where x represents the weighted sum of the inputs of each node in the first hidden layer;

[0422] o the activation function of each node in the second hidden layer is defined by the following equation:

[0423] f2(x) = max(0, x)

[0424] where x represents the weighted sum of the inputs of each node in the second hidden layer; and

[0425] o the activation function of the third hidden layer is defined by the following equation:

[0426]

[0427] where x i represents the weighted sum of inputs to the i-th node in the third hidden layer, K is the total number of nodes in the third hidden layer.

[0428] In one embodiment, the Alzheimer's disease model is generated by the artificial intelligence module using a plurality of the aforementioned parameters (i.e. a plurality of parameters selected from the clinical dementia rating, the mini-mental state examination, the normalized total brain volume, the estimated intracranial total volume, the atlas scaling factor and the age). Preferably, all of the aforementioned parameters (i.e. the clinical dementia rating, the mini-mental state examination, the normalized total brain volume, the estimated intracranial total volume, the atlas scaling factor and the age) are used to generate the Alzheimer's disease model.

[0429] Alzheimer's disease is a progressive disease, with symptoms of dementia gradually worsening over years. In the early stages, memory loss is mild, but in the late stages, patients lose the ability to converse and react to their surroundings.

[0430] The Alzheimer's disease model is generated by the artificial intelligence module, which is based on a neural network construction. When the neural network receives input parameters at the input layer, it begins forward propagation.

[0431] In the three hidden layers of the neural network, a pre-activation and activation process is performed. The pre-activation function is a mathematical function that is applied to the weighted sum of inputs to the hidden layer. The activation function is applied to this weighted sum, introducing a bias that makes the data flow of the neural network exhibit non-linear characteristics.

[0432] The first activation function used in the first hidden layer is a simple mathematical function applied to the output of the input layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data. If the input value is positive, the function outputs the value; if the input value is negative, it outputs zero.

[0433] The second activation function used in the second hidden layer is a simple mathematical function applied to the output of the first hidden layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data. If the input value is positive, the function outputs the value; if the input value is negative, it outputs zero.

[0434] The third activation function used in the third hidden layer is a complex mathematical function applied to the output of the second hidden layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data.

[0435] The output layer has only one node, representing the probability of the subject having Alzheimer's disease, whose input is the output of the third hidden layer.

[0436] In one embodiment, the artificial intelligence module, after 100 rounds of training, obtained an accuracy of over 94% in subsequent artificial intelligence predictions.

[0437] In the present document, the "mini-mental state examination" is a parameter for measuring the level of cognitive function and is presented in the form of a scale. This scale ranges from 0 to 30 points, with a higher score indicating better cognitive function in the subject.

[0438] In the present document, the "clinical dementia rating" is a parameter for expressing the severity of dementia symptoms using a pre-established scale. Preferably, this scale ranges from 0 to 3, with a higher score indicating more severe dementia symptoms.

[0439] In one embodiment, the parameters are measured in units of:

[0440] - the normalized total brain volume is measured in cubic centimeters (cc) or milliliters (ml), and / or

[0441] - the estimated total intracranial volume is measured in cubic centimeters (cc) or milliliters (ml), and / or

[0442] - the age is measured in years.

[0443] In one embodiment, the parameters normalized total brain volume, estimated total intracranial volume and atlas scaling factor are based on X-ray studies of the subject.

[0444] In one embodiment, the artificial intelligence module uses one or more of the aforementioned parameters (i.e. clinical dementia rating, mini-mental state examination, normalized total brain volume, estimated total intracranial volume, atlas scaling factor and age) as inputs to an Alzheimer's disease model to determine the probability of the subject having Alzheimer's disease.

[0445] In one embodiment, the artificial intelligence module uses a plurality of the aforementioned parameters (i.e. clinical dementia rating, mini-mental state examination, normalized total brain volume, estimated total intracranial volume, atlas scaling factor and age) in the input layer. Preferably, the artificial intelligence module uses all of the aforementioned parameters (i.e. clinical dementia rating, mini-mental state examination, normalized total brain volume, estimated total intracranial volume, atlas scaling factor and age).

[0446] In one embodiment, the artificial intelligence module is further trained to generate a plurality of disease severity models, including an Alzheimer's disease severity model generated by the artificial intelligence module using the same parameters as the Alzheimer's disease model. In addition, the activation functions of the three hidden layers of the Alzheimer's disease severity model are the same as those of the Alzheimer's disease model.

[0447] In one embodiment, the artificial intelligence module uses a plurality of the aforementioned parameters (i.e. clinical dementia rating, mini-mental state examination, normalized total brain volume, estimated total intracranial volume, atlas scaling factor, and age) in the input layer of the Alzheimer’s disease severity model. Preferably, the artificial intelligence module uses all of the aforementioned parameters (i.e. clinical dementia rating, mini-mental state examination, normalized total brain volume, estimated total intracranial volume, atlas scaling factor, and age).

[0448] In one embodiment, the plurality of disease models includes an anxiety model and / or a depression model generated by the artificial intelligence module using one or more of the following parameters of a subject having anxiety and / or depression:

[0449] - heart rate variability;

[0450] - galvanic skin activity;

[0451] - respiratory rate and breathing pattern;

[0452] - blood pressure levels;

[0453] - activity level;

[0454] - sleep pattern;

[0455] - social interaction pattern;

[0456] - history of drug abuse;

[0457] - self-reported anxiety and / or depression symptoms;

[0458] - mental health diagnosis or treatment history;

[0459] wherein:

[0460] o the activation function of each node in the first hidden layer is defined by the following equation:

[0461] f1(x) = max(0, x)

[0462] where x represents the weighted sum of the inputs to each node in the first hidden layer;

[0463] o the activation function of each node in the second hidden layer is defined by the following equation:

[0464]

[0465] where x represents the weighted sum of the inputs to each node in the second hidden layer; and

[0466] o the activation function of the third hidden layer is defined by the following equation:

[0467]

[0468] where x represents the weighted sum of inputs to each node in the third hidden layer.

[0469] In one embodiment, the anxiety and / or depression model is generated by the artificial intelligence module using a plurality of the aforementioned parameters (i.e. a plurality of parameters selected from heart rate variability, skin electrical activity, respiratory rate and pattern, blood pressure level, activity level, sleep pattern, social interaction pattern, substance abuse history, self-reported anxiety and / or depressive symptoms, and mental health diagnosis or treatment history). Preferably, all of the aforementioned parameters (i.e. heart rate variability, skin electrical activity, respiratory rate and pattern, blood pressure level, activity level, sleep pattern, social interaction pattern, substance abuse history, self-reported anxiety and / or depressive symptoms, and mental health diagnosis or treatment history) are used to generate the anxiety and / or depression model.

[0470] Depression (also known as major depression, major depressive disorder, or clinical depression) is a common but serious mood disorder. It causes severe symptoms that affect how a person feels, thinks, and handles daily activities, such as sleeping, eating, or working.

[0471] Anxiety is more than temporary worry or fear. For a person with an anxiety disorder, the feeling is often intense. It can interfere with daily activities such as job performance, schoolwork, and personal relationships.

[0472] In one embodiment, the measurement units of the parameters are:

[0473] - the unit of heart rate variability is beats per minute (bpm), and / or

[0474] - the unit of respiratory rate is breaths per minute, and / or

[0475] - the unit of blood pressure level is millimeters of mercury (mm / HG).

[0476] The depression model and / or anxiety model is generated by the artificial intelligence module, which is based on a neural network construction. When the neural network receives input parameters at the input layer, it begins forward propagation.

[0477] In the three hidden layers of the neural network, pre-activation and activation processes are performed. The pre-activation function is a calculation on the weighted sum of the inputs to the hidden layer. The activation function acts on this weighted sum by introducing a bias, which makes the data flow of the neural network exhibit nonlinear characteristics.

[0478] The first activation function used by the first hidden layer is a simple mathematical function applied to the output of the input layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data. If the input value is positive, the function outputs that value; if the input value is negative, it outputs zero.

[0479] The second activation function used by the second hidden layer is a complex mathematical function applied to the output of the first hidden layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data.

[0480] The third activation function used by the third hidden layer is a complex mathematical function applied to the output of the second hidden layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data.

[0481] The output layer has only one node, representing the probability of the subject having depression or anxiety, and its input is the output of the third hidden layer.

[0482] In one embodiment, the artificial intelligence module, after 100 rounds of training, achieved over 95% accuracy in subsequent artificial intelligence predictions.

[0483] In this context, "social interaction patterns" will be understood as a parameter that represents the way the subject interacts with others in social settings, both face-to-face and through digital communication channels.

[0484] In one embodiment, the social interaction patterns include interaction frequency, i.e., the frequency with which the subject engages in social interactions with others. In one embodiment, this frequency can range from frequent interactions with a large number of contacts to sporadic or infrequent interactions with a smaller group of close relationships.

[0485] In one embodiment, the social interaction patterns include interaction types, such as face-to-face communication, phone calls, text messages, emails, social media interactions, and participation in group activities or events.

[0486] In one embodiment, the social interaction patterns include communication styles, such as assertive, passive, or aggressive communication styles. In one embodiment, communication styles vary depending on cultural norms, personal preferences, and the nature of the relationship.

[0487] In one embodiment, the social interaction patterns include relationship quality. In one embodiment, the determination of relationship quality takes into account at least one of the following factors: trust, mutual respect, empathy, and support in the relationship.

[0488] In one embodiment, the social interaction pattern includes a social network structure of the subject, which is determined according to at least one of the following parameters: network size, connection density, and connection diversity. For example, some subjects can have a broad and diverse social network, while others can have a smaller network but with closer connections.

[0489] In one embodiment, the social interaction pattern includes a digital interaction pattern of the subject, which is determined according to at least one of the following parameters: social media engagement, online message exchange, video call, and participation in virtual communities or forums.

[0490] In this context, the “activity level” refers to the amount and / or intensity of physical activity and / or exercise performed by the subject within a certain period of time. For example, exercise, walking, running, cycling, swimming, and participation in sports or leisure activities, etc. fall within the scope of activity level.

[0491] In one embodiment, the artificial intelligence module uses one or more of the aforementioned parameters (i.e. heart rate variability, skin conductance activity, respiratory rate and pattern, blood pressure level, activity level, sleep pattern, social interaction pattern, history of drug abuse, self-reported symptoms of anxiety and / or depression, and history of mental health diagnosis or treatment) as inputs to the anxiety and / or depression model to determine the probability of the subject having anxiety and / or depression.

[0492] In one embodiment, the artificial intelligence module uses a plurality of the aforementioned parameters (i.e. heart rate variability, skin conductance activity, respiratory rate and pattern, blood pressure level, activity level, sleep pattern, social interaction pattern, history of drug abuse, self-reported symptoms of anxiety and / or depression, and history of mental health diagnosis or treatment) at the input layer. Preferably, the artificial intelligence module uses all of the aforementioned parameters (i.e. heart rate variability, skin conductance activity, respiratory rate and pattern, blood pressure level, activity level, sleep pattern, social interaction pattern, history of drug abuse, self-reported symptoms of anxiety and / or depression, and history of mental health diagnosis or treatment).

[0493] In one embodiment, the artificial intelligence module is further trained to generate a plurality of disease severity models, including an anxiety severity model and / or a depression severity model, which are generated by the artificial intelligence module using the same parameters as the anxiety model and / or the depression model. In addition, the activation functions of the three hidden layers of the anxiety severity model and / or the depression severity model are the same as those of the anxiety model and / or the depression model, respectively.

[0494] In one embodiment, the artificial intelligence module uses a plurality of the above parameters (i.e. heart rate variability, electrodermal activity, respiratory rate and breathing pattern, blood pressure level, activity level, sleep pattern, social interaction pattern, history of drug abuse, self-reported symptoms of anxiety and / or depression, and history of mental health diagnosis or treatment) in the input layer of the anxiety severity model and / or the depression severity model. Preferably, the artificial intelligence module uses all of the above parameters (i.e. heart rate variability, electrodermal activity, respiratory rate and breathing pattern, blood pressure level, activity level, sleep pattern, social interaction pattern, history of drug abuse, self-reported symptoms of anxiety and / or depression, and history of mental health diagnosis or treatment).

[0495] In one embodiment, the plurality of disease models includes an eating disorder model generated by the artificial intelligence module using one or more of the following parameters of a subject having an eating disorder:

[0496] - age,

[0497] - gender,

[0498] - body mass index,

[0499] - number of weekly binge eating episodes,

[0500] - number of weekly purging episodes,

[0501] - body dissatisfaction score, for indicating a self-reported level of body dissatisfaction,

[0502] - eating disorder inventory score, for indicating a score of eating disorder symptoms according to a standard scale,

[0503] - depression score, for indicating a score of a depression assessment scale,

[0504] - anxiety score, for indicating a score of an anxiety assessment scale, and

[0505] - eating disorder diagnosis;

[0506] wherein:

[0507] o the activation function of each node in the first hidden layer is defined by the following equation:

[0508] f1(x) = max(0, x)

[0509] where x represents the weighted sum of the input to each node in the first hidden layer;

[0510] o the activation function of each node in the second hidden layer is defined by the following equation:

[0511]

[0512] where x represents the weighted sum of inputs to each node in the second hidden layer; and

[0513] The activation function of the third hidden layer is defined by the following equation:

[0514]

[0515] where x represents the weighted sum of inputs to each node in the third hidden layer.

[0516] In one embodiment, the eating disorder model is generated by the artificial intelligence module using a plurality of the aforementioned parameters (i.e., a plurality of parameters selected from age, gender, body mass index, number of weekly binge eating episodes, number of weekly purging behavior episodes, body dissatisfaction score, eating disorder inventory score, depression score, anxiety score, and eating disorder diagnosis). Preferably, all of the aforementioned parameters (i.e., age, gender, body mass index, number of weekly binge eating episodes, number of weekly purging behavior episodes, body dissatisfaction score, eating disorder inventory score, depression score, anxiety score, and eating disorder diagnosis) are used to generate the eating disorder model.

[0517] Eating disorder is a mental health illness in which a subject copes with emotional and other life situations by controlling food intake.

[0518] The eating disorder model is generated by the artificial intelligence module and is based on a neural network construction. When the neural network receives input parameters at the input layer, it begins forward propagation.

[0519] In the three hidden layers of the neural network, a pre-activation and activation process is performed. The pre-activation function is a calculation of the weighted sum of the inputs to the hidden layer. The activation function acts on this weighted sum by introducing a bias, causing the data flow of the neural network to exhibit non-linear characteristics.

[0520] The first activation function used by the first hidden layer is a simple mathematical function applied to the output of the input layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data. If the input value is positive, the function outputs the value; if the input value is negative, it outputs zero.

[0521] The second activation function used by the second hidden layer is a complex mathematical function applied to the output of the first hidden layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data.

[0522] The third activation function used by the third hidden layer is a complex mathematical function applied to the output of the second hidden layer. Its purpose is to introduce non-linearity, enabling the network to learn complex patterns and relationships in the data.

[0523] The output layer has only one node, representing the probability of the subject having an eating disorder, whose input is the output of the third hidden layer.

[0524] In one embodiment, the artificial intelligence module, after 100 rounds of training, obtained more than 96% accuracy in subsequent artificial intelligence predictions.

[0525] In one embodiment, the units of measurement of the parameters are:

[0526] - the unit of age is years, and / or

[0527] - the unit of body mass index is kilograms per square meter kg / m 2 .

[0528] In the present text, the "body dissatisfaction score" refers to the score obtained by assessing the degree of dissatisfaction with their body through a self-report questionnaire.

[0529] In the present text, the "eating disorder inventory questionnaire score" refers to the score obtained from the eating disorder inventory questionnaire, which is a standardized scale for assessing symptoms of eating disorders.

[0530] In the present text, the "depression score" refers to the score reported by the subject according to a depression assessment scale, such as the Beck Depression Inventory.

[0531] In the present text, the "anxiety score" refers to the score reported by the subject according to an anxiety assessment scale; for example, the 7-item Generalized Anxiety Disorder scale.

[0532] In the present text, the "eating disorder diagnosis" will be understood as a parameter that indicates whether the subject has a mental health problem and / or obsessive-compulsive disorder and / or anxiety.

[0533] In one embodiment, the artificial intelligence module uses one or more of the aforementioned parameters (i.e. age, gender, body mass index, number of episodes of binge eating per week, number of episodes of purging behavior per week, body dissatisfaction score, eating disorder inventory questionnaire score, depression score, anxiety score and eating disorder diagnosis) as input to the eating disorder model to determine the probability of the subject having an eating disorder.

[0534] In one embodiment, the artificial intelligence module uses a plurality of the aforementioned parameters (i.e. age, gender, body mass index, number of episodes of binge eating per week, number of episodes of purging behavior per week, body dissatisfaction score, eating disorder inventory questionnaire score, depression score, anxiety score and eating disorder diagnosis) in the input layer. Preferably, the artificial intelligence module uses all the aforementioned parameters (i.e. age, gender, body mass index, number of episodes of binge eating per week, number of episodes of purging behavior per week, body dissatisfaction score, eating disorder inventory questionnaire score, depression score, anxiety score and eating disorder diagnosis).

[0535] In one embodiment, the artificial intelligence module is further trained to generate a plurality of disease severity models, including an eating disorder severity model generated by the artificial intelligence module using the same parameters as the eating disorder model. Moreover, the activation functions of the three hidden layers of the eating disorder severity model are the same as the activation functions of the eating disorder model.

[0536] In one embodiment, the artificial intelligence module uses the plurality of aforementioned parameters (i.e., age, gender, body mass index, number of weekly binge eating episodes, number of weekly purging behavior episodes, body dissatisfaction score, eating disorder inventory score, depression score, anxiety score, and eating disorder diagnosis) in the input layer of the eating disorder severity model. Preferably, the artificial intelligence module uses all of the aforementioned parameters (i.e., age, gender, body mass index, number of weekly binge eating episodes, number of weekly purging behavior episodes, body dissatisfaction score, eating disorder inventory score, depression score, anxiety score, and eating disorder diagnosis).

[0537] All features described in this application, including claims, descriptions and drawings, can be combined, except where they are mutually exclusive. BRIEF DESCRIPTION OF DRAWINGS

[0538] These and other features and advantages of the present application will become apparent from a consideration of the preferred embodiments of the application as described in detail in the specification, including the claims, and the drawings. The preferred embodiments of the application are given by way of example only, and are not intended to limit the scope of the application.

[0539] Figure 1a A system according to one embodiment of the application is shown.

[0540] Figure 1b A subject is shown wearing a plurality of wearable devices containing sensors.

[0541] Figure 2 A neural network on which a disease model is based is shown.

[0542] Figures 3a-3e A decision tree used to select health and personal parameters prior to training of the artificial intelligence module is shown ( Figure 3a ) as well as an example of several selected parameters for different disease models ( Figures 3b-3e ).

[0543] Figures 4a-4b An example of a table used to train the artificial intelligence module using medical database data is shown. DETAILED DESCRIPTION

[0544] Figure 1aA system 1 for issuing an alert related to a health condition of a subject 2 is schematically illustrated according to an embodiment of the application, the system 1 comprising a processor 3 having an artificial intelligence (AI) module 4.

[0545] The processor 3 is configured to:

[0546] - receive at least one health parameter of the subject, wherein the one or more health parameters are measured by at least one sensor 8.1, and

[0547] - receive at least one personal parameter of the subject.

[0548] Examples of health parameters include a blood pressure of the subject 2, a blood sugar level of the subject 2, a sleep pattern of the subject 2, or a body temperature of the subject 2. Examples of personal parameters include a gender or an age of the subject 2.

[0549] In an embodiment, the at least one sensor 8.1 is part of the system 1, in particular located inside a smartwatch 8, which is a particular type of wearable device 8. In other embodiments, the sensor 8.1 is not part of the system 1. In other embodiments, the sensor 8.1 is not located inside a wearable device 8.

[0550] In the present embodiment, the artificial intelligence module 4 is trained to generate a plurality of disease models 5. The training process uses at least one health parameter and / or at least one personal parameter of a subject 2 suffering from a disease in at least one of the disease models 5, and at least one health parameter and / or at least one personal parameter of a subject 2 not suffering from the disease. Such training data can be taken, for example, from a medical database.

[0551] Each disease model 5 is based on a neural network 6. A schematic diagram of the neural network structure of a disease model 5 is shown in Figure 2 .

[0552] Each neural network 6 comprises:

[0553] - an input layer 6.1 consisting of N nodes, N being the number of health parameters and personal parameters used to train the disease model 5;

[0554] - three hidden layers 6.2, each comprising an activation function and a plurality of nodes configured to receive a plurality of inputs;

[0555] - an output layer 6.3 having only one node representing the probability of the subject to suffer from the disease.

[0556] The output of each node of each hidden layer 6.2 of the neural network 6 is the result of applying its corresponding activation function to the weighted sum of its respective inputs.

[0557] The artificial intelligence module 4 is configured to perform the following steps:

[0558] a) using at least one health parameter and / or at least one personal parameter as input for one or more disease models 5;

[0559] b) determining a probability of the subject having a disease; and

[0560] c) issuing an alert 7 if the probability exceeds a determined probability value.

[0561] Figure 1a The issuing of the alert 7, i.e. sending a notification to a device (e.g. a mobile phone or tablet), is shown in the form of a schematic. This is for illustrative purposes only, as different types of alerts 7 can be issued.

[0562] As shown in Figure 1b , the subject 2 is wearing a plurality of wearable devices 8 with sensors 8.1. In one embodiment, these wearable devices 8 and sensors 8.1 do not belong to the system 1. In another embodiment, at least one wearable device 8 with a sensor 8.1 belongs to the system 1.

[0563] In this particular example of Figure 1b , the wearable devices 8 comprise:

[0564] - smart glasses with GPS, vision check sensor, heart rate sensor and blood pressure sensor;

[0565] - a smart T-shirt with GPS, temperature sensor, metabolite sensor, heart rate sensor and blood pressure sensor;

[0566] - a microneedle patch with tracking sensor, blood glucose sensor, heart rate sensor, blood pressure sensor, blood oxygen saturation sensor and temperature sensor;

[0567] - a wristband with tracking sensor, blood glucose sensor, heart rate sensor, blood pressure sensor, blood oxygen saturation sensor, metabolite sensor and temperature sensor;

[0568] - a smart watch with GPS, blood glucose sensor, heart rate sensor, blood pressure sensor, blood oxygen saturation sensor, metabolite sensor and temperature sensor;

[0569] - smart shoes with tracking sensor, heart rate sensor, blood pressure sensor, nervous system disease detection sensor, motion sensor, sweat level sensor and Parkinson's disease monitoring sensor.

[0570] In the present case, the partial wearable device 8 comprises a GPS or tracking sensor 8.1. In one embodiment, such GPS and tracking sensor 8.1 is configured to track the location of the subject 2, and the artificial intelligence module 4 is further configured to raise an alarm 7 if the subject 2 leaves a pre-set area (e.g. his residence or a hospital room).

[0571] In one embodiment, the smartwatch 8 and / or the wristband 8 and / or the smart T-shirt 8 and / or the smart shoe 8 further comprises a three-axis gyroscope 8.1 and a three-axis accelerometer 8.1, and the artificial intelligence module 4 is further configured to:

[0572] - detect whether the subject 2 has fainted or fallen by analyzing the angular velocity measured by the three-axis gyroscope 8.1 and the acceleration measured by the three-axis accelerometer 8.1;

[0573] - raise an alarm 7 if the subject 2 has fainted or fallen.

[0574] Although Figure 1b Although several wearable devices are shown in the embodiments comprising several sensors, in other embodiments, different types and / or numbers of sensors can be used. Moreover, these sensors can or can not be comprised in a wearable device.

[0575] In one embodiment, the system 1 comprises at least one wearable device 8 having at least one sensor 8.1 configured to:

[0576] - monitor the vital signs of the subject 2; and / or

[0577] - monitor the activity level of the subject 2; and / or

[0578] - monitor the sleep pattern of the subject 2; and / or

[0579] - monitor the heart rate of the subject 2; and / or

[0580] - monitor the location of the subject 2; and / or

[0581] - measure the blood pressure of the subject 2; and / or

[0582] - measure the blood oxygen saturation of the subject 2; and / or

[0583] - measure the body temperature of the subject 2; and / or

[0584] - measure the blood glucose level of the subject 2.

[0585] In one embodiment, the subject 2 is registered in the system 1, and the wearable device 8 is paired with the subject’s account through the subject’s device (e.g. a smartphone, a smartwatch or a tablet).

[0586] In one embodiment, the system 1 is configured to check whether the wearable device 8 is correctly placed on the body of the subject 2 to acquire accurate data. If not correctly placed, the subject 2 is asked to re-wear the wearable device 8.

[0587] In one embodiment, the wearable device 8 is configured to transmit the health parameters it measures in real time or periodically to the processor 3 of the system 1. In one embodiment, this transmission is made through a mobile application or a cloud-based platform.

[0588] In one embodiment, the system 1 is configured to take security measures during transmission and storage to protect the health parameter measurements transmitted by the wearable device 8.

[0589] In one embodiment, the artificial intelligence module 4 is also configured to:

[0590] - recommend a personalized treatment regimen according to the probability of the subject 2 having a disease; and / or

[0591] - associate the subject 2 with at least one specific healthcare provider according to the probability of the subject 2 having a certain disease.

[0592] In one embodiment, the system 1 comprises a database containing a plurality of treatment methods aimed at treating a specific disease.

[0593] In one embodiment, the artificial intelligence module 4 is also configured to provide an explanation and justification of the recommended treatment regimen, highlighting its underlying principles, evidence-based guidelines, and factors that influenced the decision-making process.

[0594] In one embodiment, the system 1 is configured to receive a report (e.g. from a healthcare provider) indicating the subject's 2 response to a previous prescribed treatment, e.g. to a previous medication regimen. The report can also include symptoms exhibited by the subject 2 when following the prescribed treatment regimen. The artificial intelligence module 4 is also configured to select a specific treatment regimen for the subject 2 taking into account the report.

[0595] In one embodiment, the system 1 is configured to receive a report (e.g. from a healthcare provider) indicating the subject's 2 response to a recommended treatment, e.g. to a medication regimen. The report can also include symptoms exhibited by the subject 2 when following the recommended regimen. The artificial intelligence module 4 is also configured to improve the treatment regimen for the subject 2 taking into account the report.

[0596] In one embodiment, the system 1 is configured to send data to and receive data from a clinical decision support tool used by the healthcare provider.

[0597] In one embodiment, the artificial intelligence module 4 of the system 1 is further configured to generate a collaborative treatment plan followed by the healthcare provider and the caregiver, the collaborative treatment plan being based on the subject 2’s changing needs and responses to treatment.

[0598] In one embodiment, the system 1 is configured to automatically input information from new subjects 2 suffering from one or more diseases, treatment outcomes and / or medical literature updates to the artificial intelligence module 4 to train the artificial intelligence module. This enables the accuracy and effectiveness of the treatment recommendations to be continually improved and enhanced over time.

[0599] In one embodiment, the system 1 is configured to provide resources for the caregiver and family members, and provide access to online support groups.

[0600] In one embodiment, the system 1 is configured to provide access to information related to the subject’s condition to the subject’s family members, and receive updated information.

[0601] In one embodiment, the artificial intelligence module 4 is further configured to:

[0602] - receive at least one textual and / or voice input from the subject 2;

[0603] - estimate the probability of the subject 2 following a particular emotional trend according to the following formula:

[0604]

[0605] wherein:

[0606] B is the textual and / or voice input of the subject 2;

[0607] A is the emotional trend;

[0608] P(A) is the prior probability of the subject 2 following the emotional trend A;

[0609] P(B) is the prior probability of at least one predefined word being found in the textual and / or voice input of the subject 2;

[0610] P(B|A) is the conditional probability of at least one predefined word being contained in the textual and / or voice input of the subject 2 when the subject 2 is following the emotional trend A;

[0611] P(A|B) is the conditional probability of the subject 2 following the emotional trend A when at least one predefined word is contained in the textual and / or voice input of the subject 2; and

[0612] - comparing the estimated probability of the subject developing a specific emotional trend to a pre-set range of probabilities;

[0613] - determining that a health crisis is detected if the estimated probability is within the pre-set range;

[0614] - issuing an alert when a health crisis is detected.

[0615] In one embodiment, the artificial intelligence module 4 is configured to:

[0616] - receive a medical history of the subject 2;

[0617] - estimate the probability of the subject 2 following a specific emotional trend according to the following formula:

[0618]

[0619] wherein:

[0620] C is the medical history of the subject 2;

[0621] A is the emotional trend;

[0622] P(A) is the prior probability of the subject 2 following the emotional trend A;

[0623] P(C) is the prior probability of finding at least one predefined data in the medical history of the subject 2;

[0624] P(C|A) is the conditional probability of finding at least one predefined data in the medical history of the subject 2 when the subject 2 follows the emotional trend A;

[0625] P(A|C) is the conditional probability of the subject 2 following the emotional trend A when at least one predefined data is found in the medical history of the subject 2; and

[0626] - comparing the estimated probability of the subject developing a specific emotional trend to a pre-set range of probabilities;

[0627] - determining that a health crisis is detected if the estimated probability is within the pre-set range;

[0628] - issuing an alert when a health crisis is detected.

[0629] In one embodiment, the system 1 comprises an artificial intelligence-driven chatbot or virtual mental health assistant that can contact the subject 2 when a health crisis is detected.

[0630] In one embodiment, the alert 7 comprises:

[0631] - a medical report to be sent to a medical service provider; and / or

[0632] - an emergency call to emergency medical services; and / or

[0633] - a notification to the subject and / or the contact.

[0634] In one embodiment, the processor 3 of the system 1 is located within a server. The subject 2, medical service providers, family members and / or caregivers access the system 1 through a mobile or web application.

[0635] In one embodiment, the mobile or web application has an intuitive and user-friendly interface for easy navigation.

[0636] In one embodiment, the system 1 is configured to send and receive data to and from other medical service systems and electronic health records (EHR) to facilitate seamless information exchange between medical service providers.

[0637] In one embodiment, the system 1 includes security measures to protect sensitive subject 2 information; for example, data encryption, access control and regular security audits.

[0638] In one embodiment, the subject 2 must register in the mobile or web application in order to establish a connection between the subject 2 and their caregivers, medical service providers and family members. In addition, by registering, the system creates a specific health profile for the subject 2, which contains personal information, emergency contacts and medical history.

[0639] In one embodiment, the method for subject 2 identity authentication and authorization with the API of the system 1 includes API keys, OAuth tokens or certificates. In one embodiment, the system 1 provides different levels of access and permissions (e.g. access to the subject 2 personal data, health indicators) to different types of users depending on the user role (subject, medical service provider, family member or caregiver) and the capabilities of the wearable device 8, if any.

[0640] In one embodiment, the system 1 is configured to send medication reminders to the caregivers and / or the subject 2 through the mobile or web application, for example, for subjects with dementia or other mental illnesses, to ensure that the correct dose of medication is taken on time.

[0641] In one embodiment, the system 1 is configured to store records of medication plans, dosages and any side effects so that medical service providers can access it through the mobile or web application.

[0642] In one embodiment, the system 1 is configured to display schedules, dosages and daily routines to medical service providers through the mobile or web application to improve their daily assistance to the subject.

[0643] In one embodiment, the system 1 is configured to send an automatic appointment reminder to the caregiver and / or the subject through a mobile or web application to reduce the likelihood of missed appointments.

[0644] Figure 3a A decision tree 9 for selecting the health parameters and the personal parameters for each disease model 5 is shown. In particular, the processor 3 of the system 1 is configured to select the health parameters and / or the personal parameters and to train the artificial intelligence module 4 with them to generate each disease model 5. This selection process is performed using a plurality of decision trees 9, each decision tree 9 comprising a plurality of layers 9.1, 9.2 and each layer 9.1, 9.2 comprising at least one parent node, at least one left child node and at least one right child node, wherein the processor 3 is configured to perform the following steps:

[0645] - taking the health parameters and the personal parameters as input of the plurality of decision trees 9;

[0646] - determining an importance value for each health parameter and each personal parameter in each layer 9.1, 9.2 of the decision tree 9, the importance value being defined as a decrease node impurity and weighted by a probability of reaching at least one parent node of the layer 9.1, 9.2; wherein

[0647] o the probability of reaching a parent node of the layer 9.1, 9.2 is determined as the number of health parameters and personal parameters reaching the parent node divided by the total number of health parameters and personal parameters used to train the artificial intelligence module 4; and

[0648] o the decrease node impurity is determined as follows:

[0649]

[0650] There are:

[0651] N left and N right respectively the total number of health parameters and personal parameters in at least one left child node and at least one right child node of the layer 9.1, 9.2, and

[0652] I p , I left and I right respectively the impurity of at least one parent node, at least one left child node and at least one right child node; the impurity being calculated by an impurity measure, preferably a Gini impurity measure or an entropy measure;

[0653] - determining a final importance value for each health parameter and each personal parameter, the final importance value being the importance value of the health parameter or personal parameter calculated in the last layer 9.2 of the decision tree 9; and

[0654] - selecting health parameters and / or personal parameters for training the artificial intelligence module 4, the health parameters and / or personal parameters being the parameters for which the final importance value is higher than a predefined threshold.

[0655] Figure 3a The decision tree 9 in FIG. 9 comprises two layers 9.1 and 9.2. The node “a” is the parent node of the first layer 9.1, the node “b” is the left child node of the first layer 9.1, and the node “c” is the right child node of the first layer 9.1. The nodes “b” and “c” are the parent nodes of the second layer 9.2, the nodes “d” and “f” are the left child nodes of the second layer 9.2, and the nodes “e” and “g” are the right child nodes of the second layer 9.2.

[0656] In one embodiment, the plurality of disease models 5 comprises a heart disease model 5 generated by the artificial intelligence module 4. Figure 3b The final importance values of 10 health parameters and personal parameters obtained using the decision tree 9 in the example are shown, which are used to build the heart disease model 5. In this example, only the parameters higher than the predefined threshold 0.07 are selected to train the artificial intelligence module 4 to generate the heart disease model 5, the selected parameters are as follows:

[0657] - age (age),

[0658] - type of chest pain experienced by the subject (cp),

[0659] - blood pressure level at rest (trestbps),

[0660] - serum cholesterol (chol),

[0661] - maximum heart rate experienced by the subject (oldpeak),

[0662] - exercise-induced angina (exang), and

[0663] - ST segment depression induced by exercise compared to the resting state (thalach).

[0664] The rejected parameters are as follows:

[0665] - electrocardiogram measured at rest (restecg);

[0666] - fasting blood sugar (fbs); and

[0667] - gender (sex).

[0668] In the present embodiment, 10 health parameters and personal parameters are selected using a random forest classifier, where all possible health parameters and personal parameters are used as input to the classifier.

[0669] For this heart disease model 5:

[0670] o The activation function of each node in the first hidden layer 3.2 is defined by the following equation:

[0671] f1(x) = max(0, x)

[0672] where x represents the weighted sum of the inputs to each node in the first hidden layer 3.2;

[0673] o The activation function of each node in the second hidden layer 3.2 is defined by the following equation:

[0674] f2(x) = x · tanh(ln(1 + e x ))

[0675] where x represents the weighted sum of the inputs to each node in the second hidden layer 3.2; and

[0676] o The activation function of the third hidden layer 3.2 is defined by the following equation:

[0677]

[0678] where x i represents the weighted sum of the inputs to the i-th node in the third hidden layer, K is the total number of nodes in the third hidden layer 3.2.

[0679] In one embodiment, the plurality of disease models 5 comprises a dementia model 5 generated by the artificial intelligence module 4. Figure 3c The final importance values of the 13 health parameters and personal parameters obtained using the decision tree 9 for building the dementia model 5 are shown in one example. In this example, only parameters above a predefined threshold of 0.03 are selected to train the artificial intelligence module 4 to generate this dementia model 5, the selected parameters are as follows:

[0680] - subjects with diabetes (Health_Conditions),

[0681] - subjects with depression (Depression_Status),

[0682] - subjects carrying the APOE ε3 allele (APOE_ε3),

[0683] - oxygen content in the blood (BloodOxygenLevel),

[0684] - heart rate (HeartRate),

[0685] - age (Age),

[0686] - the time elapsed between the subject wearing the wearable device and the magnetic resonance imaging examination, or the time elapsed between the magnetic resonance imaging examination and the subject wearing the wearable device (MRI_Deleay),

[0687] - alcohol level in blood (AlcoholLevel),

[0688] - body temperature (BodyTemperature), and

[0689] - weight (Weight).

[0690] The parameters that were rejected are as follows:

[0691] - the subject’s medication history,

[0692] - family history, and

[0693] - gender.

[0694] In this embodiment, 13 health parameters and personal parameters were selected using a random forest classifier, wherein all possible health parameters and personal parameters were used as input for the classifier.

[0695] In this context, “family history” refers to the health information of the subject’s biological relatives, including parents, siblings, grandparents, aunts and uncles, and cousins. Family history is part of the subject’s medical history and can provide important information about the subject’s risk of developing certain diseases, as well as potential genetic predispositions or genetic factors.

[0696] For this model of dementia 5:

[0697] o the activation function of each node in the first hidden layer 3.2 is defined by the following equation:

[0698] f1(x) = max(0, x)

[0699] where x represents the weighted sum of the inputs of each node in the first hidden layer 3.2;

[0700] o the activation function of each node in the second hidden layer 3.2 is defined by the following equation:

[0701] f2(x) = x · tanh(ln(1 + e x ))

[0702] where x represents the weighted sum of the inputs of each node in the second hidden layer 3.2; and

[0703] The activation function of the third hidden layer 3.2 is defined by the following equation:

[0704] f3(x) = ln(l + e x )

[0705] where x represents the weighted sum of the inputs of each node in the third hidden layer 3.2.

[0706] In one embodiment, the plurality of disease models 5 comprises a schizophrenia disease model 5 generated by the artificial intelligence module 4. Figure 3d The final importance values of the three health parameters and personal parameters obtained using the decision tree 9 are shown, which were used to build the schizophrenia disease model 5. In this example, the parameters above a predefined threshold of 0.25 were selected to train the artificial intelligence module 4 to generate this schizophrenia disease model 5. In this example, all of the following parameters were selected:

[0707] - sleep pattern,

[0708] - walking / step motion pattern, and

[0709] - heart rate.

[0710] In this embodiment, the three health parameters and personal parameters were selected using a random forest classifier, where all possible health parameters and personal parameters were used as inputs to the classifier.

[0711] For this schizophrenia disease model 5:

[0712] The activation function of each node in the first hidden layer 3.2 is defined by the following equation:

[0713] f1(x) = max(0, x)

[0714] where x represents the weighted sum of the inputs of each node in the first hidden layer 3.2;

[0715] The activation function of each node in the second hidden layer 3.2 is defined by the following equation:

[0716] f2(x) = max(0, x)

[0717] where x represents the weighted sum of the inputs of each node in the second hidden layer 3.2; and

[0718] The activation function of the third hidden layer 3.2 is defined by the following equation:

[0719]

[0720] where x iwhere x represents the weighted sum of the inputs of each node in the first hidden layer 3.2, K is the total number of nodes in the first hidden layer 3.2, and x represents the weighted sum of the inputs of each node in the third hidden layer 3.2.

[0721] In one embodiment, the plurality of disease models 5 comprises an Alzheimer’s disease model 5 generated by the artificial intelligence module 4. Figure 3e The final importance values of the 10 health parameters and personal parameters obtained using the decision tree 9 are shown, which were used to build the Alzheimer’s disease model 5. In this case, only the parameters above a predefined threshold of 0.05 were selected to train the artificial intelligence module 4 to generate this Alzheimer’s disease model 5, which were as follows:

[0722] - the clinical dementia rating, to indicate the severity of the symptoms of dementia,

[0723] - the mini-mental state examination, to indicate the level of cognitive function,

[0724] - the normalized total brain volume,

[0725] - the estimated total intracranial volume,

[0726] - the atlas scaling factor, and

[0727] - the age.

[0728] The parameters that were rejected were as follows:

[0729] - the time interval between the subject wearing the wearable device and the magnetic resonance imaging examination being performed, or the time interval between the magnetic resonance imaging examination being performed and the subject wearing the wearable device,

[0730] - the subject’s years of education,

[0731] - the subject’s socioeconomic status, and

[0732] - the gender.

[0733] In this embodiment, the 10 health parameters and personal parameters were selected using a random forest classifier, in which all the possible health parameters and personal parameters were used as inputs to the classifier.

[0734] For this Alzheimer’s disease model 5:

[0735] o the activation function of each node in the first hidden layer 3.2 was defined by the following equation:

[0736] f1(x) = max(0, x)

[0737] where x represents the weighted sum of the inputs of each node in the first hidden layer 3.2, K is the total number of nodes in the first hidden layer 3.2, and x represents the weighted sum of the inputs of each node in the third hidden layer 3.2.

[0738] o the activation function of each node in the second hidden layer 3.2 is defined by the following equation:

[0739] f2(x) = max(0, x)

[0740] where x represents the weighted sum of the inputs to each node in the second hidden layer 3.2; and

[0741] o the activation function of the third hidden layer 3.2 is defined by the following equation:

[0742]

[0743] where x i represents the weighted sum of the inputs to the i-th node in the third hidden layer, K is the total number of nodes in the third hidden layer 3.2.

[0744] In one embodiment, the plurality of disease models 5 comprises an anxiety model 5 and / or a depression model 5 generated by the artificial intelligence module 4 using the following parameters of a subject 2 suffering from anxiety and / or depression:

[0745] - heart rate variability;

[0746] - skin galvanic activity;

[0747] - respiratory rate and breathing pattern;

[0748] - blood pressure levels;

[0749] - activity level;

[0750] - sleep pattern;

[0751] - social interaction pattern;

[0752] - history of drug abuse;

[0753] - self-reported anxiety and / or depressive symptoms;

[0754] - history of mental health diagnosis or treatment.

[0755] For these disease models 5:

[0756] o the activation function of each node in the first hidden layer 3.2 is defined by the following equation:

[0757] f1(x) = max(0, x)

[0758] where x represents the weighted sum of the inputs to each node in the first hidden layer 3.2;

[0759] o the activation function of each node in the second hidden layer 3.2 is defined by the following equation:

[0760]

[0761] where x represents the weighted sum of the inputs of each node in the second hidden layer 3.2; and

[0762] o the activation function of the third hidden layer 3.2 is defined by the following equation:

[0763]

[0764] where x represents the weighted sum of the inputs of each node in the third hidden layer 3.2.

[0765] In one embodiment, the plurality of disease models 5 comprises an eating disorder model 5 generated by the artificial intelligence module 4 using the following parameters of a subject 2 suffering from an eating disorder:

[0766] - age;

[0767] - gender;

[0768] - body mass index;

[0769] - number of weekly binge eating episodes;

[0770] - number of weekly purging behavior episodes;

[0771] - body dissatisfaction score, for indicating the degree of self-reported body dissatisfaction;

[0772] - eating disorder inventory score, for indicating a score of eating disorder symptoms according to a standard scale;

[0773] - depression score, for indicating a score of a depression assessment scale;

[0774] - anxiety score, for indicating a score of an anxiety assessment scale; and

[0775] - eating disorder diagnosis.

[0776] For this eating disorder disease model 5:

[0777] o the activation function of each node in the first hidden layer 3.2 is defined by the following equation:

[0778] f1(x) = max(0, x)

[0779] where x represents the weighted sum of the inputs of each node in the first hidden layer 3.2;

[0780] o the activation function of each node in the second hidden layer 3.2 is defined by the following equation:

[0781]

[0782] where x represents the weighted sum of the inputs of each node in the second hidden layer 3.2; and

[0783] The activation function of the third hidden layer 3.2 is defined by the following equation:

[0784]

[0785] where x represents the weighted sum of the inputs of each node in the third hidden layer 3.2.

[0786] Figures 4a-4b An example of a table created using data from a medical database for training the artificial intelligence module 4 is shown. In this specific example, a heart disease model 5 is generated by the artificial intelligence module 4.

[0787] In particular, the processor 3 is configured to process the health parameters and the personal parameters for training the artificial intelligence module 4 before the training process. Such data processing steps performed by the processor 3 require the following steps:

[0788] - creating a table in which each column represents a health parameter or a personal parameter and each row represents a different subject 2 suffering from the disease;

[0789] - assigning to the values in the table a predetermined numerical format, in this case an integer format, with the result as shown in Figure 4a ;

[0790] - replacing the values in the table that are not a number (NaN) with the average value of the values of the column, obtaining the table as shown in Figure 4b ;

[0791] - identifying the outliers in each column; for example, using the Z-score algorithm;

[0792] - deleting the rows containing outliers.

[0793] Alternatively, the processor 3 is configured to process the health parameters and the personal parameters for each disease by performing the following steps:

[0794] - creating a table in which each row represents a health parameter or a personal parameter and each column represents a different subject suffering from the disease;

[0795] - assigning to the values in the table a predetermined numerical format;

[0796] - replacing the values in the table that are not a number with the average value of the values of the row;

[0797] - identifying the outliers in each row, preferably using the Z-score algorithm; - deleting the columns containing outliers.

Claims

1. A system (1) for issuing alerts related to the health status of a subject (2), the system (1) comprising a processor (3) configured to: - Receive at least one health parameter from the subject, wherein one or more of the at least one health parameter are measured by at least one sensor (4), and - Receive at least one personal parameter from the subject. in, The processor (3) includes an artificial intelligence module (4), which generates multiple disease models (5) through training based on the following parameters: The subject (2) suffering from at least one disease in at least one disease model (5) has at least one health parameter and / or at least one personal parameter, and At least one health parameter and / or at least one personal parameter of subjects (2) who do not have the above-mentioned diseases; Each disease model (5) is based on a neural network (6), which includes: o Input layer (6.1), which consists of N nodes, where N is the number of health parameters and personal parameters used to train the disease model (5); o Three hidden layers (6.2), each hidden layer includes an activation function and multiple nodes, which are configured to receive multiple inputs; o Output layer (6.3), which has only one node, representing the probability that the subject has the disease; In this context, the output of each node in each hidden layer of the neural network (6) is the result of applying its corresponding activation function to a weighted sum of its corresponding inputs; The artificial intelligence module (4) is configured to perform the following steps: a) Use at least one health parameter and / or at least one personal parameter as input to one or more disease models (5); b) Determine the probability of the subject developing the disease; and c) If the probability exceeds a certain probability value, then issue an alarm (7).

2. The system (1) according to claim 1, wherein, The processor (3) is configured to select health parameters and / or personal parameters and thereby train the artificial intelligence module (4) to generate a disease model (5), wherein the selection is made using multiple decision trees (9), each decision tree (9) comprising multiple layers (9.1, 9.2), and each layer (9.1, 9.2) comprising at least one parent node, at least one left child node, and at least one right child node, and wherein the processor (3) is configured to perform the following steps: - Use health parameters and / or personal parameters as inputs to multiple decision trees (9); - Determine the importance value of each health parameter and each individual parameter in each layer (9.1, 9.2) of the decision tree. This importance value is defined as the reduction in node impurity and is weighted by the probability of reaching at least one parent node in that layer (9.1, 9.2). The probability of reaching the parent node of the layers (9.1, 9.2) is determined as: the number of health parameters and personal parameters reaching that parent node, divided by the total number of health parameters and personal parameters used to train the artificial intelligence module (4); and The reduction in impurity at the o-node is determined as follows: exist: N left and N right These are the total number of health parameters and personal parameters in at least one left child node and at least one right child node of layers (9.1, 9.2), respectively. I p I left and I right These are the impurities of at least one parent node, at least one left child node, and at least one right child node, respectively; the impurities are calculated using an impurity metric, preferably a Gini impurity metric or an entropy metric. - Determine the final importance value for each health parameter and each personal parameter, which is the importance value of that health parameter or personal parameter calculated in the last layer (9.2) of the decision tree (9); and - Select health parameters and / or personal parameters for training the artificial intelligence module (4), which are parameters whose final importance values ​​are higher than a predefined threshold.

3. The system (1) according to any one of the preceding claims, wherein the processor (3) is configured to process health parameters and / or personal parameters for training the artificial intelligence module (4) prior to the training process, the processor (3) being configured to perform the following steps for each disease to process the health parameters and / or personal parameters: - Create a table where each column represents a health parameter or personal parameter, and each row represents a different subject with the disease; - Assign a predefined numeric format to the values ​​in the table; - Replace any values ​​in the table that do not contain a value with the average of the values ​​in that column; - Identify outliers in each column, preferably using the standard score method; - Delete rows containing outlier values; or - Create a table where each row represents a health parameter or personal parameter, and each column represents a different subject with the disease; - Assign a predefined numeric format to the values ​​in the table; - Replace values ​​in the table that do not contain a number with the average of the values ​​in that row; - Identify outliers in each row, preferably using the standard score method; - Delete columns containing outliers.

4. The system (1) according to any one of the preceding claims, wherein, The system (1) includes at least one sensor (8.1), the at least one sensor (8.1) including a three-axis gyroscope (8.1) and a three-axis accelerometer (8.1), and wherein the artificial intelligence module (4) is further configured to: - The subject (2) was detected to have fainted or fallen by analyzing the angular velocity measured by the triaxial gyroscope (8.1) and the acceleration measured by the triaxial accelerometer (8.1); - If the subject (2) faints or falls, an alarm (7) is triggered.

5. The system (1) according to any one of the preceding claims, wherein, The system (1) includes at least one sensor (8.1) configured to track the location of the subject (2), and the artificial intelligence module (4) is also configured to issue an alarm (7) when the subject (2) goes beyond a preset area.

6. The system (1) according to any one of the preceding claims, wherein, The artificial intelligence module (4) is also configured to: - Recommend personalized treatment plans based on the probability of the subject (2) having the disease; and / or -Associate subject (2) with at least one specific healthcare provider based on the probability that subject (2) has a certain disease.

7. The system (1) according to any one of the preceding claims, wherein, The artificial intelligence module (4) is also configured to: - Receive at least one text and / or voice input from subject (2); - Estimate the probability that subject (2) follows a specific emotional trend using the following formula: in: B is the text and / or voice input of the subject (2); A represents the sentiment trend; P(A) is the prior probability that subject (2) follows emotional trend A; P(B) is the prior probability that at least one predefined word is found in the text and / or speech input of the subject (2); P(B|A) is the conditional probability that subject (2)’s text and / or speech input contains at least one predefined word when subject (2) follows emotional trend A; P(A|B) is the conditional probability that subject (2) follows emotional trend A when the text and / or voice input of subject (2) contains at least one predefined word; and - Compare the estimated probability of a subject exhibiting a specific emotional trend with a pre-defined probability range; -If the estimated probability is within the preset range, a health crisis is detected; - Issue an alert when a health crisis is detected.

8. The system (1) according to any one of the preceding claims, wherein, The artificial intelligence module (4) is configured as follows: -Receive the medical history of the subject (2); - Estimate the probability that subject (2) follows a specific emotional trend using the following formula: in: C is the medical history of subject (2); A represents the sentiment trend; P(A) is the prior probability that subject (2) follows emotional trend A; P(C) is the prior probability of finding at least one predefined data in the medical history of subject (2); P(C|A) is the conditional probability of finding at least one predefined data in the patient's medical history when the subject (2) follows the emotional trend A; P(A|C) is the conditional probability that subject (2) follows emotional trend A when at least one predefined data point is found in subject (2)'s medical history; and - Compare the estimated probability of a subject exhibiting a specific emotional trend with a pre-defined probability range; -If the estimated probability is within the preset range, a health crisis is detected; - Issue an alert when a health crisis is detected.

9. The system (1) according to any one of the preceding claims, wherein, The system (1) includes at least one wearable device (8), the wearable device having at least one sensor (8.1), the sensor being configured to: - Monitor the vital signs of the subject (2); and / or - Monitor the activity level of the subjects (2); and / or - Monitor the sleep patterns of subjects (2); and / or - Monitor the heart rate of the subject (2); and / or - Monitor the location of the subject (2); and / or - Measure the blood pressure of subject (2); and / or - Measure the blood oxygen saturation of subject (2); and / or - Measure the body temperature of the subject (2); and / or -Measure the blood glucose level of the subjects (2).

10. The system (1) according to any one of the preceding claims, wherein, The alarm (7) includes: - A medical report to be sent to the healthcare provider; and / or - A call to emergency medical services; and / or - A notification sent to the subject and / or contact person and / or healthcare provider and / or emergency medical services.

11. The system (1) according to any one of the preceding claims, wherein, The multiple disease models (5) include a heart disease model (5) generated by the artificial intelligence module (4) using one or more of the following parameters of a subject (2) suffering from heart disease: -age, - The type of chest pain experienced by the subjects - Blood pressure level at rest, - Serum cholesterol, - The maximum heart rate experienced by the subject Exercise-induced angina, and - ST segment depression caused by exercise compared to the resting state; in: The activation function of each node in the first hidden layer (3.2) is defined by the following formula: f1(x) = max(0,x) Where x represents the weighted sum of the inputs to each node in the first hidden layer (3.2); The activation function of each node in the second hidden layer (3.2) is defined by the following formula: f2(x)=x·tanh(ln(1+e x )) Where x represents the weighted sum of the inputs to each node in the second hidden layer (3.2); and The activation function of the third hidden layer (3.2) is defined by the following formula: Where x i Let K represent the weighted sum of the inputs of the i-th node in the third hidden layer, where K is the total number of nodes in the third hidden layer (3.2).

12. The system (1) according to any one of the preceding claims, wherein, The multiple disease models (5) include a dementia model (5) generated by the artificial intelligence module (4) using one or more of the following parameters of a subject (2) suffering from dementia (5): -Subjects with diabetes, -Subjects suffering from depression, - Subjects carrying the APOEε3 allele -Oxygen content in the blood, Heart rate, -age, - The time elapsed between the subject wearing the wearable device and undergoing an MRI scan, or the time elapsed between undergoing an MRI scan and the subject wearing the wearable device. - Blood alcohol content, - Body temperature, and -weight; in: The activation function of each node in the first hidden layer (3.2) is defined by the following formula: f1(x) = max(0,x) Where x represents the weighted sum of the inputs to each node in the first hidden layer (3.2); The activation function of each node in the second hidden layer (3.2) is defined by the following formula: f2(x)=x·tanh(ln(1+e x )) Where x represents the weighted sum of the inputs to each node in the second hidden layer (3.2); and The activation function of the third hidden layer (3.2) is defined by the following formula: f3(x)=ln(1+e x ) Where x represents the weighted sum of the inputs of each node in the third hidden layer (3.2).

13. The system (1) according to any one of the preceding claims, wherein, The multiple disease models (5) include a schizophrenia disease model (5) generated by the artificial intelligence module (4) using one or more of the following parameters of a subject (2) suffering from schizophrenia: -Sleep mode - Walking / steps mode, and -Heart rate; in: The activation function of each node in the first hidden layer (3.2) is defined by the following formula: f1(x) = max(0,x) Where x represents the weighted sum of the inputs to each node in the first hidden layer (3.2); The activation function of each node in the second hidden layer (3.2) is defined by the following formula: f2(x) = max(0,x) Where x represents the weighted sum of the inputs to each node in the second hidden layer (3.2); and The activation function of the third hidden layer (3.2) is defined by the following formula: Where x i Let K represent the weighted sum of the inputs of the i-th node in the third hidden layer, where K is the total number of nodes in the third hidden layer (3.2).

14. The system (1) according to any one of the preceding claims, wherein, The multiple disease models (5) include an Alzheimer's disease model (5) generated by the artificial intelligence module (4) using one or more of the following parameters of a subject (2) suffering from Alzheimer's disease: - Clinical dementia assessment, used to indicate the severity of dementia symptoms. - A brief mental status test used to indicate cognitive function levels. - Normalized whole brain volume -Estimated total intracranial volume, -Map scaling factor, and -age; in: The activation function of each node in the first hidden layer (3.2) is defined by the following formula: f1(x) = max(0,x) Where x represents the weighted sum of the inputs to each node in the first hidden layer (3.2); The activation function of each node in the second hidden layer (3.2) is defined by the following formula: f2(x) = max(0,x) Where x represents the weighted sum of the inputs to each node in the second hidden layer (3.2); and The activation function of the third hidden layer (3.2) is defined by the following formula: Where x i Let K represent the weighted sum of the inputs of the i-th node in the third hidden layer, where K is the total number of nodes in the third hidden layer (3.2).

15. The system (1) according to any one of the preceding claims, wherein, The multiple disease models (5) include an anxiety disorder model (5) and / or a depression model (5) generated by the artificial intelligence module (4) using one or more of the following parameters of a subject (2) suffering from anxiety disorder and / or depression: Heart rate variability, -Skin electrical activity, -Respiratory rate and breathing pattern - Blood pressure level -Activity level, -Sleep mode -Social interaction mode - History of drug abuse, - Self-reported symptoms of anxiety and / or depression, - History of mental health diagnosis or treatment; in: The activation function of each node in the first hidden layer (3.2) is defined by the following formula: f1(x) = max(0,x) Where x represents the weighted sum of the inputs to each node in the first hidden layer (3.2); The activation function of each node in the second hidden layer (3.2) is defined by the following formula: Where x represents the weighted sum of the inputs to each node in the second hidden layer (3.2); and The activation function of the third hidden layer (3.2) is defined by the following formula: Where x represents the weighted sum of the inputs of each node in the third hidden layer (3.2).

16. The system (1) according to any one of the preceding claims, wherein, The multiple disease models (5) include an eating disorder model (5) generated by the artificial intelligence module (4) using one or more of the following parameters of a subject (2) suffering from an eating disorder: -age, -gender, -Body mass index, -Number of binge eating episodes per week - Weekly clearing of behavioral episode counts - Body dissatisfaction rating, used to indicate the degree of self-reported body dissatisfaction. - Eating Disorders Assessment Questionnaire score, used to score eating disorder symptoms according to a standardized scale; - Depression score, used to score on a depression assessment scale. - Anxiety score, used to indicate the score of an anxiety assessment scale, and - Diagnosis of eating disorders; in: The activation function of each node in the first hidden layer (3.2) is defined by the following formula: f1(x) = max(0,x) Where x represents the weighted sum of the inputs to each node in the first hidden layer (3.2); The activation function of each node in the second hidden layer (3.2) is defined by the following formula: Where x represents the weighted sum of the inputs to each node in the second hidden layer (3.2); and the activation function of the third hidden layer (3.2) is defined by the following formula: Where x represents the weighted sum of the inputs of each node in the third hidden layer (3.2).

Images