PET HEALTH MONITORING SYSTEM WITH BASE DEVICE, CENTRAL DATA PROCESSING DEVICE AND END DEVICE AND PET HEALTH MONITORING PROCEDURES

Abstract

The invention relates to a pet health monitoring system comprising a basic device, a central data processing device and an end device, a method for setting up the system and a pet health monitoring method. A system for monitoring the health status of a companion animal is disclosed, comprising: a base unit, a central data processing unit, and an end device, wherein the base unit comprises: at least one sensor for detecting a measurable health-related characteristic of a companion animal and a base data processing unit configured to process sensor data detected by the sensor, and a base data transmission unit configured to transmit processed sensor data from the base unit; the central data processing unit comprises: a central data transmission unit configured to receive the transmitted sensor data, and an analysis unit configured, by inputting the received sensor data, to classify the health status of a companion animal located in a spatial environment where the base unit is installed and to output it as diagnostic data.and the terminal device includes: a terminal device data transmission means configured to receive the diagnostic data, and a terminal device display means configured to output the classified health status in user-readable form.

Description

[0001] The invention relates to a pet health monitoring system with a base device, a central data processing device and a terminal, and a pet health monitoring method.

[0002] When diagnosing pathological conditions in companion animals, such as Felidae, especially domestic cats, the fundamental problem is that companion animals are usually evolutionarily predisposed to avoid openly displaying symptoms. Externally expressed signals of weakness can, on the one hand, result in adverse shifts in the group's hierarchy and, on the other hand, could make the individual a target of predators as preferred prey. Furthermore, due to the inherent communication barrier, the animals are unable to articulate their naturally perceived symptoms or limit their severity. This particularly limits the medical history in the early stages of a companion animal's illness.Particularly in the case of pet owners who are not trained in veterinary medicine, possible symptoms of illness remain below the threshold of perception and only those symptoms of a pathological condition that only appear at an advanced stage are recognized, such as blood in the stool or urine, licking of the fur (bald patches), urinating outside of the usual and assigned relief areas, refusal to eat, cramps, staggering or lethargy.

[0003] What makes matters worse is that certain absolute parameters may be meaningless as isolated symptoms and can only provide evidence of a pathological condition as relative parameters compared to previous values. For example, a certain body weight may well be within the usual variance for a non-pathological condition and not be meaningful in itself, but may indicate weight loss when compared to previous values. Medical personnel, who usually only have access to a snapshot of the pet's physical condition when assessing the state of health, do not have such parameters available for diagnosis. Such symptoms, which are meaningful in a temporal context, can include changes in weight, changes in feeding habits, changes in excretion behavior, or general changes in behavior.

[0004] The prior art, for example from US Pat. No. 8,797,166 B2, discloses systems, methods, and computer code products for monitoring the behavior, health, or characteristics of an animal. The animal to be monitored is located in an animal toilet container, which is placed on a system suitable for determining the animal's weight, recording it, and transmitting it via a data network. The measured values ​​can be processed to determine trends, statistics, and changes in the animal's physiological functions. The results can be used to access the animals' health status and to issue warnings, alarms, messages, and other notifications to specific caregivers. The ability to retrieve and review data can contribute to a better understanding of an animal's health status and facilitate the early detection of diseases. Pet owners who may be...While a temporal trend in values ​​for diagnosis-relevant parameters is available in observation using state-of-the-art tools, allowing a comprehensive medical history to be taken, the necessary specialist knowledge for a targeted diagnosis is lacking. Non-professional sources such as the internet or general pet care guides are of little help here. Identified symptoms can indicate several different pathological conditions, meaning that several different diagnoses are possible without a clear trend. Furthermore, such sources do not adequately take individual characteristics such as the age, breed, or sex of the pet into account. Furthermore, particularly in non-professional literature, different terms are often used for one pathological condition, such as struvite stones or uroliths for urinary stones. This makes it difficult to understand the symptoms as a whole and ultimately to make a reliable diagnosis.

[0005] In this context, the state of the art has established that results generated by self-learning systems (machine learning, ML), no matter how powerful the solutions, are not always explainable. ML does not provide insights into the cause and effect of the phenomena underlying the data. Knowledge of the underlying causes is crucial for reliable cause-based diagnosis. Only if it can be narrowed down to which pathological conditions are the cause of certain symptoms can an appropriate recommendation for action be formulated for a desired recovery. The goal is to eliminate the respective pathological condition of a pet and thus combat the underlying disease. Otherwise, only symptomatic treatment would be possible.

[0006] The fact outlined in the problem statement, namely that the symptoms perceived by humans in companion animals essentially already imply an advanced disease status, is addressed in the current state of technology primarily through (supposedly) random veterinary checkups. Previous approaches for the early detection of pathological conditions are based on detectable changes in activity or disease testing. Smart collars already exist for this purpose, recording activity levels and sleep times and rhythms. Changes or abnormalities are then displayed. Alternatively, tests for animal diseases are offered. Ideally, testing enables a reliable diagnosis of disease, although the current focus is on worm, allergy, and DNA tests.If symptoms do become apparent to the pet owner, such as constant vomiting or diarrhea, the first step is often to consult lay sources such as the internet or pet care guides. There are also forums, blogs, and social media groups that promise to provide information. If the symptoms persist, online consultations with veterinarians via video call have recently become available as a precursor to a visit to the doctor. These veterinarians then provide remote diagnoses without guarantees, as there is no direct assessment. Depending on the diagnosis, a visit to the local veterinarian is the final step. If an illness is diagnosed there, ongoing follow-up appointments are scheduled to monitor health-related parameters.

[0007] In the field of human medicine, so-called symptom checkers have gained increasing popularity in recent years. These are based on algorithms that, based on the entered symptoms, provide an overview of possible causative diseases. Patients often interact with a chatbot, which systematically queries the patient to identify possible diseases. Such chatbots can make the medical history process more efficient through data-driven comparisons and subsequently relieve physicians of the burden of diagnosis and treatment planning.

[0008] Artificial intelligence (AI) systems, and especially self-learning systems, are playing an increasingly prominent role in scientific publications in the human medical field. ML has led to a large number of applications in the field of AI in recent years.

[0009] However, due to the problems described above, a reliable diagnosis of pathological conditions in companion animals is only inadequately possible.

[0010] It is the object of this invention to eliminate the problems of the prior art and in particular to provide a system, components and methods that enable a reliable diagnosis of pathological conditions in companion animals.

[0011] This problem is solved by the subordinate claims. Advantageous further developments are the subject of the subclaims.

[0012] A system for monitoring the health status of a pet is disclosed, comprising: a base device, a central data processing device, and a terminal, wherein the base device comprises: at least one sensor for detecting a measurable health-relevant characteristic of a pet, and a base data processing means configured to process sensor data detected by the sensor, and a base data transmission means configured to transmit processed sensor data from the base device, the central data processing device comprising: a central data transmission means configured to receive the transmitted sensor data, an analysis means configured to classify the health status of a pet located in a spatial environment in which the base device is installed, by inputting the received sensor data, and to output it as diagnostic data,and the terminal comprises: a terminal data transmission means configured to receive the diagnostic data, and a terminal display means configured to output the classified health status in user-readable form.

[0013] Disclosed is a basic device comprising: at least one sensor for detecting a measurable health-relevant characteristic of a pet and a basic data processing means configured to process sensor data detected by the sensor, and a basic data transmission means configured to transmit processed sensor data from the basic device.

[0014] It is advantageous if the basic data processing means or the analysis means of the central data processing device is configured to classify from the sensor data those sensor data which are relevant for monitoring the health status of a pet, and / or the basic data transmission means is configured to transmit only the sensor data classified as relevant.

[0015] It is advantageous if the basic data processing means comprises an MLM (machine learning model) by means of which the sensor data relevant for monitoring the health status of a pet are classified.

[0016] It is advantageous if several sensor data are combined into data sets (clustering) for the classification of the sensor data, whereby the sensor data of a data set are temporally close to one another and, in particular, the time interval between the sensor data is not longer than 30 seconds and at least the first three measurements are within 30 seconds.

[0017] It is advantageous if the data set is divided into sub-data sets if the measurements have the lowest reference weight GR p of all pets p, in particular by a factor of 0.3.

[0018] It is advantageous to expand the data set or cluster if the last measurement from the data set matches the reference weight of the pet.

[0019] It is advantageous to correct the measurement by calculating the difference with the baseline BL, where the baseline BL is the measured weight as long as the base device is not loaded with a pet. For example, if the base device consistently measures a weight of 100g for longer than 180s, 100g is set as the baseline UL. If the next measurement results in, for example, 3200g, this value is corrected to 3100g.

[0020] It is advantageous if the baseline BL is determined as the last value measured over a period of at least 30s, as the lowest value in the first 40% of measurements that remains the same for at least 3s, or, in the case of sub-datasets, as the baseline of the previous dataset.

[0021] It is advantageous if such sensor data are classified as relevant for monitoring the health status of a pet that has a relevance value R dependent on a pet p, the sensor data d and the use case a p,d,a of 1, where R p,d,a is determined as follows: R p,d,a = 1 if: Use case a = “Base device is under pet toilet” AND for a duration of > 10s at a reference weight GR p of the pet p a measured weight GM d before with 0.75 GR p < GM d < 1.2 GR p OR Use case a = “Base device is not under pet toilet” AND for a duration of > 5s at a reference weight GR p of the pet p a measured weight GM d before with 0.9 GR p < GM d < 1.12 GR p R p,d,a = 0 in all other cases.

[0022] Here, the pet-dependent reference weight GR p a weight of the pet determined and communicated by the pet owner or a weight of the pet resulting from the first or previous measurements of the pet.

[0023] The relevance value R p,d,a be determined by a relevance function R(p,d,a).

[0024] It is advantageous if such sensor data are used for the identification of the pet, which has a higher verification value S compared to other sensor data pd where the verification value S pd is formed from Sp,d=1Gp,d×log(Dp,d)×Cp,d×Rep,d×Bp,d where: • -The weight value G depending on the pet p and the data set d p,d is an indicator of the similarity of a measured weight GM d of the pet p to a reference weight GR pof the pet p, is a numerical value 0.05 ≤ G p,d ≤ 1 and is determined by: Gp,d=max(|1−GMdGRp|;0.05) where the measured weight GM d is a weight of the pet determined from the data d. Preferably, the measured weight GM d the value that is maintained in a data set or cluster within a tolerance factor of ±0.04 or a tolerance amount of 20g for the longest, preferably continuous, period. Optionally, the measured weight GM d discarded or corrected to an average value resulting from the data set if the measured weight GM d deviates too much from this average value, in particular by more than a factor of 0.05 compared to the reference weight GR p deviates. • The duration value D dependent on the pet p and the data set d p,dis a numerical value ≥ 0 and specifies the duration in seconds of the period in which the relevance value R = 1. • The correction value C depending on the pet p and the data set d p,d is a numerical value with 0 ≤ C p,d ≤ 1, is an indicator of how heavy the pet is compared to other pets and indicates the relative weight of the pet with respect to the heaviest pet as a fraction of the weight of the heaviest pet. • The reliability value Re dependent on the pet p and the data set d p,d is a numerical value with 0 ≤ R p,d ≤ 1 and is an indicator of the probability of the similarity of the data with a behaviorally typical course of the previous behavior pattern of the pet p. • The baseline value B dependent on the pet p and the data set d p,d uzst a numerical value 1 ≤ B p,d≤ 2 and an indicator of whether the difference between the measurements before loading and after completion of loading (ie, the load is fully applied) of the base device and the difference between the measurements before unloading and after completion of unloading of the base device match. It is advantageous if the baseline value B p,d is determined as the absolute difference of the baseline BL pre from before loading and the baseline BL post from after discharge deducted from 2, but at least 1: Bp,d=2−min(|BLpre−BLpost|;1)

[0025] Here, the weight value G p,d by a weight function G(p,d), the measured weight GM d by a measuring function GM(d), the duration value D p,d by a duration determination function D(p,d), the correction value C p,d by a correction function C(p,d), the reliability value Re p,dby a reliability determination function Re(p,d) and the baseline value B p,d determined by a baseline function B(p,d).

[0026] It is advantageous if, after calculating the scores S, each data point is classified and assigned to a pet in the household or to the noise category. The longest contiguous data points that achieved the highest verification score for the same pet are assigned.

[0027] It is advantageous if, after determining the relevance values ​​R p,d,a the verification value S pd is determined in a second stage if the sum of the relevance values ​​across all pets p, with the data sets d and the base devices v is at least 1: ∑p from petsR(p,d,v)≥1

[0028] It is advantageous if the base device is configured to transmit sensor data when at least one of the following conditions is met: • The last data transmission took place more than 180s ago. • The measured value changes by more than a factor of 0.004 or 20g • The measured value has not changed by more than a factor of 0.004 and 20g, but has only been transmitted once in the last 10 hours.

[0029] Disclosed is a central data processing device comprising: a central data transmission means configured to receive transmitted sensor data from a base device, an analysis means configured to classify the health status of a pet located in a spatial environment in which the base device is installed, by inputting the received sensor data, and to output said health status as diagnostic data, wherein the central data transmission means is configured to transmit diagnostic data to a terminal.

[0030] It is advantageous if the central data transmission means is configured to receive image data from a pet via the terminal, and the analysis means is configured to determine a physical characteristic of the pet from the image data.

[0031] It is advantageous if the physical characteristic is a BCS (Body Condition Score) of the pet.

[0032] It is advantageous if the analysis tool has an MLM (machine learning model) by means of which the BCS of the pet is determined.

[0033] It is advantageous if the analysis tool has an MRM (machine reasoning model) by means of which the health status of the pet is classified.

[0034] It is advantageous if the MRM comprises a causal model, wherein the causal model comprises at least one rule, wherein the rule has at least one antecedent and one consequent. Preferably, the rule further comprises a source.

[0035] The antecedent preferably reflects the following facts: • Characteristics, symptoms or observations • Comparisons (greater than, less than, equal to, not, in, exists) with values • Values ​​according to different scales (e.g. different Likert scales (never, rarely, occasionally, often, very often), absolute numbers, ratios) • Logical operation with AND or OR • Preconditions (“provided that a rule applies / does not apply”)

[0036] For example, an antecedent may show: the pet has received / not received a tick vaccination AND the body weight difference in the last 7 days is smaller / greater than a factor of 0.1 of the current weight (weight loss of 10% compared to the previous week) AND the rest period duration in the last 7 days is smaller / greater than a factor of 0.1 of the average rest period duration of the previous week.

[0037] The consequent preferably represents a classification into a diagnosis of a pathological condition and its reliability. For example, it represents that, given a specific sequence of truth values ​​from the antecedent, a diagnosis is correct with a specific probability. If, based on the previous example of an antecedent, the vaccination is denied, and the weight loss and increased need for rest are affirmed, the correct diagnosis of Lyme disease would be assigned a high reliability / probability, for example, 65%. If the increased need for rest were denied, the reliability of the Lyme disease diagnosis would drop to 43%. The reliability can also assume a negative value, thus representing a contraindication for a specific diagnosis.

[0038] Preferably, as part of the consequent, there is an injective mapping between a diagnosis of a pathological condition and an urgency recommendation for consulting a veterinarian. For example, the diagnosis of obesity is assigned a lower urgency than that of a Lyme disease infection.

[0039] Preferably, the reasoning explains why the antecedent leads to the consequent. This could, for example, be a reference to a statistical survey or a reference to a principle of action.

[0040] Preferably, the causal model is created from the knowledge base of veterinarians, from research results and from general principles.

[0041] If the antecedent is empty, the consequent represents a general probability of occurrence for a particular diagnosis.

[0042] Preferably, the causal model is modified with the help of an optimization MLM to increase the reliability of the diagnoses. An MLM is trained by collecting the anamnestics of a certain statistically significant number of source pets and using this data as input for the MLM, which is then classified according to diagnosis. The MLM is used to verify the causal model and, in particular, to adjust the reliability of the respective diagnoses. Furthermore, additional rules can be determined that represent previously unidentified causalities.

[0043] Disclosed is a terminal device comprising: a terminal data transmission means configured to receive diagnostic data from a central data processing device, and a terminal display means configured to output the classified health status in user-readable form.

[0044] It is advantageous if the terminal further comprises: a terminal input means configured to receive basic information about the pet as input, wherein the terminal data transmission means is configured to transmit the basic information from the terminal to the central data processing device.

[0045] It is advantageous if the input means is an image sensor which is configured to capture image data of a pet as basic information.

[0046] A system for monitoring the health status of a pet is disclosed, comprising: one of the previously described base devices, one of the previously described central data processing devices, and one of the previously described terminal devices.

[0047] Disclosed is a method for establishing an MRM (machine reasoning model), comprising the steps of: a) collecting initial data about a pet, b) collecting ongoing health-relevant data about the pet using the above basic device, c) capturing and normalizing veterinary expertise, d) MRM creation of a causal model from the expertise captured in step c), e) optimizing the MRM using an MLM trained with the initial data from step a) and the ongoing health-relevant data from step b).

[0048] A method for pet health monitoring is disclosed, comprising the steps of: h) collecting initial data about a pet by means of the above terminal device, i) transmitting the initial data by means of the terminal data transmission means, j) collecting ongoing health-relevant data about the pet as health data by means of the above basic device, k) processing the health data, wherein relevant data is filtered by means of an MLM, i) transmitting the processed health data by means of the basic data transmission means to the above central data processing device, m) analyzing the transmitted health data by the analysis means, wherein the health data is input to a structural causal model as data input, a causal model is input as an assumption, and a classification requirement is input as a query, n) transmitting the classification result from step m) and outputting the classification result to a user.

[0049] Causal inference is an important open problem in ML. Machine reasoning (MR) approaches aim to make causal chains accessible to ML. The combination of causal chains with ML techniques creates a system that maps fundamental cause-effect pairings in medicine better than purely statistical methods and thus surpasses their inference capabilities. MR algorithms aim to overcome the shortcomings of ML algorithms and, in particular, to reveal models of causal chains, thus enabling human understanding (looking into the black box). So-called structural causal models (SCMs) play a role in this. Corresponding MR systems essentially have three input options: the causal model (also called assumption or graphical model), the queries, and data. These generate three outputs.These are an estimate E and an estimate E. s and fit indices F. The estimate E generates guidance based on the causal model for answering queries when data are available. When the estimate E then receives data, this guidance is used to generate an actual estimate E s for the answer. This is accompanied by estimated statistical parameters that reflect the confidence in the answer. These parameters account for the limited size of the data set, as well as possible measurement errors or missing data. Finally, the system creates a list of deviation indicators F that indicate how well the data fit the causal model. In ML optimization, the deviation indicators F are then successively minimized by adapting the causal model. An embodiment of the invention is described with reference to the figures. Fig. 1 shows a system for monitoring the health status of a pet according to an embodiment. Fig. 2 shows a data flow diagram of a machine reasoning algorithm of the embodiment. Fig. 3 shows a detailed data flow diagram for the creation of a machine reasoning algorithm of the embodiment. Fig. 4 shows a flowchart of the creation of an optimized machine reasoning algorithm of the embodiment. Fig. 5 shows a flowchart of a diagnostic method of the embodiment.

[0050] Based on Fig. 1 describes an embodiment of a system for monitoring the health status of a pet.

[0051] A health status monitoring system 1 comprises a base device 2, a central data processing device 3 and a terminal device 4.

[0052] The base device 2 is configured to detect a measurable health-relevant characteristic of a pet by means of a sensor, in the exemplary embodiment a dead weight of the pet or a BCS (Body Condition Score), and has a base housing 201 with a load surface 203. The base device 2 is configured to be used in a specific orientation with respect to the direction in which gravity acts on the base device, hereinafter referred to as the vertical direction. In the exemplary embodiment, information such as top and bottom, vertical and horizontal refer to this orientation of use. In the exemplary embodiment, the load surface 203 is arranged on the upper side of the base housing 201 and is oriented substantially horizontally. The load surface 203 is suitable and intended for storing objects whose weight is to be determined.The base device 2 further comprises four weight sensor surfaces 205 at locations where the base device 2 is intended to be supported, for example, on a table or on a room floor. In the exemplary embodiment, the base housing is provided with four legs, at the lower end of which spring-loaded piezoelectric modules are provided. These modules emit an electrical voltage proportional to the weight force acting on the weight sensor surfaces 205, which can be detected and evaluated.

[0053] The base device 2 further comprises a base data processing means 207. This is a device that is suitable and intended to receive digital data as input, process it, and output or store digital data based on the input and processing. Analog data can also be input here, which is converted into digital form by the base data processing means for data processing. In the exemplary embodiment, the base data processing means 207 receives a sensor signal from the weight sensor surfaces 205 and, by subtracting the weight, determines a weight force acting on the load surface 203. The base device 2 is set up in a location where the pet to be monitored regularly stays. For example, the base device 2 can be set up under a preferred sleeping place of a house cat to be monitored.Alternatively or additionally, the or another base device 2 can be placed, for example, under a litter box for a domestic cat to be monitored. The base data processing means determines a time series of the weight force acting on the load surface and determines diagnostically relevant changes in the weight force F. g For example, the weight of the sleeping area or toilet container is determined from the weight force, which remains unchanged over a longer period and is detected by the weight sensor 205, and subtracted from the other determined weights. If the pet to be monitored now enters the base device 2, this is indicated by a sudden increase in the weight force F g and leaving by a sudden drop in the weight force F g By comparing the weight force F gBefore and after entering, the mass of the excreta can be determined for the toilet container, and for the sleeping area, whether a toy was brought in and left behind, which can be taken into account for calculating the pet's own weight. The magnitude of the sudden increase in the weight force F gThe self-weight of the pet to be monitored can be determined. This weight recording data is stored in a permanent data memory 209 in the basic data processing device. This allows the pet's weight and excretion mass to be recorded in individual measurements. As a time series, this allows the recording of self-weight changes, changes in the amount of excretion, and the time and duration of stay on the base device. Furthermore, changes resulting from temperature fluctuations and sensor drift of the weight sensor 205 itself, which can occur due to the continuous activation of the sensors, must be taken into account and compensated for in addition to the above framework conditions. The basic data processing device has a machine learning model (MLM) for data processing, which receives the time series data as input and classifies it according to: -Anamnesis-relevant body weight is present -there is a change in body weight relevant to the history -Anamnesis-relevant excretory mass is present -there is a change in excretory mass relevant to the history -Anamnesis-relevant type of excretion is present -history-relevant duration of use is present -Anamnesis-relevant time of use is present -frequency of use relevant to the history is present -history-relevant usage behavior is present -there is a change in usage behavior relevant to the history

[0054] The machine learning model is an artificial neural network (ANN), in particular a recurrent ANN. This makes it possible to detect over time whether one of the classifications described above occurs at a specific point in time during the usage process. In particular, a reservoir ANN and advantageously an LSM (Liquid State Machine) is used as the ANN, since this allows a spatiotemporal pattern to be mapped, which can be used, in particular, to determine a classification occurring during the time series based on the type and temporal progression of the feature values. If one or more of the classifications is determined, this state is stored together with the associated measurement data in the data memory 209 of the basic data processing means.The machine learning model operated by the basic data processing means 207 functions, in a first function, as a filter that determines the relevant measured values ​​and compensates for the influences caused by environmental influences and pet behavior. In a second function, the machine learning model assigns the respective pets to the recorded parameters. However, the second function can also be performed by a machine learning model on the central data processing device 3, particularly if data on different individuals of pets for a household is stored there.

[0055] The MLM is previously trained in a conventional manner using comparison data sets, whereby the MLM receives sensor profile data of normal pet behavior and atypical pet behavior as input for training.

[0056] The base device 2 further comprises a basic data transmission means 211, for example a mobile radio interface configured to establish a connection to a wide area network, in particular the Internet, or a WLAN interface or an Ethernet interface configured to establish a connection to a wide area network, in particular the Internet, via a local network and to transmit the recorded health-relevant characteristics of the pet.

[0057] The central data processing device 3 is a device suitable and intended to receive digital data as input, process it, and output or store digital data based on the input and the processing.

[0058] The central data processing device 3 has a central data transmission means 301, for example a network interface such as a WLAN interface or an Ethernet interface, which is configured to establish a connection to a wide area network, in particular the Internet, via a local network and to receive recorded health-relevant properties of the pet transmitted by the base device 2.

[0059] The central data processing device 3 further comprises an analysis means 303, which is configured to classify the health status of the pet to be monitored by inputting the received health-relevant sensor data and to output it as diagnostic data. The central data processing device 3 can be located at a central location, for example, a data center. The analysis means 303 comprises a machine reasoning model (MRM), which receives the received health-relevant sensor data as input and classifies a pathological condition of the pet and, based on Fig. 2 and Fig. 3. The central data processing device 3 is configured to transmit the classified health status via the central data transmission means. Furthermore, the central data processing device 3 can perform the second function of assigning the respective pets to the recorded parameters. For this purpose, the analysis means 303 comprises a machine learning model configured to assign measured parameters to individual pets.

[0060] The central data processing device 3 further comprises a central memory 305. This memory stores health-related data and data that enable identification of individual pets.

[0061] Terminal 4 is a data processing device suitable and intended to receive digital data as input, process it, and output or store digital data based on the input and processing. In particular, it is configured to visualize digital data and record and transmit environmental data such as sound data or image data. In the exemplary embodiment, terminal 4 is a smartphone.

[0062] The terminal 4 has a terminal data transmission means 401, for example a mobile radio interface configured to establish a connection to a wide area network, in particular the Internet, or a WLAN interface or an Ethernet interface configured to establish a connection to a wide area network, in particular the Internet, via a local network and to receive the data of the classified health status of the pet.

[0063] The terminal 4 further comprises a terminal display means 403, for example a data viewing area, which is configured to display information in user-readable form, which enables the user to detect the classified health status of the pet.

[0064] The terminal device 4 has a terminal input means 405 that enables the terminal device 4 to receive input data. In the exemplary embodiment, the smartphone has a camera and text input means such as a keyboard with a corresponding user interface for data input as the terminal input means 405. For the initial configuration of the analysis system, initial data of the pet to be monitored, such as breed, age, initial weight, previous illnesses, identification information regarding basic devices, and the like, can be acquired here. Several different pet individuals of the respective household can be entered, each transmitted via the terminal data transmission means 401 and the central data transmission means 301 to the central data processing device 3, and taken into account when classifying the health status of the pet to be monitored by the MRM.Alternatively, a photograph of the pet to be monitored can be transmitted and evaluated by the analysis means 303 using image recognition, and from this, initial health-relevant information about the pet to be monitored can be determined.

[0065] Based on Fig. 2 describes a rough structure of the machine reasoning model. Fig. 2 shows a data flow diagram of an MR algorithm of the embodiment.

[0066] Machine reasoning algorithms and machine reasoning models form a structural causal model 5 (Structural Causal Models - SCM), in which input data is combined with the causal models to obtain the above-mentioned classifications as output. On the input side of the Fig. The data flow diagram of a structural causal model 5 shown in Figure 2 contains the causal model 501 (also called assumption or graphical model), which maps causal relationships between parameters and classification. The input side of the structural causal model 5 also contains the parameters measured on the pet as data input 503. In the exemplary embodiment, these are the weight time series. The input side of the structural causal model also contains the query 505, which transmits a classification request to the structural causal model.

[0067] On the output side of the structural causal model 5, there is an estimate 507 (Estimand) E. The estimate E is a guide generated on the basis of the causal model 501 for performing the classification according to the query, taking into account the data input 503. If the estimate E is served with the parameters of the data input, an actual estimate 509 (Estimate) E is generated for the query 505 based on the guide. sfor classification. On the output side of the structural causal model 5 there are also deviation indicators 511 (fit indicators) F. These deviation indicators F are statistical parameters that estimate the probability of error in the classification and thus indicate the reliability of the answer. This takes into account disturbances that can arise from a limited size of the data set of the data input 503, measurement errors, or data gaps. A list of deviation indicators F is created that indicate the extent of congruence of the data 503 with the causal model 501. In the context of conventional machine learning model optimization, the deviation indicators are successively minimized by adapting the causal model 501.

[0068] Based on Fig. 3, the data flow of the embodiment of the system for monitoring the health status of a pet is described, wherein the rough structure consists of Fig. 2 is implemented.

[0069] The machine reasoning model 601 at the center is based on scientific findings on relationships between symptoms and diseases. These are collected in a knowledge base 603. The knowledge base 603 contains the knowledge of veterinarians 607 and the literature. The content of the knowledge base 603 was processed and normalized in exchange with veterinarians. For this purpose, a uniform ontology and grammar for expressing relationships was created for the knowledge base 603. For this purpose, a homonym list with meanings and context is defined. Furthermore, a synonym list is defined in which synonyms are assigned to a main synonym. Furthermore, heteronym lists are created and, if necessary, diversified using synonyms. The knowledge is transferred using a language such as the Semantic Web Rule Language (SWRL). The purpose of this language is to formulate rules consisting of premises, conclusions, and justifications.These rules then initially determine the causal model required for the machine reasoning algorithm, i.e., a directed acyclic graph in the form of a Bayesian network. To improve the initial causal model generated from expert and literature knowledge, a causal model optimization machine learning model 609 is used. This model uses data from various sources, which are aggregated in a data pool 605. The data sources in the example are: • Transferred weight data of the base device 2. • Data collection performed by a pet keeper 606 via the terminal 4. • Data donations by veterinarians 607 without weight history, which are entered via a knowledge conversion interface 611. • Data donations through social media campaigns without weight history entered via the knowledge conversion interface 611.

[0070] Based on Fig. 4 describes a method for optimizing a machine reasoning algorithm.

[0071] In an initial data collection step S101, the pet keepers 606 enter master data of the individual pets to be monitored via their terminal devices 4. In the exemplary embodiment, for domestic cats as the pets to be monitored, this data includes: • Age and gender (once) • Race (unique) • Indoor cat / outdoor cat (one-time) • Sterilization (one-time, but adjustment possible) • Amount of food in portions and amount of water in administered quantities (ongoing) • Vaccinations (ongoing)

[0072] The collected data is assigned to an individual pet and stored as part of the individual health data 613.

[0073] In a health data monitoring step S103, health-related data is collected by the base device 2. In the exemplary embodiment, this is a time series of weight data, excretion amounts, rest periods, and excretion times. The health-related data is assigned to an individual pet and stored as part of the individual health data 613.

[0074] In a knowledge acquisition step S105, veterinarians 607 aggregate knowledge and enter it via the knowledge conversion interface 611, convert it, normalize it and store it as a knowledge base 603.

[0075] In a collective data collection step S107, data is collected through collective measures such as social media campaigns and entered, converted, normalized and stored as a knowledge base 603 via the knowledge conversion interface 611.

[0076] In a data aggregation step S109, the data from steps S101, S103, S105, S107 and S109 are combined in a data pool 605.

[0077] In a machine reasoning model creation step S111, the machine reasoning model 601 is created from the knowledge base 603. For this purpose, a uniform ontology and grammar for expressing relationships are created for the knowledge base 603. The knowledge is transferred using a language such as the Semantic Web Rule Language (SWRL).

[0078] In a machine reasoning optimization step S113, an optimization machine learning model is created from data pool 605. The optimization machine learning model optimizes the initially created causal models. To do this, new causal relationships are discovered and existing relationships are validated. To this end, the probabilities of the conclusions are "learned" by the optimization machine learning model and further causal relationships are discovered (causal discovery) that were not yet or not adequately represented in the initial model. This is all the more important because, for example, changes in weight data or toilet visits can already be detected at a time when a layperson does not yet perceive any signs of disease. For the evaluation, a focus was placed on the diseases obesity, diabetes mellitus, and FLUTD (Feline Lower Urinary Tract Disease). Case data on these diseases were specifically collected.The quality of the machine reasoning model was measured by whether it actually identified known diseases. Detection is considered successful if the actually known disease appears in the top three differential diagnoses of the machine reasoning model.

[0079] Additionally, an algorithm can be used to generate new, simulated case data based on real case data and add it to data pool 605. For this purpose, the input parameters are first grouped into clusters using principal component analysis (PCA). Different clusters of different animals are mixed together, and individual input parameters are varied. This procedure is derived from evolutionary algorithms and maps the phases of selection, recombination, and mutation. The variation (mutation) of the input parameters is always limited so that no "unnatural" case data is created. The new case data is then presented to veterinarians who make diagnoses. This procedure ensures diversity in the database. Edge cases can be specifically generated in this way.The proportion of simulated case data should be kept small if possible and only serve to increase the data basis.

[0080] Based on Fig. 5 describes a pet health monitoring procedure.

[0081] In an initial data collection step S201, the pet keepers 606 enter master data of the individual pets to be monitored via their terminal devices 4. In the exemplary embodiment, for domestic cats as the pets to be monitored, this data includes: • Age and gender (once) • Race (unique) • Indoor cat / outdoor cat (one-time) • Sterilization (one-time, but adjustment possible) • Amount of food in portions and amount of water in administered quantities (ongoing) • Vaccinations (ongoing)

[0082] In an initial data transmission step S203, the initial data is transmitted to the central data processing device 3 by means of the terminal data transmission means 405 and stored in the central memory 305.

[0083] In a health data monitoring step S205, health-related data is collected by the base device 2. In the exemplary embodiment, this is a time series of weight data, excretion amounts, rest periods, and excretion times. The data is determined by monitoring the output signals of the weight sensor system 205.

[0084] In a data preparation step S207, the basic data processing means 211 uses the data preparation machine learning model to determine health-relevant data from all data.

[0085] In a health data transmission step S209, the determined health-relevant data are transmitted via the basic data transmission means 211 to the central data processing device 3 and stored in the central memory 305.

[0086] In an analysis step S211, the initial data and the health-relevant data are entered into the analysis means 303 as data input 503, the causal model 501 obtained in the process for optimizing a machine reasoning algorithm is entered as a causal model, and the question about the presence of a specific disease, for example, diabetes mellitus, is entered as query 505. As Estimate E s the classification of whether this disease is present is displayed.

[0087] In a diagnosis display step S213, the classification is displayed to an assigned veterinarian after a corresponding data transmission or to the pet keeper 606 via the terminal device 4 via the display means.

[0088] The invention has been described using exemplary embodiments. These exemplary embodiments are merely illustrative and do not limit the invention as defined by the claims. Those skilled in the art will recognize that deviations from the exemplary embodiment are possible without departing from the scope of the claims.

[0089] In the exemplary embodiment, domestic cats were used as the pets to be monitored. The pet health monitoring system comprising the base device, central data processing device and terminal, and pet health monitoring method can be applied to other pets, such as those from the Canidae family or small rodents, without modification or with modifications within the knowledge of the person skilled in the art.

[0090] In the exemplary embodiment, the weight sensor system of the base device is designed as sensor surfaces outside the housing on the underside as piezoelectric modules that emit an electrical voltage linearly or non-linearly proportional to a weight force. Alternatively, sensor surfaces can be arranged within the housing in a multi-part housing, wherein the parts are movable relative to one another. Furthermore, alternatively, the sensor surfaces can be arranged or configured such that, if a corresponding force deflection of the weight force is carried out, they can detect forces in a non-vertical direction. Furthermore, alternatively, instead of a voltage change according to a force acting on the weight surface, the sensor surfaces can cause an electrical resistance change, which is detected and evaluated in order to determine the weight force.Furthermore, alternatively, the sensor surfaces can be replaced by sensors that detect a change in position proportional to the weight force by means of Hall sensors, ultrasonic sensors, optical sensors or the like.

[0091] The connectives "and," "or," and "either ... or" are used in a sense similar to logical conjunction (logical AND), logical adjunction (logical OR, often "and / or"), and logical contravalence (logical exclusive OR), respectively. In particular, in contrast to "either ... or," the connective "or" can imply the simultaneous presence of both operands.

[0092] A list of process steps in the description and claims merely serves to enumerate the required process steps. It does not imply a necessary order or sequence of the process steps unless such an order or sequence is explicitly stated or is obvious to a person skilled in the art. Furthermore, such a list does not imply its completeness.

[0093] The term “comprising” does not imply an exhaustive list in the claims; the presence of further elements and steps is possible.

[0094] The use of the indefinite article “ein” or “eine” does not exclude the presence of a plural, but is to be understood as “at least one” or “at least one”, unless it is qualified as “exactly one” or “exactly one”.

[0095] Where the agreement between two values ​​is mentioned or it is mentioned that a value remains the same, values ​​which do not differ by more than a tolerance factor of 0.01 or a tolerance weight of 20g are also considered to be the same or the same.

[0096] Furthermore, in the context of this invention, the following terms are understood in the following meaning: Artificial intelligence (AI), also known as artificial intelligence (AI), is a branch of computer science that encompasses all efforts aimed at making machines intelligent. Intelligence is defined as the ability to act appropriately and proactively in its environment; this includes the ability to perceive and respond to sensory impressions, to absorb, process, and store information as knowledge, to understand and generate language, to solve problems, and to achieve goals.

[0097] Machine learning (ML) is a generic term for the "artificial" generation of knowledge from experience: An artificial system learns from examples and can generalize them after the learning phase is completed. To achieve this, machine learning algorithms build a statistical model based on training data, which is then tested against the test data. This means that the examples are not simply memorized (see overfitting), but rather patterns and regularities are recognized in the learning data. This allows the system to evaluate unknown data (learning transfer) or fail to learn unknown data (overfitting). Among the wide range of possible applications, these include: automated diagnostic procedures, credit card fraud detection, stock market analysis, nucleotide sequence classification, speech and text recognition, and autonomous systems.

[0098] In information technology, a machine reasoning (MR) system is a software system that draws conclusions from existing knowledge using logical techniques such as deduction and induction. Reasoning systems play an important role in the implementation of artificial intelligence and knowledge-based systems. In typical usage in the field of information technology, the term is usually reserved for systems that perform more complex types of reasoning. For example, not for systems that draw fairly simple types of reasoning, such as calculating a sales tax or a customer discount, but for systems that draw logical conclusions about a medical diagnosis or a mathematical theorem. There are two types of reasoning systems: interactive and batch.

[0099] Interactive systems engage with the user to ask clarifying questions or otherwise allow them to guide the reasoning process. Batch processing systems take in all available information at once and generate the best possible answer without user feedback or guidance.

[0100] Body Condition Score (BCS) is a body condition assessment for animals. It is classified on a scale of 1 to 5. The categories are very poor condition (emaciated), skeleton clearly visible, balanced skeleton-to-meat ratio, predominant meat-fat content, and excessively fat.

[0101] Classification describes the assignment of an object, for example a pet, to a group of objects with one or more predetermined properties. List of reference symbols 1 health status monitoring system 2 base device 201 base housing 203 load area 205 Weight sensors 207 Basic data processing tools 209 data storage 211 Basic data transmission means 3 central data processing device 301 central data transmission means 303 analytical tools 305 central storage 4 End device 401 Terminal data transmission means 403 Terminal display means 405 Terminal input devices 5 Structural causal model 501 Causal model 503 health-related pet parameters, weight time series, data entry 505 query 507 Estimated value, Estimand E 509 Estimate, Estimate E s 511 Deviation indicators, Fit indicators F 6 Health status monitoring system 601 Machine Reasoning Model 603 Knowledge base 605 Data pool 606 pet sitters 607 veterinarians 609 Causal model optimization-machine learning model 611 Knowledge Conversion Interface 613 individual health data QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] US 8,797,166 B2

[0004]

Claims

[1] A system for monitoring the health status of a pet animal, comprising: a basic device, a central data processing device and a terminal device, where the base device has: at least one sensor for detecting a measurable health-related characteristic of a pet and a basic data processing means configured to process sensor data acquired by the sensor, and a base data transmission means configured to transmit processed sensor data from the base device, the central data processing device has: a central data transmission means configured to receive the transmitted sensor data, an analysis means configured to classify the health status of a pet that is in a spatial environment in which the base device is installed, using the received sensor data, and to output it as diagnostic data, and the terminal device has: a terminal data transmission means arranged to receive the diagnostic data, and a terminal display means configured to output the classified health status in user-readable form. [2] A base device comprising: at least one sensor for detecting a measurable health-related characteristic of a pet and a basic data processing means configured to process sensor data acquired by the sensor, and a base data transmission means configured to transmit processed sensor data from the base device. [3] The base device according to the preceding claim, wherein the basic data processing means is designed to classify from the sensor data those sensor data which are relevant for monitoring the health status of a pet, and the basic data transmission means is configured to transmit only the sensor data classified as relevant. [4] The basic device according to the preceding claim, wherein the basic data processing means comprises an MLM (machine learning model) by means of which the relevant sensor data are classified. [5] The base device according to any one of the preceding claims, wherein such sensor data are classified as relevant for monitoring the health status of a pet if they have a relevance value R dependent on a pet p, the sensor data d and the use case a p,d,a of 1, or such sensor data are classified as relevant for monitoring the health status of a pet that has a verification value S pd of 0.85, whereby the verification value S pd is formed from Sp,d=1Gp,d×log(Dp,d)×Cp,d×Rep,d×Bp,d where G p,d a weight value dependent on the pet and data set, D p,d a duration value dependent on the pet and data set, C p,d a correction value depending on the pet and data set, Re p,d a reliability value dependent on the pet and data set and B p,d is a pet and dataset dependent baseline value. [6] A central data processing device comprising: a central data transmission means configured to receive transmitted sensor data from a base device, an analysis means configured to classify the health status of a pet that is in a spatial environment in which the base device is installed, using the received sensor data, and to output it as diagnostic data, wherein the central data transmission means is configured to transmit diagnostic data to a terminal device. [7] The central data processing device of the preceding claim, wherein the central data transmission means is configured to receive image data from a pet via the terminal, and the analysis means is set up to determine a physical characteristic of the pet from the image data. [8] The central data processing device of the preceding claim, wherein the physical characteristic is a BCS of the pet. [9] The central data processing device of the preceding claim, wherein the analysis means comprises an MLM (machine learning model) by means of which the BCS of the pet is determined. [10] The central data processing device of any one of the preceding claims, wherein the analysis means comprises an MRM (machine reasoning model) by means of which the health status of the pet is classified. [11] The central data processing device of any preceding claim, wherein the MRM comprises a causal model, the causal model comprising at least one rule, the rule comprising: at least one prerequisite (antecedent), and a consequence (Consequent), and preferably a justification (source), where: the requirement includes characteristics of the pet to be examined, and the conclusion includes a diagnosis and a confidence value for the diagnosis and preferably a consultation urgency. [12] A terminal device that has: a terminal data transmission means configured to receive diagnostic data from a central data processing device, and a terminal display means configured to output the classified health status in user-readable form. [13] The terminal according to the preceding claim, further comprising: a terminal input means arranged to receive basic information about the pet as input, wherein the terminal data transmission means is configured to transmit the basic information from the terminal to the central data processing device. [14] The terminal according to the preceding claim, wherein the input means is an image sensor configured to capture image data of a pet as basic information. [15] A system for monitoring the health status of a pet animal, comprising: a base device according to any one of claims 2-5, a central data processing device according to any one of claims 6-11 and a terminal according to any one of claims 12-14. [16] A method for establishing an MRM (Machine Reasoning Model) comprising the steps of: a) collecting initial data on a pet animal; b) collecting continuous health-related data about the pet by means of a base device according to any one of claims 2-5, c) recording and standardising veterinary expertise, d) MRM creation of a causal model from the expertise captured in step c), e) Optimization of the MRM using an MLM trained with the initial data from step a) and the ongoing health-relevant data from step b). [17] A method for pet health monitoring comprising the steps of: h) collecting initial data about a pet by means of a terminal according to any one of claims 12-14; i) transmitting the initial data by means of the terminal data transmission means, j) collecting continuous health-related data about the pet as health data by means of a base device according to any one of claims 2-5, k) Preparation of health data, whereby relevant data is filtered using an MLM, I) transmitting the processed health data by means of the basic data transmission means to a central data processing device according to one of claims 6-11, m) analysis of the transmitted health data by the analysis means, wherein the health data are input to a structural causal model as data input, a causal model obtained according to claim 16 is input as an assumption and a classification requirement is input as a query, n) Transferring the classification result from step m) and outputting the classification result to a user.

Images