Feline diabetes prognosis

Abstract

The present disclosure relates to a computer-implemented method for determining the risk of developing diabetes in a feline, the method comprising the steps of: determining the risk of developing diabetes in the feline in order to assess whether the feline is likely to be pre-diabetic; and when the feline is determined to be pre-diabetic, determining the risk of developing diabetes in the feline within a time interval. The present disclosure also relates to a method for generating a machine learning system adapted to determine the risk of developing diabetes in a feline, and a computer-implemented system for determining the risk of developing diabetes in a feline.

Description

[0001] Cross-references to related applications This application claims priority to European Patent Application No. 23213725.7, filed on 1 December 2023, the contents of which are incorporated herein by reference in their entirety, and claims protection of that priority. Technical Field

[0002] This disclosure relates to the field of diabetic prognosis in felines. Background Technology

[0003] As is well known, diabetes (also known as diabetes mellitus) is a condition in mammals where the body is unable to properly produce the hormone insulin (as in, for example, type 1 diabetes) or is unable to respond to the hormone insulin (as in, for example, type 2 diabetes).

[0004] Diabetic felines most commonly suffer from a form of diabetes similar to type II diabetes found in humans. It is estimated that between 0.2% and 1% of cats will eventually be diagnosed with diabetes in their lifetime.

[0005] Typically, clinical signs of diabetes in felines may include weight loss, excessive thirst and urination, and in rare cases, nerve and hindlimb damage.

[0006] Prediabetes is a physiological condition that precedes diabetes.

[0007] In humans, prediabetes is a metabolic state that lies between normal glucose homeostasis and diabetes, and is diagnosed by demonstrating impaired glucose tolerance (IGT) and / or impaired fasting glucose (IFG).

[0008] Conversely, in felines, particularly cats, a prediabetic state has not been defined, and reference values ​​used to determine prediabetic state in humans are largely irrelevant in felines such as cats. Illustratively, blood glucose levels alone cannot serve as a valid biomarker for diabetes in felines, as cats exhibit elevated glucose levels, particularly as a response to stress. Therefore, while validated cutoff values ​​may eventually be used in veterinary clinical practice to identify altered glucose metabolism and felines at risk of developing diabetes, these concepts remain largely confined to human medical practice.

[0009] This is also why the diagnosis of diabetes in felines (including domestic cats) has been generally delayed to date, and it is often only diagnosed when clinical signs are obvious.

[0010] Unfortunately, there is no cure for diabetes in felines, making its diagnosis a late-stage event, even though insulin therapy or dietary therapy can provide slight, temporary relief and, in some cases, induce remission. However, as is known in the art, approximately 25-30% of cats in remission relapse and require intensive, long-acting insulin therapy. Furthermore, most (76%) diabetic felines in remission have impaired glucose tolerance, and some (19%) have impaired fasting glucose, indicating a lack of normal glucose metabolism or clearance in these felines.

[0011] Unlike diabetes, prediabetes is mostly reversible, such as with appropriate treatment, and therefore worth diagnosing in felines to prevent the disease from developing.

[0012] Therefore, there is a need in the field for methods that allow for the prediction of the likelihood of diabetes occurring in felines, particularly cats. Summary of the Invention

[0013] A first aspect of this disclosure relates to a computer-implemented method for assessing the risk of a feline becoming a prediabetic feline, the method comprising: a) Provide a set of at least six physiological characteristics of a feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level; b) Generate feature scores from the set of physiological features to predict the risk of developing diabetes in felines by operating a first machine learning model trained on a set of physiological features; c) Based on this characteristic score, determine whether the feline is at risk of developing diabetes; d) When the feline is at risk of developing diabetes, designate it as a prediabetic feline; and e) Optionally, generate a diet for the prediabetic feline.

[0014] In some implementations of the computer-implemented method, step a) may include providing (i) a set of one or more characteristics of the feline, selected from the feline's age, weight, and breed, and (ii) a set of three or more characteristics measured from a previously obtained sample from the feline, selected from blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0015] In some implementations of the computer-implemented method, step a) may include providing a set of features consisting of age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0016] The study found that physiological characteristics of felines selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level were statistically significantly correlated with the status of diabetes in felines.

[0017] Furthermore, it has been demonstrated that the use of at least six physiological features selected from the above list provides sufficient predictive accuracy while minimizing the amount of data required. In some embodiments of the computer-implemented method, step a) may also include providing second-order or higher-order cross features computed from the selected features.

[0018] Another aspect of this disclosure relates to a computer-implemented method for assessing the risk of developing diabetes in prediabetic felines over a time interval, the method comprising: a') By ​​operating a second machine learning model trained on a second set of at least six physiological characteristics of felines, a second characteristic score is generated from the set of physiological characteristics that predicts the development of diabetes in prediabetic felines over a period of time. These physiological characteristics are selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. b') Based on the second feature score, determine the risk of developing diabetes in prediabetic felines over a time interval; c') Optionally, a diet for the prediabetic feline is generated based on the time when the risk of developing diabetes occurs.

[0019] In some implementations of the computer-implemented method, step a') may include providing (i) a set of one or more characteristics of the prediabetic feline, selected from the feline's age, weight, and breed, and (ii) a set of three or more characteristics measured from a previously obtained sample from the prediabetic feline, selected from blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0020] In some implementations of the computer-implemented method, step a') may include providing a second set of features consisting of age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0021] In some implementations of the computer-implemented method, step a') may also include providing second-order or higher-order cross features computed from the selected features in step e).

[0022] In some implementations of the computer-based method, the cross features may consist of third-order cross features.

[0023] In some implementations of the computer-implemented method, the selected features of step a) and step a') may be the same.

[0024] In some implementations of the computer-based method, each of the first and second machine learning models used in steps b) and a') can be a trained neural network model.

[0025] In some implementations of the computer-based method, the trained neural network model may be a trained multilayer perceptron neural network model.

[0026] Neural network models, and especially multilayer perceptron neural network models, are well-suited for this type of prediction.

[0027] Predictions made by many neural networks can be interpreted as Bayesian posterior probabilities, see, for example, M. D. Richard et al. Neural Network Classifiers Estimate Bayesian a posteriori Probabilities During the training phase, the neural network learns to minimize risk based on the trained dataset, typically following the principle of gradient descent.

[0028] In some implementations of the computer-implemented method, step b') may include determining the risk of a prediabetic feline developing diabetes within a six-month time interval following the measurement of the feline's characteristics, preferably within a three-month time interval following the measurement of the feline's characteristics.

[0029] Another aspect of this disclosure relates to a computer-implemented method for assessing the risk of developing diabetes in felines, the method comprising: a) Provide a set of at least six physiological characteristics of a feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level; b) Generate a first feature score from the set of physiological features to predict the risk of developing diabetes in felines by operating a first machine learning model trained on the set of physiological features; c) Based on the first feature score, determine whether the feline is at risk of developing diabetes; d) When the feline is at risk of developing diabetes, designate the feline as a prediabetic feline; e) By operating a second machine learning model trained on a second set of at least six physiological characteristics of a feline, generate a second characteristic score from the second set of physiological characteristics that predicts the development of diabetes in the prediabetic feline over a period of time. These physiological characteristics are selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. The second set of physiological characteristics may be the same as or different from the set of physiological characteristics in step b). f) Based on the second feature score, determine the risk of developing diabetes in prediabetic felines over a time interval; g) Optionally, a diet for the prediabetic feline is generated based on the time when the risk of developing diabetes occurs.

[0030] Another aspect of this disclosure relates to a method for generating a machine learning system suitable for assessing the risk of felines becoming prediabetic felines, the method comprising the following steps: a) Generate a model suitable for determining the risk of diabetes in felines, including the following steps: i) Provide computer-implemented machine learning devices, ii) The machine learning device is trained by providing a dataset for each of a plurality of felines, wherein the dataset for each of the plurality of felines comprises: - A set of at least six physiological characteristics of the feline, selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content, and - Clinical data related to the occurrence of diabetes in the aforementioned felines. This generates a machine learning model suitable for determining the risk of felines becoming prediabetic felines.

[0031] Another aspect of this disclosure relates to a method for generating a machine learning system suitable for assessing the risk of developing diabetes in prediabetic felines over a time interval, the method comprising the following steps: a') Generate a second model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval, including the following steps: i) Provide a second computer-implemented machine learning device, ii) The second machine learning device is trained by providing a dataset of each of a plurality of prediabetic felines, wherein the dataset of each of the plurality of prediabetic felines comprises: - A set of at least six physiological characteristics of the prediabetic feline, selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content, and - Clinical data related to the development of diabetes in the aforementioned prediabetic felines. This generates a second machine learning model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval.

[0032] Another aspect of this disclosure relates to a method for generating a machine learning system suitable for assessing the risk of diabetes in felines, the method comprising the following steps: a) Generate a first model suitable for determining the risk of diabetes in felines, including the following steps: i) Provide the first computer-implemented machine learning device, ii) Training the first machine learning device by providing a dataset for each of a plurality of felines, wherein the dataset for each of the plurality of felines comprises: - A set of at least six physiological characteristics of the feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level, and - Clinical data related to the occurrence of diabetes in the aforementioned felines. This generates the first machine learning model suitable for assessing the risk of felines becoming prediabetic felines. b) Generate a second model suitable for assessing the risk of developing diabetes in prediabetic felines over a time interval, including the following steps: i) Provide a second computer-implemented machine learning device, ii) The second machine learning device is trained by providing a dataset of each of a plurality of prediabetic felines, wherein the dataset of each of the plurality of prediabetic felines comprises: - A set of at least six physiological characteristics of the prediabetic feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level, and - Clinical data related to the development of diabetes in the aforementioned prediabetic felines. This generates a second machine learning model suitable for assessing the risk of developing diabetes in prediabetic felines over a time interval.

[0033] In some implementations of the method for generating a machine learning system, steps a)-ii) may include training the first machine learning device with the following: -A dataset provided from felines tagged as not having diabetes, and - A dataset provided from felines tagged with prediabetes.

[0034] In some implementations of the method for generating machine learning systems, steps b)-ii) or a')-ii) may include training the second machine learning device with a dataset from felines labeled as prediabetic.

[0035] Another aspect of this disclosure relates to a computer-implemented system for determining the risk of a feline becoming a prediabetic feline, the computer-implemented system comprising: -processor, - A tangible, computer-readable medium operatively connected to a processor and comprising computer code configured to: a) Generate models suitable for determining the risk of felines becoming prediabetic felines, including: i) Record multiple datasets, each dataset being measured for each feline in a plurality of felines, and each dataset containing multiple feline physiological characteristics selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content. ii) Record the diabetes incidence value for each of the multiple felines in step i). iii) By using a database to train a machine learning algorithm to generate a machine learning model, a trained model for the risk of developing diabetes is developed, which is configured to generate a feature score that predicts the risk of a feline becoming a prediabetic feline.

[0036] Another aspect of this disclosure relates to a computer-implemented system for determining the risk of developing diabetes in prediabetic felines over a time interval, the computer-implemented system comprising: -processor, - A tangible, computer-readable medium operatively connected to a processor and comprising computer code configured to: a') Generate a second model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval, including: i) Record multiple datasets, each dataset being measured for each feline in a plurality of prediabetic felines, and each dataset containing multiple prediabetic feline physiological characteristics selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content. ii) For each of the multiple prediabetic felines in step i), record the time of diabetes occurrence. iii) By using a database to train a machine learning algorithm to generate a second machine learning model, thereby developing a trained model for the risk of developing diabetes over a time interval, the model being configured to generate feature scores that predict the risk of developing diabetes in the prediabetic feline over a time interval.

[0037] Another aspect of this disclosure relates to a computer-implemented system for assessing the risk of developing diabetes in felines, the computer-implemented system comprising: -processor, - A tangible, computer-readable medium operatively connected to a processor and comprising computer code configured to: a) Obtain a set of at least six physiological characteristics of a feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level; b) Generate a first feature score from the set of physiological features to predict the risk of developing diabetes in felines by operating a first machine learning model trained on the set of physiological features; c) Based on the first feature score, determine whether the feline is at risk of developing diabetes; d) When the feline is at risk of developing diabetes, designate the feline as a prediabetic feline; e) By manipulating a second machine learning model trained on a set of physiological features, a second feature score is generated from the set of physiological features to predict the occurrence of diabetes in prediabetic felines over a time interval. f) Based on the second feature score, determine the risk of developing diabetes in prediabetic felines over a time interval; g) Optionally, a diet for the prediabetic feline is generated based on the time when the risk of developing diabetes occurs.

[0038] In one implementation, the set of physiological characteristics in step b) is different from the set of physiological characteristics in step e), and each set includes at least six physiological characteristics of felines selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. The physiological characteristics used in both steps are obtained in step a).

[0039] Tangible computer-readable media may include computer code configured to: a) Generate a first-hand model suitable for assessing the risk of felines becoming prediabetic felines, including: i) Record multiple data sets in a first database, each data set being measured for each feline among multiple felines, and each data set containing multiple feline physiological characteristics selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. ii) Record the incidence of diabetes for each of the multiple felines in step i) in the first database. iii) Use the first database to train the machine learning algorithm to generate the first machine learning model. An augmented model was thus developed to assess the risk of developing diabetes. This augmented model was configured to generate a feature score that predicts the risk of a feline becoming a prediabetic feline.

[0040] In some implementations, the tangible computer-readable medium may also additionally include computer code configured to: b) Generate a second model suitable for assessing the risk of developing diabetes in felines over a time interval, including: i) Record multiple datasets in a second database. Each dataset is measured for each prediabetic feline in a plurality of prediabetic felines, and each dataset contains multiple physiological characteristics of prediabetic felines selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. ii) In the second database, record the time of diabetes occurrence for each of the multiple prediabetic felines from step i). iii) Use a second database to train the machine learning algorithm to generate a second machine learning model. An enhanced model was developed to assess the risk of developing diabetes over a time interval. This enhanced model was configured to generate feature scores that predict the risk of developing diabetes in the prediabetic feline over a time interval.

[0041] The risk of a feline becoming a prediabetic feline is assessed by performing at least steps a) to d) of the computer-implemented method as described above, and the set of characteristics is recorded as a data set in the first database, and the risk of developing diabetes is recorded as a diabetes occurrence value, which can enrich the first database.

[0042] The risk of developing diabetes in prediabetic felines over a time interval is assessed by performing at least steps a) to f) or a') to c') of the computer-implemented method as described above; and the set of features is recorded as a data set in the second database, and the risk of developing diabetes over a time interval is recorded as a diabetes time occurrence value, which can enrich the second database.

[0043] The first database and the second database can be the same database.

[0044] Another aspect of this disclosure relates to a method for preventing diabetes in felines, the method comprising the following steps: 1) Assess the risk of developing diabetes in felines by performing a computer-implemented method according to the instructions. 2) If the risk of developing diabetes in a feline has been determined in step 1), then the feline is provided with an appropriate diet.

[0045] In some implementations of the above method, step 2) may also include providing the feline with a preventative or dietary regimen.

[0046] Another aspect of the present invention relates to a method for preventing the development of diabetes in prediabetic felines over a period of time, the method comprising the following steps: 1') Assess the risk of developing diabetes in the prediabetic feline over a period of time by performing the computer-implemented method described above, and 2') If the risk of developing diabetes in the prediabetic feline has been determined in step 1) over a period of time, then preventive or therapeutic treatment is provided to the prediabetic feline.

[0047] In some implementations of the above method, step 2') may include administering a preventative or dietary regimen to the prediabetic feline. Attached Figure Description

[0048] Figure 1 An implementation scheme for a computer-based method for assessing the risk of felines becoming prediabetic felines is described; Figure 2 An implementation scheme for a computer-based method for assessing the risk of developing diabetes in felines over a time interval is described; Figure 3 Examples of machine learning models suitable for implementing methods for assessing the risk of felines becoming prediabetic and / or for assessing the risk of felines developing diabetes over a time interval are described; and Figure 4 A flowchart illustrating a computer-implemented method for assessing the risk of a feline becoming a prediabetic feline, according to one implementation scheme. Detailed Implementation

[0049] This document provides a method for assessing the risk of developing diabetes in felines, which will be described in further detail in this specification.

[0050] In one or more instances, the described techniques can be implemented using hardware, software, firmware, or any combination thereof. If implemented in software, these functions can be stored as one or more instructions or code on a computer-readable medium and executed by a hardware-based processing unit. Computer-readable media can include non-transitory computer-readable media, which corresponds to tangible media such as data storage media (e.g., RAM, ROM, EEPROM, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer).

[0051] Instructions can be configured to be executed by one or more processors, such as one or more digital signal processors (DSPs), general-purpose microprocessors, application-specific integrated circuits (ASICs), field-programmable arrays (FPGAs), or other equivalent integrated or discrete logic circuits. Therefore, the term "processor" as used herein can refer to any of the foregoing structures or any other physical structure suitable for implementing the described techniques. Similarly, these techniques can be implemented entirely within one or more circuit or logic elements.

[0052] For example, a machine learning model can be built based on selected physiological data that must be provided to a specific subject. Then, a training process can be performed using the selected physiological data to obtain a trained machine learning model. This trained model can then be invoked at runtime based on selected feline physiological data input by the user. This feline physiological data can encompass feline contextual information such as age, weight, and breed, as well as feline physiological data obtained from samples previously obtained from tested felines, such as blood glucose levels, white blood cell count (WBC), blood cholesterol levels, blood phosphorus levels, blood albumin levels, and blood alkaline phosphatase levels.

[0053] The computer-implemented methods disclosed herein can exist, for example, in web-based applications and mobile applications. The machine learning model can be implemented / deployed on a backend server that runs the machine learning model as part of a web-based application and transmits information back to the user's phone or other terminal device, and / or the machine learning model can also be implemented as part of a mobile application that runs on the user's phone or other terminal device and transmits information back to other terminal devices. Regardless of the implementation, the computer-implemented system can send push notifications to feline owners or alternatively to veterinarians to alert them in the event of an estimated prediabetes state, can receive and provide accurate and consistent responses to requests from users, such as feline owners or veterinarians, and can also provide a pathway for inputting feline physiological data, which can be used to track the feline's history and improve predictive estimates.

[0054] definition Within the context of this subject matter and in the specific context in which each term is used, the terms used herein generally have their common meaning in the art. Certain terms are defined below to provide additional guidance in describing compositions and methods of the disclosed subject matter and how to prepare and use such compositions and methods.

[0055] In the following detailed description, reference is made to the accompanying drawings, which form part of the detailed description. In the drawings, like symbols generally identify like components unless the context otherwise indicates. The illustrative embodiments described in the detailed description, drawings, and claims are not intended to be limiting. Other embodiments may be used, and other variations may be made, without departing from the spirit or scope of the subject matter presented herein. It will be readily understood that, as generally described herein and illustrated in the drawings, aspects of this disclosure can be arranged, substituted, combined, and designed in a variety of different configurations, all of which are expressly considered and constitute a part of this disclosure.

[0056] In this description, references to "implementation scheme," "an implementation scheme," "one implementation scheme," "in various implementation schemes," etc., indicate that the described implementation scheme may include specific features, structures, or characteristics; however, each implementation scheme may not necessarily include specific features, structures, or characteristics. Furthermore, such phrases do not necessarily refer to the same implementation scheme. Moreover, when a specific feature, structure, or characteristic is described in connection with an implementation scheme, whether explicitly stated or not, it is assumed that the influence of such feature, structure, or characteristic on other implementation schemes is within the knowledge of those skilled in the art. After reading this specification, those skilled in the art will understand how to implement this disclosure in alternative implementation schemes.

[0057] As used herein, unless the context clearly indicates otherwise, the singular forms “a / an” and “the” include plural indicators. Thus, for example, a reference to “a compound” includes a mixture of compounds.

[0058] As used herein, the terms “include,” “including,” “comprises,” “comprising,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article of manufacture, or apparatus that comprises a list of elements includes not only those elements but may include other elements not expressly listed or inherent to such process, method, article of manufacture, or apparatus.

[0059] The terms "about" or "approximately" mean within an acceptable margin of error for a particular value, as determined by a person skilled in the art, which will depend in part on how the value was measured or determined, i.e., the limitations of the measurement system. For example, "about" may mean within three or more standard deviations according to practice in the art. Alternatively, "about" may mean a range up to 20%, preferably up to 10%, more preferably up to 5%, and even more preferably up to 1%, of a given value. Furthermore, particularly for systems or processes, the term may mean within an order of magnitude, preferably within five times the value, and more preferably within two times.

[0060] Furthermore, the terms "at least" and "ranging from" encompass the values ​​referenced below. For example, "at least 40 ppm" must be understood to also encompass "40 ppm". Unless otherwise specified, amounts (specifically weight percentages, amounts in parts per million (ppm), or milliequivalents / kg (mEq / kg) of fat) are expressed herein as weights of products or compositions referenced (e.g., preservative food compositions according to this disclosure). In this disclosure, ranges are stated in abbreviated form to avoid having to state and describe every value within a range in detail. Where appropriate, any suitable value within the range may be chosen as the upper limit, lower limit, or terminal value of the range. For example, a range of 1 to 10 represents the terminal values ​​of 1 and 10, and intermediate values ​​of 2, 3, 4, 5, 6, 7, 8, 9, as well as all intermediate ranges covered within 1-10, such as 2 to 5, 2 to 8, 7 to 10, etc.

[0061] As used herein, depending on the context, the term "blood" encompasses whole blood, plasma, and serum. Illustratively, several procedures for determining selected blood physiological parameters can be performed indiscriminately on plasma or serum samples, in all cases yielding determined values ​​of the selected physiological parameters from previously collected blood samples.

[0062] As used herein, the term "effective amount" refers to the amount of a component, when included in a composition, sufficient to achieve the intended compositional or physiological effect. It should be understood that various biological factors can influence a substance's ability to perform its intended task. Therefore, in some cases, the "effective amount" may depend on such biological factors. Furthermore, while those skilled in the art can use evaluations known in the art to measure the realization of physiological effects, it should be recognized that individual differences can make the realization of physiological effects a subjective determination. The determination of effective amounts is entirely within the general technical scope of the field of nutritional science.

[0063] As used herein, “administration” and “administering” refer to the manner in which the composition is presented to a subject. Preferably, the food composition is administered orally.

[0064] As used herein, "oral administration" refers to a route of administration that can be achieved by swallowing or sucking a food composition.

[0065] As used herein, the term "food composition" encompasses all foods; diets; food; or materials containing at least protein, carbohydrates, and / or fats; which serve in living organisms to maintain growth, repair, and life processes; and / or to provide energy for companion animals. Food compositions as described herein may also contain supplements or additives, such as minerals, vitamins, and flavorings (see Merriam-Webster's Collegiate Dictionary, 10th edition, 1993). Food compositions according to this disclosure may consist of nutritionally complete and balanced food compositions or functional supplements. Food compositions as described herein may preferably be cooked products.

[0066] As used herein, the term "nutritionalally complete" refers to animal food products containing all known nutrients required by the intended recipient in appropriate amounts and proportions, such as based on recommendations from recognized competent authorities in the field of animal nutrition. Therefore, such foods can be used as a source of dietary intake to sustain life without the need for supplemental nutritional sources.

[0067] As used herein, the term “diabetes” refers to an incurable chronic disease in which the feline body is unable to properly produce the hormone insulin (such as, for example, type 1 diabetes) or is unable to respond to the hormone insulin (such as, for example, type 2 diabetes).

[0068] As used herein, the term "prediabetes," or sometimes "pre-diabetes," refers to a physiological state in felines where the rate of progression to type 1 or type 2 diabetes is higher than the expected rate in the absence of any therapeutic intervention (diet, exercise, medication, or others). Prediabetes can also refer to those felines that will or are expected to progress to type 1 or type 2 diabetes within a given time period or time range. It can be stated in terms of the relative risk between risk quartiles relative to normal values, or in terms of the likelihood of diabetes development based on a characteristic score provided by the computer-implemented method described herein. Unless otherwise stated and without limitation, when a category of prediabetes status in a feline is affirmatively determined according to the computer-implemented method described herein, it can also be limited by a predicted time period prior to actual progression to type 1 or type 2 diabetes, based on a given threshold. Therefore, the computer-implemented method described herein provides for determining the risk of developing diabetes in felines, i.e., for determining prediabetes status in felines, and also covers the prediction of the expected or predicted annual conversion rate for any time period. Therefore, felines identified as being in a prediabetic state during the steps of the computer-implemented method described herein have a predictable risk of developing diabetes compared to felines not yet identified as having prediabetes. In a preferred embodiment, prediabetic felines are considered to have a risk of developing diabetes within approximately 12 months. In some embodiments, prediabetic felines may be considered to have a risk of developing diabetes within approximately 12 months, approximately 11 months, approximately 10 months, approximately 9 months, approximately 8 months, approximately 7 months, approximately 6 months, approximately 5 months, approximately 4 months, approximately 3 months, approximately 2 months, or approximately 1 month.

[0069] As used herein, the term "time interval" refers to a time interval of approximately 6 months or less. The time interval should encompass periods of approximately 6 months or less, approximately 5 months or less, approximately 4 months or less, approximately 3 months or less, approximately 2 months or less, or approximately 1 month or less. In some embodiments, the time interval refers to a period of approximately 3 months or less.

[0070] As used herein, the term "feline" encompasses animals selected from cheetahs, cougars, jaguars, leopards, lions, lynxes, ligers, tigers, black panthers, lynxes, leopard cats, saber-toothed cats, caracals, servals, and cats, including pets. As used herein, cats include wildcats and domestic cats. In certain embodiments, a cat may be a domestic cat.

[0071] Typically, a feline is categorized as not having diabetes at the given time if it is not diagnosed with diabetes for the next 12 months or longer from that time.

[0072] Typically, felines are marked as having prediabetes within 12 months prior to the date of diagnosis.

[0073] As used herein, the term "risk" relates to the probability of an event occurring, possibly within a specific time period, such as in the conversion to overt diabetes, and can refer to either the subject's "absolute" or "relative" risk. Absolute risk can be measured with reference to actual observations in a relevant time cohort after measurement, or with reference to an index developed from a statistically valid historical cohort tracked over the relevant time period. Relative risk is the ratio of a subject's absolute risk to the absolute risk of a low-risk cohort or the population average risk, which may vary depending on how clinical risk factors are assessed. The odds ratio, i.e., the proportion of positive to negative events in a given test outcome, is also commonly used for no conversion (the odds is calculated according to the formula p / (1-p), where p is the probability of the event occurring and (1-p) is the probability of no event occurring). Alternative continuous measurements that can be assessed in the context of this invention include the time to diabetes conversion and the rate of reduction in the risk of therapeutic diabetes conversion.

[0074] As used herein, “risk assessment” or “risk evaluation” encompasses the prediction of the probability, preponderance, or likelihood of an event or disease state occurring, the rate of occurrence of the event, or the rate of transition from one disease state to another, i.e., from a non-diabetic state to a prediabetic symptom state or prediabetes, or from a prediabetic state to diabetes. The computer-implemented methods of this disclosure can be used to continuously or categorically measure the risk of transitioning to type 1 or type 2 diabetes, thereby diagnosing and defining the risk spectrum of felines identified as having prediabetes. The computer-implemented methods described herein can be used to distinguish between non-prediabetic and prediabetic felines. The computer-implemented methods according to this disclosure can be used to differentiate felines based on a predicted time period prior to the actual occurrence of a diabetic state. Such different uses may require different combinations of feline physiological characteristics, mathematical algorithms, and / or threshold points, but must conform to the same accuracy measurements intended for the above purposes.

[0075] As used herein, “physiological characteristics” of a feline refers to information relating to the feline, encompassing contextual information such as age, weight, and breed, and physiological parameters obtained from the sample, such as blood glucose levels, white blood cell count (WBC), blood cholesterol levels, blood phosphorus levels, blood albumin levels, and blood alkaline phosphatase levels. Physiological parameters can be quantitative or qualitative. According to some embodiments, physiological parameters can be obtained from physiological signals using feature extraction techniques, and may include, for example, combining multiple extracted features and / or parameters using nonlinear regression techniques. In the context of this disclosure, the terms “feature extraction,” “feature processing,” and “signal processing” can refer to procedures, manipulations, and signal processing measures performed for analyzing physiological signals previously obtained from a feline sample. Non-limiting examples of suitable physiological parameters are depicted in Table 1 below.

[0076] As used herein, “sample” means a biological sample isolated from a subject and may include, but is not limited to, whole blood, serum, plasma, blood cells, endothelial cells, tissue biopsy, lymph, ascites, interstitial fluid (also known as “extracellular fluid” and encompassing fluids found in the spaces between cells, particularly including gingival crevicular fluid), bone marrow, cerebrospinal fluid (CSF), saliva, mucus, sputum, sweat, urine, or any other secretions, excretions, or other bodily fluids. “Blood sample” means whole blood or any fraction thereof, including blood cells, serum, and plasma; serum is a preferred blood sample.

[0077] As used herein, “measuring,” “measure,” or “measurement” means assessing the presence, absence, quantity, or amount of a given substance or cell type in a sample of feline origin, including obtaining qualitative or quantitative concentration levels of such substances or cells.

[0078] As used in this article, "statistically significant" means a change greater than that expected to occur by chance. Statistical significance can be determined by any method known in the art. Commonly used significance measures include the p-value, which represents the probability of obtaining an outcome at least as extreme as that data point if the given data point were assumed to be a mere fluke. Results with p-values ​​of 0.05 or less are generally considered highly significant.

[0079] As used in this article, the term "TN" means true negative, which, in the context of disease state testing, means correctly classifying a non-disease subject or a normal subject.

[0080] As used in this article, the term "TP" means true positive, which in the context of disease state testing means that the subject of the disease was correctly classified.

[0081] As used in this article, the term "FN" refers to a false negative, which, in the context of disease state testing, means the incorrect classification of a diseased subject as non-disease or normal.

[0082] As used in this article, the term "FP" refers to a false positive, which in the context of disease state testing means the incorrect classification of a normal subject as having a disease.

[0083] As used herein, the term “sensitivity” refers to the sensitivity value of a prognostic method and is calculated based on the formula TP / (TP+FN) or the true positive score of disease subjects.

[0084] As used herein, the term “specificity” refers to the specificity of a prognostic method and is calculated based on the formula TN / (TN+FP) or the true negative score of non-disease subjects or normal subjects.

[0085] As used herein, a "trained machine learning model" means that the machine learning model has been provided with labeled data to learn from, that is, the weights and / or biases of the machine learning model have been determined to minimize the loss. Once trained, the machine learning model can determine the output from a given input dataset. In this disclosure, the given input dataset includes a set of physiological features corresponding to a set of at least six physiological characteristics of the feline, and an output of feature scores predicting the risk of developing diabetes in the feline or predicting the risk of developing diabetes in the prediabetic feline over a time interval.

[0086] As used in this paper, the term "neural network" refers to various configurations of classifiers used in machine learning, including multilayer perceptrons with one or more hidden layers, support vector machines, and dynamic Bayesian networks. These methods share the ability to be trained, to evaluate the quality of training, and to perform class classification or classification of a continuous number of classes in regression patterns. Typically, a neural network consists of an input layer, an output layer, and one or more hidden layers, each containing multiple neurons. The number of hidden layers limits the depth of the neural network. Neural networks can be fully connected or not fully connected, supervised or unsupervised, and can use forward propagation or backpropagation.

[0087] As used herein, the terms "computer memory" and "computer storage device" refer to any storage medium readable by a computer processor. Examples of computer memory include, but are not limited to, RAM, ROM, computer chips, digital video optical discs (DVDs), compact optical discs (CDs), hard disk drives (HDDs), and magnetic tape.

[0088] As used herein, the term "computer-readable medium" means any device or system used to store information (e.g., data and instructions) and provide that information to a computer processor. Examples of computer-readable media include, but are not limited to, DVDs, CDs, hard disk drives, magnetic tapes, and servers used for network streaming media.

[0089] As used herein, the terms “processor” and “central processing unit” or “CPU” are used interchangeably and refer to a device capable of reading a program from computer memory (e.g., ROM or other computer memory) and executing a set of steps according to that program.

[0090] describe This article provides a computer-implemented method for assessing the risk of felines becoming prediabetic. In other words, the computer-implemented method allows for the assessment of the risk of developing diabetes within approximately 12 months or less.

[0091] This article provides a computer-implemented method for assessing the risk of developing diabetes in prediabetic felines over a time interval. In other words, the computer-implemented method allows for the assessment of the risk of developing diabetes within approximately 6 months or less, preferably within approximately 3 months or less.

[0092] This article provides a computer-implemented method for assessing the risk of developing diabetes in felines. The computer-implemented method includes two main steps: a first main step of assessing the risk of developing diabetes in the feline, followed by a second main step of assessing the risk of developing diabetes over a time interval when the risk of developing diabetes has been determined at the end of the first step, that is, when the feline has been designated as having prediabetes.

[0093] This document provides a computer-implemented method for assessing the risk of developing diabetes in felines over a time interval. The computer-implemented method includes two main steps: a first main step assessing the risk of a feline becoming a prediabetic feline; and a second main step assessing the risk of developing diabetes over a time interval, after the risk of developing diabetes has been determined at the end of the first step, i.e., when the feline has been designated as prediabetic. According to some embodiments, the first main step of the computer-implemented method, including steps a) to d), allows assessment of whether the tested feline is likely to have prediabetes, while the second main step of the computer-implemented method, including steps b') to c'), allows assessment of the time interval from the start of the test after which the feline is likely to develop diabetes.

[0094] The methods disclosed in this paper all stemmed from an unexpected discovery that demonstrated that a combination of various feline physiological characteristics, when used in combination, are associated with and indicate the future development of diabetes in felines.

[0095] Surprisingly, it has been found that the risk of developing diabetes in felines can be assessed by a combination of at least six physiological characteristics, namely, age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0096] Of these possible combinations of physiological characteristics, age, weight, and breed are inherent to the history of felines and do not require any testing methods to assess.

[0097] Blood glucose levels in felines can be determined by any known method, such as by following the guidelines of the Clinical Chemistry Standard Operating Procedures published by the World Health Organization in 2020. Blood glucose levels in plasma samples can be determined. Typically, determining blood glucose levels involves collecting a blood sample from the feline, such as in a heparinized container, and determining the blood glucose level in its plasma fraction by using glucose oxidase to convert glucose into gluconic acid. This enzymatic reaction generates hydrogen peroxide, which is subsequently converted to water by peroxidase. 4-Aminopyrine acts as an oxygen acceptor and forms a pink chromogen with phenol, which can be measured at a wavelength of 515 nm. Skilled technicians can also use commercially available blood glucose meters, such as those conforming to EN ISO 15197, according to the manufacturer's instructions.

[0098] White blood cell count (WBC) is a routine physiological characteristic by which the number of white blood cells in a blood sample is measured and routinely reported as part of a standard whole blood count. WBC determination can be performed in a whole blood sample, preferably in a heparinized whole blood sample. WBC determination encompasses polymorphonuclear neutrophils, eosinophils, basophils, monocytes, B lymphocytes, and T lymphocytes. White blood cell count can be determined by any known method. Illustratively, white blood cell count is determined by a method that involves collecting a blood sample from a feline and determining the WBC using an automated hematology analyzer, such as, for example, the Beckman Coulter LH 750® commercial hematology analyzer.

[0099] Blood cholesterol level is a routine physiological characteristic by which the amount of total cholesterol in a blood sample is measured and routinely reported as part of a standard blood analysis. Blood cholesterol levels are typically determined from serum or plasma samples. Blood cholesterol levels encompass additional amounts of low-density lipoprotein (LDL) and high-density lipoprotein (HDL). Blood cholesterol levels can be determined by any known method, such as by following the guidelines for standard operating procedures in clinical chemistry published by the World Health Organization in 2020. Illustratively, a skilled technician can use a cholesterol test called CHOD-PAP, performed using ATCS, commercially available from Dialab.

[0100] Blood phosphorus content is a routine physiological characteristic used to determine the amount of inorganic phosphorus in a blood sample. Blood phosphorus content can be determined from serum samples. Blood phosphorus content can be determined by any known method, such as by following the guidelines of the World Health Organization's Standard Operating Procedures for Clinical Chemistry published in 2020. Illustratively, blood phosphorus content can be determined by a method involving the use of ammonium molybdate as a colorimetric reagent in a feline blood sample. The inorganic phosphate contained in the blood sample forms an ammonium phosphomolybdate complex with ammonium molybdate in the presence of sulfuric acid. The blood phosphorus content in the resulting product is then assessed by measuring the DO value of the resulting sample at a wavelength of 340 nm (subwavelength 700 nm). When using an automated biochemical analyzer sold by Roche under the name "Cobas 6000 C 501", technicians can refer to, for example, laboratory procedures and the manufacturer's instructions.

[0101] Blood albumin content is a routine physiological characteristic used to determine the amount of albumin in a blood sample. Blood albumin content can be determined from plasma or serum samples. It can be determined by any known method, such as by following the guidelines of the World Health Organization's Standard Operating Procedures for Clinical Chemistry published in 2020. The well-known BCG binding method, published in the 2020 WHO guidelines, can be used to determine blood albumin content. When using an automated biochemistry analyzer sold by Roche under the name "Cobas 6000 C 501," technicians can refer to, for example, laboratory procedures and the manufacturer's instructions.

[0102] Blood alkaline phosphatase (ALP) level is a routine physiological characteristic used to determine the amount of alkaline phosphatase in a blood sample. Blood ALP samples can be determined from serum or heparinized plasma. Blood ALP levels can be determined by any known method, such as by following the guidelines of the World Health Organization's Standard Operating Procedures for Clinical Chemistry published in 2020. Blood ALP levels can be determined using p-nitrophenyl phosphate, which is colorless and hydrolyzed by ALP at 37°C and pH 10.5 to form yellow free p-nitrophenol. The addition of NaOH terminates enzyme activity, and the final color shows maximum absorbance at 410 nm. Illustratively, blood ALP levels can be determined using the Unicel DxC800® Synchron automated analyzer, commercially available from Beckman Coulter.

[0103] Methods for assessing the risk of felines becoming prediabetic felines This disclosure relates to a computer-implemented method for assessing the risk of a feline becoming a prediabetic feline, the method comprising: a) Provide a set of at least six physiological characteristics of a feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level; b) Generate feature scores from the set of physiological features to predict the risk of developing diabetes in felines by operating a first machine learning model trained on a set of physiological features; c) Based on this characteristic score, determine whether the feline is at risk of developing diabetes; d) When the feline is at risk of developing diabetes, designate it as a prediabetic feline; and e) Optionally, generate a diet for the prediabetic feline.

[0104] A method for assessing the risk of developing diabetes in prediabetic felines over a time interval. This disclosure relates to a computer-implemented method for assessing the risk of developing diabetes in prediabetic felines over a time interval, the method comprising: a') By ​​operating a second machine learning model trained on a second set of at least six physiological characteristics of felines, a second characteristic score is generated from the set of physiological characteristics that predicts the development of diabetes in prediabetic felines over a period of time. These physiological characteristics are selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. b') Based on the second feature score, determine the risk of developing diabetes in prediabetic felines over a time interval; c') Optionally, a diet for the prediabetic feline is generated based on the time when the risk of developing diabetes occurs.

[0105] Methods for assessing the risk of developing diabetes in felines This disclosure relates to a computer-implemented method for assessing the risk of developing diabetes in felines, the computer-implemented method comprising the following steps: a) Provide a set of at least six physiological characteristics of a feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level; b) Generate a first feature score from the set of physiological features to predict the risk of developing diabetes in felines by operating a first machine learning model trained on the set of physiological features; c) Based on the first feature score, determine whether the feline is at risk of developing diabetes; d) When the feline is at risk of developing diabetes, designate the feline as a prediabetic feline; and optionally... e) By manipulating a second machine learning model trained on a set of physiological features, a second feature score is generated from the set of physiological features to predict the occurrence of diabetes in prediabetic felines over a time interval. f) Based on the second feature score, determine the risk of developing diabetes in prediabetic felines over a time interval.

[0106] g) Designate prediabetic felines as being at risk of developing diabetes over a period of time.

[0107] This disclosure relates to a computer-implemented method for assessing the risk of developing diabetes in felines, the computer-implemented method comprising the following steps: a) Provide a set of at least six physiological characteristics of a feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level; b) Generate a first feature score from the set of physiological features to predict the risk of developing diabetes in felines by operating a first machine learning model trained on the set of physiological features; c) Based on the first feature score, determine whether the feline is at risk of developing diabetes; d) When the feline is at risk of developing diabetes, designate the feline as a prediabetic feline; e) By manipulating a second machine learning model trained on a set of physiological features, a second feature score is generated from the set of physiological features to predict the occurrence of diabetes in prediabetic felines over a time interval. f) Based on the second feature score, determine the risk of developing diabetes in prediabetic felines over a time interval; g) Optionally, a diet for the prediabetic feline is generated based on the time when the risk of developing diabetes occurs.

[0108] First main steps: Steps a) to d) of this method The first major step of this method (including steps a) through d) allows for the assessment of the risk of developing diabetes in felines. Therefore, this first major step allows for the determination of whether the tested feline is likely to have prediabetes.

[0109] Steps a) through d) are computer-implemented steps that can be performed by implementing a trained machine learning model, which, once provided to the machine learning model with a set of at least six physiological characteristics of the feline being tested, thereby determines the risk of developing diabetes in the feline.

[0110] In some implementations, in step b), the risk of developing diabetes is specified by a trait score that, independent of its numerical value, indicates the probability of developing diabetes in the future.

[0111] As will be described in further detail in this specification, the second major step (including steps e) to f) is performed only if the risk of developing diabetes in the tested feline has been determined at the end of step d), that is, only if the tested feline is designated as a prediabetic feline.

[0112] Step a) of the method In step a) of the method, a set of at least several characteristics of the feline being tested is provided. The actual representation of the provided characteristics is not important. Typically, (i) age and (ii) weight can be expressed as numerical values, such as (i) in days, months, or years, or (ii) in grams, hectares, or kilograms. Typically, breed characteristics can be represented as alphanumeric symbols, such as the common name used for the breed.

[0113] In some embodiments of step a), providing a set of at least six characteristics includes providing (i) a set of one or more characteristics of the feline animal selected from the age, weight, and breed of the feline animal, and (ii) a set of three or more characteristics measured from a sample previously obtained from the feline animal selected from blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0114] According to this method, regardless of the implementation scheme considered, the feline characteristics provided should be contemporaneous, meaning that one or more selected characteristics should reflect the feline tested at a defined time point. Illustratively, one or more characteristics selected from blood glucose level, white blood cell count (WBC), blood cholesterol level, and blood phosphorus level should be measured from one or more samples collected at the age characteristics of the provided feline. Similarly, one or more characteristics selected from blood glucose level, white blood cell count (WBC), blood cholesterol level, and blood phosphorus level should be measured from the same samples, or alternatively from multiple samples collected over approximately the same time period, such as within the same three-month period, advantageously within the same two-month period, better within the same one-month period, such as during the same seven-day period.

[0115] In some embodiments of the method, step a) may include providing at least six physiological characteristics selected from two different sets of physiological characteristics: (i) a first set of physiological characteristics that are available through documentation and do not require testing, such as age, weight, and breed; and (ii) a second set of physiological characteristics that are available primarily as a result of testing previously performed on samples obtained from felines, such as blood glucose levels, white blood cell count (WBC), blood cholesterol levels, blood phosphorus levels, blood albumin levels, and blood alkaline phosphatase levels.

[0116] Therefore, in some embodiments of the method, step a) may include providing (i) a set of one or more characteristics of the feline, selected from the age, weight and breed of the feline, and (ii) a set of three or more characteristics previously obtained from a sample of the feline, selected from blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level and blood alkaline phosphatase level.

[0117] In some preferred embodiments of the method, step a) may include providing a set of features consisting of age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level and blood alkaline phosphatase level.

[0118] In some embodiments of the method, step a) may further include providing second-order or higher-order cross features computed from selected features. Preferably, the cross features consist of third-order cross features.

[0119] In some implementations, step a) may include second- or higher-order cross features and residual blocks computed from selected features.

[0120] Second-order cross features combine two selected features to generate a new feature. Third-order cross features combine three selected features to generate a new feature.

[0121] In one implementation, step a) provides all second-order cross features obtained by making selected features cross in pairs.

[0122] In a preferred embodiment, step a) provides all second-order cross features obtained by making selected features cross in pairs and all third-order cross features obtained by making selected features cross every three times.

[0123] In some of these implementations, step a) may include a second-order cross feature selected from the following: (i) The second-order crossover characteristic between blood glucose level and (ii) blood phosphorus level yields blood glucose; blood phosphorus; blood glucose 2 Blood phosphorus 2 Blood glucose x blood phosphorus.

[0124] In some of these implementations, step a) may include a third-order cross feature.

[0125] In some preferred embodiments, the selected features in steps a) and e) can be the same.

[0126] Step b) of the method In step b) of the method, a first trained machine learning model is used.

[0127] The first trained (i.e., pre-trained) machine learning model can be of various kinds and can include deep learning architectures. Deep learning architectures consist of multiple hidden layers. The more hidden layers a deep learning architecture has, the deeper it is considered.

[0128] This article covers machine learning models such as neural networks, whose applications can be designed with various connection patterns. In feedforward neural networks, information is passed from lower layers to higher layers, with each neuron in a given layer communicating with neurons in higher layers. Neural networks can also have backflow or feedback connections, also known as top-down connections. In backflow connections, the output from a neuron in a given layer can be fed to another neuron in the same layer. Backflow architectures can help identify patterns across more than one block of input data passed sequentially to the neural network. Connections from neurons in a given layer to neurons in lower layers are called feedback connections, also known as top-down connections. Networks with many feedback connections can be helpful when the recognition of higher-level concepts can help discern specific lower-level features of the input.

[0129] Specifically, deep learning architectures can perform physiological parameter integration tasks by learning to represent the inputs of successive higher levels of abstraction in each layer, thereby constructing useful feature representations of the input data.

[0130] A machine learning model can be trained using data from a very large population of felines to learn various relationships between physiological traits, and preferably learn second- or higher-order cross features obtained by crossing physiological traits, as well as various data categories. Then, based on feline data related to physiological traits and the model, predictions about felines can be performed by training the learned relationships.

[0131] The machine learning prediction models disclosed in this paper are very powerful in performing clinical predictions and helping veterinary practitioners generate decision support (such as dietary plan decision support).

[0132] The trait score can be of different kinds. The trait score can include a probability or a category selected from predetermined categories. In one embodiment, the first trait score can include the probability of developing diabetes and / or the probability of not developing diabetes. The first trait score can be selected from categories such as very low risk of developing diabetes, low risk of developing diabetes, moderate risk of developing diabetes, high risk of developing diabetes, and very high risk of developing diabetes, or from categories selected from sets of values ​​representing a relatively high or low risk of developing diabetes.

[0133] In some implementations, the feature score may be a value representing the probability of diabetes occurring in the feline.

[0134] Step c) of this method Determining the risk of developing diabetes based on the feature score values ​​generated in step b) depends solely on how those score values ​​are represented.

[0135] In all cases, for a particular implementation of the representation of the feature score value, it is associated with a threshold used as a reference.

[0136] Therefore, in some implementations of the feature scoring, when the feature score value is higher than the threshold, it can be predicted that diabetes may occur in the tested feline.

[0137] In some other implementations of the feature scoring, when the feature score value is below the threshold, it can be predicted that diabetes may occur in the tested feline.

[0138] Illustratively, in an embodiment where the feature score represents the probability of developing diabetes in the feline, the threshold may be fixed at 0.5 and a score value greater than 0.5 predicts that the feline may develop diabetes in the future, while a score value of 0.5 or less predicts that the feline may not develop diabetes in the future.

[0139] Step d) of this method If the risk of developing diabetes has not been determined in step c), the feline is not classified as having prediabetes, and the method ends at this step.

[0140] If the risk of developing diabetes has been determined in step c), the tested feline is classified as prediabetic because it has not yet been affected by the symptoms of diabetes but may develop diabetes in the future.

[0141] Once the risk of developing diabetes has been determined in step c), and the feline being tested has been designated as a prediabetic feline, the method continues by performing steps e) and f) as described in more detail below.

[0142] The second main step: Steps e) to f) of this method The second major step, including steps e) and f) of the method, aims to assess the time interval at which the feline designated as having prediabetes in step d) is likely to actually develop diabetes thereafter.

[0143] Step e) of this method In step e) of the method, at least six physiological characteristics of the prediabetic feline or those previously obtained from the prediabetic feline are provided, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0144] In some embodiments of the method, step e) may include providing (i) a set of one or more characteristics of the prediabetic feline, selected from the age, weight, and breed of the feline, and (ii) a set of three or more characteristics measured from a previously obtained sample from the prediabetic feline, selected from blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0145] In some preferred embodiments of the method, step e) may include providing a set of features consisting of age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level and blood alkaline phosphatase level.

[0146] In some other preferred embodiments of the method, step e) may further include providing second-order or higher-order cross features calculated from the selected features of step e). Preferably, the cross features may consist of third-order cross features.

[0147] In step e) of the method, a second trained machine learning model is used.

[0148] The second trained (i.e., pre-trained) machine learning model can be of a different kind and can include deep learning architectures. The second machine learning model can be of the same kind as the first machine learning model. Typically, the first and second machine learning models can differ in terms of the number of hidden layers and / or the number of neurons per layer and / or the type of inter-layer connections and / or the type of input data and / or the type of feature scores used as output data.

[0149] The feature scores can be of different types. The second feature score can be of the same type as the first feature score, or a different type of the first feature score. Preferably, the second feature score is a probability, representing the probability of developing diabetes within a certain time interval or the probability of not developing diabetes within a certain time interval. The second feature score can be selected from categories such as very low risk of developing diabetes within a certain time interval, low risk of developing diabetes within a certain time interval, moderate risk of developing diabetes within a certain time interval, high risk of developing diabetes within a certain time interval, and very high risk of developing diabetes within a certain time interval, or selected from categories representing sets of values ​​representing relatively high or low risks of developing diabetes within a certain time interval.

[0150] In some implementations, the feature score may be a value representing the probability that diabetes will occur in the feline after a defined time period.

[0151] Step f) of the method Determining the predicted time interval for the occurrence of diabetes based on the feature score values ​​generated in step e) depends solely on how the score values ​​are represented.

[0152] In all cases, for a particular implementation of the representation of the feature score value, it is associated with one or more thresholds that serve as a reference for each of the multiple possible time periods during which the feline may develop diabetes.

[0153] Therefore, in some implementations of the eigenvalue scoring, when the eigenvalue score is close to a threshold indicating a specific time interval in which diabetes occurs, the possible time interval for the occurrence of diabetes in the tested feline can be predicted.

[0154] In some embodiments of the method, step f) may include assessing the risk of diabetes in the feline over a six-month time interval, more preferably a three-month time interval, following the measurement of the feline's characteristics.

[0155] Other implementations of this method In some implementations of the method, each of the first and second machine learning models used in steps b) and e) can be a trained neural network model.

[0156] In an implementation where each of the first and second machine learning models used in steps b) and e) is a trained neural network model, the trained neural network model may be a trained multilayer perceptron neural network model.

[0157] A method for generating machine learning systems suitable for assessing the risk of felines becoming prediabetic felines.

[0158] This disclosure also relates to a method for generating a machine learning system suitable for assessing the risk of felines becoming prediabetic felines, the method comprising the following steps: a) Generate a model suitable for determining the risk of diabetes in felines, including the following steps: i) Provide computer-implemented machine learning devices, ii) The machine learning device is trained by providing a dataset for each of a plurality of felines, wherein the dataset for each of the plurality of felines comprises: - A set of at least six physiological characteristics of the feline, selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content, and - Clinical data related to the occurrence of diabetes in the aforementioned felines. This generates a machine learning model suitable for determining the risk of felines becoming prediabetic felines.

[0159] In some implementations of the method for generating a machine learning system, steps a)-ii) may include training the first machine learning device with the following: -A dataset provided from felines tagged as not having diabetes, and - A set of data provided from felines tagged as either prediabetic or non-diabetic.

[0160] Machine learning systems can be trained using two or more sets of data that can distinguish between two or more states.

[0161] A method for generating machine learning systems suitable for assessing the risk of developing diabetes in prediabetic felines over a time interval.

[0162] This disclosure also relates to a method for generating a machine learning system suitable for assessing the risk of developing diabetes in prediabetic felines over a time interval, the method comprising the following steps: a') Generate a second model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval, including the following steps: i) Provide a second computer-implemented machine learning device, ii) The second machine learning device is trained by providing a dataset of each of a plurality of prediabetic felines, wherein the dataset of each of the plurality of prediabetic felines comprises: - A set of at least six physiological characteristics of the prediabetic feline, selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content, and - Clinical data related to the development of diabetes in the aforementioned prediabetic felines. This generates a second machine learning model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval.

[0163] In some implementations of the method for generating machine learning systems, steps a')-ii) may include training the second machine learning device with a dataset from felines labeled as having prediabetes.

[0164] In some implementations of the method for generating a machine learning system, steps a)-ii) may include training the second machine learning device with the following: -A dataset provided from prediabetic felines labeled as non-diabetic within a time interval of more than six months, preferably three months, following data collection, and - A dataset of prediabetic felines labeled as having diabetes within six months of data collection, at three-month intervals.

[0165] Machine learning systems can be trained using two or more sets of data that can distinguish between two or more states.

[0166] A method for generating machine learning systems suitable for assessing the risk of diabetes in felines.

[0167] This disclosure also relates to a method for generating a machine learning system suitable for assessing the risk of diabetes in felines, the method comprising the following steps: a) Generate a first model suitable for assessing the risk of diabetes in felines, including the following steps: i) Provide the first computer-implemented machine learning device, ii) Training the first machine learning device by providing a dataset for each of a plurality of felines, wherein the dataset for each of the plurality of felines comprises: - A set of at least six or more physiological characteristics of the feline or previously obtained from a sample of the feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level, and - Clinical data related to the occurrence of diabetes in the aforementioned felines. This generates the first machine learning model suitable for assessing the risk of diabetes in felines. b) Generate a second model suitable for assessing the risk of developing diabetes in felines over a time interval, including the following steps: i) Provide a second computer-implemented machine learning device, ii) The second machine learning device is trained by providing a dataset of each of a plurality of prediabetic felines, wherein the dataset of each of the plurality of prediabetic felines comprises: - A set of at least six physiological characteristics measured in the prediabetic feline or from samples previously obtained from the prediabetic feline, selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level, and - Clinical data related to the development of diabetes in the aforementioned prediabetic felines. This generates a second machine learning model suitable for assessing the risk of developing diabetes in felines over a period of time.

[0168] In some implementations of the method for generating a machine learning system, steps a)-ii) may include training the first machine learning device with the following: -A dataset provided from felines tagged as not having diabetes, and - A dataset provided from felines tagged with prediabetes.

[0169] In some implementations of the method for generating machine learning systems, steps b)-ii) may include training the second machine learning device with a dataset from felines labeled as having prediabetes.

[0170] In some implementations of the method for generating a machine learning system, step b)-ii) may include training the second machine learning device with the following: -A dataset provided from prediabetic felines labeled as non-diabetic within a time interval of more than six months, preferably three months, following data collection, and - A dataset of prediabetic felines labeled as having diabetes within six months of data collection, at three-month intervals.

[0171] Machine learning systems can be trained using two or more sets of data that can distinguish between two or more states.

[0172] Computer-implemented systems This disclosure also relates to a computer-implemented system for determining the risk of a feline becoming a prediabetic feline, the computer-implemented system comprising: -processor, - A tangible, computer-readable medium operatively connected to a processor and comprising computer code configured to: a) Generate models suitable for determining the risk of felines becoming prediabetic felines, including: i) Record multiple datasets, each dataset being measured for each feline in a plurality of felines, and each dataset containing multiple feline physiological characteristics selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content. ii) Record the diabetes incidence value for each of the multiple felines in step i). iii) By using a database to train a machine learning algorithm to generate a machine learning model, a trained model for the risk of developing diabetes is developed, which is configured to generate a feature score that predicts the risk of a feline becoming a prediabetic feline.

[0173] Tangible computer-readable media may include computer code configured to: a) Generate a first-hand model suitable for assessing the risk of felines becoming prediabetic felines, including: i) Record multiple data sets in a first database, each data set being measured for each feline among multiple felines, and each data set containing multiple feline physiological characteristics selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. ii) Record the incidence of diabetes for each of the multiple felines in step i) in the first database. iii) Use the first database to train the machine learning algorithm to generate the first machine learning model. An augmented model was thus developed to assess the risk of developing diabetes. This augmented model was configured to generate a feature score that predicts the risk of a feline becoming a prediabetic feline.

[0174] The risk of a feline becoming a prediabetic feline is assessed by performing at least steps a) to d) or steps a') to c') of the computer-implemented method as described above, and the set of characteristics is recorded as a data set in the first database, and the risk of developing diabetes is recorded as a diabetes occurrence value, which can enrich the first database.

[0175] This disclosure also relates to a computer-implemented system for determining the risk of developing diabetes in prediabetic felines over a time interval, the computer-implemented system comprising: -processor, - A tangible, computer-readable medium operatively connected to a processor and comprising computer code configured to: a') Generate a second model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval, including: i) Record multiple datasets, each dataset being measured for each feline in a plurality of prediabetic felines, and each dataset containing multiple prediabetic feline physiological characteristics selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content. ii) For each of the multiple prediabetic felines in step i), record the time of diabetes occurrence. iii) By using a database to train a machine learning algorithm to generate a second machine learning model, thereby developing a trained model for the risk of developing diabetes over a time interval, the model being configured to generate feature scores that predict the risk of developing diabetes in the prediabetic feline over a time interval.

[0176] Tangible computer-readable media may include computer code configured to: a') Generate a second model suitable for assessing the risk of developing diabetes in felines over a time interval, including: i) Record multiple datasets in a second database. Each dataset is measured for each prediabetic feline in a plurality of prediabetic felines, and each dataset contains multiple physiological characteristics of prediabetic felines selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. ii) In the second database, record the time of diabetes occurrence for each of the multiple prediabetic felines from step i). iii) Use a second database to train the machine learning algorithm to generate a second machine learning model. An enhanced model was developed to assess the risk of developing diabetes over a time interval. This enhanced model was configured to generate feature scores that predict the risk of developing diabetes in the prediabetic feline over a time interval.

[0177] The risk of developing diabetes in prediabetic felines over a time interval can be assessed by performing at least steps a) to f) or a') to c') of the computer-implemented method as described above, and the set of features can be recorded as a data set in the second database, and the risk of developing diabetes over a time interval can be recorded as a diabetes time occurrence value, which can enrich the second database.

[0178] This disclosure also relates to a computer-implemented system for assessing the risk of developing diabetes in felines, the computer-implemented system comprising: -processor, - A tangible, computer-readable medium operatively connected to a processor and comprising computer code configured to: a) Generate a first model suitable for assessing the risk of diabetes in felines, including: i) Record multiple datasets, each dataset being measured for each feline in a plurality of felines, and each dataset containing multiple feline physiological characteristics selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. ii) Record the diabetes incidence value for each of the multiple felines in step i). iii) By using a database to train a machine learning algorithm to generate a first machine learning model, thereby developing a trained model for the risk of developing diabetes, the model being configured to generate a feature score that predicts the risk of developing diabetes in the feline.

[0179] b) Generate a second model suitable for assessing the risk of developing diabetes in felines over a time interval, including: i) Record multiple datasets, each dataset being measured for each feline in a plurality of prediabetic felines, and each dataset containing multiple prediabetic feline physiological characteristics selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. ii) For each of the multiple prediabetic felines in step i), record the time of diabetes occurrence. iii) By using a database to train a machine learning algorithm to generate a second machine learning model, thereby developing a trained model for the risk of developing diabetes over a time interval, the model being configured to generate feature scores that predict the risk of developing diabetes in the prediabetic feline over a time interval.

[0180] Methods for preventing diabetes in felines This disclosure also relates to a method for preventing diabetes in felines, the method comprising the following steps: 1) Assess the risk of developing diabetes in felines by performing a computer-implemented method according to the instructions. 2) If the risk of developing diabetes in a feline has been determined in step 1), then the feline is provided with an appropriate diet.

[0181] In some implementations of the above method, step 2) may also include providing the feline with a preventative or dietary regimen.

[0182] This disclosure also relates to a method for preventing the development of diabetes in prediabetic felines over a period of time, the method comprising the following steps: 1') Assess the risk of developing diabetes in the prediabetic feline over a period of time by performing the computer-implemented method described above, and 2') If the risk of developing diabetes in the prediabetic feline has been determined in step 1) over a period of time, then preventive or therapeutic treatment is provided to the prediabetic feline.

[0183] In some implementations of the above method, step 2') may include administering a preventative or dietary regimen to the prediabetic feline.

[0184] This disclosure also relates to a method for preventing diabetes in felines, the method comprising the following steps: 1) Assess the risk of developing diabetes in the feline animals by performing the computer-implemented method described above, and 2) If the risk of developing diabetes in the feline has been determined in step 1), then the feline is provided with an appropriate diet.

[0185] In some implementations of the above method, step 2) may include administering a preventative or dietary regimen to the feline.

[0186] According to some implementation schemes, the preventative or dietary programs can be specifically formulated to meet their dietary requirements. Such dietary programs may include diets containing low levels of total sugars (<62g of monosaccharides and disaccharides per kg of complete feed with 12% moisture), as detailed in European Commission Regulation 2020 / 354 (4 March 2020). Dietary programs may additionally include diets containing high levels of protein (metabolizable energy >40%), as recommended by the American Animal Hospital Association (Behrend et al. 2018), such as in wet or dry kibble form, Hill's prescription diet m / d, Purina d / m, Purina o / m, Royal Canin Diabetic / Glycobalance, or Royal Canin Satiety. Dietary programs may also include diets containing low levels of dietary starch (<20%), such as in wet or dry kibble form, Hill's prescription diet m / d, Purina d / m, or Royal Canin Diabetic / Glycobalance. Finally, a diet plan may include diets that have two or more of the aforementioned properties.

[0187] Example Materials and methods Figure 1 The implementation of a method for assessing the risk of developing diabetes in felines is described.

[0188] feature To predict the risk of developing diabetes in felines, specific characteristics must be provided. Specifically, as mentioned above, at least six physiological characteristics selected from the following must be provided: age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0189] Features can be of different kinds. Categorical features and continuous features can undergo different preprocessing.

[0190] Categorical characteristics can be age or breed, while weight, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level are continuous characteristics.

[0191] Preferably, continuous features can be normalized.

[0192] Categorical features can be transformed using column embedding. The transformed categorical features can then be normalized.

[0193] It can calculate second-order or higher-order cross features of features, and preferably, it can calculate second-order or higher-order cross features of normalized features.

[0194] For example, if we consider six physiological characteristics (f1, f2, f3, f4, f5, f6), the second-order cross features of all six preprocessed physiological characteristics yield the following features: f1, f2, f3, f4, f5, f6, f1*f1, f1*f2, f1*f3, f1*f4, f1*f5, f1*f6, f2*f1, f2*f2, f2*f3, f2*f4, f2*f5, f2*f6, f3*f1, f3*f2, f3*f3, f3*f4, f3*f5, f3*f6, f4*f1, f4*f2, f4*f3, f4*f4, f4*f5, f4*f6, f5*f1, f5*f2, f5*f3, f5*f4, f5*f5, f5*f6 f6*f1, f6*f2, f6*f3, f6*f4, f6*f5,f6*f6. Feature selectors can be applied to cross features.

[0195] In a preferred embodiment, the machine learning model includes residual blocks, such as Figure 3 As shown.

[0196] Preferably, after preprocessing, after finally calculating the cross-order features, and optionally after applying a feature selector, at least six physiological features are submitted to the first trained machine learning model.

[0197] The first trained machine learning model The first machine learning model can be a neural network, especially a multilayer perceptron neural network.

[0198] A multilayer perceptron neural network includes at least one input layer, one output layer, and one or more hidden layers.

[0199] The following is a detailed example of a first machine learning approach that can be used to predict the risk of developing diabetes in felines. This example is for illustrative and not limiting purposes only, and each layer comprises several neurons.

[0200] Categorical features, such as cat breed, are embedded into numerical features that the algorithm can understand. Numerical features, such as blood parameters, are normalized using mean and standard deviation. All features are then concatenated into a fully connected feature vector. From this feature vector with at least six features, new feature vectors are generated, which are composed of all third-order polynomials of the original feature vector, resulting in a new feature vector with 83 features. For example, if the input sample is two-dimensional and in the form [a, b], the third-order polynomial produces [a, b, a...]. 2 , ab, b 2 , a 3 , b 3 , a 2 b, b 2 a).

[0201] An instance of the MLP architecture will consist of an input layer with at least 83 features, feeding information into a neural network with four hidden layers of eight neurons each. Each layer feeds into the next layer and directly into layers approximately 2-3 hops apart (an instance of a neural network with residual blocks is disclosed in the paper "Deep Residual Learning for Image Recognition" by Kaiming He et al., arXiv.1512.03385, https: / / doi.org / 10.48550 / arXiv.1512.03385). As part of the optimization algorithm, it is preferable to repeatedly estimate the error of the current state of the model. This requires choosing an error function, conventionally called a loss function, which can be used to estimate the loss of the model so that the weights can be updated to reduce the loss in the next evaluation. The Focal Loss function has been used to address class imbalance during training (see, for example, the paper "Focal Loss for Dense Object Detection" by Tsung-YiLin et al., arXiv.1708.02002, https: / / doi.org / 10.48550 / arXiv.1708.02002). Several other model architectures and loss functions can be used to achieve similar performance.

[0202] The first trained machine learning model outputs feature scores that predict the risk of developing diabetes in felines.

[0203] Typically, this feature score takes the form of a prediction, which can be interpreted as a Bayesian posterior probability.

[0204] Determine the risk of developing diabetes The risk of developing diabetes in felines is determined based on feature scores provided by a first trained machine learning model.

[0205] For example, a trait score can be compared to a threshold. If the trait score exceeds the threshold, there is a risk of developing diabetes, and if the trait score is below or equal to the threshold, there is no risk, and vice versa.

[0206] The threshold can be 0.5, and the feature score is contained in the interval [0; 1].

[0207] If a feline is determined to be at risk of developing diabetes, it is labeled as having prediabetes.

[0208] Figure 2 The implementation of a method for assessing the risk of developing diabetes in felines over a time interval is described.

[0209] Unlike methods used to assess the risk of developing diabetes, this method, which assesses the risk of felines developing diabetes over a period of time, is applicable to felines classified as having prediabetes.

[0210] Felines can be classified as having prediabetes in other ways.

[0211] Preferably, a method for assessing the risk of developing diabetes is applied to determine whether the feline to be tested is at risk, and if the feline to be tested is at risk, a method for assessing the risk of the feline to be tested developing diabetes over a period of time is applied.

[0212] Similar to methods used to assess the risk of developing diabetes in felines, methods for assessing the risk of developing diabetes in felines over a time interval are based on specific... feature Acquisition, processing, and analysis.

[0213] It is worth noting that, as mentioned above, at least six physiological characteristics selected from the following must be provided: age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level.

[0214] Categorical features and continuous features can be preprocessed differently.

[0215] Categorical characteristics can be age or breed, while weight, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level are continuous characteristics.

[0216] Preferably, continuous features can be normalized.

[0217] Categorical features can be transformed using column embedding. The transformed categorical features can then be normalized.

[0218] It can calculate second-order or higher-order cross features of features, and preferably, it can calculate second-order or higher-order cross features of normalized features.

[0219] In a preferred embodiment, the machine learning model includes residual blocks, such as Figure 3 As shown.

[0220] Preferably, after preprocessing, after finally calculating the cross-order features, and optionally after applying a feature selector, at least six physiological features are submitted to the first trained machine learning model.

[0221] The second trained machine learning model Categorical features, such as cat breed, are embedded into numerical features that the algorithm can understand. Numerical features, such as blood parameters, are normalized using mean and standard deviation. All features are then concatenated into a fully connected feature vector. From this feature vector with at least six features, new feature vectors are generated, which are composed of all third-order polynomials of the original feature vector, resulting in a new feature vector with 83 features. For example, if the input sample is two-dimensional and in the form [a, b], the third-order polynomial produces the following cross-features [a, b, a...]. 2 , ab, b 2 , a 3 , a 2 b, b 3 , b 2 a).

[0222] An instance of an MLP architecture will consist of an input layer with at least 83 features, feeding information into a neural network with four hidden layers of eight neurons each. Each layer feeds into the next layer and directly into layers approximately 2-3 hops apart (neural networks with residual blocks https: / / doi.org / 10.48550 / arXiv.1512.03385). As part of the optimization algorithm, the error of the model's current state must be repeatedly estimated. This requires choosing an error function, conventionally called a loss function, which can be used to estimate the model's loss so that the weights can be updated to reduce the loss in the next evaluation. The focus loss function has been used to address class imbalance during training (https: / / doi.org / 10.48550 / arXiv.1708.02002). Several other model architectures and loss functions can be used to achieve similar performance.

[0223] The second trained machine learning model outputs a feature score that predicts the risk of developing diabetes in felines over a time interval, preferably longer than two months and shorter than six months, particularly about three months.

[0224] Typically, this feature score takes the form of a prediction, which can be interpreted as a Bayesian posterior probability.

[0225] To determine the risk of developing diabetes in felines over a period of time.

[0226] Based on feature scores provided by a second trained machine learning model, the risk of developing diabetes in felines over a time interval is determined.

[0227] For example, a trait score can be compared to a threshold. If the trait score exceeds the threshold, the prediabetic feline being tested is at risk of developing diabetes within that time interval, and if the trait score is below or equal to the threshold, there is no risk, and vice versa.

[0228] The threshold can be 0.5, and the feature score is contained in the interval [0; 1].

[0229] Figure 3 This diagram illustrates an example of a framework that can be used to assess the risk of developing diabetes in felines and / or the risk of developing diabetes in felines over a time interval.

[0230] An example of this architecture is the focus loss function. This function allows for the resolution of class imbalance by adjusting the weights of cases that are easy to classify relative to cases that are difficult to classify.

[0231] In addition, the machine learning model used can be an MLP (Multilayer Perceptron Neural Network), which is particularly suitable for classification problems, especially nonlinear problems.

[0232] Preferably, the architecture used may include residual blocks. These blocks allow information to be retained from the initial layer to the final layer.

[0233] Figure 4 A preferred embodiment of method 3 is disclosed, which is used to determine whether a feline to be tested has prediabetes, and in the case that the feline to be tested has prediabetes, to determine the risk that the feline to be tested will develop diabetes over a period of time.

[0234] This method 3 includes implementing method 1 according to the invention for assessing the risk of developing diabetes. If the feline to be tested is designated as having prediabetes, then method 2 according to the invention for assessing the risk of said feline to be tested developing diabetes over a period of time is applied.

[0235] Both methods can use the same features.

[0236] Specifically, when applied, preprocessing, cross-featureization, and / or feature selection can be applied to both methods once.

[0237] Features can be evaluated from a database that stores features.

[0238] result The results and performance data from three machine learning models, trained to determine the risk of felines becoming prediabetic (prediabetes model), the risk of prediabetic felines developing diabetes over a three-month time interval (3-month time interval prediabetes model), and the risk of felines developing diabetes over a three-month time interval (3-month time interval model), respectively. These three machine learning models are multilayer perceptron neural networks, consisting of an input layer with eight neurons, an output layer with eight neurons, and four hidden layers with eight neurons each.

[0239] Each model is trained using a different dataset.

[0240] Two clinical datasets from two different clinical groups (Clinical 1 and Clinical 2) were used to train and test the machine learning model, first training and testing separately, and then combining the two sets. The datasets included visits to multiple individuals. The visits and individuals selected for the training and testing sets may differ depending on the expected output of the machine learning model and ultimately the input.

[0241] Therefore, nine trained models were evaluated: prediabetes model - clinical 1; prediabetes model - clinical 2; prediabetes model - clinical 1 and clinical 2; prediabetes model with a 3-month time interval - clinical 1; prediabetes model with a 3-month time interval - clinical 2; prediabetes model with a 3-month time interval - clinical 1 and clinical 2; model with a 3-month time interval - clinical 1; model with a 3-month time interval - clinical 2; model with a 3-month time interval - clinical 1 and clinical 2.

[0242] For each dataset, training and testing sets are created based on a 70% / 30% split of the unique individuals in that dataset, respectively. This split ensures that an individual found in one set cannot be found in another.

[0243] Tables 1 and 2 below summarize the distribution of the dataset between the training and test sets. The training and test sets include individuals labeled "prediabetes" and other individuals labeled "healthy." A dataset may include several visits to an individual.

[0244] Table 1 shows the distribution of the datasets Clinical 1 and Clinical 2 corresponding to the training and testing phases of the prediabetes model.

[0245] In Table 1, individuals are marked as “prediabetes” during visits 0 to 12 months prior to the visit in which they were diagnosed with diabetes. Strictly speaking, visits longer than 12 months prior to the visit in which an individual is marked as “prediabetes” are not considered. Visits after the visit in which an individual marked as “prediabetes” was diagnosed with diabetes are also not considered. If an individual has not been diagnosed with diabetes during the visit, that individual is marked as “healthy.” In Table 1, for healthy individuals, i.e., those indicated as “false” in the “prediabetes 0-12 months” column, only visits longer than 12 months are considered.

[0246] Table 1 Table 2 shows the distribution of the datasets for Clinical 1 and Clinical 2 corresponding to the training and testing phases of the prediabetes model performed at 3-month time intervals.

[0247] In Table 2, all individuals are classified as "prediabetes". That is, only individuals who have been diagnosed with diabetes are considered, and only visits 0 to 12 months prior to the visit in which the individual was diagnosed with diabetes are considered.

[0248] In the 0-12 month visits for these individuals with prediabetes, the "Prediabetes 0-3 months" column was set to "True" for visits 0-3 months prior to the individual's diagnosis of diabetes. For the "Prediabetes 0-3 months" column, visits prior to visits occurring more than 3 months or less than 12 months after the individual's diagnosis of diabetes were set to "False". Visits occurring after the individual's diagnosis of diabetes were not considered.

[0249] Table 2 Training phase The training phase comprises multiple periods during which the model is trained using a training set by defining and updating the weights between the layers of the model.

[0250] The prediabetes model has been trained using the training set of the Clinical 1 dataset as shown in Table 1.

[0251] The prediabetes model has been trained using the training set of the Clinical 2 dataset as shown in Table 1.

[0252] The prediabetes model has been trained using the training sets of the Clinical 1 and Clinical 2 datasets, which are split as shown in Table 1.

[0253] The prediabetes model with a 3-month time interval has been trained using the training set of the Clinical 1 dataset as shown in Table 2.

[0254] The prediabetes model with a 3-month time interval has been trained using the training set of the Clinical 2 dataset as shown in Table 2.

[0255] The prediabetes model with a 3-month time interval has been trained using the training sets of Clinical 1 and Clinical 2 datasets as shown in Table 2.

[0256] The three-month time interval model - Clinical 1 is a combination of a prediabetes model trained on the training set of the Clinical 1 dataset and a three-month time interval prediabetes model trained on the training set of the Clinical 1 dataset.

[0257] The three-month time interval model - clinical 2 is a combination of a prediabetes model trained on the training set of the clinical 2 dataset and a three-month time interval prediabetes model trained on the training set of the clinical 2 dataset.

[0258] The three-month time interval model - Clinical 1 and Clinical 2 are a combination of a prediabetes model trained with the training set of Clinical 1 and Clinical 2 datasets and a three-month time interval prediabetes model trained with the training set of Clinical 1 and Clinical 2 datasets.

[0259] The combination of a prediabetes model and a 3-month time interval prediabetes model means that the two models are implemented in parallel or preferably sequentially. If both return "true," meaning that the individual is predicted to have prediabetes and is at risk of developing diabetes within the 3-month time interval, then the 3-month time interval model also returns "true." Otherwise, the 3-month time interval model will return "false."

[0260] Preferably, when testing an individual using a three-month time interval model, a prediabetes model is first implemented. If the prediabetes model returns "false," the individual is assumed to be healthy, and the three-month time interval model returns "false." If the prediabetes model returns "true," a three-month time interval prediabetes model is implemented. If the three-month time interval prediabetes model returns "false," the three-month time interval model also returns "false." If the three-month time interval prediabetes model returns "true," the three-month time interval model returns "true."

[0261] In certain implementations, the three-month time interval does not return binary output. For example, if the prediabetes model returns "true" while the three-month time interval prediabetes model returns "false", the three-month time interval model may return "possibly".

[0262] Testing phase The test set is not involved in the training phase and is used only for performance evaluation once the model has been trained.

[0263] The three tables below summarize the performance metrics representing each of the nine trained machine learning models.

[0264] Accuracy refers to the number of correct test results relative to the total number of test results. Mathematically, accuracy is defined as: (TP+TN) / (TP+TN+FP+FN), where TP is the number of true positive test results, TN is the number of true negative test results, FP is the number of false positive test results, and FN is the number of false negative test results.

[0265] Sensitivity refers to a model's ability to predict positive test results. Mathematically, sensitivity is defined as: TP / (TP+FN) Specificity refers to a model's ability to predict negative test results. Mathematically, specificity is defined as: TN / (TN+FP) Precision refers to the score of correct positive test results relative to the total number of positive test results. For example, considering a machine learning model trained to determine whether an individual has prediabetes, precision corresponds to the number of individuals correctly labeled as "prediabetes" relative to the total number of individuals labeled as "prediabetes". Mathematically, precision is defined as: (TP) / (TP+FP).

[0266] The F1 score is a weighted average of precision and recall. Mathematically, the F1 score is defined as follows: (TP) / [TP+1 / 2(FN+FP)] Prevalence refers to the total number of individuals who have been, or should have been, predicted to be, positive relative to the total number of individuals. Mathematically, prevalence is defined as: (TP+FP) / (TP+TN+FP+FN) Prediabetes model Table 3 Prediabetes model with a 3-month time interval Table 4 3-month time interval model (Prediabetes model + 0-3 month model) Table 5 The positive predictive value (PPV) and negative predictive value (NPV) relative to Table 5 are shown in Table 6 below: Table 6 The positive predictive value (PRF) is the fraction of correct positive test results relative to the total number of positive test results. Mathematically, the PRF is defined as: (TP) / (TP+FP). The PRF is equivalent to precision.

[0267] The negative predictive value is the score of correct negative test results relative to the total number of negative test results. Mathematically, the negative predictive value is defined as: TN / (TN+FN).

[0268] References: [1] M. D. Richard, and al. Neural Network Classifiers Estimate Bayesian a posteriori Probabilities " [2] Merriam-Webster's Collegiate Dictionary, 10 th Edition, 1993 [3] Guidelines on Standard Operating Procedures for ClinicalChemistry published by the World Health Organization in 2020 [3] “AAHA Diabetes Management Guidelines for Dogs and Cats”, Behrendet al 2018, American Animal Hospital Association, [4] Kaiming He and al. "Deep Residual Learning for ImageRecognition", arXiv.1512.03385, https: / / doi.org / 10.48550 / arXiv.1512.03385foo [5] Tsung-Yi Lin and al., “Focal Loss for Dense Object Detection”,arXiv.1708.02002, https: / / doi.org / 10.48550 / arXiv.1708.02002

Claims

1. A computer-implemented method for assessing the risk of a feline becoming a prediabetic feline, the method comprising: a) Provide a set of at least six physiological characteristics of the feline, the physiological characteristics being selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level; b) Generate a feature score from the set of physiological features that predicts the risk of developing diabetes in felines by operating a first machine learning model trained on the set of physiological features; c) Based on the feature score, determine whether the feline is at risk of developing diabetes; d) When the feline is at risk of developing diabetes, the feline is designated as a prediabetic feline; and e) Optionally, generate a diet for the prediabetic feline.

2. A computer-implemented method for assessing the risk of developing diabetes in felines over a time interval, the method include: The method according to claim 1 determines whether the feline is a prediabetic feline. a') By ​​operating a second machine learning model trained on a second set of at least six physiological characteristics of the feline, a second characteristic score is generated from the second set of physiological characteristics to predict the development of diabetes in the prediabetic feline over a time interval. The physiological characteristics are selected from age, weight, breed, blood glucose level, white blood cell count (WBC), blood cholesterol level, blood phosphorus level, blood albumin level, and blood alkaline phosphatase level. The second set of physiological characteristics may be the same as or different from the set of physiological characteristics in step b). b') Based on the second feature score, determine the risk of the prediabetic feline developing diabetes over a time interval; c') Optionally, a diet for the prediabetic feline is generated based on the time when the risk of developing diabetes occurs.

3. The computer-implemented method according to any one of the preceding claims, wherein the time interval is 6 months, preferably 3 months.

4. The computer-implemented method according to any one of the preceding claims, wherein step a) comprises providing (i) a set of one or more characteristics of the feline, the characteristics being selected from the group consisting of the age, weight and breed of the feline, and (ii) a set of three or more characteristics measured from a sample previously obtained from the feline, the characteristics being selected from the group consisting of plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content.

5. The computer-implemented method according to any one of the preceding claims, wherein step a) comprises providing a set of features consisting of age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content.

6. The computer-implemented method according to any one of the preceding claims, wherein step a) further comprises providing a second-order or higher-order cross feature calculated based on the selected feature.

7. The computer-implemented method according to any one of claims 2 to 6, wherein step a') comprises providing (i) a set of one or more characteristics of the prediabetic feline, the characteristics being selected from the group consisting of the age, weight, and breed of the feline, and (ii) a set of three or more characteristics measured from a sample previously obtained from the prediabetic feline, the characteristics being selected from the group consisting of plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content.

8. The computer-implemented method according to any one of claims 2 to 7, wherein step a') comprises providing a set of features consisting of age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level and plasma alkaline phosphatase content.

9. The computer-implemented method according to any one of claims 2 to 8, wherein step a') further comprises providing a second-order or higher-order cross feature calculated based on the selected feature.

10. The computer-implemented method according to any one of claims 6 and 9, wherein the cross features consist of third-order cross features.

11. The computer-implemented method according to any one of claims 1 to 6 and 10, wherein the selected features in steps a) and a') are identical.

12. The computer-implemented method according to any one of claims 1 to 11, wherein each of the first machine learning model and the second machine learning model used in steps b) and b') is a trained neural network model.

13. The computer-implemented method of claim 12, wherein the trained neural network model is a trained multilayer perceptron neural network model.

14. A method for generating a machine learning system suitable for assessing the risk of felines becoming prediabetic felines, the method comprising the steps of: a) Generate a model suitable for determining the risk of diabetes in felines, including the following steps: i) Provide computer-implemented machine learning devices, ii) The machine learning device is trained by providing a dataset for each of a plurality of felines, wherein the dataset for each of the plurality of felines includes: - A set of at least six physiological characteristics of the feline, wherein the physiological characteristics are selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content; and - Clinical data related to the occurrence of diabetes in the aforementioned felines. This generates a machine learning model suitable for determining the risk of felines becoming prediabetic felines.

15. A method for generating a machine learning system suitable for assessing the risk of developing diabetes in prediabetic felines over a time interval, the method comprising the steps of: a') Generate a first machine learning system suitable for assessing the risk of felines becoming prediabetic felines, as claimed in claim 14: b') Generate a second model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval, including the following steps: i) Provide a second computer-implemented machine learning device, ii) Training the second machine learning device by providing a dataset of each of a plurality of prediabetic felines, wherein the dataset of each of the plurality of prediabetic felines comprises: - A set of at least six physiological characteristics of the prediabetic feline, wherein the physiological characteristics are selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content; and - Clinical data related to the development of diabetes in the aforementioned prediabetic felines. This generates a second machine learning model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval.

16. The method of claim 14; wherein steps a)-ii) comprise training the machine learning device with the following: -A dataset provided from felines tagged as not having diabetes, and - A dataset provided from felines tagged with prediabetes.

17. The method of claim 15, wherein steps b')-ii) comprise training the second machine learning device with a dataset from felines tagged as prediabetic.

18. A computer-implemented system for determining the risk of a feline becoming a prediabetic feline, the computer-implemented system comprising: -processor, - A tangible computer-readable medium operatively connected to the processor and comprising computer code configured to: a) Generate models suitable for determining the risk of felines becoming prediabetic felines, including: i) Record multiple datasets, each dataset being measured for each feline in a plurality of felines, and each dataset containing multiple feline physiological characteristics selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content. ii) Record the diabetes incidence value for each of the plurality of felines in step i). iii) Develop a trained model for the risk of developing diabetes by using a database to train a machine learning algorithm to generate a machine learning model, the model being configured to generate feature scores that predict the risk of a feline becoming a prediabetic feline.

19. A computer-implemented system for determining the risk of developing diabetes in a prediabetic feline over a time interval, the computer-implemented system comprising: -processor, - A tangible computer-readable medium operatively connected to the processor and comprising computer code configured to: a') Generate a first model suitable for determining the risk of a feline becoming a prediabetic feline according to claim 18. b') Generate a second model suitable for determining the risk of developing diabetes in prediabetic felines over a time interval, including: i) Record multiple datasets, each dataset being measured for each feline in a plurality of prediabetic felines, and each dataset containing multiple prediabetic feline physiological characteristics selected from age, weight, breed, plasma glucose level, white blood cell count (WBC), plasma cholesterol level, plasma phosphorus content, plasma albumin level, and plasma alkaline phosphatase content. ii) Record the time of diabetes occurrence value for each of the plurality of prediabetic felines in step i). iii) By using a database to train a machine learning algorithm to generate a second machine learning model, thereby developing a trained model for the risk of developing diabetes over a time interval, the model being configured to generate feature scores that predict the risk of developing diabetes in the prediabetic feline over a time interval.

Images