Method and system for estimating calorie sources of energy expenditure of a user during physical activities, and, non-transitory computer-readable storage medium

Abstract

The present invention refers to a method for estimating calorie sources of energy expenditure, EE, stamina, training load, and recovery time of a user during physical activities, comprising: obtaining user data including user profile data, bioelectrical impedance analysis, BIA, data and physical activity data, wherein the physical activity data is obtained from a wearable device during a physical activity session; determining a respiratory exchange ratio, RER, time series of the user during the physical activity with a RER prediction model based on the user data, wherein the RER prediction model is trained with a dataset comprising physical activities data to predict RER time series; determining an energy expenditure, EE, time series of the user during the physical activity with an EE prediction model based on the user data, wherein the EE prediction model is trained with a dataset comprising physical activities data to predict EE time series; and estimating the calorie sources during the physical activity based on the user data, RER time series and EE time series (308), wherein the calorie sources comprise carbohydrate, fat, and protein. estimating stamina, training load, and recovery time during physical activity based on the user data and calorie sources. The invention also refers to a corresponding system and a non- transitory medium.

Description

METHOD AND SYSTEM FOR ESTIMATING CALORIE SOURCES OF ENERGY EXPENDITURE OF A USER DURING PHYSICAL ACTIVITIES, AND, NON-TRANSITORY COMPUTER-READABLE STORAGE MEDIUMFIELD OF THE DISCLOSURE

[0001] The present invention refers to systems and methods based on wearable technologies to estimate continuously and noninvasively the individualized metabolic sources of energy expenditure during resting and physical activity.DESCRIPTION OF RELATED ART

[0002] Energy Expenditure (EE) is the amount of energy spent, generally in terms of calories, to maintain the body homeostasis. The EE is supplied by the consumption of substrates, such as carbohydrate (CHO), fat, and proteins. Identifying the energetic sources of EE during rest and exercise is an important biomarker for metabolic health and has fitness applications, including optimized weight loss and aerobic training. During continuous submaximal exercises, fat and CHO are the primary sources of energy (>95%), while protein (amino acids) contributes with only up to 5% of the EE under normal conditions. The oxidation of CHO, fat, and protein are metabolically associated with O2 uptake and CO2 output, which is the basis of the “indirect calorimetry” methods. At the cellular level, the ratio between the volumes of CO2 produced and O2 consumed is referred to as the respiratory quotient (RQ). RQ follows biochemical equations similar (but not restricted) to the following ones:7-» 63O2+ 38H2O + SO34- 9CO(NH2)24- Energy RQ =63C°2 / 77O2 =0.8

[0003] Equations (1), (2), and (3) show that RQ provides essential information to map the origin of EE in terms of CHO, fat, and protein. For instance, a RQ of 1 implies that 100% of the energy was produced by the oxidation of CHO, while a RQ of 0.7 indicates that 100% of EE was produced by the oxidation of fat RQ for fat ranges between 0.69 and 0.73, depending on the oxidized fatty acid). Although cellular-level RQ measurements are determined by biochemical analyses, in vivo evaluation of RQ should require invasive techniques such as muscle biopsy, arterial and venous blood sampling, and blood gas analysis. These limitations preclude the direct estimation of the RQ with a single-cell resolution. Even if possible, different cells vary the substrate due to differences in the metabolic pathways of choice for a given energetic demand. In addition, for the same cell, the energetic sources may also change depending on the duration and intensity of the task. Therefore, solutions should check the general body RQ to account for the overall calorie sources as the metabolic product from agglomerates of cells.

[0004] Recent advances in physiological carts allowed breath-by- breath gas analysis with high accuracy, so there is a unique opportunity to infer the overall body respiration, thus, RQ, by the measurement of the oxygen uptake (VO2) and carbon dioxide (VCO2) released at the mouth level. The whole body “RQ” is calculated as the ratio between VCO2and VO2(or RER However, cellular respiration is connectedwith the lungs by the circulatory system that has to deliver the O2 and remove the CO2 for each cell, so circulation distortions may change the interpretation of the RER as the actual product of the RQ from multiple cells mainly during exercise when the circulation plays a key role in the aerobic response. Theimpact of these distortions over the aerobic response measured at the month level are vastly studied since the 80 ’s, and the current state-of-art regarding this topic will be discussed below.

[0005] During steady states of exercise, RER is a superposition of theRQ values of individual cells. Thus, the proportion of substrate consumption for energy generation can be defined by the following set of equations:(4) PCHOx1 + PFATX0.7 + (3PrX 0.8 = RER(5) PcHO + PFAT + Ppr=1, where / 3CH0, PFATand Pprrepresent the individual oxidative contributions as calorie sources of CHO, fat, and protein, respectively, with value ranging from O to 1.

[0006] Compared to other methods that derivate calorie sources from the time spent within some fixed HR zones, the estimation of calorie sources with RER is more precise because it is directly related to how the active cells are burning calories from different substrates to supply the energetic demand from exercise or to support the body metabolism at rest. The RER measurement during exercise is currently the state-of-the-art of non-invasive methodology to assess calories source in vivo. Therefore, assessing RER is of great interest for different applications across health domains, including the continuous monitoring of fitness levels, personalized training intensities and health level evaluation. However, direct RER measurements require expensive metabolic carts, such as the K5 and Quark from COSMED (Rome, Italy), as well as laboratory-based exercise protocols to collect the data. RER measurements are not accessible by general population, and cannot be incorporated in one’s daily life due to device and environmental restrictions.

[0007] Currently, there is no previous work that estimates calorie sources throughout the estimation of RER in the technical field related to wearable devices, such as smartwatches. The solutions found in the prior art focus on using HR zones, which is very limited.

[0008] The source of the EE during rest and physical activity providevaluable information about an individual's health and fitness level. Different proportions of CHO, fat, and protein consumption are an indicative of physical performance or presence of metabolic abnormalities.

[0009] State-of-art technologies for laboratorial measurements are based on indirect calorimetry obtained with metabolic carts. Despite precise, these technologies require specialized and expensive devices in laboratory settings. Technological alternatives include handheld devices for gas exchange measurements and sensors for sweat analysis.

[0010] State-of-art technologies for measurement of calorie sources of energy used in physical exercise demand specialized labour and expensive equipment. Those requirements limit their use only to specialized sports facilities.

[0011] With the current advances in the field of artificial intelligence and cutting-edge technologies of embedded sensors at wearable devices, new methods can be developed to monitor biological signals and predict important health and fitness indicators in a non-invasive, unobtrusive and constant approach. In the broad intersection of human metabolism and wearable technologies, most of the previous works have focused on quantifying the total body EE instead of also estimating the calorie sources of such EE (CHO, fat or protein) which have many health-related applications. Thus, there are only a few published papers and patents that relate to the present innovation.

[0012] In the paper entitled, “A wearable electrochemical biosensor for the monitoring of metabolites and nutrients”, the authors introduce a novel wearable non-invasive biosensor that can continuously monitor and trace different levels of multiple metabolites and nutrients, including amino acids and vitamins. The system proposed in this document works for vigorous exercises and with a sweat-based detector. However, this document relies on sweat-based detectors and cannot be applied for exercises of any intensity, as well as during resting.

[0013] The report entitled “A Handheld Metabolic Device (Lumen) to Measure Fuel Utilization in Healthy Young Adults: Device Validation Study”, presents Lumen, a handheld device from Metaflow Ltd designed to determine metabolic fuel usage by C02exhaled. In this study, the capacity of this device to detect changes in metabolic fuel usage was validated in healthy young adults. The device was able to detect changes in metabolism due to dietary intake, presenting results compatible with standard metabolic carts. The user can interact with Lumen by pairing it with their smartphone. However, Lumen is based on a breath sensor that requires the user to actively breathe into the device and cannot be use continuously.

[0014] In the paper entitled “Validation of Polar Grit X Pro for Estimating Energy Expenditure during Military Field Training: A Pilot Study”, the authors compare the performance of a commercial metabolic cart (COSMED, Rome, Italy) and the Polar HR monitor chest strap combined with a Polar Grit X Pro watch. The metabolic cart uses directly measures of VO2and VCO2to estimate the calorie sources and was used as ground truth for the calorie source estimation from the HR measured by Polar. This paper describes how Polar’s solution uses HR zone to estimate the calorie sources, but by a limited methodology that does not properly individualize the predictions by only considering fixed HR zones for their estimation.

[0015] In the paper entitled, “Low Cost Metabolic Fuel Sensor for On-Demand Personal Health and Fitness Tracking”, the authors describe a prototype metabolic fuel sensor designed for personal tracking of EE and fuel substrate utilization. This system employs indirect calorimetry, similar to metabolic carts, to determine EE from the direct measurement of VO2and VCO2at the mouth. As stated above, our solution estimates VO2and VCO2indirectly via anthropometric data and biological signals measured by the smartwatches, such as wrist-worn bioelectrical impedance analysis (BIA), HR, and personalized exercise intensity.

[0016] The publication “Model and Personal Sensor for Metabolic Tracking and Optimization”, presents a metabolic fuel model and a low-cost prototype breath sensor for measuring, tracking and enhancing the metabolism. The model is based on the relationship between insulin and blood glucose levels, fat and glycogen stores, and nutrient substrate utilization to predict fat accumulation or consumption. The data can be processed on the device and transferred via Bluetooth to an Android device. The information output includes respiration rate, instantaneous flow, tidal volume, minute volume, VO2, VCO2, RQ, and EE. The major difference from our innovation is that their solution employs a breath sensor to perform indirect calorimetry, while our solution uses only a smartwatch that has many advantages as explained above.

[0017] The patent “US 6.540.686 B2” proposes a method to describe EE in terms of CHO, fat, and proteins. This patent assumes that “only after 15 minutes from the outset of the exercise is it possible to utilize fatty acids”. This assumption is not precise as low-intensity exercises can bum fat in a proportion similar to (or even higher than) CHO. This limitation decreases the degree of personalization (thus, precision) of the estimated calorie sources, while the correct proportion can be precisely inferred by RER, thereby our methodology outperforms this patent. Another difference is that our innovation uses BIA data from smartwatch to estimate the initial storage of glycogen allowing the computation of protein oxidation with higher precision, and we use RER instead of exclusively HR zones to estimate the calorie sources.

[0018] The patent “US 2024 / 0065575 Al” proposes a wearable system to quantify metabolic and cardiorespiratory parameters of a subject, including RQ. One major difference of our solution is that we do not measure gases directly. We infer RER, which reflects RQ during steady states of exercise, through anthropometric data and bio-signals measured by smartwatch, such as BIA, HR, and individualized exercise intensity. We do not employ breath sensors that limit the calorie source estimation as stated above (practicallimitations).

[0019] The prior art does not comprise a solution based on wearable technologies to estimate continuously and noninvasively the individualized metabolic sources of EE during resting and physical activity based on RER. In addition, the prior art does not comprise a solution that uses machine-learning models to predict the metabolic sources of EE during resting and physical activity based on RER.

[0020] Thus, there is a need of a solution that quantifies more precisely the proportion and amount of CHO, fat, and protein consumed as well as the stamina, training load, and recovery time of each user.OBJECTIVES OF THE INVENTION

[0021] It is an objective of the invention to provide a system and method based on wearable technologies to estimate continuously and noninvasively the individualized metabolic sources of EE (CHO, fat, and protein), thus, enabling the estimation of energy-related parameters (stamina, training load, and recovery time during aerobic exercises) during resting and physical activity via the prediction of RER.

[0022] It is another objective of the invention to provide a system and method that quantify the proportion and amount of CHO, fat, and protein consumed as well as the stamina, training load, and recovery time of each user.

[0023] It is another objective of the present invention to provide a solution that uses wearable devices to predict the calorie sources of EE during resting and physical activities allowing the non-specialized user from general public to track their metabolic health status.

[0024] It is another objective of the present invention to enable the estimation of RER using commercially available wearable technologies, such as the smartwatches, eliminating the need for sophisticated metabolic carts.

[0025] It is another objective of the present invention to accurately estimate calorie sources (i.e., solve equations (4) and (5)), including proteinoxidation that may occur during prolonged exercises, by using wearable devices alongside bioelectrical impedance analysis (BIA) information. Specifically, individual’s skeletal muscle mass obtained from BIA and the predicted glycogen storage decay estimation from predicted RER can be used to compute the percentage of protein oxidation, stamina, training load, and recovery time of the users.SUMMARY OF THE INVENTION

[0026] The present invention proposes a method for estimating calorie sources of energy expenditure, EE, of a user during physical activities, comprising: obtaining user data including user profile data, bioelectrical impedance analysis (BIA) data, and physical activity data, wherein the physical activity data is obtained from a wearable device during a physical activity session; determining respiratory exchange ratio, RER, time series of the user during the physical activity with a RER prediction model based on the user data, wherein the RER prediction model is trained with a dataset comprising physical activities data to predict RER time series; determining an energy expenditure, EE, time series of the user during the physical activity with an EE prediction model based on the user data, wherein the EE prediction model is trained with a dataset comprising physical activities data to predict EE time series; and estimating the calorie sources during the physical activity based on the user data, RER time series and EE time series, wherein the calorie sources comprise carbohydrate, fat, and protein.

[0027] The present invention is also related to a system and a non- transitory computer-readable storage medium for estimating calorie source of energy expenditure of a user during physical activities.BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The invention is explained in greater detail below and references the drawings and figures attached herewith, when necessary. Attached herewith are the following:

[0029] Figure 1 presents a user case scenario of one embodiment of the invention, including the main displays of the application.

[0030] Figure 2 illustrates the input data and the data processing steps according to the present invention.

[0031] Figure 3 illustrates a workflow according to the present invention.

[0032] Figure 4 illustrates a user case scenario of an alternative embodiment of the present invention.

[0033] Figure 5 illustrates a workflow according to the second embodiment of the invention.DETAILED DESCRIPTION OF THE INVENTION

[0034] The present invention proposes a method for estimating calorie sources of energy expenditure, EE, of a user during physical activities, comprising: obtaining user data including user profile data, bioelectrical impedance analysis (BIA) data and physical activity data, wherein the physical activity data is obtained from a wearable device during a physical activity session; determining respiratory exchange ratio, RER, time series of the user during the physical activity with a RER prediction model based on the user data, wherein the RER prediction model is trained with a dataset comprising physical activities data to predict RER time series; determining an energy expenditure, EE, time series of the user during the physical activity with an EE prediction model based on the user data, wherein the EE prediction model is trained with a dataset comprising physicalactivities data to predict EE time series; and estimating the calorie sources during the physical activity based on the user data, RER time series and EE time series, wherein the calorie sources comprise carbohydrate, fat, and protein.

[0035] In general, the present invention refers to a method and system based on wearable technologies that estimates the calorie sources of energy expenditure (EE) during exercise through the prediction of respiratory exchange ratio (RER) from wearable devices. The present application is grounded on the human physiology and metabolism documented in decades of research in the fields of experimental and applied physiology.

[0036] The method and system according to the present invention use wearable technologies to estimates the calorie sources of EE, stamina level, training load, and recovery time during resting and exercise through the prediction of RER.

[0037] The preferred embodiment of the invention uses wearable devices, such as smartwatches, to quantify, on near real-time (1 Hz after an initial buffer of 30 seconds), the proportion [%] and absolute [g] amount of carbohydrate (CHO), fat, and protein consumed during resting and exercise (indoor and outdoor). In addition, by tracking the cumulative consumption of CHO and glycogen decay dynamics, the present invention estimates the stamina level (0%=about to fatigue, 100%=fully charged with glycogen), training load (100%=about to fatigue, 0%=fully charged with glycogen), and recovery time (time of rest requested to fully charge the glycogen levels, decreasing training load to zero and increasing stamina to 100%). It is important to note that, here, the stamina means the physical capacity to sustain the physical activity. These features provide a way for users to adjust and plan their workout sessions to best fit their individual objectives.

[0038] Figure 1 depicts a user case scenario of the preferred embodiment. A user 101 performs an exercise while wearing a smartwatch 102.The smartwatch displays (for a specific interval for example, every second), a screen 103 with the user’s stamina and training load in percentage, and detailed information about the substrate utilization and time to recover. The consumption of CHO, fat, and protein are shown in percentages in screen 104, and the total utilization in weight (grams) is displayed in screen 105. With this information, users can adjust their workout intensity to meet their goals in realtime. For example, exercise intensity can be adjusted to bum more fat if the goal is weight lost and avoid protein consumption to stay away from overtraining.

[0039] Figure 2 shows a mid-level description of the model according to the preferred embodiment. The system comprises prediction models 201 which acquire and process user data to predict the physiological targets (i.e., RER and EE). The physiological targets are then inputted into the biophysical model 202 that converts RER and EE into calorie sources, stamina, training load, and recovery time.

[0040] Figure 3 presents a detailed diagram of the preferred embodiment. Input data are related to user profile data 301 and data obtained by wearable sensors, which may be BIA data 302 or exercise data 303. The profile data comprises demographic information of the user, which may comprise: age, gender, weight, height, and body-mass index (as weight / height2). The sensor data carry information about the body composition from wrist-worn BIA, including body fat [%], body fat mass [kg], skeletal muscle mass [%], skeletal muscle mass [kg], fat free mass [kg], basal metabolic rate [Kcal], and total body water [kg]. Time-series are also acquired from the sensors during the exercise sessions which includes HR, walking running / walking cadence and / or speed. Both profile data 301 and sensor data are combined to create a vector of attributes 304. The vector of attributes has a quasi-static component (comprising the profile data 301 and BIA data 302) and a dynamic component (comprising the exercise data 303). The quasi-staticcomponent does not change during the exercise. The dynamic component is updated every second during the workout after an initial buffer of 30 seconds. The vector of attributes 304 is used on a trained machine-learning model to predict RER and EE. The system comprises a RER prediction model 305, trained to predict the RER(t) 306, and an EE prediction model 307, trained to predict the EE(t) 307. In the preferred embodiment, the machine learning of prediction models 305 and 307 are light models that can run in the microprocessor of the smartwatch, such as linear regression or random forests models.

[0041] The system further comprises a biophysical model 309 that is a component that merges specific features (such as skeletal muscle mass, see Biophysical Model section below) from the vector of attributes 304 with the predicted EE and RER from the RER machine-learning model 305 and EE machine-learning model 307 to estimate the time series of calorie source consumption 310, including CHO, fat and proteins consumption in percentages and in grams. Figure 3 also illustrate a workout summary 311 after the exercise session by aggregating EE, CHO, fat, and protein time series. The biophysical model will be described in detail below.Biophysical Model

[0042] Carbohydrate (CHO) and fat are the main sources of energy, but when the exercise time is long and the intensity is high enough, protein can account for up to 5% of the total EE. The source of energy (i.e., substrate utilization) is dependent on the exercise intensity and duration. For example,1. During short exercises at 25% of maximal intensity, most of the energy is provided by fat oxidation.2. During short exercises at 50% of maximal intensity, around 50% of the energy is provided by CHO and the other 50% by fat.3. During short exercises at 85% of maximal intensity, about 2 / 3 of the energy is provided by CHO and 1 / 3 by fat.4. During prolonged exercises, the use of protein starts to happen particularly when CHO sources are depleted.

[0043] The aforementioned dependence of substrate utilization on exercise duration and intensity allows us to use indirect physiological signals, such as heart rate (HR), to estimate the calorie sources since HR is influenced by exercise intensity and duration. This is the reason why prior art solutions use HR zones to estimate calorie sources. However, HR zones is an oversimplification because users with similar HR may have different metabolic profiles so the substrate utilization can be different. HR can only partially explain the individual metabolic demand (exercise intensity) as the exercise intensity is determined by the whole body VCO2and V'O2> and their ratio (RER).

[0044] During steady states of exercise, RER is a superposition of substrate utilization from the metabolism of all cells in the body. Hence, the proportion of substrate consumption for energy generation is defined by the following set of equations:(6) PCHOx1 + PFATX0.7 + pPrX 0.8 = RER(7) PcHO + PFAT + Ppr=1 where fie HO, PFAT and Pprrepresent the individual oxidative contributions as calorie sources of CHO, fat, and protein, respectively, with values ranging from O to 1.

[0045] For instance, a 0FAT = 0.1 indicates that 10% of EE comes from fat which is a substrate exclusively utilized in the aerobic metabolism. In general, during short duration exercises (preferably, running and walking), most of the energy is supplied by the aerobic pathway, while longer exercises require a combination of both aerobic and anaerobic systems. The aerobic system is about 14 times more efficient than the anaerobic one, but at the same time, slower to supply a new energetic demand as these two different systems differ on how they handle the substrates across the different metabolicpathways.

[0046] The major challenge of using RER to estimate the calorie source is solving equations (6) and (7) when protein oxidation occurs after long duration exercise because we will have three variables but only two equations. To overcome this problem, the protein oxidation is modeled as a function of the CHO availability stored as glycogen in the skeletal muscle mass (SMM). Therefore, considering ftPr= ftPr(glycogen) from the expected glycogen stored at the active SMM, equations (6) and (7) can be rearranged to:(8) CHO[%]= 100CAT 1 nn xz1~RER-0.2xfrT(glycogen)(9) r / U [o / o] = 1UU X0.3

[0047] Hence, the percentages of CHO (equation 8) and fat (equation 9) can be directly inferred for any given RER when the amount of glycogen (J3Pr(glycogen)) is known a priori. It should be noted that the percentage of protein is computed by multiplying ftPr(glycogen) by 100. The (3Pr(glycogen) time-series dynamics was modeled as the exercise progresses considering the following physiological grounds:1. CHO is limited and stored in the form of glycogen in the liver and SMM.2. A contracting muscle can only use its own local available glycogen. Therefore, @Pr(glycogen) can be estimated from the amount of glycogen stored in SMM.3. Under a regular diet, glycogen concentration in SMM is 85 mmol / kg, which is equivalent to 16.5 g / kg of wet weight SMM, on average.4. Fatigue state occurs when glycogen is lower than 4.5 g / kg of wet weight SMM.5. During most common physical activities, such as walking and running, the active portion of the total SMM corresponds to 56.3% of the whole body SMM.6. Protein oxidation does not occur in the beginning of the exercise because local glycogen stores are high, thus the EE comes from CHO and fat.7. We can trigger the protein utilization by assuming that protein oxidation increases as a linear function when glycogen is below 40% of its initial state that is considered as glycogen depletion state.8. Protein oxidation can only account for a maximal of 5% of the total EE.9. The average amount of energy produced by the oxidation of each gram of CHO, fat, and protein is 4.1, 9.3, and 4.2 kcal, respectively.

[0048] More recently, bioelectrical impedance analysis (BIA) was ported to wrist-worn wearables, so body composition analysis became more accessible to general population. As SMM is among the variables measured by BIA, the initial glycogen store is estimated by considering:(10) Glycogen (mmol) = 0.563 * SMM[kg* 85, or(11) Glycogen (g) = 0.563 * SMM[kg* 16.5 where 56.3% comes from physiological ground 5 while factors 85 and 16.5 are derived from physiological ground 3.

[0049] Thus, by combining the aforementioned points with wearablebased BIA (specifically SMM), we can estimate the initial glycogen state that will start to decrease as soon as the exercise begins, and the glycogen decay dynamics will be further determined by RER and EE. From the glycogen decay dynamics, the calorie source including protein oxidation, can be finally computed as follows:1. At the beginning of the workout, defined by t = t0, start with the initial glycogen level = G(to).2. Predict RER(t) and EE(t) at latter time t (for example, 30 seconds).3. Compute the ratio of protein oxidation (flPr(t)) as a function of G(to).4. Insert ftPr(t) and RER(t) in equations (8) and (9) to compute CHO[o / 0] and FAT[o / o].5. Use EE(t), CHO[o / o], FATpq, and ftPr(t) to compute the amount in grams of consumed CHO (CHO[g]),fat (FAT[g]), and protein (PRO[g]) such that:a. CHOla]= — xCH°(° / o\xEE(t)\kcal\,-100 4.1[Kcal]v y L J'6. Update the glycogen storage such that G(t) = G(t0) — CHO(g') / (0.563 * SMM[Kg]y)7. t -» t0and repeat steps 2 to 6.

[0050] The above construction is referred in the present invention as biophysical model component or simply as biophysical model. As the fatigue state occurs when glycogen is lower than 4.5 g / kg of wet weight SMM, Errol Imiiewior nSo definido. stamina levels can be described based on the current level of glycogen (G(t)), the fatigue onset (Gfatigue = 4.5 g / kg ), and full glycogen storage (G&U, (11)):(12) Stamina[%] = 100

[0051] F ollowing the same rationale, the Training Load of each session can be defined as a function of the glycogen spent in each session:(13) Training Load Session [%] = 100whereis the glycogen concentration consumed during an exercise session (i.e., total amount of consumed CHO divided by the active skeletal muscle mass).

[0052] The accumulated training load (or just Training Load for simplicity) considers the sum of the consumed glycogen of the last session with the deficit of glycogen due to previous sessions:^session ,r rtotal(14) Training Load [%] = 100 xde'““ = 100 xde'“uGfull Gfatigue Gfuu Gfatigue where Gdeficitis a deficit of glycogen in the beginning of the exercise due to previous exercise sessions, and G^fi^1= G™££n+ Gdeficit.

[0053] After exercise, there is an increase in the glycogen storage due to glycogen resynthesis which is on average 0.27 g / kg of wet weight SMM. Therefore, the deficit of glycogen and Training Load decrease in parallel andat the same rate:0.27 x hour(16) Training Load [%] (hour) = 100

[0054] Under this framework, the recovery time defined by the time in which Training Loadwill approach zero can be calculated:

[0055] The EE time-series and RER time-series predictions used in the above equations are determined by the RER prediction model 305 and EE prediction model 307, which are supervised machine learning models trained to estimate the VO2and VCO2from the features obtained from wearable d b(metabolic power = 16.89 x VO2+ 4.84 x VCO2) and then used to calculate the calorie sources and stamina in association with BIA measures, as explained above. The main inputs from the wearable devices are (but not limited to):1. Time series of: a. Step frequency b. HR2. BIA: a. Total body water b. Body fat mass c. SMM d. Soft lean mass e. Fat-free mass3. Anthropometries: a. Gender b. Age c. Height d. WeightDatasets

[0056] Experimental data from a standard metabolic cart were obtained along with a smartwatch during indoor and outdoor physical activities such as walking and running. Specifically, we employed the Quark and K5 from COSMED (Rome, Italy) for the indoor and outdoor data acquisitions, respectively. Both devices measure continuously and noninvasively breath-by- breath VO2and VCO2that were used to label the smartwatch data.

[0057] After data validation, the final dataset had data from 532 participants (369 males and 163 females, average age of 33 years old, weight of 71.3 kg, and height of 171.5 cm), at rest, walking, and running. The experimental protocols were developed by physiologists and included maximal graded intensity and sub-maximal exercises for the complete physiological evaluation of each participant. The data was acquired simultaneously and noninvasively by smartwatches and metabolic carts to assess total body VO2and VCO2(thus, RER). The dataset was used to train supervised machine learning models to estimate the V O2and VCO2. Next, V O2and V CO2were used to compute RER and EE as described above.

[0058] This preferred embodiment uses wearable devices to predict the calorie sources of EE during resting and physical activities allowing the nonspecialized user from general public to track their metabolic health status. This allows on-the-fly adjustment of the user activities toward their goals. The absence of costly equipment and multiple actions requirements from the user allows the diffusion of knowledge and adoption of better practices in physical activities.Alternative embodiment

[0059] As shown by the workout summary 311 in Figure 3, it would be possible to process the data only when the users finish the workout session, so only an information summary will be displayed, not in real-time during theexercise.

[0060] Although this alternative embodiment limits the iteration of the user with the application during the exercise, the machine learning algorithms can consume too much computation time from the microprocessor embedded into the smartwatch, limiting the real-time solution depending on the amount of available computational power. Although simplified, and applicable to wearable devices with lower processing capacities, this alternative embodiment is still very competitive and match with approaches of prior art that only display the calorie source when the exercise session is finished. This scenario precludes the real-time adjustments in exercise intensity to increase the training outcomes, but it can contribute for planning exercise training programs.

[0061] Figure 4 shows a high-level illustration of this alternative embodiment. The user 401 performs a physical activity, while wearing the smartwatch 402. After the workout session, the user 401 has access to a workout summary 403 of the calorie sources. The summary shows the exercise time, EE in kcal, training load, recovery time, and the calorie sources as CHO, fat, and proteins, both in percentage and in grams.

[0062] Figure 5 presents a detailed description of the alternative embodiment of the present invention. Similar to the preferred embodiment, input data are related to user profile data 501 and wearable sensors data, including BIA data 502 and exercise data 503. The profile data 501 has the following information: age, gender, weight, height, and body-mass index (as weight / height2). The sensor data carry information about the body composition from wrist-worn BIA, including body fat [%], body fat mass [kg], skeletal muscle mass [%], skeletal muscle mass [kg], fat free mass [kg], basal metabolic rate [Kcal], and total body water [kg]. Time-series are also acquired from the sensors during the exercise sessions which includes HR, walking running / walking cadence and / or speed. The time-series of the whole workout session are aggregated and combined with the profile and BIA data to create avector of attributes. Both profile data 501 and sensor data 502 and 503 are combined to create a vector of attributes 504. This vector of attributes 504 is used in trained machine learning models.

[0063] As the processing of the data is not done in near real-time as the preferred embodiment, the machine learning algorithms can be more sophisticated and time consuming, such as recurrent neural networks. After the exercise session, the machine learning models 505 predict RER and EE timeseries that feeds the biophysical model 506. The machine learning models including a RER prediction model and an EE prediction model as in the preferred embodiment. The biophysical model 506 is the same as the one described above in the preferred embodiment. The final output of this solution is a workout summary 507.

[0064] As seen, the present invention provides a unique method to estimate the sources of calorie consumption, stamina, training load, and recovery time by the estimation of RER through machine-learning models and data acquired with wearable devices. While other state-of-art techniques need the use of devices for respiration and sweat analysis, our method employ only information acquired by a wearable device. The invention proposes the use of a more precise and effective methodology than other techniques that use HR solely, delivering a solution with improved accuracy and higher temporal resolution.

[0065] Furthermore, as it will be clear for a person skilled in the art, the exemplificative embodiments described herein may be implemented using hardware, software, or any combination thereof and may be implemented in one or more computer systems or other processing systems. Additionally, one or more of the steps described in the example embodiments herein may be implemented, at least in part, by machines. Examples of machines that may be useful for performing the operations of the example embodiments herein include general purpose digital computers, client computers, portablecomputers, mobile communication devices, tablets, smartphones, notebooks or wearable electronic devices, such as smartwatches.

[0001] For instance, one illustrative example system for performing the operations of the embodiments herein may include one or more components, such as one or more microprocessors, for performing the arithmetic and / or logical operations required for program execution, and storage media, such as one or more disk drives or memory cards (e.g., flash memory) for program and data storage, and random-access memory, for temporary data and program instruction storage.

[0002] Therefore, the present invention is also related to a system for estimating calorie source of energy expenditure of a user during physical activities comprising a processor and a memory comprising the computer- readable instructions that, when performed by the processor, cause the processor to perform the method steps previously described in this disclosure.

[0003] The system may also include software resident on a storage media (e.g., a disk drive or memory card), which, when executed, directs the microprocessor(s) in performing transmission and reception functions. The software may run on an operating system stored on the storage media, such as, for example, UNIX or Windows, Linux, Android, and the like, and can adhere to various protocols such as the Ethernet, ATM, TCP / IP protocols and / or other connection or connectionless protocols.

[0004] As well known in the art, microprocessors can run different operating systems and contain different software types, each type being devoted to a different function, such as handling and managing data / information from a particular source or transforming data / information from one format into another format. The embodiments described herein are not to be construed as being limited for use with any particular type of server computer, and any other suitable device for facilitating the exchange and storage of information may be employed instead.

[0005] Software embodiments of the illustrative example embodiments presented herein may be provided as a computer program product or software that may include an article of manufacture on a machine-accessible or non- transitory computer-readable medium (also referred to as "machine-readable medium") having instructions. The instructions on the machine-accessible or machine-readable medium may be used to program a computer system or other electronic device. The machine-readable medium may include, but is not limited to, floppy diskettes, optical disks, CD-ROMs, magneto-optical disks, or another type of media / machine-readable medium suitable for storing or transmitting electronic instructions.

[0006] Therefore, the present invention also relates to a non-transitory computer-readable storage medium for estimating calorie source of energy expenditure of a user during physical activities, comprising computer-readable instructions that, when performed by the processor, cause the processor to perform the method steps previously described in this disclosure.

[0066] The techniques described herein are not limited to any particular software configuration. They may be applicable in any computing or processing environment. The terms "machine-accessible medium," "machine- readable medium" and "computer-readable medium" used herein shall include any non-transitory medium that is capable of storing, encoding, or transmitting a sequence of instructions for execution by the machine (e.g., a CPU or other type of processing device) and that cause the machine to perform any one of the methods described herein. Furthermore, it is common in the art to speak of software in one form or another (e.g., program, procedure, process, application, module, unit, logic, and so on) as taking action or causing a result. Such expressions are merely a shorthand way of stating that the execution of the software by a processing system causes the processor to act to produce a result.

[0067] As described above, the present invention proposes a method that estimates the calorie sources at rest and during exercise (walking and running) by predicting the RER with individualized physiological anddemographic data, including heart rate (HR), velocity-related time-series, BIA, weight, age, and gender. Ground truth RER time-series for training machinelearning models were obtained as the ratio between the volumes of CO2 release and O2 uptake measured at the mouth level by a mask attached toa standard metabolic cart. This invention allows users to check calorie sources (CHO, fat or protein), stamina (performance reserve), training load, and recovery time in near-real time so that these features can be used as a physical activities coach, such as a running or walking coach, for guiding individualized exercise intensity. Training load and Recovery time will also be shown to the user after the exercise so that they can adjust and plan their workouts accordingly.

[0068] The advantages and effect of the invention are associated to the prediction of RER with wearable devices, and the use of the predicted RER coupled with BIA data to predict other physiological parameters. These two elements allow the estimation of the calorie sources in terms of CHO, fat, and protein as well as to estimate the decay and dynamics of glycogen’s storage during physical activities. The dynamics of glycogen decay during exercise is used to estimate users’ stamina, training load, and recovery time. Stamina indicates when the muscular fatigue is about to happen. Training load shows the cumulative amount of stress placed on an individual from a single or multiple training sessions over a period of time. Recovery time indicates the requested time for achieving a full recover (i.e., training load equals to zero).

[0069] The major limitation of the prior art solutions is the oversimplification of the phenomenon by only considering the time that users spent in some predefined HR zones, instead of taking into account many individualized metabolic variables related to energy uptake during exercise. While the main competitor assumes fixed HR zones for each of the calorie sources, the actual phenomenon is a continuum (i.e., there are unlimitedpossible combination of the proportions of CHO, fat, and protein for different HR within the same HR zone). More specifically, the competitor’s solution is based on simulations that consider the following relationships between HR and stimulus duration:From one to two hours at the anaerobic threshold: the consumptions of CHO, fat, and protein are within the ranges: 68 - 86%, 10 - 22%, and 4 - 9%, respectively.From one to two hours at the aerobic threshold: the consumptions of CHO, fat, and protein are within the ranges: 48 - 58%, 41 - 49%, and 1 - 3%, respectively.30 minutes at 100 beats per minute: the consumptions of CHO, fat, and protein are within the ranges: 27 - 40%, 60-73%, and 0%, respectively.30 min at 95% of maximal HR: the consumptions of CHO, fat, and protein are within the ranges: 89 - 91%, 6 - 7%, and 3%, respectively.

[0070] To extract the calorie sources without limiting the present invention to fixed HR zones, a machine learning approach is proposed to continuously predict individualized values of RER which increased the accuracy of the calories source and stamina predictions.

[0071] Compared to other solutions, another advantage of the present invention is the use of BIA to estimate the amount of glycogen stored and available for performing the exercise. While other techniques use the body weight to estimate CHO reserves, the present invention uses the skeletal muscle mass acquired with BIA data. This is a major advance as a contracting muscle can only use its own local available glycogen.

[0072] This invention presents a unique method for decomposing EE at rest and during exercise into CHO, fat, and protein components using machine-learning algorithms trained on individual physiological time-series and demographic data. These algorithms predict RER obtained from advanced metabolic carts, allowing for accurate estimation of the calorie sources, user’sstamina, training load, recovery time, and glycogen depletion, without the need of respiratory or sweat analysis; According to the preferred embodiment, the invention requires only a wearable device (such as Galaxy Watch4, Galaxy Watch5, Galaxy Watch6, and Galaxy Watch7). Additionally to the fitness domain applications, the calorie sources provide important biomarkers for metabolic disorders, such as prediabetes and diabetes. In addition, this solution has applicability in weight loss programs. In comparison to our primary competitor in the smartwatch market that only considers HR, the present invention offers a significant advantage by displaying personalized calorie sources in near real-time, empowering users to adapt their workout intensity to achieve the desired fitness outcomes. As the present invention can be encoded into wearable devices, personalized training plans (such as fat bum) can be individualized and optimized for each user. Thus, users can reach their fitness goals faster and with more effectiveness, while avoiding overtraining, exhaustion, and injuries.

[0073] While various exemplary embodiments have been described above, it should be understood that they have been presented by example, not limitation. It is apparent to persons skilled in the relevant art(s) that various changes in form and detail can be made therein.

Claims

CLAIMS1. Method for estimating calorie sources of energy expenditure, EE, of a user (101) during physical activities, characterized by comprising: obtaining user data including user profile data (301), bioelectrical impedance analysis, BIA, data (302) and physical activity data (303), wherein the physical activity data (303) is obtained from a wearable device (102) during a physical activity session; determining a respiratory exchange ratio, RER, time series (306) of the user during the physical activity with a RER prediction model (305) based on the user data, wherein the RER prediction model (305) is trained with a dataset comprising physical activities data to predict RER time series (306); determining an energy expenditure, EE, time series (308) of the user during the physical activity with an EE prediction model (307) based on the user data, wherein the EE prediction model (307) is trained with a dataset comprising physical activities data to predict EE time series (308); and estimating the calorie sources during the physical activity based on the user data, RER time series (306) and EE time series (308), wherein the calorie sources comprise carbohydrate, fat, and protein.

2. The method of claim of 1, characterized in that the user profile data (301) comprises the user age, gender, weight, height and body-mass index; the BIA data (302) comprises the user body fat percentage, body fat mass, skeletal muscle percentage, skeletal muscle mass, fat free percentage, fat free mass, basal metabolic rate and total body water; and the physical activity data (303) comprises the heart rate time series, velocity time series and cadence obtained through the wearable device (102) during the physical activity session.

3. The method of claims 1 or 2, characterized by further comprising:estimating the decay of glycogen storage based on the RER time series (306); and estimating the protein oxidation, stamina, training load and recovery time of the user based on the BIA data (302) and the estimated decay of glycogen storage.

4. The method of any one of claims 1 to 3, characterized in that the calorie sources are estimated in percentages by computing:PCHO X 1 4- PFAT X 0.7 T Ppf X 0.8=RER,'PcHO + PFAT + Ppr=1 / where PCHO> PFAT> and Ppr represent the individual oxidative contributions as calorie sources of CHO, fat, and protein, respectively, with value ranging from 0 to 1 ; and determining the ratio of protein oxidation, / 3Pr, as function of the carbohydrate stored as glycogen in the skeletal muscle mass G(t).

5. The method of claim 4, characterized by further comprising: determining the percentage of carbohydrate and fat by:RER — 0.7 — 0.1 x BPr(glycogen) CH0[%]= 100 x - n kJ / V AT mo FAT[%]= 100wherein CHO|%| is the percentage of carbohydrate, FAT[%] is the percentage of fat and / 3Pr(glycogen) is the proportion of protein contribution, percentage of protein is computed by multiplying / 3Pr(glycogen) by 100.

6. The method of claim 5, characterized by further comprising: computing the amount calorie sources consumed in weight by:1 CH0(%)z x riCH°^]~ 100X4.1[Kcal]X EE^kcal^1 FAT(° / o)z s ri„100X9.3[Kcal]X EE^kcal^ BPr(t)PR0^ = ^^X Em cal]’wherein CHO[g] is the amount of carbohydrate in weight, FAT[g] is the amount of fat in weight and PRO[g] is the amount of protein in weight; and updating the concentration of glycogen storage by:G(t) = G(t0) - CHO(W(0.563 * SMM[Kg])) wherein SMM|Kg| is the amount of skeletal muscle mass in weight.

7. The method of claim 6, characterized by further comprising: determining the stamina level based on the current level of glycogen G(t), the fatigue onset G fatiguean^ fall glycogen storage GfUn by:Stamina [%] = 1008. The method of claim 6 or 7, characterized by further comprising: determining the training load of the physical activity as a function of the glycogen:Training Load Session [%] = 100wherein GJpent°nis the glycogen concentration consumed during a physical activity session determined by the ratio between the total amount of consumed carbohydrate and the active skeletal muscle mass, and Gfatigueis the glycogen amount in the fatigue state; and computing the accumulated training load by:Training Load [%] = 100where Gdeficitis a deficit of glycogen in the beginning of the exercise due to previous exercise sessions, and= G“££n+ Gdeficit.

9. The method of claim 8, characterized by further comprising: computing the recovery time by:

10. The method of any one of claims 1 to 9, characterized in thatthe RER prediction model (305) and the EE prediction model (307) are supervised machine learning models trained to predict oxygen uptake and the carbon dioxide output; wherein the RER prediction model (305) is further configured to convert the predicted oxygen uptake and the carbon dioxide output into the RER time series (306); and wherein the EE prediction model (307) is further configured to convert the predicted oxygen uptake and the carbon dioxide output into the EE time series (308).

11. The method of any one of claims 1 to 10, characterized in that the RER prediction model (305) and the EE prediction model (307) are trained with a dataset comprising data acquired simultaneously and noninvasively by smartwatches and metabolic carts to assess total body oxygen uptake and carbon dioxide output.

12. The method of any one of claims 1 to 11, characterized in that estimating the calorie sources during the physical activity further comprises: continuously estimating the calorie sources during the physical activity session; and continuously displaying in the wearable device (102) the estimated calorie sources to the user (101).

13. The method of any one of claims 1 to 11, characterized in that estimating the calorie sources during the physical activity further comprises: estimating the calorie sources after the physical activity session; and displaying in the wearable device (102) the estimated calorie sources to the user in a workout summary (403) after the physical activity session.

14. System for estimating calorie sources of energy expenditure, EE, of a user (101) during physical activities, characterized by comprising: a processor; and a memory comprising computer readable instructions that, when executed by the processor, causes the processor to perform the method as defined in any of claims 1 to 13.

15. Non-transitory computer-readable storage medium characterized by comprising computer-readable instructions that, when performed by a processor, cause a computer to perform the method as defined in any claims 1 to 13.

Images