Mobile intelligent injury minimization system and method

Abstract

A system 300 and method for biologically monitoring an athlete's health status and providing alerts 338 when an overtraining state is determined to reduce injuries. Through the implementation of an efficient system architecture, the use of micro-artificial intelligence is practically useful in mobile situations where internet coverage is poor or non-existent.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. Provisional Patent Application No. 62 / 982,660 and U.S. Patent Application No. 16 / 863,285, the entire contents of each of which are incorporated herein by reference.

[0002] FIELD OF THE INVENTION The present disclosure relates to improved systems and methods for physically identifying biological conditions, such as overtraining and injury that alter fitness training activities, and the architecture and design of the system configured to reduce memory storage requirements for certain artificial intelligence-based applications on mobile devices. [Background technology]

[0003] Background of the Invention Some athletes choose a finish line event (e.g., the New York City Marathon) and attempt to implement a training program with little consideration for the possibility of injury. Athletes are often hyper-focused and miss cues, or they simply do not notice or are unable to recognize the signs of overtraining or impending injury. Athletes would be helped if they could receive an overtraining indication from a device (e.g., a fitness tracker or similar device) that they carry with them during their training activities. Summary of the Invention [Problem to be solved by the invention]

[0004] Artificial intelligence (“AI”) and machine learning algorithms are becoming more robust in today's society to solve many problems. Traditional AI techniques utilize large data sets to identify patterns and draw conclusions. Traditional AI techniques are simply not suitable for use on the types of mobile devices athletes are likely to carry, for a variety of reasons. First, the typical memory requirements needed to accommodate the large data sets often used by traditional AI are prohibitive. Second, the processing power needed to generate the conclusions that can be obtained from traditional AI techniques is equally large. Third, when athletes train, they typically venture into areas without Wi-Fi networks, and even areas without internet. Even if athletes' devices are equipped to support cellular coverage, there are still areas where cellular coverage is nonexistent or impaired. Even when cellular coverage is available, the cost of transmitting large amounts of data would make the use of such cellular coverage prohibitively expensive and impractical for typical AI methodologies. What is needed is a system and method that incorporates the ability to analyze and provide alerts or indications of rising overtraining conditions so that modifications to the training program can be made before it is too late and before an overtraining or injury situation occurs. [Means for solving the problem]

[0005] overview The present disclosure, in one aspect, provides a less complex system and method for identifying overtraining conditions and providing advanced alerts to athletes / users. In examples utilizing artificial intelligence ("AI"), the system and methodology provides an architecture that reduces memory, communication, and processing requirements, allowing "micro-AI" to become a reality. In one aspect, the system and method utilizes dichotomous data separated into "latent data" and "current data." The latent data is preferably further separated into general athlete historical data (which may be categorized across various non-specific individuals) and historical data specific or particular to an individual athlete. The range of data usable by the micro-AI is reduced and separated compared to traditional AI systems to increase efficiency, making mobile applications a practical possibility.

[0006] In one preferred aspect, the present disclosure provides a fitness tracker wearable around a portion of a user's body. The fitness tracker includes a heart rate sensor and a memory. The memory includes a latent memory component configured to hold latent data, the latent data including non-personal data not specifically related to the user and personal data specific to the user. The latent memory is updatable only when the tracker is located within an internet service area. The memory also includes a current memory component configured to hold current data that can be updated while the user is exercising. The tracker also includes a microprocessor including a classifier. The microprocessor is configured to determine the presence of an overtraining condition based on a predictive model using only the data in the memory, regardless of whether the tracker is located within an internet service area. The microprocessor is configured to provide a warning to the user after determining that an overtraining condition exists according to the output of the classifier.

[0007] In another aspect, the present disclosure provides a system for biologically monitoring an athlete's health status and providing an alert to the athlete when the biological status indicates the athlete is overtraining. The system includes a remote processor accessible over the internet, a remote database accessible over the internet, and a wearable fitness monitoring device having a heart rate sensor, an on-board processor, and a memory. The memory includes a latent memory that can be updated only when the device is in internet communication with the remote processor, and a current memory that can be updated with current training data measured by the monitoring device while the athlete is training. The latent memory is configured to store non-personal data that is not specific to a wearer of the fitness device. The non-personal data includes historical data of a general athlete. The latent memory is also configured to store personal data that is specific or unique to a wearer of the fitness device. The on-board processor includes at least one classifier. The on-board processor is configured to analyze current data and compare the current data with the latent data. The on-board processor identifies the likelihood that the athlete is overtraining or is injured based on the output of the classifier. The on-board processor provides an alert to the athlete upon determining that the athlete is overtraining based on the output of the classifier. The system may further include a temperature sensor.The system may further include a moisture sensor configured to measure a sweat index.

[0008] In another aspect, the present disclosure describes a method for generating a latent feature set related to past training patterns and results; periodically refreshing a wearable fitness device worn by the athlete with the latent features; recording the athlete's current training data while the athlete is training to generate a current feature set, the current features including heart rate; performing statistical applications on the latent features and the current features to generate a feature vector indicative of an overtraining state; feeding the feature vector to a neural network or any other selected classifier resident on the wearable fitness device; and obtaining a result from the neural network regarding whether an overtraining state exists.

[0009] The following terms are defined for clarity as you read this disclosure: References herein to "training" generally include exercise, and the person participating in the training activity need not be a professional athlete, but can be anyone with an interest in exercise.

[0010] "Latent data" is data that is not constantly updated, but is updated occasionally (if at all) at convenient times, such as when located in a Wi-Fi zone. Latent data can include general historical data and specific historical data, both of which are described below.

[0011] "General historical data" is historical data that is not focused on a single individual, but relates to a grouping or segmentation of the human population.

[0012] "Specific historical data" is historical data that is specific to a particular person or individual and that relates to that person's personality and / or experiences.

[0013] "Current data" is data that is constantly or dynamically updated when the data collection device is active.

[0014] "Training intensity" is a measure of an individual's athletic effort or perceived athletic effort or exertion during exercise.

[0015] As used herein, "configuring" includes creating, changing, or modifying a program on a microprocessor, computer, or network of computers so that the processor, computer, or network of computers behaves according to a set of instructions. The programming for carrying out the various embodiments described herein will be apparent to those skilled in the art after reviewing this specification, and for the sake of brevity, will not be described in detail herein. The programming can be stored on a computer-readable medium, such as, but not limited to, a non-transitory computer-readable storage medium. The system can be implemented on a field-programmable gate array and a graphics processing unit.

[0016] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not intended to limit the invention as claimed unless otherwise stated. As used in this specification and claims, the term "comprising" and its derivatives, including "comprises" and "comprises," includes each of the recited integers but does not exclude the inclusion of one or more additional integers. The claims filed hereby are incorporated herein by reference. The entire disclosures of U.S. Patent Nos. 10,013,638, 10,124,234, and 10,322,314 are incorporated herein by reference.

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate several embodiments of the invention and, together with the description, serve to explain the principles of one or more aspects of the invention. [Brief explanation of the drawings]

[0018] Brief description of the diagram [Figure 1]FIG. 1 is a flow diagram of a method for determining an overtraining condition, according to an embodiment of the present disclosure. [Figure 2] 1 is a pictorial representation of an athlete wearing a fitness tracking device. [Figure 3] FIG. 1 is a flow diagram of a method for classifying overtraining conditions for use in neural networks. [Figure 4] FIG. 1 is a flow diagram illustrating the interaction between remote and local storage and processing of general population historical data and individual historical data, and how different inputs are handled to generate outputs according to preferred aspects of the present disclosure. [Figure 5] FIG. 1 is a flow diagram showing how a latent model is created. DETAILED DESCRIPTION OF THE INVENTION

[0019] Detailed Description of the Drawings Reference will now be made in detail to the exemplary embodiments of the invention, some of which are illustrated in the accompanying drawings.

[0020] FIG. 1 illustrates a flow diagram of a method and system 100 for identifying and providing an advanced alert to an athlete regarding an overtraining condition or a condition in which certain training is not appropriate. An exemplary situation in which training may be inappropriate is when the athlete has asthma and weather conditions are not conducive to the athlete's condition. While the preferred embodiment is described in relation to an overtraining condition, it will be understood that the present disclosure has broader application, as explained further below. Briefly, in a fitness device 124 (FIG. 2) carried by an athlete, the device 124 is configured to have dichotomous data separated into "latent data" 104 and "current data" 106. The latent data 104 is preferably further separated into general athlete historical data (which may be categorized across various non-specific or non-specific individuals) and specific historical data specific to the individual athlete. The latent data can be updated at convenient times, such as when the device enters a public Wi-Fi network. Current data 106 is updated while the user is working out during a training activity, regardless of internet service area, and can later be automatically uploaded to a remote server when Wi-Fi is available (or via any configured internet connection) to update any potential data stored on the remote server. Using predefined parameters, a fitness device with an on-board processor with micro-artificial intelligence capabilities can utilize a combination of data groupings to feed on-board neural networks and / or other classifiers, depending on the intended application, to identify the presence of an overtraining condition, provide an alert to the athlete that an overtraining condition exists, and offer alternatives for modifying the training plan to avoid potential overtraining situations and injuries. The scope and amount of data available through micro-AI is reduced compared to traditional AI systems, making mobile applications a realistic possibility.

[0021] As previously discussed, the latent data 104 may include general athlete historical data and specific or unique historical data for an individual athlete. The general historical data may relate, for example, to demographic data for the general human population, and may be categorized or subcategorized. For example, if the athlete is a 40-year-old male runner, the general historical data may focus on the subset of the human population that targets male runners in the 40-44 age range. Other factors may be taken into account: geographic location, type of activity (e.g., running, whether marathon runner or sprinter), demographic data (age, gender), experience with a particular activity (e.g., novice or long-time activity), terrain (e.g., hilly, flat), and climate (e.g., hot, cold, humid).

[0022] The specific historical data relates to the athlete's own personality and experiences. For example, such data items may include injury susceptibility factors (how susceptible is the person to injury or how tolerant is they to strenuous exercise), family history / genetics, specific activity experience, and training goals (fitness level or upcoming competition). For training organization, the athlete's goals and time (e.g., time between exercise sessions, sleep time, and time remaining until a competitive finish (number of days or weeks until the competition date)) can be addressed.

[0023] Other indicators and factors indicative of an overtraining state include measuring and recording fatigue factor (a subjective factor relating to how tired a person feels), muscle soreness, sweat index (which takes into account how much an athlete is sweating, the amount of water lost through the skin, and preferably humidity levels), and VO2 max data (maximal oxygen uptake, which relates to the maximum amount of oxygen a person can utilize during hard exercise). VO2 max data is typically obtained by testing an athlete running on a treadmill while wearing a breathing mask. As such, all of the features can be subjective, objective, or a combination of subjectively and objectively obtained features.

[0024] Current data 106 is data updated and stored on the athlete's fitness device (e.g., fitness device 124 (FIG. 2)) while the athlete is training, tracking aspects of the athlete's training activity, such as exercise type (e.g., core workout, yoga, swimming, cycling, running), exercise duration (cumulative time and time for different exercise segments), exercise location (geography), elevation change, and heart rate. Such data is stored for a predetermined length of time (e.g., two weeks), after which such data is automatically deleted from the fitness device as "stale" current data. Storing current data for a predetermined period of time frees up memory for newer data. Depending on the intended use, other time periods, such as only a few days or even a few hours, can also be accommodated. Meanwhile, the current data can be automatically or manually uploaded from device 124 to a remote server when the device is within internet service area for merging or integration with potential data residing on the remote server, as described in more detail below.

[0025] Referring to FIG. 2 , device 124 may be a global positioning system (GPS) fitness tracker with processing capabilities and memory configured to log or record athlete training data. Device 124 may be a wearable device (e.g., a fitness tracker worn around a user's wrist, chest, or head). For example, device 124 may be incorporated as part of a watch, swim goggles, bicycle helmet, chest strap, or running shoes. Device 124 may include a Wi-Fi wireless transceiver and / or a Bluetooth wireless transceiver. Device 124 may be paired with a mobile communication device, such as a smartphone, to take advantage of the smartphone's memory capabilities. Some types of exercise, such as swimming, do not lend themselves to the athlete carrying a smartphone while training.

[0026] The time frame for automatic deletion of current data can be based on the volume of current data. For example, 14 days of training data can be stored regardless of the actual date of training. That way, even if it's been a week since you last used the device, you'll still have two weeks' worth of current data on the device.

[0027] The system can be configured such that, when internet is available, current data is merged or integrated with latent data residing on a remote server to update the latent data. For example, when the device 124 enters an internet service area, the latent data is refreshed in the device's memory, and a portion of the current data can be added to the latent data, such as updating the athlete's specific historical data with recent activities from the current data. Refreshing or updating the latent data (e.g., refreshing or updating the athlete's specific historical data with elements of the current data) can be based on stale current data (data older than a predetermined time frame) and / or all of the athlete's training data (specific past and current data as of the most recent training activity). Optionally, archived current data can be used to update the latent data using only the stale current data, such that merging of the stale current data and latent data is completed at the remote server where such current data is archived. If archived data is utilized, a decompressor can be used prior to the data merging / integration.

[0028] In a preferred embodiment, the latent data is not continually updated or refreshed with current data, but is instead maintained separately as a distinct component of device memory, so that, for example, during a training session, the amount of latent data remains static and substantially unchanged.

[0029] Heart rate data can be of two general types: active heart rate (heart rate during workouts) and resting heart rate (e.g., heart rate measured when an athlete first wakes up at the start of the day). Active heart rate can be weighted or analyzed against historical heart rate data (general category and / or specific individual).

[0030] Resting heart rate or "wake-up heart rate" (WHR) is a very good indicator of fitness and overtraining. It is usually measured when a person wakes up, and a pulse rate in the high 30s per minute usually means that the person is in good health. A pulse rate in the 40s or even 50s per minute is a sign of good fitness. Wake-up heart rate can be measured by a fitness device with a heart rate sensor if the athlete wears the fitness device to sleep, or it can be obtained manually by taking the athlete's pulse while in a prone position.

[0031] Real-time "active heart rate" monitoring during exercise (typically in fitness tracking devices) can also indicate health and overtraining, but it is usually related to physical exertion and effort. A heart rate of 60-90 beats per minute usually means a person is engaged in low-intensity activity (e.g., walking or sitting). A range of 100-140 beats per minute usually indicates moderate activity. A range of about 140-160 beats per minute indicates vigorous activity. A heart rate of 170-210 beats per minute typically indicates peak exertion. Maximum heart rate is age-dependent. Young people (in their 20s and 30s) should be able to have a maximum heart rate of about 210 beats per minute. Maximum heart rate has been shown to decrease with age. For someone in their 40s, a typical maximum heart rate may be 190 beats per minute.

[0032] Preferably, a set of 17 features related to overtraining status is utilized for classification, including nine latent features (e.g., age, heart rate, age difference from average age within a specific sport, exercise duration, exercise volume, past injuries, past intensity measurements, variables derived from statistical analysis based on past data, parameters from statistical tests or regressions) and eight current features (e.g., heart rate (active and resting), exercise duration, exercise type, time between exercise sessions, exercise volume, intensity level, sleep time). Intensity level can be measured by heart rate, a combination of heart rate and pace (speed) (when exercise involves long distance travel), or weight for weight training.

[0033] It will be understood that various combinations of features can be used without departing from the scope of this disclosure. For example, more or fewer than nine latent features can be utilized. More or fewer than eight current features can be utilized. The feature set can include any feature that contributes to identifying an overtraining state. Some additional important features relevant to detecting an overtraining state include heart rate, training duration, and intensity (depending on the type of exercise), in combination with historical demographic data.

[0034] Referring to Figure 1, once the feature set has been generated, the set is preferably classified using one or more classifier models 108 to create secondary features. Referring to Figure 1, a neural network (NN) classifier 112 is used to distinguish between normal and abnormal training, taking into account the specific characteristics of normal populations and individuals. There are a variety of techniques suitable for use as a classifier. Suitable classifiers include, but are not limited to, statistical applications (e.g., Bayes, K-nearest neighbors, fuzzy pyramid linking, discriminant analysis (DA), logistic regression (LR), multivariate adaptive regression splines (MARS), support vector machines (SVM), and hidden Markov models), neural networks (parallel, double, deep learning, regression), decision trees, association rule mining, and case-based reasoning, or any combination of the foregoing.

[0035] Neural networks (NNs) typically involve artificial neurons that are applied with a set of inputs, each representing the output of another neuron. Each input is multiplied by a corresponding weight, analogous to the strength of a synapse in a biological neuron. The weighted inputs are summed to determine the neuron's net input. This net input is further processed by using a squashing function (activation function) to generate the neuron's output signal. This function can be linear, nonlinear (e.g., a step function), or sigmoid (S-shaped).

[0036] Examples of weighting inputs include giving a higher weight to heart rate during exercise durations and giving a lower weight to heart rate during short sleep periods (e.g., during the last 24 hours) and when awake. For example, active HR may be assigned a weight of 1.5, while WHR may be assigned a weight of 1.0 or 1.1. Other features may be assigned lower weights, such as the user's age being assigned a weight of 0.5.

[0037] 1, for NN applications 112, the current features extracted from the current data 106 and secondary features from the classifier model 108 are preferably normalized to create a feature set 110 to improve training efficiency. All features are preferably normalized as real numbers ranging from 0 to 1 based on the entire data set used for training and testing.

[0038] Feature vectors are preferably created for all states associated with overtraining and associated with the current features 106 plus the output from one or more classifier models 108 (e.g., LR, DA, MARS models). In this step, a file with vectors, containing, for example, 20 features, for all cases is created.

[0039] The features are preferably normalized before the feature file can be efficiently used by the NN 112. The normalized combined feature vector is fed to the neural network 112 for further classification.

[0040] It will be appreciated that methods other than statistical applications can be used to provide further or secondary features, for example, data mining techniques or other classifiers (such as neural networks, decision trees, association rule mining and case-based reasoning or any combination of the above) can be used to provide secondary features, which can ultimately be fed into a neural network.

[0041] Other applications can be used instead of or in addition to NNs as the final classifier, for example, NNs can be used in combination with or instead of SVMs for final classification, if desired.

[0042] If desired, micro-AI analysis can be performed in real-time athletic situations (i.e., while the athlete is training). While micro-AI analysis during training consumes more device power and processing power, it can be performed when conditions are right (e.g., when conditions exist to trigger local in-training AI analysis). Several trigger levels can be utilized. For example, a primary trigger can be active heart rate level. Heart rate is one of the most important measures in identifying the presence of an overtraining state. Secondary triggers can include one or more of training duration, short sleep periods (e.g., sleep during the last 24 hours), training intensity, and wake-up heart rate (WHR). Real-time micro-AI analysis can be initiated solely by a primary trigger (e.g., heart rate above a predetermined threshold), by a combination of secondary triggers (e.g., a combination of current training duration longer than a threshold level, a sleep level below a predetermined minimum level, and a WHR above a predetermined threshold), or by a combination of primary and secondary triggers (e.g., a combination of active HR and training intensity). When using active heart rate, the micro AI analysis compares current data of active heart rate and training intensity level (running situation, running pace) with latent data related to an individual's specific historical data (active heart rate and intensity level) to determine whether active heart rate is above a pre-defined threshold (indicating the presence of an overtraining condition that may lead to an injury condition).

[0043] Once the real-time micro-AI is triggered, a micro-AI analysis is performed while the athlete is training. If the result is that an overtraining condition exists while the athlete is training, an alert is provided to the athlete while the athlete is training. Such an alert may be provided through a fitness device 124 configured to provide one or more of an audible signal, a vibration indication, and / or a visual alert (e.g., a text message on a screen). The alert may be provided through use of a paired smartphone if the smartphone is within range.

[0044] 2, fitness monitoring device 124 preferably includes a heart rate sensor and on-board memory, the on-board memory including a latent memory component configured to hold latent data, the latent data including non-personal data not specifically associated with the user and personal data specific to the user. The heart rate sensor preferably includes a combination of light emitters and sensors for detecting pulse.

[0045] The latent memory is preferably only updatable when the tracker is located within an internet service area. One consequence of this is that the amount and type of data in the latent memory is finite compared to a cloud server, reducing the amount of data analyzed as part of any AI analysis as the AI ​​analysis is limited to only the data in the on-device memory.

[0046] The memory of device 124 further includes a current memory component configured to hold current data that can be updated while the user is training. Device 124 also includes a processor (e.g., a microprocessor) that includes a classifier. The processor is configured to determine the presence of an overtraining condition and make other decisions as needed based on a predictive model, utilizing only the data in the device's on-board memory, regardless of the presence of an internet service area. The processor is configured to provide a warning to the user after determining that an overtraining condition exists according to the output of the classifier. In a preferred embodiment, the processor is configured to compare the current data with potential data to determine the presence of an overtraining condition.

[0047] Having described the preferred components of system 100, Figure 3 illustrates a preferred method 200 for classifying an athlete's overtraining status. The method includes, in step 202, generating a latent feature set related to past training patterns and results, in step 204, periodically refreshing a wearable fitness device worn by the athlete with the latent features, in step 206, recording the athlete's current training data while the athlete is training, the current features including heart rate, to generate a current feature set, in step 206, automatically uploading the recorded training data to a server when the device is located within an internet service area to update the latent features stored on the server, in step 208, using an on-board processor of the wearable fitness device to perform statistical applications and one or more classifier models on the latent features and current features in device memory to generate a feature vector for predicting an overtraining status, in step 210, feeding the feature vector created based on the selected features to a neural network resident on the wearable fitness device, and obtaining a result from the neural network regarding whether an overtraining status exists, in step 214. The method may include refreshing the wearable fitness device only when the device is located within an internet service area. The method may include refreshing the latent data (e.g., athlete-specific historical data) at a remote server with current data that has expired and / or is not expired. The method may also include obtaining results from the neural network based solely on data stored on the wearable fitness device.

[0048] It will be appreciated that the steps described above may be performed in a different order, varied, or have steps added or omitted entirely without departing from the scope of the present disclosure. For example, a method may include monitoring an athlete's heart rate during training and, depending on the type of exercise, comparing the activity heart rate to past intensity levels. Preferably, if the heart rate exceeds a threshold, taking into account one or more of intensity, speed, and duration, an AI analysis is triggered while the athlete is training to determine whether the elevated heart rate indicates an overtraining condition and whether an impending injury is indicated. If the analysis determines that an overtraining condition exists, the method may include sending a warning to the athlete, preferably using the wearable fitness device. The warning may be communicated as an auditory message, a vibration indicator, and / or a visual indicator. The warning may instruct the athlete to reduce training intensity, slow down, walk, or stop to reduce injury risk. If the analysis determines that an overtraining condition does not exist, monitoring the athlete's heart rate continues.

[0049] 4 illustrates a system 300 having a local processor and storage configuration, preferably in the form of a micro AI module 302, such as might be present on device 124 (FIG. 2), and a remote processor and storage, preferably in the form of a remote cloud computing platform 304. The configuration and operation of micro AI module 302 has already been substantially described above in connection with system 100 (FIG. 1), but is repeated below in conjunction with a description of elements of the remote platform to enhance the overall system context in a broader environment.

[0050] The remote platform 304 includes a raw population data database 306 configured to store general population data, which may include historical data of the typical athlete mentioned above. The remote platform 304 also includes a personal history database 308 configured to store or record the athlete's own personality and experiences, such as, by way of example only, injury predisposition factors (e.g., how susceptible is the person to injury or how tolerant is he / she to strenuous exercise), family history / genetics, specific activity experience, and training goals (e.g., fitness level or upcoming competition).

[0051] The raw population data from database 306 and the individual historical data from database 308 are used to generate a latent dataset 330, a heart rate (HR) model 310, a latent model 316, and a classification model 320, which can be exported / transferred to local storage in a micro AI module carried by the athlete and used by a local microprocessor (described further below). The heart rate-related data contained in the raw population database 306 and the individual historical database 308 are used to build the HR model.

[0052] Data from the raw population database 306 is used to extract one or more features in analysis 312. Descriptive statistical analysis is then applied to create a descriptive feature set 314 based on the features extracted from the population data, such as heart rate range upon waking and heart rate range during exercise. The descriptive feature set 314 is used to form part of local latent data 330 (described further below). The features extracted in 312 can be used in a latent model 316 to generate an output set 326 (described further below).

[0053] The features extracted from the outputs 326 and 312 are utilized by a classification algorithm (such as NN, SVM, LR, BN and hybrid algorithms) 318. Individual historical data and population historical data (optional) from the database 308 can be subjected to time series analysis at 322, the output of which can be used in combination with the algorithm at 318 to build a final classification model 320. The time series analysis can be, for example, heart rate variability patterns over time and / or sweat intensity pattern changes over time.

[0054] Two or more data outputs from the remote HR model 310, latent model 316, and final classification models 320, 324, 328, which may be generated by the remote platform 304, are exported to local storage within the micro AI module 302. These models and data can only be updated when the device is within an internet-accessible area (e.g., a Wi-Fi area). When not within an internet area, the data in the latent database 330 does not change while the user is exercising or performing an exercise workout. The current data database 332 is configured to store or record current data when the device is in an active state, such as when the user is performing a workout, training, or exercising. When not exercising, the current data database 332 can actively store current data reflecting the user's resting state, such as wakeful heart rate (WHR) or sleep duration (as indicated by WHR over a measurement time interval). The current data database 332 dynamically stores data as it is being generated, whether the user is at rest or exercising, and retains that data for a predetermined time interval (e.g., two weeks), as previously described.

[0055] As shown in FIG. 4 , static data from the database of latent data 330 and dynamic data from the database of current data 332 are used in the local HR model 334 to generate an output at 336 indicating whether the user is overtraining and at risk of injury. If the local HR model 334 produces an output above a predefined threshold, a warning 338 is provided to the user or athlete and a message 340 appears on the screen of the user's local device. If the local HR model 334 produces an output below a predefined threshold, the data is analyzed by a locally stored classification model 342 that was exported from the remote platform 304. Part of the input for the local classification model 342 is the latent data imported / transferred from the remote platform 304. Model 337. Optionally, a short period of 5 or 10 minutes of current data from the current data database 332 can be used in a time series analysis at 344. The output of the time series analysis can also be used as input for a local classification model 342. The local classification model 342 generates an output to determine whether the risk of injury is high at 346. High risk can be determined by whether the heart rate is above a predefined threshold limit, combined with other features such as the amount of rest and recovery after a hard session or race, lack of sufficient sleep (from the calculated sleep score), etc. If the risk of injury is high, an injury or overtraining warning 348 is provided by, by way of example only, a message 350 on a local device, such as device 124 (FIG. 2) used by the user or athlete. The warning can be visual, audible, and / or vibrational. The visual message can be a text message on the device screen. If the risk of injury is not high, the output generated by the classification model at 342 is provided for incorporation as part of the individual's historical data in database 308 of remote processor and storage 304 if and when local processor and memory 302 is located within an internet service area.

[0056] Referring to Figure 4, the latent model preferably includes multiple classification or regression models, as shown in Figure 5. Model inputs include normalized features extracted from population data, such as each individual's heart rate recording, family history, genetic phenotype, exercise type, exercise intensity, etc.

[0057] Supervision of the training model is preferred to train one or more classification models, including those of the latent model and those for constructing the final classification model. The models (DA, LR, NN...) included in the latent model generate probability outputs (326 in FIG. 4) for use as inputs for the secondary classification model constructed by algorithm 318 in FIG. 4. Final Classification Model gives a categorical output that indicates whether or not you are at high risk of overtraining.

[0058] Example 1 Let's say Joe Smith is a 40-year-old runner who has been running for three years. He lives in the Gold Coast hinterland of Australia (a hilly area with a moderately warm climate). His father was a professional runner. His mother was an average rower. Joe's goal is to run the Gold Coast Marathon in three months.

[0059] Potential historical data For men aged 40-44, (general) historical patterns suggest that 40-50 km per week results in reasonable fitness (e.g., an average waking heart rate (WHR) of 45 beats per minute).

[0060] Joe's own (specific) historical data is based on his last three years: he averaged 40km per week and was in reasonably good health (average waking HR 50), but was injured once while running 80km per week.

[0061] Current Data Joe's training data is automatically stored in his fitness tracking device every time he exercises for up to two weeks. This data includes his heart rate (active and awake). One day, Joe logged a 15-kilometer long run to prepare for a marathon. This was the day after he ran 20 kilometers. He had only slept four hours. His HR was only 62 when he woke up.

[0062] Conclusion: Joe's waking HR is elevated compared to his individual "appropriate" HR and outside the "appropriate" HR for a 40-year-old male in the general athletic population. Another contributing factor is that he trained longer than normal for two consecutive days. Combining HR with the duration of the exercise suggests that Joe is overtraining. Joe reported muscle soreness (sharp pain) after a 15-kilometer run on Day 2. Joe has previously been injured when he overdid his workouts without adequate recovery between long or hard sessions. Based on potential historical data (general and specific) and current data (HR), the resulting conclusion from the micro-AI on Joe's device is that Joe is overtraining and needs to modify his training regimen. This conclusion is based on technical conclusions, which are explained further below. The sharp pain is a warning signal that he is at high risk of injury and should significantly modify his training by taking a day or two off.

[0063] Technical Conclusion: A conclusion of overtraining is derived from underlying historical parameters using a statistical application stored on the fitness tracker in combination with current data on the fitness tracker. Current data of HR, intensity, and training duration (time and / or distance) is utilized in combination with analysis comparing past individual trends and past trends of general population categories.

[0064] Output: Analysis indicating that training is on track towards goal (which can be presented visually as a curve) or that training needs to be modified due to overtraining and / or increased risk of injury. This can be communicated as an alert (verbal, vibration and / or visual) on the fitness tracker or via another connected device.

[0065] Example 2 Continuing with the example above, Joe decides to run 10 kilometers the day after he ran 15 kilometers. His WHR is 65, and he had 7 hours of sleep. Although tired, he decides to run at a comfortable pace. Ten minutes into his run, his active heart rate, as measured by the heart rate sensor on the fitness tracker he is wearing, reaches 180 beats per minute. The micro-AI component on Joe's fitness tracker analyzes Joe's current active heart rate and intensity level against potential specific historical data, and, given Joe's relaxed pace, determines that his heart rate is excessively elevated and activates a primary trigger. The processor on Joe's fitness tracker then sends audio, vibration, and visual alerts to Joe, warning him that he should discontinue further training or risk injury.

[0066] The system may be configured for use in activities other than athletics, such as running, cycling, swimming, or multi-sports. By way of example only, the system may be configured for use in forms of exercise such as kayaking, boat racing, hiking, yoga, weight training, core sessions, and other sports or athletics.

[0067] Other sensors can be used in combination with or in place of a heart rate sensor to help identify overtraining conditions and minimize injury risk. For example, one or more motion sensors can be used to determine proper running form, cycling form, or swimming stroke analysis to increase training efficiency and reduce injury risk. Information or data that is typically stored in a remote physical location can be stored in the cloud, thereby significantly reducing the on-site hardware required for memory requirements often associated with large volumes of data.

[0068] If the system is configured to monitor an athlete's health and provide an alert to the athlete when a biological state indicates the athlete is overtraining, the system can be configured to base the determination of overtraining on the athlete's active heart rate while the athlete is exercising and / or to base the determination on the athlete's resting heart rate when the athlete wakes up from sleep, and / or to combine heart rate and training duration. The alert provided by the system can be configured to remind the athlete to modify their training frequency and / or the alert can be configured to alert the athlete to modify their training duration.

[0069] If desired, elements of the micro AI module can be implanted (e.g., subcutaneously) to increase the device's durability against outdoor conditions.

[0070] Features described with respect to one embodiment may be applied to, combined with, or interchanged with features of other embodiments, as appropriate, without departing from the scope of the present disclosure.

[0071] The present disclosure, in one or more preferred forms, provides the advantages of being more responsive, more convenient, with reduced data transmission (saving communication resources), and usable even outside communication range (as may occur during training segments), since there is less bulk data stored for ongoing analysis. Less Memory is required.

[0072] Other embodiments of the present disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the embodiments disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

Claims

1. 1. A fitness tracker wearable about a portion of a user's body, comprising: A heart rate sensor; a latent memory component configured to hold latent data including non-personal data not specific to the user, the non-personal data including general past user data and personal data specific to the user, the latent memory component being updatable at a period defined as when the fitness tracker is located within an internet service area; and a current memory component configured to hold current data that can be updated independently of an internet connection while the user is exercising; a memory including: a microprocessor including a classifier configured to determine the existence of an overtraining condition based on an output of the classifier utilizing only data in the memory, regardless of presence within an internet service area, and configured to provide a warning to the user after determining that the overtraining condition exists according to the output of the classifier; and Including fitness trackers.

2. The fitness tracker of claim 1 , wherein the determination is based on the user's active heart rate while the user is exercising.

3. The fitness tracker of claim 1 , wherein the determination is based on a combination of active heart rate and at least one of training intensity level and training duration.

4. The fitness tracker of any one of claims 1 to 3, wherein the microprocessor is configured to compare the current data with the potential data to determine the existence of the overtraining condition.

5. The fitness tracker of any preceding claim, wherein the classifier is a neural network.

6. A fitness tracker according to any preceding claim, configured to be worn around the wrist of the user.

7. The fitness tracker of any one of claims 1 to 3, wherein the microprocessor is configured to determine the existence of an overtraining condition while the user is exercising.

8. 4. The fitness tracker of claim 1, wherein the microprocessor is configured to determine the existence of the overtraining state based on the output of the classifier while the user is exercising.

9. The fitness tracker of any one of claims 1 to 3, wherein the determination of the existence of the overtraining state is triggered by weighted trigger conditions.

10. 10. The fitness tracker of claim 9, wherein the primary trigger is heart rate and the secondary trigger is sleep duration.

Images