Activity level and fall detection using accelerometer data

Abstract

Activity level and fall detection using accelerometer data is described. In one or more embodiments, measurements of a user generated by a wearable monitoring device during an observation period are obtained, the measurements including accelerometer data. Actual steps taken by the user are detected based on peaks in the accelerometer data that are above a threshold, and a step count is generated based on the actual steps detected over a predetermined time period. Based on the step count, an activity level classification of the user over the predetermined time period is generated. The step count and activity level classification over the predetermined time period can then be output, e.g., as a notification or in a user interface. A fall prediction can be generated and output by processing the accelerometer data using a machine learning model trained to associate patterns in the accelerometer data with fall events.

Description

[0001] Related applications

[0002] This application claims priority to U.S. Provisional Application No. 63 / 703,820, filed October 4, 2024, entitled “Activity Level and Step Count Detection”; U.S. Provisional Application No. 63 / 703,783, filed October 4, 2024, entitled “Fall Detection Using a Chest Accelerometer”; U.S. Provisional Application No. 63 / 703,833, filed October 4, 2024, entitled “Constrained Sleep and Activity Prediction”; and U.S. Non-Provisional Application No. 19 / 347,425, filed October 1, 2025, entitled “Activity Level and Fall Detection Using Accelerometer Data”, the entire contents of which are incorporated herein by reference. Background Technology

[0003] In recent years, wearable activity trackers and medical monitoring devices have become increasingly prevalent, enabling users to track various aspects of their health and physical activity. These devices typically use accelerometers and other sensors to collect data on motion, heart rate, and / or other physiological parameters. However, many existing solutions rely on additional sensors such as gyroscopes, which can be power-intensive and costly. Furthermore, conventional methods often use accelerometers that sample at high frequencies (25-100 Hz) and may rely on multiple sensors to calculate body motion and position across a variety of possible body placements. These hardware configurations and sampling frequencies limit the ability to achieve longer battery life while maintaining accurate detection capabilities in wearable devices. Moreover, traditional sensor designs face challenges in distinguishing between different types of activities and events that may produce similar accelerometer characteristics, which can affect the reliability and accuracy of activity classification and event detection systems. Attached Figure Description

[0004] Figure 1 This is a block diagram of a non-limiting example of an environment operable to employ the activity level and fall detection using accelerometer data described herein.

[0005] Figure 2 Non-limiting examples of monitoring equipment are described.

[0006] Figure 3 A non-limiting system is depicted in an example implementation of activity level and fall detection using accelerometer data, which is shown in more detail. Figure 1 The operation of the prediction system.

[0007] Figure 4 An example depicting step detection using accelerometer data is shown.

[0008] Figure 5 An example of step counting during a structured treadmill activity is shown.

[0009] Figure 6 An example of fall detection using chest accelerometer data is shown.

[0010] Figure 7A and Figure 7B An example of a user interface configuration for displaying health monitoring data and clinical reports is shown.

[0011] Figure 8 A flowchart is shown that depicts the algorithm as a step-by-step process in an example implementation, which can be executed by a processing device to generate activity level and step count predictions based on accelerometer measurements.

[0012] Figure 9 A flowchart is provided that illustrates the algorithm as a step-by-step process in an example implementation, which can be executed by a processing device to generate fall predictions associated with heart rhythm classification.

[0013] Figure 10 A flowchart is provided that depicts the algorithm as a step-by-step process in an example implementation, which can be executed by a processing device to generate health status predictions. Detailed Implementation

[0014] Conventional activity tracking systems typically rely on high-frequency accelerometer sampling rates of 25–100 Hz and often incorporate additional sensors such as gyroscopes to calculate body motion and position across a variety of possible body placement locations. These conventional approaches face several limitations that affect their practical deployment and effectiveness. For example, the high sampling frequency and reliance on multiple sensors result in significant power consumption, limiting battery life and reducing the feasibility of extended monitoring periods. Furthermore, conventional systems struggle to accurately distinguish between different types of activities and events that produce similar accelerometer characteristics, leading to false positives and reduced reliability in activity classification and fall detection. The complexity of multi-sensor configurations also increases device cost and manufacturing complexity, while potentially compromising user comfort and compliance during extended wear periods.

[0015] Therefore, this paper describes techniques, methods, and systems for activity level and fall detection using accelerometer data. These techniques, methods, and systems address these limitations by providing activity level and fall detection using accelerometer data acquired at low sampling rates, while maintaining high accuracy and enabling comprehensive physiological analysis. In one or more embodiments, a wearable monitoring device acquires measurements including accelerometer data during an observation period, wherein the accelerometer data is acquired at a sampling rate of less than 5 Hz to extend battery life and reduce power consumption. Actual steps taken by the user are detected based on peak values ​​in the accelerometer data exceeding a predetermined threshold, thereby achieving accurate step counting even at lower sampling frequencies. Step counts are generated based on the actual steps detected within predetermined time periods, and activity level classifications are generated for these time periods based on the step counts within each corresponding time period. The activity level classification categorizes user activity as sedentary, mild, moderate, or vigorous, thereby providing insights into physical activity patterns over long monitoring periods. In one or more embodiments, temporal and frequency domain features are extracted from the accelerometer data to improve the accuracy of the activity level classification.

[0016] The system further processes accelerometer data using at least one machine learning model to generate fall predictions. This model is specifically trained to correlate patterns in the accelerometer data with fall events. The model is trained using historical accelerometer data and historical outcome data from a user group, enabling it to accurately distinguish genuine falls from other activities that may produce similar accelerometer characteristics. The system analyzes factors such as impact force, changes in body orientation, and post-fall movement patterns to minimize false positives while maintaining high sensitivity to actual fall events.

[0017] In implementations where the measurements also include electrical potential measurements of the user's heart, the system processes the potential measurements using one or more additional machine learning models trained to associate patterns in the potential measurements with the heart rhythm classification. The heart rhythm classification identifies specific types of arrhythmias, such as atrial fibrillation, bradycardia, ventricular arrhythmias, heart block, premature ventricular contractions, supraventricular tachycardia, or normal sinus rhythm. The system correlates fall predictions with contemporaneous heart rhythm classifications and outputs notifications when a fall event is temporally associated with a detected arrhythmia, thereby providing healthcare providers with a comprehensive analysis of the potential causal relationship between cardiac events and falls.

[0018] The described technique offers significant advantages over conventional methods, enabling accurate activity detection and fall prediction using accelerometer sensors that operate at low sampling frequencies to acquire measurements related to user movement and position. This extends battery life and reduces hardware complexity in wearable monitoring devices compared to conventional multi-sensor systems. For example, low-power operation and simplified hardware design allow wearable monitoring devices to be suitable for extended continuous monitoring periods (e.g., two weeks) without charging or battery replacement, while maintaining high accuracy in both activity classification and fall detection tasks. Furthermore, combining accelerometer-based fall detection with concurrent cardiac monitoring provides comprehensive physiological analysis results that are not available when analyzing individual sensor modalities separately. For instance, the correlation between fall events and cardiac arrhythmias allows healthcare providers to make informed clinical decisions without relying solely on patient recollection, particularly in cases where cardiac events may lead to falls. Additionally, the integration of activity level detection with fall detection enables the system to identify correlations between activity patterns and fall events and predict which activity intensities are likely to induce falls. Further discussion of these and other examples and advantages is included in the following sections, illustrated with corresponding figures.

[0019] In some aspects, the technology described herein relates to a method implemented by a processing device, the method comprising: acquiring measurements of a user generated by a wearable monitoring device during an observation period, the measurements including accelerometer data; detecting actual steps taken by the user based on peak values ​​in the accelerometer data that exceed a threshold; generating a step count based on the actual steps detected within a predetermined time period; generating an activity level classification of the user within the predetermined time period based on the step count corresponding to the predetermined time period; and outputting the step count and activity level classification within the predetermined time period.

[0020] In some respects, the techniques described herein relate to a method in which accelerometer data is acquired at a sampling rate of less than 5 Hz.

[0021] In some respects, the technology described herein relates to a method in which activity level classification includes categorizing a user’s activity during a predetermined time period into one of sedentary, light, moderate, or vigorous activities based on the number of steps taken within that time period.

[0022] In some respects, the techniques described herein relate to a method that also includes extracting time-domain and frequency-domain features from accelerometer data, wherein generating an activity level classification is additionally based on the extracted features.

[0023] In some respects, the techniques described herein relate to a method that also includes: processing accelerometer data using a machine learning model to generate a fall prediction, the machine learning model being trained to associate patterns in the accelerometer data with fall events; and outputting the fall prediction.

[0024] In some respects, the techniques described herein relate to a method that further includes: training a machine learning model for fall prediction using historical accelerometer data and historical outcome data from a user group as training data, wherein the historical accelerometer data is sampled at the same sampling rate as the accelerometer data acquired by the wearable monitoring device.

[0025] In some respects, the technology described herein relates to a method in which the measurements also include electrical potential measurements of a user's heart, and the method further includes: processing the electrical potential measurements using another machine learning model to generate a heart rhythm classification, the other machine learning model being trained to associate patterns in the electrical potential measurements with the heart rhythm classification.

[0026] In some respects, the technology described herein relates to a method in which heart rhythm classification includes an indication of arrhythmia, and the method further includes: outputting a notification in response to a fall prediction being temporally associated with arrhythmia.

[0027] In some respects, the techniques described herein relate to a method in which the heart rhythm classification is one of atrial fibrillation, bradycardia, ventricular arrhythmia, cardiac conduction block, premature ventricular contractions, supraventricular tachycardia, or normal sinus rhythm.

[0028] In some aspects, the technology described herein relates to a processing device comprising: one or more processors; and a memory storing computer-readable instructions executable by the one or more processors to perform operations including: acquiring measurements of a user generated by a wearable monitoring device during an observation period, the measurements including accelerometer data; detecting actual steps taken by the user based on peak values ​​above a threshold in the accelerometer data; generating a step count based on the actual steps detected within a predetermined time period; generating an activity level classification of the user within the predetermined time period based on the step count corresponding to the predetermined time period; and outputting at least one of the step count and activity level classification within the predetermined time period to a user interface.

[0029] In some respects, the technology described herein relates to a processing device in which accelerometer data is acquired by the accelerometer of a wearable monitoring device at a sampling rate of less than 5 Hz.

[0030] In some respects, the technology described herein relates to a processing device in which activity level classification includes categorizing a user’s activity during a predetermined time period as sedentary, mild, moderate, or vigorous based on the number of steps taken within that time period.

[0031] In some respects, the technology described herein relates to a processing device in which operation further includes processing accelerometer data using a machine learning model to generate fall predictions, the machine learning model being trained to associate patterns in the accelerometer data with fall events.

[0032] In some respects, the technology described herein relates to a processing device in which the measurements also include electrical potential measurements of a user's heart, and the operation further includes: processing the electrical potential measurements using another machine learning model to generate a heart rhythm classification, the other machine learning model being trained to associate patterns in the electrical potential measurements with the heart rhythm classification; and outputting the heart rhythm classification in a user interface.

[0033] In some respects, the technology described herein relates to a processing device in which the operation further includes: associating an activity level classification with one or both of a fall prediction and a heart rate classification; and outputting a health status prediction based on the association.

[0034] In some aspects, the technology described herein relates to a system comprising: a wearable monitoring device capable of being worn by a user to detect one or more measurements of the user during an observation period, the one or more measurements including accelerometer measurements and electrical potential measurements of the user's heart; and a computing device configured to: receive one or more measurements from the wearable monitoring device; generate an activity level classification of the user within a predetermined time period based on the accelerometer measurements corresponding to that time period; generate a fall prediction by processing the accelerometer measurements using a machine learning model trained to associate patterns in the accelerometer measurements with fall events; and output the activity level classification and the fall prediction.

[0035] In some respects, the technology described herein relates to a system in which accelerometer measurements are acquired at a sampling rate of less than 5 Hz.

[0036] In some respects, the technology described herein relates to a system in which a computing device is also configured to process potential measurements using another machine learning model to generate a heart rhythm classification, the other machine learning model being trained to associate patterns in the potential measurements with the heart rhythm classification.

[0037] In some respects, the technology described herein relates to a system in which a computing device is further configured to: correlate fall prediction with concurrent heart rate classification; and output notifications regarding fall prediction and concurrent heart rate classification.

[0038] In some respects, the technology described herein relates to a system in which, in order to generate an activity level classification, a computing device is further configured to: detect steps taken by a user based on peak values ​​exceeding a threshold in accelerometer measurements; generate a number of steps within a given predetermined time period based on the steps detected within that time period; and indicate a given activity level classification within the given predetermined time period as one of sedentary, mild, moderate, or vigorous, based on the number of steps within that time period.

[0039] Figure 1 This is a block diagram of a non-limiting example 100 of an environment operable to perform activity level and fall detection using accelerometer data as described herein. Example 100 includes a person 102 depicted wearing monitoring device 104. The environment also includes an analytics platform 106. Analytics platform 106 may be directly connected to monitoring device 104 via one or more wireless connections, or via one or more wired and / or wireless connections and one or more intermediate devices, such as computing devices, network routing devices and equipment, server equipment, and / or the Internet associated with person 102, to name a few.

[0040] Monitoring device 104 can be used to monitor one or more aspects of person 102, such as generating measurements 108. For example, in some cases, monitoring device 104 can be provided to record the electrical activity of person 102's heart over an observation period, such as several seconds or minutes, or several days. For example, person 102 may have the amplitude of the electrical potential of his or her heart monitored over time to generate one or more electrocardiograms (ECGs), which can be used to predict various events. In one or more embodiments, alternatively or additionally, monitoring device 104 can be provided to record accelerometer data over an observation period, such as acquiring motion and activity data for step counting, activity level classification, and fall detection. In one or more embodiments, accelerometer data can be sampled at a low sampling rate (e.g., in the range between 1 and 5 Hz, such as 1.6 Hz) to extend battery life while maintaining accurate detection capabilities. The monitoring device 104 can output measurement values ​​108 (e.g., time series of measurement values, such as potential measurements, acceleration measurements, and / or other types of physical and / or physiological measurements), which can indicate observations or be used to generate predictions for one or more events.

[0041] As used in this document for the purposes of this activity, the term "step" refers to body movement. A step can include a variety of types of body movement or action, including but not limited to a single foot landing during walking or running, heel strike followed by toe lift during a gait cycle, half of a full stride cycle involving movement of the left and right feet, climbing movements during going up or down stairs, stepping movements during lateral movements or changes of direction, stepping during stationary movements, dance steps or rhythmic movement patterns, stepping movements during balance recovery or posture adjustment, and / or any repetitive lower limb movements that produce a characteristic acceleration pattern detectable by an accelerometer sensor.

[0042] In conjunction with the monitoring device, instructions can be provided to person 102, instructing them on how to operate the monitoring device 104 and / or how to act while wearing the monitoring device 104 (e.g., sleep, engage in activities). In one or more embodiments, the instructions may be provided as part of a kit, such as as written instructions. Alternatively or additionally, the analysis platform 106 may transmit and output the instructions via a computing device associated with person 102 (e.g., for display and / or audio output). In one or more embodiments, the analysis platform 106 may provide the instructions for output only after the person 102 has worn the monitoring device 104 for a predetermined amount of time (e.g., two days) of the observation period and / or based on patterns observed in the person 102.

[0043] Monitoring device 104 can be configured in various ways to monitor one or more aspects of personnel 102. Furthermore, Figure 1 and Figure 2 The depicted external dimensions are merely an example, and the external dimensions of monitoring device 104 may vary in variations. It should be understood that monitoring device 104 may be configured with one or more sensors, examples of which include one or more of the following: multiple electrodes (e.g., capable of being placed on the skin of person 102), an accelerometer, and a pulse oximeter (e.g., to measure and record blood oxygen saturation (SpO2) and / or generate a photoplethysmogram of person 102), to name just a few. Of course, monitoring device 104 may be configured with various types of sensors without departing from the described technology.

[0044] While the monitoring device 104 may be configured in a manner similar to that used for clinical patient monitoring, in one or more embodiments, the monitoring device 104 may be configured differently from devices used for clinical patient monitoring and / or diagnosis. For example, and not limitingly, the monitoring device 104 may be configured as a ring, watch, patch, and / or band, to name just a few form factors. Alternatively or additionally, the monitoring device 104 may have a similar form factor to those used in a clinical setting but may have different functions, such as preventing the wearer from viewing the measurement value 108.

[0045] As used herein, the term “continuous” in conjunction with any signal associated with monitoring and person 102 (e.g., acceleration data and / or electrical activity of the heart of person 102) can refer to the ability of a device to generate measurements substantially continuously, such that the device can be configured to generate measurements 108 at time intervals (e.g., every hour, every 30 minutes, every 5 minutes, every minute, every 30 seconds, every second, every half second, etc.), in response to events (e.g., an inflection point in an electrical signal such as a peak or trough). Without departing from the spirit or scope of the described technology, the functionality of the monitoring device 104 in generating measurements 108 along with other measurements and / or recording various signals can vary.

[0046] In one or more embodiments, the monitoring device 104 may be configured to unload the measurement value 108 during the observation period. For example, the monitoring device 104 may unload the measurement value 108 by transmitting it to an external computing device via a wired or wireless connection, for example, at predetermined time intervals and / or in response to establishing or re-establishing a connection with the computing device. In one or more embodiments, the measurement value 108 and / or other data from the monitoring device 104 may be compressed by the monitoring device 104 for wireless transmission, for example, using one or more of a variety of data compression techniques. Compressing sensor data in this way can reduce the battery consumption of the monitoring device 104 during the observation period and facilitate wearing the monitoring device during activities such as assessing activities, arrhythmias, and fall detection. For example, the monitoring device 104 may lightly process the measurement value 108 during wear, for example, by averaging or encoding at least a portion of the measurement value 108, to reduce the amount of data stored and / or transmitted while retaining information for post-wear analysis. In some such embodiments, more comprehensive data processing may be performed after the wear period to conserve battery power and reduce memory storage usage.

[0047] In one or more embodiments, the monitoring device 104 may also implement other power management strategies, wherein data from one type of sensor is used to selectively trigger measurements captured by other sensors. As an illustrative example, additionally or alternatively, accelerometer data indicating a potential fall event may trigger enhanced cardiac monitoring data during and immediately after the fall event, thereby enabling analysis of whether arrhythmias such as atrial fibrillation, cardiac conduction block, or ventricular arrhythmias occurred before or after the fall event.

[0048] In cases where monitoring device 104 can be configured to store measurement values ​​108 over the entire observation period, in one or more embodiments, monitoring device 104 can be configured without wireless transmission means; for example, without any antenna for wirelessly transmitting measurement values ​​108, and without hardware or firmware for generating data packets for such wireless transmission. Instead, monitoring device 104 can be configured with hardware for transmitting measurement values ​​108 via physical wired coupling. In such cases, monitoring device 104 can be "plugged in" to retrieve measurement values ​​108 from the device's storage.

[0049] Therefore, monitoring device 104 may be configured with one or more ports to transmit measurement values ​​108 via a wired connection to an external computing device. Examples of such physical coupling may include a micro universal serial bus (USB) connection, a mini USB connection, and a USB-C connection, to name just a few. Although monitoring device 104 may be configured to retrieve measurement values ​​108 via a wired connection as discussed above, alternatively or additionally, in different situations, monitoring device 104 may be configured to offload measurement values ​​108 via one or more wireless connections.

[0050] Once the monitoring device 104 generates a measurement value 108, it provides the measurement value 108 to the analysis platform 106. As described above, the measurement value 108 can be transmitted to the analysis platform 106 via a wired and / or wireless connection.

[0051] For example, where the analysis platform 106 is partially or fully implemented on the monitoring device 104, the measured value 108 can be transferred from the device's local storage to the device's processing system via a bus. Where the monitoring device 104 is configured to generate one or more predictions 110 by processing the measured value 108, the monitoring device 104 can also be configured to provide the generated one or more predictions 110 as output, for example, by transmitting the one or more predictions 110 to an external computing device. In other cases, the measured value 108 can be processed by an external computing device configured to generate one or more predictions 110. For example, the measured value 108 can be processed by a smartphone associated with a user, a smartphone or other dedicated device associated with the monitoring device 104, and / or one or more server computers located in a data center or other location accessible to entities associated with the monitoring device 104, to name just a few. In other words, these other devices can implement at least a portion of the analysis platform 106 and / or the prediction system 114.

[0052] In one or more embodiments, the monitoring device 104 is configured to transmit measurement values ​​108 to an external device via a wired connection (e.g., via USB-C or some other physical communication coupling). Here, a connector can be inserted into the monitoring device 104, or the monitoring device 104 can be inserted into a device having a socket having corresponding contacts that engage with the monitoring device 104. Measurement values ​​108 can be retrieved from the storage device of the monitoring device 104 via this wired connection, for example, by transmitting the measurement values ​​to the external device via the wired connection. Such a connection can be used when personnel 102 mail the monitoring device 104 to, for example, a healthcare provider, a telemedicine provider, the provider of the monitoring device 104, or a medical testing laboratory after the observation period.

[0053] Alternatively or additionally, monitoring device 104 may provide measurement value 108 to analysis platform 106 by transmitting the measurement value 108 via one or more wireless connections. For example, monitoring device 104 may wirelessly transmit measurement value 108 to external computing devices, such as mobile phones, tablets, laptops, smartwatches, other wearable health trackers, etc. Therefore, monitoring device 104 may be configured to communicate with external devices using one or more wireless communication protocols or technologies. For example, monitoring device 104 may use one or more of Bluetooth (e.g., Bluetooth Low Energy link), Near Field Communication (NFC), Long Term Evolution (LTE) standards such as 5G to communicate with external devices. In the case of transmitting measurement value 108 to external devices for processing, monitoring device 104 may be configured with corresponding antennas and other wireless transmitting devices. In these situations, the measurement value 108 can be transmitted to the analysis platform 106 in various ways, such as at predetermined time intervals (e.g., daily, hourly, or every five minutes), in response to the occurrence of an event (e.g., the storage buffer of the monitoring device 104 fills up), or in response to the end of the observation period, to name just a few.

[0054] Therefore, regardless of where the analysis platform 106 is implemented (e.g., at monitoring device 104, at a smartphone associated with person 102, or at a server device), the analysis platform 106 acquires the measurements 108 generated by monitoring device 104. In one or more embodiments, the analysis platform 106 also acquires additional measurements generated by monitoring device 104 and / or any other device used during the observation period (e.g., a smartwatch, chest strap, etc.). As mentioned above, examples of such additional measurements include, but are not limited to, blood oxygen saturation (SpO2) measurements. In one or more embodiments, the combination of accelerometer data and cardiac measurements can enable enhanced fall detection capabilities, including the ability to associate fall events with contemporaneous arrhythmias (e.g., arrhythmias that may increase fall risk or are triggered by fall events and / or specific activity levels, such as atrial fibrillation, ventricular arrhythmias, or bradycardia episodes). In one or more embodiments, the analysis performed by the analysis platform 106 may include correlation analysis between fall events detected by the accelerometer and concurrent cardiac arrhythmias to identify patterns where specific types of arrhythmias may induce user falls or fall events may trigger arrhythmias. Additionally or alternatively, the analysis platform 106 may correlate exercise intensity with cardiac events to identify which activity intensities may induce arrhythmias or other physiological responses. Activity intensity can provide context for detected falls, supporting analytical results that cannot be obtained through separate analyses, such as determining which exercise intensities might lead to falls in specific users or patient groups.

[0055] In one or more embodiments, the analysis platform 106 may be implemented wholly or partially at the monitoring device 104. Alternatively or additionally, the analysis platform 106 may be implemented wholly or partially using one or more computing devices external to the monitoring device 104, such as one or more computing devices associated with person 102 (e.g., mobile phone, tablet, laptop, desktop computer, or smartwatch), or one or more computing devices associated with a service provider (e.g., healthcare provider, telemedicine provider, service provider corresponding to the provider of monitoring device 104, medical testing laboratory service provider, etc.). In the latter case, the analysis platform 106 may be implemented at least partially on one or more server devices.

[0056] In the illustrated example 100, the analytics platform includes a storage device 112. According to the technology, the storage device 112 can be configured to store measurement values ​​108 and / or other measurement values ​​or information processed by the prediction system 114 to generate one or more predictions 110. The storage device 112 can represent one or more databases and / or other types of storage devices capable of storing measurement values ​​108 and / or other types of measurement values. The storage device 112 can also store various other data, such as personal information, demographic information describing the person 102, information about healthcare providers, information about insurance providers, payment information, prescription information, determined health indicators, account information (e.g., username and password), etc. The storage device 112 can also store data of other users within the user group.

[0057] In one or more embodiments, storage device 112 may also store activity thresholds, step detection parameters, fall detection algorithms, activity classification algorithms, historical activity patterns, baseline accelerometer data for comparison purposes, body angle calculation parameters, and a multimodal data fusion algorithm that determines how to combine accelerometer data with other physiological measurements such as ECG or SpO2 data to improve prediction accuracy. One or more algorithms may be machine learning algorithms or may include machine learning algorithms. For example, at least one machine learning algorithm may be trained to identify the temporal relationship between a cardiac event and a subsequent fall. In one or more embodiments, storage device 112 may also store arrhythmia classification model, a fall-arrhythmia correlation database, cardiac event time points and activity time points relative to the fall occurrence, and / or algorithms for identifying causal relationships between specific arrhythmia types and fall risk factors.

[0058] In the illustrated example 100, the analysis platform 106 also includes a prediction system 114. The prediction system 114 represents the functionality to process the measurements 108 to generate one or more predictions 110. Alternatively or additionally, the prediction system 114 may output one or more time series indicating observations or predictions of one or more events over time. It should also be understood that, in variations, the prediction system 114 may output different combinations of multiple predictions.

[0059] In at least one embodiment, prediction system 114 uses machine learning to generate at least a portion of one or more predictions 110. For example, and not limitingly, prediction system 114 may include one or more neural networks trained on historical measurements and historical outcome data of a user group. Prediction system 114 may include one or more machine learning models (e.g., an ensemble of models). Alternatively or additionally, prediction system 114 may include logic for preprocessing measurements 108 (machine learning models and / or other types of logic), such as extracting various cardiovascular features, motion features, and / or other features from the measurement sequence. One or more predictions 110 may, for example, be the output of prediction system 114.

[0060] In various examples, prediction system 114 may represent and / or may include activity detection system 116 and fall detection system 118, and one or more predictions 110 may include and / or may represent step count 120, activity level classification 122, fall prediction 124, heart rate classification 126, and / or health status prediction 128. For example, as described in more detail below, activity detection system 116 may operate to implement accelerometer-based techniques to generate step count 120 and activity level classification 122. Activity detection system 116 may receive accelerometer data and detect peaks above a threshold to identify steps as part of step count 120, count steps over a predetermined time period, and classify activity intensity based on the step count. For example, activity level classification 122 may categorize activity level as sedentary, mild, moderate, or vigorous based on the step count over the predetermined time period and features extracted from accelerometer data. Activity detection system 116 may dynamically adjust the threshold to improve step detection accuracy and may incorporate temporal and frequency domain features to enhance activity classification.

[0061] The fall detection system 118 can operate to analyze accelerometer data patterns to detect fall events and correlate fall events with cardiac data to generate a fall prediction 124, which can indicate the number and timing of falls, including their potential correlation with arrhythmias. In one or more embodiments, the fall detection system 118 can utilize accelerometer data of measurement 108 to identify fall events and correlate fall events with cardiac data of measurement 108, thereby providing context for arrhythmia diagnosis and patient care. The fall detection system 118 can employ one or more algorithms (e.g., machine learning algorithms and / or conventionally programmed algorithms) to analyze accelerometer data, distinguishing genuine falls from other activities by considering factors such as impact force, changes in body orientation, and post-fall movement patterns. The fall detection system 118 can also employ one or more algorithms (e.g., machine learning algorithms and / or conventionally programmed algorithms) to analyze cardiac data of measurement 108 (e.g., ECG waveform data) to classify heart rhythms, resulting in a heart rhythm classification 126.

[0062] For example, the fall detection system 118 can be configured to identify a specific type of arrhythmia preceding an event. For instance, rhythm classification 126 can indicate a bradycardia episode that may cause dizziness, an atrial fibrillation episode that may affect cardiac output and balance, and / or a ventricular arrhythmia that may cause sudden weakness or syncope leading to a fall. The fall detection system 118 can also analyze whether the fall event itself triggered subsequent arrhythmias due to physical stress or emotional responses associated with the fall, and / or analyze the correlation between activity levels (e.g., activity levels detected by the activity monitoring system 116) and the fall event. The fall detection system 118 can also correlate exercise intensity with the fall event and / or arrhythmia to determine which activity intensities are most likely to induce a fall, thereby providing analytical results regarding activity-related fall risk patterns that cannot be obtained through separate analyses of fall detection and activity monitoring. By combining fall detection with simultaneous cardiac data analysis, the fall detection system 118 can reveal the relationship between falls and cardiac events, thus providing healthcare providers with a comprehensive tool to assess patient health and fall risk.

[0063] Health Prediction 128 provides comprehensive physiological analysis by combining exercise analysis with cardiac context, enabling the detection of relationships between activity patterns, cardiac health measures, and overall health. Health Prediction 128 can include correlations between daily activity levels and cardiac health indicators that suggest patterns of cardiovascular fitness, exercise tolerance, or potential health risks. For example, Health Prediction 128 can indicate the relationship between steps and heart rate variability, the correlation between activity intensity and arrhythmia frequency, or the association between exercise patterns and overall cardiovascular health. Health Prediction 128 can also provide analysis of how different activity levels affect cardiac function over time, including information on exercise capacity, recovery patterns, and the impact of physical activity on cardiac health. By combining accelerometer-derived activity data with contemporaneous cardiac measurements, Health Prediction 128 can identify optimal activity levels for individual users, indicate early signs of declining cardiovascular function, and / or provide personalized recommendations for maintaining or improving overall health.

[0064] Health Prediction 128 can also include the relationship between fall events and overall health patterns, the correlation between fall frequency and cardiovascular health, the association between fall risk and activity levels, and how fall data can be combined with activity and cardiac measurements to provide healthcare providers with comprehensive health analytics results. Health Prediction 128 can also include a comprehensive health score that combines multiple physiological parameters, enabling healthcare providers to assess patient health holistically, rather than through isolated measurements. This multimodal approach can provide clinical benefits by supporting preventative care strategies and enabling early intervention when certain patterns are detected in the combined activity and cardiac data.

[0065] Example 100 also illustrates accessory device 130 and healthcare provider 132. For example, accessory device 130 may include one or more devices associated with person 102 and / or monitoring device 104, such as those described above. For example, accessory device 130 may include display devices (e.g., smartphones and / or personal display devices) to display one or more predictions 110 and / or the functionality of controlling monitoring device 104. Accessory device 130 may also display notifications 134, as will be further detailed below.

[0066] For example, healthcare provider 132 may represent one or more additional processing devices associated with an authorized healthcare system, such as healthcare personnel devices, electronic health record systems, diagnostic imaging equipment, laboratory information systems, telemedicine platforms, clinical decision support tools, etc. In various embodiments, notification 134 may be generated based on one or more predictions 110, such as alerts for detected fall events, changes in activity levels, changes in health status predictions 128, and / or other detected events, and notification 134 may be displayed by accessory device 130 to deliver information to personnel 102, healthcare providers, caregivers, and / or emergency contacts. As a non-limiting example, prediction system 114 may generate reports and / or alerts based on fall predictions 124, which can lead to more targeted interventions and improved fall prevention strategies.

[0067] In various examples, one or more operations of the analysis platform 106, the prediction system 114, the activity detection system 116 and / or the fall detection system 118 can be performed by one or more of the monitoring device 104, the accessory device 130, the device and system of the healthcare provider 132 and / or one or more additional devices not shown.

[0068] Figure 2 A non-limiting example 200 of a monitoring device is depicted. The example 200 depicts a monitoring device 104.

[0069] According to the described technology, the monitoring device 104 includes one or more sensors 202, examples of which include, but are not limited to, a pair or more pairs of electrodes, an accelerometer, a pulse oximeter, and a sweat sensor, to name just a few. The monitoring device 104 may also include a transmitter 204. In this example 200, the monitoring device 104 also includes one or more adhesive portions 206. In operation, the monitoring device 104 is configured to be attached to the skin via one or more adhesive portions 206, such that, for example, one or more sensors 202 are positioned to detect and record the electrical activity of the heart of the person 102, for example, generating an electrocardiogram (ECG and / or EKG). In at least one embodiment, the monitoring device 104 can be removed by peeling one or more adhesive portions 206 off the skin.

[0070] It should be understood that the monitoring device 104 and its components are only one form factor, and the monitoring device 104 and its components may have different form factors without departing from the spirit or scope of the technology.

[0071] In one or more embodiments, monitoring device 104 may include a processor and / or memory (not shown). By utilizing the processor, monitoring device 104 can generate measurement values ​​108 based on communication with one or more sensors 202 that indicate an aspect of person 102 (e.g., the electrical activity of person 102's heart). In one or more embodiments, the processor also generates one or more communicable data packets including measurement value 108 and / or one or more other measurements such as blood oxygen saturation (SpO2) measurements. Alternatively or additionally, the processor generates and / or stores other data that can be used to predict the classification of physiological conditions (e.g., arrhythmias, activity levels, etc.).

[0072] In embodiments where monitoring device 104 is configured for wireless transmission, transmitter 204 can wirelessly transmit measurement value 108 as a data stream to a computing device. For example, in one or more embodiments, monitoring device 104 is configured to transmit (e.g., transmit and / or receive) information (e.g., potential measurements) via a Bluetooth Low Energy (BLE) connection. Alternatively or additionally, monitoring device 104 can buffer measurement value 108 (e.g., buffer it in memory) and have transmitter 204 later transmit the buffered measurement value at various intervals (e.g., time intervals (per second, every thirty seconds, per minute, every five minutes, per hour, etc.), storage intervals (when the buffered measurement value reaches a threshold data amount), etc.).

[0073] Activity level and fall detection using accelerometer data

[0074] Figure 3 A non-limiting system for activity level and fall detection using accelerometer data is described in example implementation 300, which is shown in more detail. Figure 1 The operation of the prediction system 114.

[0075] In this example, the prediction system 114 first receives sensor data 302, which may include electrical data 304 (e.g., potential measurements and / or ECG data), accelerometer data 306, SpO2 data 308 (e.g., blood oxygen saturation data), and / or various supplementary data 310. In various examples, sensor data 302 is acquired by one or more devices and / or sensors (e.g., wearable monitoring device 104). Sensor data 302 may include time-ordered data instances, such as continuous data, data acquired at predetermined intervals (e.g., every half-second, every minute, every five-minute interval, etc.) over the duration of an observation period, such as a day, multiple days of a week, a month, etc. For example, sensor data 302 may be a combination of... Figure 1The measurement value 108 is described. In one or more embodiments, accelerometer data 306 is acquired at a sampling rate of less than 5 Hz (e.g., 1.6 Hz). Typically, sensor data 302 is processed by prediction system 114 to generate one or more predictions 110 according to techniques described in more detail below, which may include one or more of the following: step count 120, activity level classification 122, fall prediction 124, heart rate classification 126, and health status prediction 128.

[0076] For example, prediction system 114 includes a training module 312 operable to train a machine learning model 316 using training data 314 to perform one or more activity detection and fall detection tasks. Activity detection and fall detection tasks may involve, for example, generating one or more predictions 110, such as predicting steps, activity levels, and fall events based on patterns in accelerometer data 306; associating activity patterns with cardiac data from electrical data 304 to obtain a health status prediction 128; monitoring body position based on electrical data 304 and monitoring cardiac data based on accelerometer data 306 for fall recovery analysis; and detecting fall events based on patterns in accelerometer data 306, which may be associated with arrhythmias based on electrical data 304. These tasks may include: classifying activity levels as sedentary, mild, moderate, or vigorous; counting steps over a predetermined time period; detecting fall events; and associating falls with cardiac data (e.g., based on electrical data 304). In one or more embodiments, these tasks may further include identifying cardiac arrhythmias that may occur before or after a fall event based on electrical data 304, thereby enabling analysis of a causal relationship between a specific type of cardiac arrhythmia and the occurrence of a fall. Therefore, machine learning model 316 is trained to associate patterns in sensor data 302 (e.g., various potential measurements from electrical data 304 and / or accelerometer measurements and body angle calculations from accelerometer data 306) with one or more predictions 110. It should be understood that more than one machine learning model 316 may be trained separately, for example, separately for one or more different predictions 110.

[0077] The foregoing examples describe various instances of artificial intelligence (“AI”) models and / or machine learning models. In one or more examples, an AI model (e.g., a machine learning model) refers to a computer representation capable of being adjusted (e.g., through training and retraining) based on inputs without being actively programmed by the user to automatically and without user intervention approximate an unknown function. For example, the term “machine learning model” includes models that utilize algorithms to learn from and make predictions about known data by analyzing training data for learning and relearning, thereby generating outputs that reflect the patterns and properties of the training data. In a context using accelerometer data for activity level and fall detection, a machine learning model can achieve (e.g., by one or more processing devices of prediction system 114) analyzing accelerometer data patterns and correlating these patterns with cardiac data to identify activity levels, detect fall events, and generate comprehensive physiological analysis results.

[0078] The training module 312 is also operable to initialize various parameters of the machine learning model 316, which can be used as internal variables to represent and process information during training. These parameters can also be used to represent inferences acquired through training. In one or more embodiments, the training data 314 is divided into multiple batches to improve the efficiency of parameter processing and optimization of the machine learning model 316 during training, which is particularly beneficial for model accuracy when processing large amounts of accelerometer time-series data acquired at low sampling rates and correlating motion patterns with concurrent cardiac measurements.

[0079] In this example, training data 314 may include historical electrical potential measurements (e.g., ECG) and accelerometer measurements from the user group, along with corresponding historical outcome data such as arrhythmia classification, activity classification, step count, fall events, and / or overall health status classification. This data may be collected from clinical studies, activity monitoring studies, or other sources that simultaneously record electrical data 304 and accelerometer data 306. In some cases, training data 314 may also include additional physiological measurements such as blood oxygen saturation levels (e.g., SpO2 data 308) and / or additional relevant biomarkers. Training data 314 can be labeled with various activity and fall-related information, such as the presence of activity periods, activity types (e.g., walking, running, sleeping), step count, fall events, body positions, etc.

[0080] Training data 314 can also be labeled with various heart rate-related information. That is, for training a machine learning model for activity level and fall detection, training data 314 can provide examples of what the machine learning model "will learn," such as serving as a basis for learning patterns from data. For activity and fall detection applications, training data 314 can include labeled datasets of accelerometer measurements and cardiac data from users with known activity patterns, fall events, and heart conditions, as well as measurements from individuals with normal activity levels and no history of falls. For example, training module 312 can acquire and preprocess training data 314, which includes input features (e.g., accelerometer waveforms, body angle calculations, gait patterns, ECG data, heart rate patterns) and corresponding target labels (e.g., "sedentary activity," "vigorous activity," "fall event detected," "fall associated with arrhythmia," or specific activity and fall classifications). Training data 314 can also be labeled with physiological features such as normal sinus rhythm, atrial fibrillation episodes, ventricular arrhythmias, bradycardia events, cardiac conduction block, premature ventricular contractions, supraventricular tachycardia, and / or other arrhythmias that may be temporally associated with a fall event or changes in activity level.

[0081] Training data 314 can be received as input by machine learning model 316 and used as the basis for generating predictions based on the current state of the model's layers and corresponding nodes. The results are output as output data, such as activity level classification, step count, fall event detection, and the correlation between falls and cardiac events.

[0082] For example, a machine learning model can be configured using multiple layers, each with multiple nodes. These multiple layers can be configured to include an input layer, an output layer, and one or more hidden layers. In the context of activity level and fall detection, the input layer can receive sensor data 302 or features thereof, including amplitude calculations, body angle measurements, step detection peaks, movement patterns, activity level indices, and cardiac features. The hidden layer processes these inputs, for example, through weighted connections, to identify complex patterns indicative of activity levels, fall events, and correlations between movement and cardiac status, patterns that would not be detected using conventional analysis modalities relying solely on basic activity tracking. The output layer can produce one or more predictions 110, including step count 120, activity level classification 122, fall prediction 124, and / or health status prediction 128, as well as correlations between fall events and cardiac arrhythmias. Nodes within a layer perform computations via hidden states through a weighted connection system “learned” during the training of the machine learning model 316 to achieve various activity detection and fall detection tasks.

[0083] In some implementations, different training schemes and / or model architectures are employed based on what one or more predictions 110 are to include. For example, the composition and structure of training data 314 can vary depending on the specific type of prediction to be generated. In an example where one or more predictions 110 are to indicate a binary classification of activity presence (e.g., whether the user is active or inactive), training data 314 can be labeled with a yes / no metric. In an additional or alternative example where one or more predictions 110 are to include fine-grained predictions such as specific activity levels, step counts, and / or fall severity, training data 314 includes detailed annotations related to the fine-grained predictions. In an example where one or more predictions 110 are to indicate the correlation between activity patterns from accelerometer data 306 and cardiac health measurements from electrical data 304, training data 314 can include health status analysis results associated with various activity levels and cardiac measurements. In one or more of the predictions 110, training data 314 may include multimodal input data to capture the complex relationships between movement, cardiac status, and physiological outcomes, such as using postural and cardiac data for fall recovery monitoring, cardiac-activity health status correlation, and comprehensive analysis of fall event contextualization by combining arrhythmia data from electrical data 304.

[0084] In some examples, training data 314 is constructed to support multi-task learning, where machine learning model 316 can simultaneously predict multiple aspects of activity level and / or fall detection, such as a combination of activity level and fall risk, as well as activity-cardiac correlations and fall detection with physiological context, such as health analysis results based on activity-heart rate relationships and fall recovery patterns based on a combination of postural-cardiac data. In additional or alternative examples, training module 312 trains machine learning model 316 task-by-task, for example, implementing a first round of training to train machine learning model 316 for activity-cardiac correlation analysis and a second round of training to train machine learning model 316 for fall detection using cardiac contextualization. In this way, the techniques described herein support task-specific targeted training of machine learning model 316, which improves the performance and efficiency of the model in performing individual aspects of activity and fall detection, as well as the performance and efficiency of the model in performing complex multimodal physiological analysis results that combine accelerometer data 306 with electrical data 304, and not just basic activity classification.

[0085] In one or more implementations, training module 312 trains machine learning model 316 using an iterative process that adjusts weights and learning parameters to minimize the loss function. For example, training module 312 may use backpropagation and / or gradient descent algorithms to update the model's parameters based on the differences between predicted activity and fall classifications in training data 314 and actual activity and fall classifications. The learning rate, batch size, and / or number of rounds may be adjusted to optimize the performance of machine learning model 316.

[0086] Training the machine learning model 316 may include calculating a loss function to quantify the loss associated with the operations performed by the nodes of the machine learning model 316. The loss function can be configured in various ways to control the operation and / or functionality of the machine learning model 316. For example, the loss function may be designed to prioritize the accuracy of fall event detection while minimizing false positives and optimizing the accuracy of the correlation between activity patterns and cardiac health metrics. Calculating the loss function may include, for example, comparing the differences between predictions specified in the output data (e.g., predicted activity levels, fall events, or cardiac-activity correlations) and target labels specified in the training data 314 (e.g., clinically confirmed activity classifications and recorded fall events). Examples of the various configuration options for the loss function include quadratic loss functions as part of a least-squares technique for continuous activity parameters, cross-entropy loss for classification tasks such as activity level classification, custom loss functions incorporating clinical risk factors specific to fall detection and cardiac-activity correlations, and so on.

[0087] Furthermore, various architectures / types of the machine learning model 316 are considered. In one or more implementations, the machine learning model 316 may include neural networks, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), or combinations thereof. In some instances, the machine learning model 316 incorporates one or more U-Net and / or ResNet architectures, features, or components. The model can also be implemented as an ensemble of different algorithms that combine one or more decision trees, random forests, and / or gradient boosters with neural network methods. For example, CNNs can be used to analyze accelerometer waveform patterns and detect fall features. As another example, long short-term memory (LSTM) neural networks can be used to analyze temporal motion patterns and correlate activity sequences with cardiac events. In other examples, generative adversarial networks (GANs), decision trees (e.g., for activity level classification and fall risk assessment), support vector machines, linear regression, logistic regression for binary fall detection, Bayesian networks, random forest learning for feature importance in the correlation between accelerometer and cardiac data, dimensionality reduction algorithms, boosting algorithms, deep learning neural networks, etc., can be employed. It should be understood that the above examples are given by way of illustration, and other configurations may be used without departing from the spirit or scope of the technology described.

[0088] In some examples, training module 312 may employ techniques such as dropout or regularization to prevent overfitting. The training process can continue until the model reaches the desired level of accuracy on the validation dataset and / or until a predetermined number of iterations are completed. This method enables machine learning model 316 to learn complex patterns in sensor data 302, indicating relationships between activity patterns from accelerometer data 306 and cardiac health measures from electrical data 304, correlations between postural changes and heart rate responses during fall recovery, temporal correlations between arrhythmias and fall events, and multimodal analysis results that combine motion and cardiac data to identify subtle features and form physiological understandings that cannot be obtained using conventional analysis methods. For example, manually analyzing sensor data 302 to obtain such analysis results may be impractical or impossible. This is merely an example and not a limitation, and various suitable training techniques are considered.

[0089] For example, multiple attention heads can allow the machine learning model 316 to allocate resources to focus on different aspects of the input data, thereby making different predictions. Continuing with the example above where prediction 110 indicates activity-cardiac correlation, each analysis head can be trained to detect specific relationships between motion patterns from accelerometer data 306 and physiological responses from electrical data 304. This enables the prediction system 114 to provide a comprehensive multimodal analysis of sensor data 302 for health status analysis results and fall recovery monitoring. In this way, the techniques described herein support the adaptability of the prediction system 114 to efficiently provide targeted activity-cardiac related information and physiologically contextualized fall detection as a complement or alternative to step count 120, activity level prediction 122, and fall prediction 124.

[0090] Once the machine learning model 316 is trained, the sensor data 302 is processed by the feature extraction module 318, which generates data features 320. For example, the feature extraction module 318 preprocesses the sensor data 302 to generate usable (e.g., processable) input for the trained machine learning model 326. The feature extraction module 318 may generate at least a portion of the data features 320 based on various properties of the accelerometer signal present in the accelerometer data 306, such as amplitude calculations, body angle measurements, step detection peaks, motion patterns, and / or activity level indices. These extracted features may include time-domain measurements, frequency-domain measurements, and statistical measurements that capture relevant information about motion and body position. It should be understood that, at least in some embodiments, the training data 314 may include accelerometer data acquired at the same sampling frequency as the accelerometer data 306.

[0091] The feature extraction module 318 is also operable to perform body angle calculations as part of the output data features 320. In some implementations, body angle techniques can be used to extract positional information from the accelerometer data 306 to determine changes in body position. The feature extraction module 318 can analyze changes in accelerometer orientation, such as changes in the gravity vector component, which can be correlated with and / or influenced by body position and movement. These positional properties of the accelerometer data 306 can be used to derive body angle measurements and activity patterns. Features derived from body angles can also be combined with other data features 320 to provide a comprehensive set of inputs to the trained machine learning model 326. In this way, the prediction system 114 is able to simultaneously capture motion and positional information based on a single accelerometer signal, which improves the detection of correlations between body position and activity events—something that cannot be achieved using conventional methods.

[0092] In one or more embodiments, the feature extraction module 318 may also extract cardiac features from the electrical data 304, such as heart rate variability, QRS complex morphology, RR interval pattern, and arrhythmia features that may be associated with fall risk or changes in activity level.

[0093] The feature extraction module 318 can also implement various additional techniques, such as wavelet decomposition, principal component analysis, peak detection, statistical analysis, and / or other signal processing methods, to isolate and quantify relevant aspects of the electrical data 304 and / or accelerometer data 306. For example, a step detection algorithm can be applied to identify peaks above a predetermined threshold, while amplitude calculation can provide information about the overall activity level. The feature extraction module 318 can incorporate both time-domain features (e.g., mean, variance, and peak characteristics) and frequency-domain features (e.g., spectral energy and dominant frequency) to enhance activity classification. The feature extraction module 318 can also optimize and / or refine the data features 320, for example, based on the discriminative ability of the data features 320 to detect specific activities or fall events.

[0094] Once data features 320 are extracted, analysis module 322 configures them for input into a trained machine learning model 326 (e.g., machine learning model 316 output by training module 312). In one or more embodiments, analysis module 322 uses encoder 324. Encoder 324 can be configured to process input data and compress it into a compact representation, and may include one or more of convolutional encoders, recurrent encoders, transformer encoders, one or more autoencoder variants, etc. For example, encoder 324 generates a compressed representation based on data features 320 that can be efficiently processed by trained machine learning model 326. In the example, encoder 324 reduces the dimensionality of data features 320 while preserving relevant information, thereby creating a compact representation suitable for use as input to trained machine learning model 326.

[0095] The analysis module 322 processes the encoded data features 320 using a trained machine learning model 326 to generate one or more predictions 110 for output. Prediction 110 may include various information. For example, step count 120 may indicate the number of steps taken, activity level classification 122 may indicate different levels of physical activity, fall prediction 124 may indicate the detection of a fall event related to cardiac activity and / or physical activity, and health status prediction 128 may provide a comprehensive physiological analysis result combining motion analysis with the cardiac context and / or additional predictions providing supplementary analysis results. In one or more examples, one or more predictions 110 may include confidence intervals, such as confidence levels for associated values ​​or conditions.

[0096] In at least one implementation, step count 120 may include the total number of steps detected within a predetermined time period during the observation period. Step count 120 may be generated by identifying individual steps by detecting peaks above a threshold in accelerometer data 306, counting steps detected within a predetermined time period (e.g., per minute, per hour, or per day), and providing accurate step counts while operating at a lower sampling frequency compared to conventional systems. Activity detection system 116 may dynamically adjust the threshold based on user-specific patterns or environmental conditions to improve step detection accuracy. Activity level classification 122 may distinguish activity levels including, but not limited to, sedentary, light activity, moderate activity, vigorous activity, or combinations thereof. Activity level classification 122 may incorporate features such as time-domain and frequency-domain characteristics extracted from accelerometer data 306 to classify activity intensity based on the number of steps within the predetermined time period. As a non-limiting example, exercise equivalent to a walking speed of 2 miles per hour (mph) or higher may indicate light activity, while speeds below 2 mph may indicate inactivity / rest. Fall prediction 124 can, for example, indicate whether a user associated with sensor data 302 has never fallen, the risk of falling in the future (e.g., low risk of falling, moderate risk of falling, high risk of falling, etc.), the timing of the fall, and any correlation with arrhythmia during the detected fall, to name just a few.

[0097] For example, fall prediction 124 can include indications of the correlation between a fall event and additional physiological events such as cardiac events, activity events, etc., for example, when a trained machine learning model 326 is capable of identifying and analyzing the relationship between fall occurrence and various levels of cardiac or physical activity. For example, fall prediction 124 can indicate that, for a particular user, a fall event is more likely to occur during periods of increased activity. Combining fall prediction 124 with heart rhythm classification 126 can indicate the temporal correlation between a fall event and a specific arrhythmia, such as an episode of atrial fibrillation, a bradycardia event that may cause dizziness or syncope, or a ventricular arrhythmia that may cause sudden weakness leading to a fall. Such analysis results can provide a comprehensive view of the physiological response to a fall event and the causes of the fall event, enabling healthcare providers to better understand whether the fall is primarily due to mechanical causes or may be induced by an underlying cardiac condition. Furthermore, fall prediction 124 and heart rhythm classification 126 can provide clinicians with detailed information to make informed decisions without relying solely on patient recollection.

[0098] Furthermore, accelerometer data 306 can be used to detect postural changes during daily activities, which can influence the occurrence and / or severity of fall events. By incorporating such location information, a trained machine learning model 326 can generate an accurate fall prediction 124 that takes into account the relationship between posture and fall risk. For example, the trained machine learning model 326 can process accelerometer data 306 to generate a fall prediction 124 that indicates the correlation between user behavior (e.g., posture during activity) and fall events. In various embodiments, the trained machine learning model 326 can also analyze electrical data 304 to generate a heart rhythm classification 126 that identifies arrhythmias common during specific postures or activity levels, thereby enabling fall risk prediction based on both movement patterns and contemporaneous cardiac status.

[0099] In various examples, sensor data 302 also includes SpO2 data 308 and / or various additional data 310. These additional measurements can be fed together with electrical data 304 and / or accelerometer data 306 into a machine learning model 316 to predict one or more predictions 110. Thus, the techniques described herein support multimodal prediction, which provides analytical results that conventional techniques relying solely on accelerometer data 306 or solely on electrical data 304 cannot obtain.

[0100] In some implementations, SpO2 data 308 can be used to improve the accuracy of activity and fall predictions and / or validate these predictions. For example, analysis module 322 can process SpO2 data 308 in conjunction with electrical data 304 and / or accelerometer data 306 to identify potential activity periods or fall events. A trained machine learning model 326 can be configured to detect changes in blood oxygen saturation levels during activity periods (which may coincide with increased physical exertion) and correlate these changes with motion patterns in the accelerometer signal. Additionally or alternatively, the trained machine learning model 326 can be configured to detect changes in blood oxygen saturation levels and correlate these changes with arrhythmias identified based on electrical data 304. By combining these data sources, prediction system 114 is able to operate to distinguish different types of activity and fall events in a more accurate manner.

[0101] For example, different types of activity events can be characterized by different accelerometer patterns; walking shows regular periodic peaks, while running shows higher frequency and amplitude variations. Similarly, heart rhythm classification 126 can identify different types of arrhythmias that may be associated with different fall patterns; for example, falls associated with bradycardia are characterized by a gradual onset, while falls associated with ventricular arrhythmias are characterized by a sudden collapse pattern. This multimodal approach enables nuanced and accurate activity and fall prediction, which reduces false positives and provides additional context for detected activity levels and fall risk, including the ability to distinguish between falls caused by mechanical factors and falls that may be induced by underlying heart disease.

[0102] The following discussion describes techniques that can be implemented using the systems and devices described above. References will be made to these techniques in the following sections of the discussion. Figures 1 to 3 .

[0103] Figure 4 Example 400 depicting step detection using accelerometer data is shown. Example 400 includes an acceleration amplitude 402 plotted over time, where the acceleration amplitude 402 represents the total amplitude of the acceleration force detected by the accelerometer (e.g., sensor 202) of monitoring device 104. In one or more embodiments, the acceleration amplitude 402 may be calculated as the vector sum of acceleration components in three orthogonal directions (x, y, and z axes). The acceleration amplitude 402 may, for example, be a processed signal freed from the effects of gravity. The acceleration amplitude 402 depicts periodic variations corresponding to the periodicity of human walking patterns, with distinct peaks appearing at regular intervals corresponding to individual step events.

[0104] Example 400 includes a threshold 404 for identifying individual step events based on acceleration amplitude 402. In one or more embodiments, the prediction system 114 may dynamically adjust the threshold 404 based on observed changes in acceleration amplitude 402 to improve step counting accuracy across different walking speeds, terrains, or user movement patterns. A dynamic threshold can allow the prediction system 114 to adapt to individual user characteristics and / or environmental conditions that may affect the amplitude of the acceleration signal during walking. However, in at least one variant, the threshold 404 is a static threshold.

[0105] Multiple step peaks 406 can be observed in the acceleration amplitude 402, where each step peak 406 corresponds to a local maximum value of the acceleration amplitude 402 exceeding a threshold 404, for example, due to foot impact or body movement during a walking cycle. Step peaks 406 correspond to step events detected by the activity detection system 116. The activity detection system 116 can generate a step count 120 by counting the number of peaks exceeding the threshold 404 within a predetermined time period. The analysis platform 106 can apply additional filtering or verification algorithms to the detected step peaks 406 in at least one or more embodiments to eliminate false positives caused by non-walking activity or sensor noise.

[0106] Figure 5 An example 500 of steps count during structured treadmill activity is shown. Example 500 includes a step count graph 502 containing step measurements over time (e.g., in elapsed seconds), providing a visual representation of how the step count 120 can be organized and analyzed for activity classification purposes. Step count graph 502 includes multiple vertical dashed lines representing stage boundaries 504, each stage boundary 504 representing a predetermined time interval during which the treadmill maintains a predetermined speed and incline. For example, example 500 could be a cardiac stress test following the Bruce protocol.

[0107] Step count graph 502 shows the number of steps 120 accumulated within measurement intervals. Measurement intervals can have a length of seconds (e.g., 5 seconds, 10 seconds, 30 seconds), a length of minutes (e.g., one minute, five minutes), or another predetermined duration based on the accelerometer sampling rate that contributes to the analysis of step count 120. In one or more embodiments, measurement intervals can be, or can be further grouped, into time periods for activity level (e.g., intensity) classification, allowing the analysis platform 106 to process step count data in discrete segments rather than a continuous stream. In one or more embodiments, the prediction system 114 can analyze the step count 120 within each time period to generate an activity level classification 122, thereby classifying physical activity into different intensity levels. For example, when the step count 120 within a time period is below a first minimum activity threshold, the activity detection system 116 can classify the activity as sedentary, indicating minimal physical movement. When the step count 120 within a time period exceeds the first activity threshold but remains below a second activity threshold, the activity detection system 116 can classify the activity as light activity, indicating light movement, such as a slow walk or basic daily activity. When the number of steps (120) within a time period is greater than or equal to the second activity threshold but less than the third maximum activity threshold, the activity detection system 116 can classify the activity as moderate activity, indicating activities such as brisk walking or light exercise. When the number of steps (120) within a time period is greater than or equal to the third activity threshold, the activity detection system 116 can indicate vigorous activity, indicating high-intensity activity such as running or high-intensity physical exercise.

[0108] In one or more embodiments, for example, in combination Figure 3 As described, the feature extraction module 318 can extract both time-domain and frequency-domain features from the accelerometer data 306 to improve the accuracy of activity intensity classification. For example, time-domain features may include statistical measurements such as the mean, variance, and standard deviation of acceleration amplitude within each time interval, providing information about movement patterns and intensity variations. Frequency-domain features can be derived through spectral analysis of the accelerometer data 306, which can provide periodic patterns and frequency characteristics corresponding to different types of physical activity. The analysis module 322 can combine the step count 120 with these extracted features to generate a more accurate activity level classification 122 that considers both the quantity and quality characteristics of movement within each time interval.

[0109] Figure 6Example 600 of fall detection using chest accelerometer data is shown, illustrating how a fall detection system 118 can combine multiple measurements from monitoring device 104 to identify a potential fall event. Example 600 includes a voltage graph 602 that displays potential measurements captured over an elapsed time period, providing measurements of cardiac activity acquired by monitoring device 104 during the monitoring period. A magnified voltage graph 604 presents a detailed view of voltage changes within a specific time window corresponding to a potential cardiac arrest event, indicated as a suspected arrest region 606. The suspected arrest region 606 corresponds to an area where cardiac electrical activity appears to be interrupted or significantly altered. In one or more embodiments, the suspected arrest region 606 may be indicated via a heart rhythm classification 126 output by prediction system 114.

[0110] However, due to noise at the end of a cardiac arrest, clinicians may find it difficult to distinguish whether it is a true cardiac arrest, a cardiac arrest that caused the fall, or the duration of the arrest (if it is a true cardiac arrest). Detecting the additional background of the fall can inform clinicians without relying on the patient's memory.

[0111] like Figure 6 As further shown, acceleration amplitude plot 608 displays changes in acceleration measurements during a time period overlapping with the suspected cardiac arrest region 606, and in this example, this time period begins before the suspected cardiac arrest region 606. Acceleration amplitude plot 608 enables a direct temporal correlation between cardiac events and movement patterns. Acceleration amplitude plot 608 can capture various types of movement, including normal daily activities, sudden movements, and potential fall events.

[0112] Example 600 also includes a body angle graph 610, which tracks the orientation of the monitoring device 104 relative to gravity measured by an accelerometer, providing information about the position and posture of the person 102 during the overlapping time period. The body angle graph 610 can display angle measurements ranging from approximately 0 to 90 degrees, where lower values ​​may indicate a more upright posture and higher values ​​may indicate a more backward or horizontal posture.

[0113] A fall event 612 can be identified based on the acceleration amplitude diagram 608 alone or in combination with the body angle diagram 610. For example, the acceleration amplitude diagram 608 can indicate a sudden free fall followed by an impact and then no movement, and the body angle diagram 610 can indicate a corresponding sudden change in body orientation, such as a rapid change from a more upright posture to a more horizontal posture.

[0114] The fall detection system 118 can simultaneously analyze multiple parameters to distinguish genuine fall events from other activities that may produce similar accelerometer characteristics. Impact force analysis may involve examining the amplitude and rate of change of acceleration measurements to identify patterns consistent with a fall, as well as other activities such as rapidly sitting down or intentionally lying down. Changes in body orientation can be assessed by monitoring the body angle graph 610 for sudden shifts in both amplitude and rate of change exceeding predetermined thresholds. Post-fall movement patterns can be assessed by analyzing accelerometer data following a suspected fall event to determine whether the person 102 maintains a horizontal posture for an extended period or exhibits movement patterns consistent with recovery from a fall.

[0115] The correlation between the fall event 612 and the suspected cardiac arrest area 606 can provide clinical context for healthcare providers. For example, Example 600 illustrates the temporal correlation between the suspected cardiac arrest area 606 in voltage graph 602 and the fall event 612. In this example, the fall event 612 occurs during a period of cardiac arrhythmia identified in the suspected cardiac arrest area 606. This correlation can help clinicians determine that the suspected cardiac arrest area 606 corresponds to a true cardiac arrest, without relying on patient memory, because the fall was caused by suspected cardiac arrest (e.g., dizziness, lightheadedness, and / or syncope during cardiac arrest).

[0116] In one or more embodiments, when the fall detection system 118 processes at least a portion of the measurements 108 during an observation period and detects a fall event 612 accompanied by a cardiac arrhythmia such as a suspected pause area 606, the prediction system 114 can generate an automatic alarm. Figure 1 The notification 134 depicted is an example. Notification 134 may include detailed information about the timing and characteristics of both the fall event 612 and the associated cardiac arrhythmia, and may be sent, for example, to person 102, healthcare provider 132, and / or person 102's emergency contact. Notification 134 can provide an alert for potential medical events that may require medical attention or follow-up evaluation.

[0117] In at least one implementation, notification 134 may not be a real-time notification. For example, there may be a delay (e.g., seconds, minutes, hours, days) between acquiring measurement 108 and analyzing measurement 108 by fall detection system 118 to detect fall event 612. However, in at least one variation, measurement 108 may be analyzed by fall detection system 118 at least partially at the time of acquisition. Therefore, notification 134 may indicate a past / recent fall event 612 during the observation period (e.g., when notification 134 is delayed), or it may indicate a current fall event 612 (e.g., when notification 134 is real-time).

[0118] Figure 7Aand Figure 7B An example 700 of a user interface 702 configuration for displaying health monitoring data and clinical reports is shown. For example, example 700 illustrates one implementation of how activity level and fall detection data collected by monitoring device 104 during an observation period can be displayed after analysis by analysis platform 106. User interface 702 provides a platform for healthcare providers and / or users (e.g., personnel 102) to view activity patterns, fall events, and associated cardiac data. User interface 702 can be implemented on various computing devices, including smartphones, tablets, desktop computers, or dedicated medical workstations.

[0119] Figure 7A A user interface 702 is depicted configured to display activity-related health data via an activity report section 704, which reports the activities of person 102 detected during an observation period. In this example, when the "Activity" tab is selected, the activity report section 704 displays activity data for the time period from July 16 to July 30. For example, receiving input to select the "Activity" tab at the user interface 702 causes the prediction system 114 to display the activity report section 704.

[0120] The activity report section 704 includes multiple graphical representations of activity data that provide visual correlations between different activity intensity levels detected by accelerometer data 306 and processed by the activity detection system 116, including a mild activity graph 706, a moderate activity graph 708, and a vigorous activity graph 710. In this example, each activity graph displays a daily measurement of the corresponding physical activity level within a monitoring period, indicating how much of the corresponding activity was performed each day. While example 700 shows activity levels associated with the time spent at each activity level, variations are possible. For example, steps, distance traveled, or another type of measurement may be used as a supplement or alternative to time. Furthermore, activity levels may be categorized differently from those shown. The mild activity graph 706, moderate activity graph 708, and vigorous activity graph 710 can provide a visual representation of activity patterns and / or physical activity levels over time, and in at least some examples, can be further correlated with other analyses.

[0121] For example, the activity graph in the activity report section 704 can enable healthcare providers to identify periods of increased or decreased activity that may be associated with a health event or changes in the condition of person 102. For instance, a significant decrease in moderate or vigorous activity levels may indicate a decline in health, while a consistent activity pattern may suggest a stable condition. (As shown in the example, combining...) Figure 3As described, the prediction system 114 can process accelerometer data 306 to generate activity classifications using a machine learning model trained on labeled activity data, thereby achieving accurate differentiation between different activity intensities.

[0122] Figure 7B A user interface 702 is shown configured to display fall detection and cardiac event correlation data via a patient event reporting section 712. For example, receiving input selecting the "Patient Events" tab causes the prediction system 114 to display the patient event reporting section 712. In this configuration, the patient event reporting section 712 presents event-based information for the same observation period as the activity reporting section 704. The patient event reporting section 712 allows the user to analyze the relationship between fall events and cardiac arrhythmias, which can be further correlated with the activity level of the activity reporting section 704. This correlation capability addresses clinical situations where fall events may be associated with cardiac events, providing context for diagnostic and treatment decisions.

[0123] The patient event reporting section 712 includes a fall event graph 714 displaying the occurrence (e.g., count) and time of occurrence (e.g., by day in this example) of detected fall events, and a cardiac event graph 716 displaying cardiac event data for the same period. In at least one embodiment, in response to a user selecting a specific date, specific event, etc., the user interface 702 may display more granular information to present the temporal correlation between the occurrence of a fall and cardiac arrhythmias (e.g., arrhythmias, arrests, or other cardiac arrhythmias detected by analyzing electrical data 304 acquired by monitoring device 104).

[0124] For example, if a user selects a date (e.g., "07 / 25"), the user interface 702 can output the same data as the date displayed on the screen. Figure 6 A similar visual presentation is shown in Example 600, demonstrating the visual correlation between fall prediction 124 and heart rhythm classification 126. Therefore, by utilizing both accelerometer and potential data to detect both fall events and cardiac events, prediction system 114 can provide enhanced analytical results regarding the complex interaction between movement patterns and cardiovascular activity during daily activities.

[0125] Such analytical results cannot be obtained using conventional techniques that rely on separate analyses of fall detection, activity detection, and cardiac monitoring. In some examples, this further conserves computational resources that would otherwise be used to process input data types from multiple different monitoring devices and / or analysis platforms. Therefore, by generating multimodal analysis results based on correlated accelerometer and cardiac data, the technique described herein improves the operation of the device implementing the predictive system 114.

[0126] In one or more embodiments, user interface 702 can generate a comprehensive report that includes a timeline of detected fall events, associated cardiac data for each fall event, and fall prevention recommendations based on the activity graph of activity report section 704. This report can combine information from mild activity graph 706, moderate activity graph 708, vigorous activity graph 710, fall event graph 714, and cardiac event graph 716 to provide the user with a detailed analysis of the patient's health. In one or more embodiments, as part of the comprehensive report, user interface 702 can display steps 120, heart rate classification 126, and / or health status prediction 128. Therefore, user interface 702 can present a multifaceted view of activity, fall, and cardiac data, along with other physiological observations. This comprehensive presentation can help interpret complex movement and cardiac data, potentially leading to earlier detection of fall risk factors and more effective management of related health problems.

[0127] The following sections describe example processes for activity level and fall detection using accelerometer data in one or more embodiments. Aspects of each process can be implemented in hardware, firmware, software, or a combination thereof. These processes are shown as a set of blocks specifying operations that can be performed by one or more devices, and are not necessarily limited to the order in which these operations are performed by the respective blocks shown. For example, one or more blocks of the process specify operations that can be programmed into instructions by hardware (e.g., a processor, microprocessor, controller, firmware) to create a dedicated machine for executing the algorithm illustrated in the flowchart. Therefore, these instructions can be stored on a computer-readable storage medium that enables the hardware to execute the algorithm. References will be made in the following discussion sections. Figures 1 to 6 , Figure 7A and Figure 7B .

[0128] Figure 8 A flowchart is shown, depicting the algorithm as a step-by-step process 800 in an example implementation. This algorithm can be executed by a processing device to generate activity level and step count predictions based on accelerometer measurements. Process 800 can be executed by the prediction system 114 or other components of the analysis platform 106 to analyze movement patterns and classify the physical activity level of the person 102.

[0129] In this example, firstly, measurements of the user generated by the wearable monitoring device during the observation period are acquired (box 802). For example, measurement 108 may include accelerometer data 306, which may be sampled at a rate less than 5 Hz (e.g., 1.6 Hz). In various examples, monitoring device 104 uses one or more sensors 202 (e.g., accelerometers) to detect changes in motion and orientation of person 102 and generates accelerometer data 306 based on the detected motion. Monitoring device 104 may process the accelerometer data 306 locally, at least partially, before sending it to analysis platform 106. Alternatively, analysis platform 106 may be included as part of monitoring device 104. In yet another example, monitoring device 104 may send the raw accelerometer data 306 to analysis platform 106 for processing. Additionally or alternatively, monitoring device 104 may use various data compression techniques to compress sensor data to reduce battery consumption during transmission (e.g., wireless or wired transmission) to an external computing device.

[0130] The actual steps taken by the user are detected based on peak values ​​in the accelerometer data that exceed a threshold (box 804). For example, the activity detection system 116 analyzes the accelerometer data 306 to identify characteristic patterns corresponding to individual steps taken by the person 102. The step detection process may include examining the acceleration amplitude 402 and comparing the measurement to a threshold 404 to identify step peak values ​​406 that exceed a predetermined threshold level. The threshold 404 may be dynamically adjusted based on changes in the accelerometer data 306 to accommodate different walking speeds, terrain, and / or individual movement patterns of the person 102. Alternatively, the threshold 404 may be a fixed value.

[0131] The step count is generated based on the actual steps detected within a predetermined time period (box 806). The activity detection system 116 can organize the detected step peaks 406 into discrete time periods or segments to facilitate step counting. For example, each time period can represent a specific duration, such as one minute, five minutes, one hour, one day, or another suitable time interval, during which the total number of steps detected is calculated and recorded. The step count 120 can be stored in the storage device 112 for subsequent analysis and correlation with other physiological measurements, and / or output for display, for example via the accessory device 130.

[0132] Extracting time-domain and frequency-domain features from accelerometer data (box 808). For example, feature extraction module 318 processes accelerometer data 306 to identify various features that may indicate different types of physical activity. Time-domain features may include statistical measurements such as the mean, variance, and standard deviation of the acceleration signal, while frequency-domain features may include spectral analysis to identify dominant frequencies and patterns in the motion data. These data features 320 can provide additional information beyond the step count 120, which can be used to distinguish different activity intensities and types of movement.

[0133] Based on the number of steps and the extracted features, an activity level classification (box 810) is generated for a predetermined time period. For example, a trained machine learning model 326 receives steps 120 and data features 320 as input to classify the activity level of person 102 within each time period. The activity level classification 122 can classify physical activity into discrete intensity levels, such as sedentary, mild, moderate, or vigorous, based on a combination of steps and feature analysis. The machine learning model 316 can be trained using historical accelerometer data and corresponding activity classifications to accurately distinguish between different intensity levels of physical activity. In at least one variant, the number of steps 120 within a time period is compared to one or more thresholds to classify the activity level and output the activity level classification 122.

[0134] Output the number of steps and activity level classification for a predetermined time period (box 812). For example, the prediction system 114 may display the number of steps 120 and / or activity level classification 122 via a user interface 702, or the prediction may be incorporated into another type of health report for review by healthcare provider 132 and / or personnel 102. The output may include detailed charts and visualizations showing activity patterns over time, such as displaying daily activity minutes as a light activity chart 706, a moderate activity chart 708, and a vigorous activity chart 710. In various examples, a notification 134 or another type of alert related to the activity level classification 122 may be sent to a computing device associated with personnel 102 and / or healthcare provider 132 to facilitate the monitoring and assessment of physical activity levels, alone or in combination with other physiological measurements acquired by monitoring device 104.

[0135] Figure 9 A flowchart is depicted illustrating the algorithm as a step-by-step process 900 in the example implementation, which can be executed by a processing device to generate fall predictions associated with heart rhythm classification. Process 900 integrates accelerometer data 306 and electrical data 304 acquired by monitoring device 104 from person 102 during the observation period to provide a comprehensive analysis of the fall event and its relationship to cardiac activity.

[0136] In this example, firstly, measurements of the user generated by the wearable monitoring device during the observation period are acquired, including accelerometer measurements and electrical potential measurements (box 902). For example, measurement 108 is generated by monitoring device 104 during continuous wear by person 102. In various examples, monitoring device 104 detects motion patterns based on accelerometer data 306 while simultaneously capturing cardiac electrical activity using one or more sensors 202. Accelerometer data 306 can be sampled at a lower frequency than conventional activity trackers typically operate at 25-100 Hz, thereby extending the battery life of monitoring device 104 and conserving its memory resources. Electrical data 304 (e.g., electrical potential measurements) can be acquired substantially continuously to provide heart rate information that can be correlated with detected motion patterns.

[0137] To generate fall predictions, machine learning models are used to process accelerometer measurements. These models are trained to associate patterns in the accelerometer data with fall events (box 904). For example, machine learning model 316 is trained using historical accelerometer measurements from a user group and labeled fall event data to perform a fall detection task. The fall detection task may include analyzing impact force, changes in body orientation, and post-fall movement patterns to distinguish genuine falls from other activities with similar accelerometer characteristics. For example, trained machine learning model 326 receives accelerometer data 306 as input and generates a fall prediction 124 based on identified patterns indicative of fall events. Machine learning model 316 may analyze features such as sudden changes in acceleration amplitude, the duration of the impact event, and subsequent movement patterns to differentiate falls from activities such as sitting down or intentionally lying down. Machine learning model 316 may also consider location information determined based on accelerometer data 306.

[0138] Another machine learning model is used to process the potential measurements, which is trained to associate patterns in the potential measurements with heart rhythm classifications (box 906). For example, a trained machine learning model 326 for cardiac analysis is trained using historical potential measurements and corresponding heart rhythm labels to identify various arrhythmias and cardiac events. In some cases, the trained machine learning model 326 processes electrical data 304 to generate a heart rhythm classification 126, which may include identifying conditions such as atrial fibrillation, cardiac arrest, or other arrhythmias. The trained machine learning model 326 for cardiac analysis can examine features such as heart rate variability, heart rhythm patterns, and electrical signal morphology to classify different types of cardiac events that may occur before, during, or after a fall accident.

[0139] The fall prediction is correlated with a concurrent heart rhythm classification (box 908). For example, the fall detection system 118 performs a temporal analysis to identify a relationship between a fall prediction 124 occurring within a specified time window and a heart rhythm classification 126. In various examples, the correlation process can identify cardiac events occurring seconds or minutes before the fall, suggesting a causal relationship where a heart rhythm abnormality may increase the risk of a fall. The fall detection system 118 can examine the time point of the fall prediction 124 relative to the heart rhythm classification 126 to determine whether a cardiac event, such as cardiac arrest or arrhythmia, occurred concurrently with or immediately before the detected fall. This correlation analysis can provide healthcare providers with clinically relevant insights into the underlying causes of fall events and the potential cardiac-related fall risk.

[0140] Output a notification regarding fall prediction and concurrent cardiac rhythm classification (box 910). For example, prediction system 114 prompts the generation of notification 134, which includes information about both the detected fall event and any associated cardiac rhythm abnormalities. Notification 134 can be sent to healthcare provider 132 or displayed via a user interface associated with analytics platform 106 (e.g., on accessory device 130). In various examples, when a fall event is detected along with a cardiac rhythm abnormality, notification 134 may include detailed time-point information, severity assessment, and recommendations for further evaluation. Notification 134 can also be incorporated into a comprehensive report to provide healthcare providers with context for clinical decision-making without relying on the patient's memory of the event, particularly in cases where cardiac arrest or other abnormalities could have led to the fall.

[0141] Figure 10 A flowchart is depicted, illustrating the algorithm as a step-by-step process 1000 in an example implementation, which can be executed by a processing device to generate health status predictions. Process 1000 represents a comprehensive approach to health status analysis that combines multiple physiological measurements and predictions to generate a comprehensive health assessment. In some cases, process 1000 may be implemented by a prediction system 114 to correlate various outputs of one or more predictions 110 and provide a healthcare provider with a holistic view of the patient's health status.

[0142] Obtain activity level classification, fall prediction, and heart rhythm classification (box 1002). For example, activity level classification 122, fall prediction 124, and heart rhythm classification 126 can be combined... Figure 8 and Figure 9 The process described is used to generate the predictions. In various implementations, one or more predictions 110, including activity level classification 122, fall prediction 124, and heart rhythm classification 126, are generated during or after the observation period of the monitoring device 104.

[0143] Activity level classification is correlated with one or both of fall prediction and heart rhythm classification (box 1004). For example, analytics platform 106 can perform temporal analysis to identify relationships between different physiological events (e.g., heart rhythm abnormalities and / or fall events) and activity states. For instance, analytics platform 106 can determine whether fall events occur more frequently during a particular activity level or occur concurrently with a particular heart rhythm abnormality. The correlation can also examine patterns where heart rhythm abnormalities precede changes in activity level or fall accidents. In one or more embodiments, analytics platform 106 can identify sedentary periods that coincide with the onset of heart rhythm abnormalities or periods of vigorous activity that trigger a specific cardiac response.

[0144] Based on the correlation, a health status prediction is output (box 1006). For example, a health status prediction 128 may represent a comprehensive assessment that integrates multiple health indicators to provide actionable analytical results for healthcare providers 132 and / or personnel 102. For instance, a health status prediction 128 may indicate an increased risk of falls during periods of arrhythmia, or it may recommend adjustments to activity levels based on the observed correlation between exercise intensity and arrhythmia events. The health status prediction 128 may be formatted as part of a comprehensive health report that includes recommendations for lifestyle modifications, medical interventions, or monitoring adjustments, and may be output, for example, via a user interface 702 and / or via an accessory device 130.

[0145] The analysis platform 106 can employ advanced signal processing techniques to synchronize data from multiple sensors and ensure accurate temporal correlation of events. The machine learning model within the prediction system 114 (e.g., a trained machine learning model 326) can be trained on multimodal datasets including concurrent accelerometer measurements, cardiac measurements, and other physiological measurements to improve the accuracy of health status predictions. The storage device 112 can maintain historical data across multiple observation periods to enable longitudinal health status trend analysis and personalized health analysis results. In various examples, health status prediction 128 can incorporate demographic factors, medical history, and environmental conditions to provide context-relevant health assessments, supporting clinical decision-making and optimization of patient care.

[0146] The integrated approach represented by process 1000 enables analysis platform 106 to provide multimodal health analysis results that cannot be achieved by analyzing a single physiological parameter alone. Monitoring device 104 can be configured with various combinations of one or more sensors 202 (e.g., accelerometer sensors, electrocardiogram electrodes, pulse oximeters, etc.) to support this integrated analysis approach, thereby capturing different physiological signals.

[0147] The various functional units shown in the accompanying drawings and / or described herein are implemented in any of a variety of different ways, such as hardware circuitry, software or firmware executing on a programmable processor, or any combination of two or more of hardware, software, and firmware. The provided methods are implemented in any of a variety of devices, such as a general-purpose computer, a processor, or a processor core. Suitable processors include, for example, general-purpose processors, special-purpose processors, conventional processors, digital signal processors (DSPs), graphics processing units (GPUs), parallel accelerator processors, multiple microprocessors, one or more microprocessors associated with a DSP core, controllers, microcontrollers, application-specific integrated circuits (ASICs), field-programmable gate array (FPGA) circuitry, any other type of integrated circuit (IC), and / or state machines.

[0148] In one or more embodiments, the methods and processes provided herein are implemented in a computer program, software, or firmware, which includes a non-transitory computer-readable storage medium for execution by a general-purpose computer or processor. Examples of non-transitory computer-readable storage media include read-only memory (ROM), random access memory (RAM), registers, cache memory, semiconductor storage devices, magnetic media such as internal hard disks and removable disks, magneto-optical media, and optical media such as CD-ROM discs and digital versatile optical discs (DVDs).

[0149] It should be understood that many variations are possible based on the disclosure herein. Although features and elements have been described above in specific combinations, each feature or element can be used alone without other features and elements, or in various combinations with or without other features and elements.

[0150] Clause 1. A method implemented by a processing device, the method comprising: acquiring measurements of a user generated by a wearable monitoring device during an observation period, the measurements including accelerometer data; detecting actual steps taken by the user based on peak values ​​above a threshold in the accelerometer data; generating a number of steps based on the actual steps detected within a predetermined time period; generating an activity level classification of the user within the predetermined time period based on the number of steps within the corresponding predetermined time period; and outputting the number of steps and the activity level classification within the predetermined time period.

[0151] Clause 2. The method according to Clause 1, wherein the accelerometer data is acquired at a sampling rate of less than 5 Hz.

[0152] Clause 3. The method according to Clause 1 or Clause 2, wherein the activity level classification includes classifying the user's activity during the predetermined time period as one of sedentary, light, moderate or vigorous based on the number of steps taken within the predetermined time period.

[0153] Clause 4. The method according to Clause 3 further comprises: extracting time-domain features and frequency-domain features from the accelerometer data, wherein the activity level classification is generated additionally based on the extracted features.

[0154] Clause 5. The method according to any one of Clauses 1 to 4 further comprises: processing the accelerometer data using a machine learning model to generate a fall prediction, the machine learning model being trained to associate patterns in the accelerometer data with fall events; and outputting the fall prediction.

[0155] Clause 6. The method according to Clause 5 further comprises: training the machine learning model to perform the fall prediction using historical accelerometer data and historical outcome data of the user group as training data, wherein the historical accelerometer data is sampled at the same sampling rate as the accelerometer data acquired by the wearable monitoring device.

[0156] Clause 7. The method according to Clause 5 or Clause 6, wherein the measurement further includes a potential measurement of the user's heart, and the method further includes: processing the potential measurement using another machine learning model to generate a heart rhythm classification, the other machine learning model being trained to associate patterns in the potential measurement with the heart rhythm classification.

[0157] Clause 8. The method according to Clause 7, wherein the heart rhythm classification includes an indication of arrhythmia, and the method further includes: outputting a notification in response to the fall prediction being temporally associated with the arrhythmia.

[0158] Clause 9. The method described in Clause 7 or Clause 8, wherein the heart rhythm classification is one of atrial fibrillation, bradycardia, ventricular arrhythmia, cardiac conduction block, premature ventricular contractions, supraventricular tachycardia, or normal sinus rhythm.

[0159] Clause 10. A processing device comprising: one or more processors; and a memory storing computer-readable instructions executable by the one or more processors to perform operations including: acquiring measurements of a user generated by a wearable monitoring device during an observation period, the measurements including accelerometer data; detecting actual steps taken by the user based on peak values ​​above a threshold in the accelerometer data; generating a step count based on the actual steps detected within a predetermined time period; generating an activity level classification of the user within the predetermined time period based on the step count corresponding to the predetermined time period; and outputting at least one of the step count and the activity level classification within the predetermined time period in a user interface.

[0160] Clause 11. The processing device according to Clause 10, wherein the accelerometer data is acquired by the accelerometer of the wearable monitoring device at a sampling rate of less than 5 Hz.

[0161] Clause 12. The processing device according to Clause 10 or Clause 11, wherein the activity level classification includes classifying the user's activity during the predetermined time period as one of sedentary, mild, moderate, or vigorous based on the number of steps taken within the predetermined time period.

[0162] Clause 13. The processing apparatus according to any one of Clauses 10 to 12, wherein the operation further comprises processing the accelerometer data using a machine learning model to generate a fall prediction, the machine learning model being trained to associate patterns in the accelerometer data with fall events.

[0163] Clause 14. The processing device according to Clause 13, wherein the measurement further includes a potential measurement of the user's heart, and the operation further includes: processing the potential measurement using another machine learning model to generate a heart rhythm classification, the other machine learning model being trained to associate patterns in the potential measurement with the heart rhythm classification; and outputting the heart rhythm classification in the user interface.

[0164] Clause 15. The processing apparatus according to Clause 14, wherein the operation further comprises: associating the activity level classification with one or both of the fall prediction and the heart rhythm classification; and outputting a health status prediction based on the association.

Claims

1. A method implemented by a processing device, the method comprising: Acquire user measurements generated by wearable monitoring devices during the observation period, including accelerometer data; The actual steps taken by the user are detected based on peak values ​​above a threshold in the accelerometer data. The number of steps is generated based on the actual steps detected within a predetermined time period; Based on the number of steps taken within the corresponding predetermined time period, an activity level classification for the user within the predetermined time period is generated. as well as Output the number of steps and the activity level classification within the predetermined time period.

2. The method according to claim 1, wherein, The accelerometer data is acquired at a sampling rate of less than 5 Hz.

3. The method according to claim 1 or 2, wherein, The activity level classification includes categorizing the user's activities during the predetermined time period into one of sedentary, mild, moderate, or vigorous activities based on the number of steps taken within that time period.

4. The method according to claim 3, further comprising: Time-domain and frequency-domain features are extracted from the accelerometer data, and the activity level classification is generated additionally based on the extracted features.

5. The method according to any one of claims 1 to 4, further comprising: The accelerometer data is processed using a machine learning model to generate fall predictions, the machine learning model being trained to associate patterns in the accelerometer data with fall events. as well as Output the predicted fall.

6. The method according to claim 5, further comprising: The machine learning model is trained using historical accelerometer data and historical result data from the user group as training data to perform the fall prediction, wherein the historical accelerometer data is sampled at the same sampling rate as the accelerometer data acquired by the wearable monitoring device.

7. The method according to claim 5 or 6, wherein, The measurements also include the electrical potential measurements of the user's heart, and the method further includes: The potential measurements are processed using another machine learning model to generate a heart rhythm classification, which is trained to associate patterns in the potential measurements with the heart rhythm classification.

8. The method according to claim 7, wherein, The heart rhythm classification includes an indication of arrhythmias, and the method further includes: A notification is output in response to the fall prediction being temporally correlated with the arrhythmia.

9. The method according to claim 7 or 8, wherein, The heart rhythm classification is one of the following: atrial fibrillation, bradycardia, ventricular arrhythmia, cardiac conduction block, premature ventricular contractions, supraventricular tachycardia, or normal sinus rhythm.

10. A processing apparatus, comprising: One or more processors; as well as A memory storing computer-readable instructions executable by the one or more processors to perform operations, including: Acquire user measurements generated by wearable monitoring devices during the observation period, including accelerometer data; The actual steps taken by the user are detected based on peak values ​​above a threshold in the accelerometer data. The number of steps is generated based on the actual steps detected within a predetermined time period; Based on the number of steps taken within a predetermined time period, an activity level classification for the user within that predetermined time period is generated; and The user interface outputs at least one of the number of steps and the activity level category within the predetermined time period.

11. The processing apparatus according to claim 10, wherein, The accelerometer data is acquired by the accelerometer of the wearable monitoring device at a sampling rate of less than 5 Hz.

12. The processing apparatus according to claim 10 or 11, wherein, The activity level classification includes categorizing the user's activities during the predetermined time period into one of sedentary, mild, moderate, or vigorous activities based on the number of steps taken within that time period.

13. The processing apparatus according to any one of claims 10 to 12, wherein, The operation further includes: processing the accelerometer data using a machine learning model to generate fall predictions, the machine learning model being trained to associate patterns in the accelerometer data with fall events.

14. The processing apparatus according to claim 13, wherein, The measurements also include the electrical potential measurements of the user's heart, and the operation further includes: The potential measurements are processed using another machine learning model to generate a heart rhythm classification, said other machine learning model being trained to associate patterns in the potential measurements with heart rhythm classifications; and The heart rhythm classification is output in the user interface.

15. The processing apparatus according to claim 14, wherein, The operation also includes: Associate the activity level classification with one or both of the fall prediction and the heart rate classification; and Based on the aforementioned correlation, a health status prediction is output.

Images