Wearable device and associated methods using real-time body temperature data

Abstract

Wearable devices, and associated methods for measuring body temperature, are described herein. In one aspect, a wearable device can include: a biocompatible shell; a thermal sensor configured to measure temperature data of a user; and a microcontroller disposed in the biocompatible shell and in electronic communication with the thermal sensor; the microcontroller configured to: cause the thermal sensor to measure the temperature data of the user at a frequency of approximately one cycle per second; determine whether the temperature data deviates from a baseline profile of the user; and cause a notification to be sent to the user indicative of an underlying condition of the user if the temperature data deviates from the baseline profile.

Description

WEARABLE DEVICE AND ASSOCIATED METHODS USING REAL-TIME BODY TEMPERATURE DATACROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims benefit of U.S. Provisional Application No. 63 / 584,740 titled “Wearable Device and Associated Methods Using Real-Time Body Temperature Data'’ filed September 22, 2023, the content of which is hereby incorporated by reference in its entirety for any and all purposes.TECHNICAL FIELD

[0002] The disclosure provided herein relates to wearable devices, and in particular, wearable devices that can collect body temperature.BACKGROUND

[0003] Wearable devices can be worn by a user and can collect biometric data of the user through various sensors. This biometric data can then typically be displayed back to the user for health tracking purposes. For example, wearable device can monitor a user’s sleep cycle, heart rate, and the like, which can then be displayed to a user, for example, for determining the quality of the user’s sleep, the quality of a user’s exercise activity, etc.

[0004] However, these wearable devices include various deficiencies. For example, conventional devices are unable to collect data at a granular level. Wearable devices are able to collect, for example, temperature data approximately once every minute. Wearable devices also do not currently create baseline data for a user, nor identifying existing health conditions of a user. Additionally, wearable devices do not include integrated machine learning algorithms into the device itself, and instead are forced to outsource advanced processing (e.g., via the cloud or other network). These deficiencies, and others, are addressed herein.SUMMARY

[0005] Wearable devices for collecting and processing real-time body temperature data is described herein. The wearable device can include: electronic components to measure skin temperature; a bio- compatible casing to enclose the electronic device that comes in contact with the skin; and the software for the electronics and both mobile and cloud based applications to read, store, convert, and analyze the temperature values. The electroniccomponents can be configured to maximize the battery life of the device and minimizing the burden of monitoring and reporting. A circular buffer and Bluetooth Low Energy (BLE) transmission can be used to minimize storage limitations, given the high frequency data collection, and to maximize data transmission flexibility and lessen the demands placed on a user’s capacity for providing information. The casing can be a biocompatible device with a high- resolution surface that can contact the user’s skin.

[0006] For software, a BLE mobile application can be configured for easy detection and connection to the wearable device. Transmission of data can be made to endpoint devices for secure storage and rendering in a cloud application. A web frontend can allow users to securely create accounts, assign wearable devices, view summary' analytics of their data, and download their temperature data. Data preparation and analysis can filter and prepare the temperature data for analysis. The wearable device described herein, including hardware, software, cloud storage, and computing resources for machine learning, can provide a cost- effective, and thus, scalable solution.BRIEF DESCRIPTION OF THE DRAWINGS

[0007] For the purpose of illustrating the invention, there is shown in the drawings a form that is presently preferred; it being understood, however, that this invention is not limited to the precise arrangements and instrumentalities shown.

[0008] FIG. 1 shows various perspectives of a wearable device according to the present disclosure. Panel (A): White arrow points to the RedBear BLE Nano v2 - nRF52832 (64KB RAM, 512KB Flash, and 8-bit ADC) with a voltage divider as two fixed resistors (upper left box); Panel (B): RedBear was flashed with the Nordic firmware / software using the Nordic nRF52 Development Kit (white arrow); Panel (C): RedBear wearable prototy pe powered by a CR2032 coin cell battery' with thermistor (white arrow); Panels (D) and (E): custom 3D printed, biocompatible casing with thermistor (white arrow); Panel (F): comparison of size of Edge version 2 with the commercial Fever Scout (white arrow is the thermistor).

[0009] FIG. 2 shows various perspectives of a wearable device according to the present disclosure. Panel (A): comparison of size of RedBear nRF52 with a Nordic dongle; Panel (B): comparison of size of the RedBear case (thermistor at white arrow) with the Nordic dongle; Panel (C): comparison of size of the Nordic dongle with the commercial Fever Scout; Panel (D): comparison of size of the RedBear case with the dongle case; Panel (E): custom3D printed, biocompatible dongle casing; Panel (F): casing with dongle inside; Panel (G): location of thermistor (white arrow) for body temperature measurements.

[0010] FIG. 3 shows various perspectives of a wearable device according to the present disclosure. Panel (A): comparison of size of RedBear case (top) with a coin cell battery surface mounted onto the Nordic dongle with the USB portion removed (middle) and printed circuit board (bottom) containing the Fanstel nRF52 chip (right-facing arrow) and thermistor (up-facing arrow); Panel (B): enlarged photo of printed circuit board with Fanstel nRF52 chip (up-facing arrow) and location of thermistor (left-facing arrow); Panel (C): comparison of size of the printed circuit board with Fanstel chip (bottom, upward-and-left-facing arrow), thermistor (top, upward-and-left-facing arrow), and coin cell battery (upward-and-right- facing arrow); Panels (D) and (E): custom 3D printed, biocompatible casing with visible coin cell battery.

[0011] FIG. 4 shows a linear representation of an array concerted into a ring showing, conceptually, a circular buffer (left). This visually shows that the buffer has no real end and data can loop around the buffer (right).

[0012] FIG. 5 shows a communication flow between different components of the wearable device, which supports data transmission between sending samples via the ADC to the ring buffer and upon trigger to the ‘nus.c’ service.

[0013] FIG. 6 shows a communication flow between different components of the wearable device. The code is structured using multiple functions, each designed to perform specific tasks. The main function 200 initializes necessary peripherals and timers and sets up the advertising timer that periodically advertises the device’s presence to potential hosts. The code then enters a continuous loop, waiting for events and handling them. The codebase provided appears represents a system for acquiring data through an Analog to Digital Converter (ADC), buffering the acquired data, and then sending it over Bluetooth Low Energy (BLE) via the Nordic UART Service (NUS). The 'main.c' file sets up the system, initializing the Real Time Clock (RTC), the ADC, and the ring buffer. Once initialization is complete, the main loop of the program begins. Within this loop, the ‘adc.c‘ schedules events to sample data and stores these data samples, along with a timestamp from the ‘rtc.c\ into the ring buffer (‘ringbuf.c‘) 201 and 202. This ring buffer serves as a temporary storage for data samples, allowing for smooth data handling especially when data production and consumption rates differ. The ‘nus.c‘ file handles the transmission of data over BLE 203. Itschedules events that fetch data from the ring buffer and sends this data over the NUS. When new data is received over BLE, it updates the RTC timestamp based on this received data. There are also checks and operations in place to handle connection events, such as starting and stopping data transmission when the BLE connection starts and stops. The 'misc.c' file provides auxiliary functions to help with various tasks like fetching the device ID, power management, and error handling. Overall, the interaction between these files allows for efficient and timely data acquisition, storage, and transmission in this embedded system, with a clear separation of concerns that contributes to maintainability and scalability of the codebase.

[0014] FIG. 7 shows a microcontroller nRF52 module (Fanstel BT832F) mounted on the copper clad PCB by hand, for implementing in a wearable device according to the present disclosure.

[0015] FIG. 8 shows PCB designs for implementing with a wearable device according to the present disclosure. Panel (A): front of empty PCB without mounted electronics; Panel (B): back of empty PCB without mounted electronics; Panel (C): front of PCB with professionally surface mounted electronics (capacitors Cl - C3); reference resistor Rl; crystal oscillator Yl, and the nRF52 module except thermistor; Panel (D): thermistor hand soldered; Panels (E) and (F): heat gun used to attach the coin cell holder to the back of the PCB; Panel (G): front of final PCD with all electronics mounted; Panel (H): side profile of final, two layer PCB with all electronics mounted.

[0016] FIG. 9 shows a sizing diagram for a PCB for the wearable device described herein.

[0017] FIG. 10 shows various perspective of wearable device described herein. Panel (A): Adafruit Feather nRF52 Bluefruit LE; Panel (B): custom 3D printed casing; Panel (C): arm strap; Panel (D): collection of skin temperature from the arm.

[0018] FIG. 11 shows various perspective of a wearable device described herein.

[0019] FIG. 12 shows Actual (blue line) vs Prediction (orange line) using an ARIMA model, for temperature-differenced data over a 1.5-hour snapshot, with quantiles set at l-99th percentiles (shaded orange area).

[0020] FIG. 13 shows Bland- Altman plot for ARIMA model predictions vs actual for temperature data. Mean is denoted by the blue dashed line, and the outer orange lines denote + / - 1.96 standard deviation (SD) from the mean.

[0021] FIG. 14 shows Actual (red line) vs Prediction (blue line) using ARIMA model, for temperature in degrees centigrade over a 1.5-hour snapshot.

[0022] FIG. 15 shows Actual (blue line) vs Prediction (orange line) using LSTM model, for temperature- differenced data over a 1.5-hour snapshot, with quantiles set at 1 -99th percentiles (shaded orange area).

[0023] FIG. 16 shows Bland-Altman plot for LSTM model predictions vs actual for temperature data. Mean is denoted by the blue dashed line, and the outer orange lines denote + / - 1.96 standard deviation (SD) from the mean.

[0024] FIG. 17 shows Actual (red line) vs Prediction (blue line) using LSTM model, for temperature in degrees centigrade over a 1.5-hour snapshot.

[0025] FIG. 18 shows Actual (blue line) vs Prediction (orange line) using LSTM HPO model, for temperature-differenced data over a 1.5-hour snapshot, with quantiles set at l-99th percentiles (shaded orange area).

[0026] FIG. 19 shows Bland- Altman plot for LSTM HPO model predictions vs actual for temperature data. Mean is denoted by the blue dashed line, and the outer orange lines denote + / - 1.96 standard deviation (SD) from the mean.

[0027] FIG. 20 shows Actual (red line) vs Prediction (blue line) using LSTM HPO model, for temperature in degrees centigrade over a 1.5-hour snapshot.

[0028] FIG. 21 shows Actual (blue line) vs Prediction (orange line) using TFT model, for temperature- differenced data over a 1.5-hour snapshot, with quantiles set at l-99th percentiles (shaded orange area).

[0029] FIG. 22 shows Bland-Altman plot for TFT model predictions vs actual for temperature data. Mean is denoted by the blue dashed line, and the outer orange lines denote + / - 1.96 standard deviation (SD) from the mean.

[0030] FIG. 23 shows Actual (red line) vs Prediction (blue line) using TFT model, for temperature in degrees centigrade over a 1.5-hour snapshot.

[0031] FIG. 24 shows Actual (blue line) vs Prediction (orange line) using TFT HPO model, for temperature-differenced data over a 1.5-hour snapshot, with quantiles set at l-99th percentiles (shaded orange area).

[0032] FIG. 25 shows Bland-Altman plot for TFT HPO model predictions vs actual for temperature data. Mean is denoted by the blue dashed line, and the outer orange lines denote + / - 1.96 standard deviation (SD) from the mean.

[0033] FIG. 26 shows Actual (red line) vs Prediction (blue line) using TFT HPO model, for temperature in degrees centigrade over a 1.5-hour snapshot.

[0034] FIG. 27 shows a table of performance metrics of a TempTraq device implementing algorithms for forecasting temperature data described herein.

[0035] FIG. 28 shows a table of performance metrics of a wearable device described herein, implementing the algorithms described herein and provided in FIG. 27. In comparison to the performance metrics of FIG. 27, the algorithms implemented by the wearable device results in more accurate temperature forecasting data.

[0036] FIG. 29 shows a printed circuit board diagram according to the present disclosure.

[0037] FIG. 30 shows a process flow for a wearable device described herein.DETAILED DESCRIPTION

[0038] Wearable technology has become increasingly popular over the last decade, with a wide range of devices now available. These devices can potentially provide valuable insights into various aspects of human health, including physical activity, sleep patterns, and even vital signs. One area of health where wearable devices have been beneficial is monitoring body temperature. Temperature monitoring is an essential tool for early detection of infectious diseases. The capacity to track early indicators of infectious diseases without direct clinical staff interaction and exposure is invaluable for controlling disease transmission. Identifying ill patients earlier through continuous, localized temperature monitoring and digitally tracing their contacts back to a single “patient zero” source can significantly enhance resource and time usage. This is particularly advantageous in contexts with limited staffing and resources. Nevertheless, algorithms to detect biometric anomalies are only as good as the data they are trained on, and temperature data has historically been poorly collected.

[0039] The market for temperature monitoring offers an array of commercial wearable devices. However, their reliability, especially for prolonged, continuous monitoring, is often questionable. This situation underscores the necessity for wearable devices capable of continuously delivering reliable temperature data. Scalability and global applicability are also significant considerations, acknowledging that the threat of disease transmission spans all socioeconomic boundaries. While conventional solutions allow for biometric data collection, their cost implications limit their usage as universal solutions. Furthermore, these devices lack dedicated functionality for continuous temperature monitoring.

[0040] To address these needs, a wearable device for continuous temperature monitoring, is described herein. The wearable device provides enhanced device operation lifespan, improved reliability, and cost-effectiveness. The wearable device is capable of high wearability and ambulatory use, and possessed substantial operating time before recharging or data offloading. Crucially, to achieve disease detection at the the device supported high- frequency data collection with adequate sensitivity', an attribute fundamental for effective biometric anomaly detection.Hardware

[0041] The wearable device can include a sensor configured to collect body temperature data of a user. The sensor can be, for example, a thermistor used to capture the temperature data. The wearable device can also include a voltage divider. The voltage divider can convert the thermistor value into an analog voltage value. The analog value can be then converted into a digital voltage value by an analog-to-digital converter (ADC).

[0042] The wearable device can also include a circular buffer. The circular buffer can store the digital data. The digital data can be transmitted to a mobile application via BLE. A microcontroller can manage the overall control of the wearable device, connectivity via BLE, data storage, and timekeeping.

[0043] The main active electronic components can include the microcontroller and the temperature sensor. The primary' passive electronic components can include a resistor of the voltage divider, capacitors for stable operation of the device, and a crystal oscillator that provides a stable frequency source for the microcontroller’s timer.Microcontroller

[0044] The microcontroller can include a broad range of support for different programming languages and libraries, which makes the microcontroller adaptable and flexible for various development needs. This versatility is crucial when it comes to prototyping and development as it allows for swift modifications and adjustments as needed. The microcontroller can also support the transmission type of the wearable device, such as BLE. BLE can allow for the low-power transmission of physiological information. Given that wearable devices ty pically have small batteries and are expected to run for extended periods, using BLE can significantly enhance the device’s lifespan between charges. Additionally, BLE can enable the secure and efficient transfer of data to the paired mobile health application.

[0045] The microcontroller can also include an ADC to convert the analog signals from the temperature sensor and a BLE to establish a connection to other BLE-enabled devices. The microcontroller can also include low-power modes that help to extend the device's battery life. An example microcontroller can be Arduino nRF52 microcontroller, although one skilled in the art will understand that other microcontrollers can be implemented in the wearable device operations. The BLE services, app timer, ADC, and real-time counter (RTC) are software / firmware functionalities built into the microcontroller and used in the program to enable temperature sampling and storage, BLE connectivity, timekeeping, and scheduling of events.

[0046] In some embodiments the temperature data of user of the wearable device can be sampled at a frequency that is higher or lower than about one cycle per second. For example, the sampling frequency can be once every two seconds, once every three seconds, once every five seconds, once every 10 seconds, once every 20 seconds, once every 30 seconds, once every 50 seconds or even once every 100 seconds. Typically the sampling time period is less than any of the aforementioned time periods. The sampling frequency can, in some embodiments, also be faster than about once every second, for example twice every second (2 / s), 3 / s, 4 / s, 5 / s, 10 / s, 20 / s, or even faster.Batter Life

[0047] One of the main challenges to tackle is the energy7consumption of the sensors, since most devices today have limited battery lifespan and require frequent charging. Wearable sensors with direct WiFi can still require high power consumption to properly function. Components of the wearable device can include low power consumption mechanisms for maximizing power savings.

[0048] For example, the microcontroller can include a System ON sleep mode that allows the CPU to be turned off while keeping peripherals (like the RTC and BLE radio) running. The wearable device can sleep, for example, for 1 minute and is awake for 1 minute when no tasks need to be processed. This can significantly reduce power consumption.

[0049] The RTC, used for timekeeping, can be a low power peripheral that can continue to keep time even in sleep mode. The BLE can also be low power consuming. The microcontroller can include a high-efficiency, on-chip BLE radio for data transmissions. Further, the ADC can be turned off when not sampling data, which can further reduce powerconsumption. The circular buffer can also be configured for batch processing of tasks and data transmission, which can reduce the amount of time the CPU and BLE radio are active.

[0050] As an example, the microcontroller can be in one of three states: idle, advertising, and connected. Assuming a current consumption of 2.6 pA in idle state, a current consumption of 20 mA in advertising state (for 24ms for every 60 s), and 5 mA current consumption during active state, a 3V coin cell battery' of the wearable device can last approximately 36 on a single charge.Temperature Sensor

[0051] The wearable device can include a temperature sensor. The temperature sensor can in some cases be a thermistor. Thermistors can be characterized as having fast response times, are affordable, easily obtainable, and have a small form factor for device comfort. An example thermistor of the wearable device can be the PR103J2 by Littelfuse. which can include an accuracy of 0.05 degrees Celsius.Voltage Divider

[0052] The voltage divider can be used to measure the resistance of the temperature sensor. The voltage divider can include two or more resistors in series. One of the resistors can be the temperature sensor, and another can be a fixed resistor (e.g., 10k Ohms). By measuring the voltage across the fixed resistor, the resistance of the temperature sensor can be calculated. The reference resistor can include a tolerance of 0.05%. The microcontroller can include tolerance of 0.2% for the internal reference voltage. The analog voltage output across the fixed resistor in the voltage divider circuit can be sampled by the ADC of the microcontroller and converted into a digital ADC value that represents the temperature.

[0053] The Steinhart-Hart formula can be used to convert analog data from the temperature sensor into values to be read by the onboard ADC converter. Where R0 = 10000 is the reference (standard) resistance value in ohms of the thermistor at TO, TO = 25°C is the temperature at which R0 was measured in Kelvin, and (3 = 3890 is the beta coefficient of the temperature sensor in Kelvin.ADC

[0054] The analog-to-digital converter (ADC) receive analog data coming from the temperature sensor and convert this into a digital value which can then be turned into a temperature. The accuracy of the temperature reading can be dependent on the resolution of the ADC, since a linear approximation can be used to map the digital ADC values back toresistance values and, ultimately, to temperature readings via a look-up table. For example, a resolution of 12-bits can yield temperature reading within 0. 15°C of the theoretical value.

[0055] A Successive Approximation Analog-to-Digital Converter (SAADC) can serve as the ADC peripheral on the microcontroller, which can capture analog signals from external sensors and convert them into digital signals for the microcontroller to process. In this context, the SAADC can sample data every7one second from a temperature sensor via an analog input from a voltage divider circuit connected to the microcontroller. An acquisition time of the SAADC can be set (e.g.. 40ps), which defines how long the SAADC takes to capture a reading.

[0056] Once the SAADC is initialized, the SAADC conversion can be triggered in the adc sched event function. The function can be responsible for initializing the ADC. triggering a conversion, and waiting for its completion. The function can initialize the ADC with adc init, then the function can trigger the conversion with a nRFx saadc sample and waited for the conversion to complete. The saadc event handler function can serve as the event handler for the SAADC, where the handler function can log the completion of the conversion and update the flag waiting conv to false to signal that the conversion is complete. The while(waiting conv) loops can ensure that the function waited for the ADC conversion to finish before proceeding. If the ADC fails for any reason, the program can stay in the loop until a value is successfully read. This mechanism can serve as a first level of redundancy, ensuring the data is properly sampled before moving forward.

[0057] Once the conversion is complete, the result can be stored as adc record t and pushed into a ring buffer for further use. The record can include a timestamp and the converted ADCvalue. The timestamp can be derived from event data and the ADC value can be obtained from adc buf. An instance of this record can be for example, “timestamp: 1622232000. value: 5127’ In this example, the timestamp can be the Unix time expressed in seconds, and the value can be an unsigned integer from the result of the ADC conversion. The function can conclude by de-initializing the SAADC with nRFx saadc uninit. The rate at which adc sched event function is called (temperature sampling rate) can be set to one sample per second in the main loop.RTC

[0058] Timestamps can be generated by the RTC. The RTC can keep track of time even when the device is powered off. The RTC and timestamps can allow the device to 1) takesamples at consistent time intervals, 2) associate sample values with an absolute time, and 3) synchronize the RTC time to an external source as follows: the ADC sampling events can be scheduled based on timestamps from the RTC. This can ensure samples are taken at consistent intervals; and the timestamp of an ADC sample can be stored along with the sample value. This can allow correlating sample values to absolute times.

[0059] When the device is connected to BLE, the wearable device can send a timestamp which can then be set as the RTC time. This can synchronize the device’s RTC to an external time source.Circular Buffer

[0060] The circular buffer can be designed to be an efficient, fixed-size, cyclic data structure. The circular buffer design can directly address the critical considerations of reliable and secure information flow, and data transmission flexibility for wearable devices. In terms of reliability, the buffer can be designed for systems where there is a producer (in this case, the ADC data acquisition) and a consumer (in this case, the BLE transmission) that may operate at different rates. If data production outpaces consumption, rather than causing an error or requiring a resize, the circular buffer can handle overflow by overwriting the oldest data. The inherent robustness of the circular buffer can ensure a safe and secure exchange of information from the device to intended locations. Furthermore, the ring buffer can ensure data transmission flexibility. Additionally, the circular buffer design can provide an efficient solution to the storage limitations often encountered with loT devices. By continuously sampling and storing data in a non-blocking manner, the circular buffer can enable compressed storage of information, maximizing the use of available memory resources.

[0061] The buffer can store temperature data received from the ADC every second on the CPU of the microcontroller. A ringbuf put(rec) function can be responsible for appending new records to the buffer, maintaining the head and tail indices, and overwriting old data in the event of buffer overflow. The buffer can be configured for a predefined time (e.g., 6 hours) of storage before being overwritten.BLE Connection and Transmission

[0062] The wearable device can transmit the data after a predefined time period (e.g., every 6 hours) to collect the stored data before the data is overwritten overwritten in the circular buffer. The implementation of BLE as the transmission protocol can maximize data transmission flexibility. The wearable device can operate on a predefined awake-sleep cycle(e.g., 24.2 ms awake every 60 s), which can conserve power when data transmission was not required. Data (e.g., timestamp) can be sent to the wearable device. Once received, the RTC timestamp can be updated using the received data. If a connection is made between the wearable device and a central device, the data can be dequeued from the circular buffer and stored in a transmit buffer such that if there is a connection issue, the circular buffer can retain the data which had not been transmitted. If the connection is not active or if the connection is lost, the ADC data can remain stored in the ring buffer until a connection was re-established. At that point, the system can resume transmission where it left off. This provided an implicit layer of redundancy and fault tolerance to the system. In the event of connectivity issues, no data was lost, just delayed, as long as the ring buffer did not overflow in the meantime. When the transmit buffer is full or all records had been fetched (circular buffer was empty), the data can be sent over BLE.Casin

[0063] The outer case of the wearable device is composed of biocompatible materials. Biocompatibility can be characterized as highly wearable for long-term, real-time monitoring, which included the size and the material used in these sensors to reduce any adverse side effects. For example, the outer case can be composed of Nylon 12, polylactic acid (PLA), and the like. The outer case can be 3-D printed. Multiple 3-D processing techniques can be implemented for forming the outer shell. As one example, fused deposition modeling (FDM) can generate the outer shell by extruding thermoplastic fdament layer by layer. As another example, stereolithography (SLA) can implement a UV laser for curing liquid photopolymer resin. In another example, selective laser sintering (SLS) can implement a laser to sinter powdered material layer by layer to form the outer case.

[0064] The outer case can also include internal support structures for disposing the internal electronic components within. These internal support structures can also be formed via 3-D printing, or can alternatively be placed within the outer case after the case is formed.PCB

[0065] FIG. 29 shows an example PCB for the wearable device. The wearable device can include a PCB to ensure proper functioning of the device (Fanstel BC840 / M / E, Compact BLE 5.3, Thread, Zigbee Modules). For example, the PCB can include a 32.768 kHz crystal 103 and 2 capacitors 103 added on the host board. A 4.7 uF capacitor can be added between positive supply voltage (VCC) and ground 102. Additionally, a pin connection 101 can beprovided to allow flashing of the board with the appropriate firmware, while the circuit 104 can provide the ADC and voltage divider circuit to read temperature from the thermistor device. The circuit 103 can facilitate maintaining the device in a lower sleep state as when not transmitting or receiving data.Process Example

[0066] FIG. 30 shows an example process for temperature data processing. Data for the wearable device can be stored in a database sequentially at 301. Data can be sent into the database service at 304, which can include the embedded analytics layer which collects location, activity and temperature via the wearable device 302 into the mobile application. Whenever the application is run and connected to the wearable device 302, the application can auto pull data from the device every 5 minutes; and stores up to 6 hours’ worth of data on the device. The app layer 304 can also synchronize time with the wearable device 302 whenever the two connect with each other. The wearable device has the ability to connect to other services that are not the application layer provided they provide Bluetooth support. Data can be analyzed both at the device level 303 and at the app 304 and 306 cloud level. This information can then be provided back to the 305 user, government and medical community, if the user has a temperature that is unexpected or characteristic of a disease they are notified in 305, similarly are government and medical community individuals.Software

[0067] The temperature data collected can be inputted into a trained machine learning model. The model can create a baseline temperature profile for a user, which can serve as a reference point to detect any changes in health status of the user, such as those caused by disease or therapeutic treatments. In some cases, the model can forecast future temperature states of the user, such as forecasted out to an hour. The model can either be stored locally at the wearable device, or externally (e.g., via a cloud network). Further, the data inputted into the model can be the temperature data collected by the wearable device. For example, the wearable can be tuned to collect temperature data of the user at approximately a 1 Hz cycle. In some cases, the model can determine the baseline temperature profile, potential deviations from the baseline profile, solely from the temperature data. Examples of performance of different models are described below. The user data discussed below was fed to the particular model with an 80% training and 20% test split.ARIMA

[0068] FIG. 12 compares the predicted (orange line) to the actual (blue line) temperature differences. The orange band indicates the degree of uncertainty associated with the predictions. The Bland-Altman plot is a method to visualize the agreement between two different measurements. The Bland-Altman analysis (FIG. 13) indicates a significant level of agreement between the actual vs. ARIMA predicted for most of the data, as evidenced by the high con- centration of data points around between the mean of -0.04°C and the +1.96 SD of 0.01°C. However, with such tight tolerances around the bands of acceptance, there is limited insight to be gained outside of higher temperatures leading to higher disagreement as seen by the concentration of biases near 36.2°C closer to the -1.96 SD.

[0069] Model performance was evaluated through two metrics: MAPE (Mean Absolute Percentage Error) and RMSE (Root Mean Square Error). A calculated MAPE value of 0. 11% yielded a “highly accurate forecast”. An RMSE value of 0.04°C was determined; the FDA standard for RMSE is 0. 1°C. Further scrutiny of the results (FIG. 14) showed that the model exhibited high accuracy until it began to underfit as temperature inverted around the 520 minute marker. After this segment, a growing discrepancy or offset in the model’s performance was observed over time. Despite this divergence, the model maintained its ability to replicate the subtle nuances of a healthy individual’s baseline temperature profile, offering valuable insights into the system’s behavior.LSTM

[0070] FIG. 15 compares the predicted (orange line) to the actual (blue line) temperature differences. The orange band indicates the degree of uncertainty associated with the predictions. The Bland-Altman analysis (FIG. 16) indicates a significant level of agreement between the actual vs LSTM predicted for most of the data, as evidenced by the high concentration of data points around the mean of 0.02°C. As with prior analysis, the tight tolerances limit insight, however, there is a consistent bias of 0.02°C throughout the temperature range and an almost even data distribution throughout both bounds of the confidence interval, with a concentration of biases around 35.9 - 36°C closer to the - 1.96 SD and a concentration of biases around 36.15 - 36.25°C closer to the + 1.96 SD.

[0071] A calculated MAPE value of 0.07% yielded a “highly accurate forecast”. An RMSE value of 0.03°C was determined. FIG. 17 showed that the model exhibited high accuracy in predicting the general characteristic of the profile with the exception of the window 500 - 530 minutes, where there is a visible disagreement between prediction and actual values.Despite this divergence, the model maintained its ability to moderately replicate the subtle nuances of a healthy baseline.LSTM HPO

[0072] FIG. 18 compares the predicted (orange line) to the actual (blue line) temperature differences. The orange band indicates the degree of uncertainty associated with the predictions. The Bland-Altman analysis (FIG. 19) indicates a limited level of agreement between the actual vs LSTM HPO predicted for a majority of the data, as evidenced by the spread of data points on opposite ends of the standard deviations: concentration of biases ranging from 36.0 to 36.1 C closer to -1.96 SD and spread throughout the temperature range near the +1.96 SD.

[0073] A calculated MAPE value of 0.50% yielded a “highly accurate forecast”. An RMSE value of 0.24C was determined. The model (FIG. 20) exhibited high accuracy in predicting the initial 10 minutes of data. After this segment, a large discrepancy or offset in the model’s performance was observed over time. With this divergence, the model did not maintain its ability to replicate the subtle nuances of the baseline temperature profile, instead decreasing rapidly in temperature as time progressed.TFT

[0074] The Bland-Altman analysis (FIG. 22) indicates a significant level of agreement between the actual vs TFT predicted for most of the data. While the points are not all concentrated around the mean, there is negligible insight from the plot as the differences are below the FDA limits of concern. There is an average bias of 0.00°C throughout with a cluster of data between 36.2 - 36.25C close to -1.96 SD.

[0075] A calculated MAPE value of 0.03% yielded a “highly accurate forecast”. An RMSE value of 0.01C was determined. The model (FIG. 23) accurately predicted the entire span of data. The biggest if any discrepancy was observed as the temperature transitioned from 35.9 to 36.2C. Despite this divergence, the model maintained its ability to replicate the subtle nuances of the individual’s baseline.TFT HPO

[0076] FIG. 24 compares the predicted (orange line) to the actual (blue line) temperature differences. The orange band indicates the degree of uncertainty associated with the predictions. The Bland-Altman analysis (FIG. 25) indicates a significant level of agreement between the actual vs TFT HPO predicted for most of the data, as evidenced by the highconcentration of data points around the mean of -0.01°C. However, there is a concentration of biases ranging from 35.9 to 36.0 closer to the + 1.96 SD and a concentration of biases ranging from 36.15 to 36.25 closer to the - 1.96 SD.

[0077] A calculated MAPE value of 0.04% yielded a “highly accurate forecast”. An RMSE value of 0.02C was determined. The model (FIG. 26) exhibited high accuracy in prediction. During the transition from decreasing temperature to increasing, the model over-fits initially in some areas before under-fitting. Despite this divergence, the model maintained its ability to replicate the subtle nuances of the baseline.

[0078] The TFT model initially performed the best. Surprisingly, the HPO TFT model proved to be less effective than the initial parameters set for TFT. The Bland-Altman plots showed a strong concentration around the mean in both models. However, in the HPO TFT model, the values were below the mean, resulting in an error closer to 0 and, thus, higher performance for the model. The results support that wearable devices can be used to develop forecast of probable temperature profiles and operate inside of the range of FDA acceptance (< 0. 1C) for temperature readings given the RMSE of 0.01 C from the TFT model.

[0079] The term “approximately” as used herein can refer to manufacturing tolerances. In one example, the term “approximately” as used with reference to a size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter as well as + / - 10% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 9% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 8% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 7% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 6% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 5% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 4% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 3% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 2% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter, including + / - 1% of the size, shape, direction, or other parameter can equal the stated size, shape, direction, or other parameter.

[0080] Throughout the description above, reference made to singular elements using terms such as “a,” “an,” or “the” can also apply to plural elements or at least one of the elements. Thus, description of a singular element applies with equal force and effect to a plurality of elements, including at least one of the elements. Conversely, reference to a plurality of elements can also apply with equal force and effect to a single one of the elements or at least one of the elements. Further, description of at least one element can apply with equal force and effect to a singular element or a plurality of the elements.

[0081] “Optional” or “optionally” means that the subsequently described event or circumstance may or may not occur, and that the description includes instances where said event or circumstance occurs and instances where it does not.

[0082] Throughout the description and claims of this specification, the word “comprise” and variations of the word, such as “comprising” and “comprises,” means “including but not limited to,” and is not intended to exclude, for example, other components, integers or steps. “Exemplary” means “an example of’ and is not intended to convey an indication of a preferred or ideal embodiment. “Such as” is not used in a restrictive sense, but for explanatory purposes.

[0083] Disclosed are components that may be used to perform the disclosed methods, systems, and apparatuses. These and other components are disclosed herein, and it is understood that when combinations, subsets, interactions, groups, etc. of these components are disclosed that while specific reference of each various individual and collective combinations and permutation of these may not be explicitly disclosed, each is specifically contemplated and described herein, for all methods, systems, and apparatuses. This applies to all features of this application including, but not limited to, components and / or steps in disclosed methods, systems, and apparatuses. Thus, if there are a variety of additional steps that may be performed it is understood that each of these additional steps may be performed with any specific embodiment or combination of embodiments of the disclosed methods, systems, and apparatuses.

[0084] While the methods, systems, and apparatuses have been described in connection with preferred embodiments and specific examples, it is not intended that the scope be limited to the particular embodiments set forth, as the embodiments herein are intended in all respects to be illustrative rather than restrictive.

[0085] Unless otherwise expressly stated, it is in no way intended that any method set forth herein be construed as requiring that its steps be performed in a specific order. Accordingly, where a method claim does not actually recite an order to be followed by its steps or it is not otherwise specifically stated in the claims or descriptions that the steps are to be limited to a specific order, it is no way intended that an order be inferred, in any respect. This holds for any possible non-express basis for interpretation, including: matters of logic with respect to arrangement of steps or operational flow; plain meaning derived from grammatical organization or punctuation; the number or type of embodiments described in the specification.

[0086] It will be apparent to those skilled in the art that various modifications and variations may be made without departing from the scope or spirit. Other embodiments will be apparent to those skilled in the art from consideration of the specification and practice disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit being indicated by the following claims.EXAMPLES

[0087] The following embodiments are exemplary only and do not serve to limit the scope of the present disclosure of the appended claims. It should be understood that any part of any one or more Examples can be combined with any part of any other one or more Examples. Example 1

[0088] A wearable device comprising: a biocompatible shell; a thermal sensor configured to measure temperature data of a user; and a microcontroller disposed in the biocompatible shell and in electronic communication with the thermal sensor; the microcontroller configured to: cause the thermal sensor to measure the temperature data of the user at a frequency of approximately one cycle per second; determine whether the temperature data deviates from a baseline profile of the user; and cause a notification to be sent to the user indicative of an underlying condition of the user if the temperature data deviates from the baseline profile.Example 2

[0089] The wearable device of Example 1 , further comprising a circular buffer, and wherein the microcontroller is further configured to: cause the temperature data to be stored in the circular buffer.Example 3

[0090] The wearable device of Example 1 or Example 2, wherein the microcontroller is further configured to generate the baseline profile by: causing the thermal sensor to measure baseline temperature data of the user at a frequency of approximately 1 measurement per second and over a time period of approximately a day; inputting the baseline temperature data into a trained machine learning model; and receiving, from the trained machine learning model, the baseline profile.Example 4|0091| The wearable device of any of Examples 1-3, wherein the trained machine learning model is stored locally at the wearable device.Example 5

[0092] The wearable device of any of Examples 1-4, wherein the trained machine learning model comprises an autoregressive integrated moving average (ARIMA) model, a long shortterm memory (LSTM) model, a LSTM hunter-prey optimizer (HPO) model, a temporal fusion transformer (TFT) model, or a TFT HPO model.Example 6

[0093] The wearable device of any of Examples 1-5, wherein the biocompatible shell is composed of Nylon 12 or poly lactic acid (PLA).Example 7

[0094] The wearable device of any of Examples 1-6, wherein the thermal sensor comprises a thermistor.Example 8

[0095] The wearable device of any of Examples 1-7, further comprising: a voltage divider comprising the thermal sensor and at least one other resistor configured to convert the temperature measurements of the thermal sensor into an analog current.Example 9

[0096] The wearable device of any of Examples 1-8, further comprising a low energy communication transceiver configured to communicate with a corresponding mobile application of the wearable device.Example 10|0097| The wearable device of any of Examples 1-9, wherein the microcontroller is further configured operate in an idle state for approximately 95% while operating.Example 11

[0098] The wearable device of any of Examples 1-10, further comprising a battery configured to power the microcontroller and the thermal sensor, and wherein the battery is further configured to have a battery life of approximately 30 days based on operating conditions of the wearable device.Example 12

[0099] The wearable device of any of Examples 1-3, wherein the underlying condition of the user comprises a fever.Example 13

[0100] A method performed by a wearable device, comprising: measuring temperature data of user of the wearable device at a frequency of approximately one cycle per second; determining whether the temperature data deviates from a baseline profile of the user; and sending a notification to the user indicative of an underlying condition of the user if the temperature data deviates from the baseline profile.Example 14

[0101] The method Example 13, further comprising: measuring baseline temperature data of the user at a frequency of approximately 1 measurement per second and over a time period of approximately a day; inputting the baseline temperature data into a trained machine learning model; and receiving, from the trained machine learning model, the baseline profile. Example 15

[0102] The method of Example 13 or 14, further comprising: storing the temperature data in a circular buffer of the wearable device; and retrieving the stored temperature data upon initiation of determining whether the temperature data deviates from the baseline profile of the user.

Claims

CLAIMS1. A wearable device comprising: a biocompatible shell; a thermal sensor configured to measure temperature data of a user; and a microcontroller disposed in the biocompatible shell and in electronic communication with the thermal sensor; the microcontroller configured to: cause the thermal sensor to measure the temperature data of the user at a frequency of approximately one cycle per second; determine whether the temperature data deviates from a baseline profile of the user; and cause a notification to be sent to the user indicative of an underlying condition of the user if the temperature data deviates from the baseline profile.

2. The wearable device of claim 1, further comprising a circular buffer, and wherein the microcontroller is further configured to: cause the temperature data to be stored in the circular buffer.

3. The wearable device of claim 1, wherein the microcontroller is further configured to generate the baseline profile by: causing the thermal sensor to measure baseline temperature data of the user at a frequency of approximately 1 measurement per second and over a time period of approximately a day; inputting the baseline temperature data into a trained machine learning model; and receiving, from the trained machine learning model, the baseline profile.

4. The wearable device of claim 3, wherein the trained machine learning model is stored locally at the wearable device.

5. The wearable device of claim 3, wherein the trained machine learning model comprises an autoregressive integrated moving average (ARIMA) model, a long short-term memory (LSTM) model, a LSTM hunter-prey optimizer (HPO) model, a temporal fusion transformer (TFT) model, or a TFT HPO model.

6. The wearable device of claim 1 , wherein the biocompatible shell is composed of Nylon 12 or polylactic acid (PLA).

7. The wearable device of claim 1, wherein the thermal sensor comprises a thermistor.

8. The wearable device of claim 1, further comprising: a voltage divider comprising the thermal sensor and at least one other resistor configured to convert the temperature measurements of the thermal sensor into an analog current.

9. The wearable device of claim 1, further comprising a low energy communication transceiver configured to communicate with a corresponding mobile application of the wearable device.

10. The wearable device of claim 1, wherein the microcontroller is further configured operate in an idle state for approximately 95% while operating.

11. The wearable device of claim 1 , further comprising a battery7configured to power the microcontroller and the thermal sensor, and wherein the battery is further configured to have a battery life of approximately 30 days based on operating conditions of the wearable device.

12. The wearable device of claim 1, wherein the underlying condition of the user comprises a fever.

13. A method performed by a wearable device, comprising: measuring temperature data of user of the wearable device at a frequency of approximately one cycle per second; determining whether the temperature data deviates from a baseline profile of the user; and sending a notification to the user indicative of an underlying condition of the user if the temperature data deviates from the baseline profile.

14. The method of claim 13, further comprising: measuring baseline temperature data of the user at a frequency of approximately 1 measurement per second and over a time period of approximately a day: inputting the baseline temperature data into a trained machine learning model; and receiving, from the trained machine learning model, the baseline profile.

15. The method of claim 13. further comprising: storing the temperature data in a circular buffer of the wearable device; and retrieving the stored temperature data upon initiation of determining whether the temperature data deviates from the baseline profile of the user.

16. A wearable device for predicting heart-related stress, comprising: a) a body temperature sensor configured to measure a user's body temperature; b) a processor coupled to the body temperature sensor; c) a memory coupled to the processor, storing instructions that when executed by the processor, cause the wearable device to: i) continuously monitor the user's body temperature; ii) analyze variations in the user's body temperature over time using machine learning algorithms; hi) perform time series forecasting and anomaly detection on the body temperature data using at least one of: transformer models; generative Al models; autoregressive integrated moving average (ARIMA) models; exponential smoothing methods; prophet models; long short-term memory (LSTM) networks; isolation forests; one-class support vector machines (SVM); orGaussian mixture models;iv) predict a likelihood of a heart-related stress condition based on the time series forecasting and anomaly detection; and v) generate an alert when the predicted likelihood exceeds a predetermined threshold.

17. The wearable device of claim 16, wherein the machine learning algorithms include at least one of: transformer models for sequence-to-sequence learning on the body temperature data; generative Al models for synthesizing potential future body temperature patterns;ARIMA models for time series forecasting; exponential smoothing methods for trend and seasonality analysis; prophet models for detecting multiple seasonalities in the temperature data;LSTM networks for processing sequential body temperature data; isolation forests for detecting anomalies in body temperature patterns; one-class SVMs for novelty7detection in temperature data; orGaussian mixture models for modeling temperature distribution and detecting outliers.

18. The wearable device of claim 16, wherein the heart-related stress conditions include heat stroke and sudden cardiac episodes.

19. The wearable device of claim 16, further comprising a wireless communication module configured to transmit the alert to a remote device.

20. The wearable device of claim 16, wherein the instructions further cause the wearable device to: store historical body temperature data for the user; establish a personalized baseline body temperature profile for the user using machine learning techniques; and use the personalized baseline body temperature profile in predicting the likelihood of a heart-related stress condition.

21. The wearable device of claim 1, further comprising an environmental temperature sensor, wherein the instructions further cause the wearable device to: monitor ambient temperature; analyze the relationship between the user's body temperature and the ambient temperature using machine learning algorithms; and incorporate the relationship into the prediction of the likelihood of a heart-related stress condition.

22. A method for predicting heart-related stress using a wearable device, comprising: continuously measuring a user's body temperature using a body temperature sensor in the wearable device; analyzing variations in the user's body temperature over time using machine learning algorithms; performing time series forecasting and anomaly detection on the body temperature data using at least one of transformer models, generative Al models, ARIMA models, exponential smoothing methods, prophet models, LSTM networks, isolation forests, one- class SVMs, or Gaussian mixture models; predicting a likelihood of a heart-related stress condition based on the time series forecasting and anomaly detection; and generating an alert when the predicted likelihood exceeds a predetermined threshold.

23. The method of claim 22, wherein the machine learning algorithms include at least one of: transformer models for sequence-to-sequence learning on the body temperature data; generative Al models for synthesizing potential future body temperature patterns;ARIMA models for time series forecasting; exponential smoothing methods for trend and seasonality analysis; prophet models for detecting multiple seasonalities in the temperature data;LSTM networks for processing sequential body temperature data; isolation forests for detecting anomalies in body temperature patterns; one-class SVMs for novelty detection in temperature data; orGaussian mixture models for modeling temperature distribution and detecting outliers.

24. The method of claim 22, wherein the heart-related stress conditions include heat stroke and sudden cardiac episodes.

25. The method of claim 22, further comprising transmitting the alert to a remote device using a wireless communication module in the wearable device.

26. The method of claim 22. further comprising: a. storing historical body temperature data for the user; b. establishing a personalized baseline body temperature profde for the user; and c. using the personalized baseline body temperature profile in predicting the likelihood of a heart-related stress condition.

27. The method of claim 22, further comprising: monitoring ambient temperature using an environmental temperature sensor in the wearable device; analyzing the relationship between the user's body temperature and the ambient temperature; and incorporating the relationship into the prediction of the likelihood of a heart-related stress condition.

28. A non-transitory computer-readable medium storing instructions that, when executed by a processor in a wearable device, cause the wearable device to perform operations comprising: continuously measuring a user's body temperature using a body temperature sensor; analyzing variations in the user's body temperature over time; comparing the analyzed variations to predetermined patterns associated with heart- related stress conditions; predicting a likelihood of a heart-related stress condition based on the comparison; and generating an alert when the predicted likelihood exceeds a predetermined threshold.

29. The non-transitory computer-readable medium of claim 28, wherein the operations further comprise: monitoring the user's heart rate using a heart rate sensor; correlating the heart rate data with the body temperature data; and refining the prediction of the likelihood of a heart-related stress condition based on the correlation.

30. The non-transitory computer-readable medium of claim 28. wherein the operations further comprise: storing historical body temperature data for the user; establishing a personalized baseline body temperature profile for the user; and using the personalized baseline body temperature profile in predicting the likelihood of a heart-related stress condition.

Images