Method and system for body temperature estimation using a wearable biosensor

1. A method to determine patient body temperature using a wearable sensor, comprising:

measuring at least one of ambient, instantaneous, or patient condition sensor data of a patient;

extracting from the at least one of ambient, instantaneous, or patient condition sensor data of the patient,

instantaneous parameters that are entities that are derived based on the sensor data of the patient;

processing contextual parameters and the instantaneous parameters by utilizing a plurality of trained models by a plurality of estimators to determine a plurality of body temperature estimates, respectively, wherein the contextual parameters are learned based on at least one of: personalized prior physiological data of the patient and temporal dynamics of physiological, physical, and skin-surface/microclimatic thermal parameters of the patient;

implementing bias and trend correction procedures to correct the plurality of body temperature estimates by improving trend matching and accuracy of the plurality of body temperature estimates;

determining a confidence level of each of the plurality of body temperature estimates based on meta-features of a feature set that each of the plurality of estimators uses; and

determining a final body temperature based on a combination of the plurality of body temperature estimates weighted based on the confidence level of each of the body temperature estimates.

2. The method of claim 1 , wherein the ambient, instantaneous, or patient condition sensor data includes physiological, physical, skin-surface/microclimatic thermal parameters, temperature, humidity, vitals, and activity metrics.

3. The method of claim 1 , wherein the contextual features parameters include at least one of: learned baseline, short-term dynamics, and long-term dynamics.

4. The method of claim 3 , further comprising smoothing the contextual parameters with different kernels.

5. The method of claim 1 , further comprising utilizing domain knowledge of the contextual parameters to obtain derived features.

6. The method of claim 5 , wherein the domain knowledge to obtain derived features include at least one of: an estimate of change of heart rate from the basal heart or respiratory rate, physiological metrics normalized by accelerometer-based activity metrics, or personalized parameters reflective of a patient's learned baseline physiological state.

7. The method of claim 1 , wherein the plurality of trained models utilized by the plurality of estimators include machine learning models, recursive filters, and mathematical forward models.

8. A non-transitory computer-readable medium, associated with a wearable sensor to determine patient body temperature, storing instructions that, when executed, cause one or more processors to perform operations comprising:

measuring at least one of ambient, instantaneous, or patient condition sensor data of a patient; extracting from the at least one of ambient, instantaneous, or patient condition sensor data of the patient, two classes of parameters including:

instantaneous parameters that are entities that are derived based on the sensor data of the patient, and

contextual parameters that are learned based on time series of the instantaneous parameters; processing the contextual parameters and the instantaneous parameters by utilizing a plurality of trained models by a plurality of estimators to determine a plurality of body temperature estimates, respectively;

implementing bias and trend correction procedures to correct the plurality of body temperature estimates by improving trend matching and accuracy of the plurality of body temperature estimates;

determining a confidence level of each of the plurality of body temperature estimates based on meta-features of a feature set that each of the plurality of estimators uses; and

determining a final body temperature based on a combination of the plurality of body temperature estimates weighted based on the confidence level of each of the body temperature estimates.

9. The non-transitory computer-readable medium of claim 8 , wherein the time series of the instantaneous parameters are determined by smoothing or filtering with different kernals including at least one of: low pass filtering, band pass filtering, high pass filtering, or wavelets to determine short-term dynamics of the instantaneous parameters.

10. The non-transitory computer-readable medium of claim 9 , wherein as increased sensor data is accumulated, the smoothing is performed over longer time frames than the smoothing for the short-term dynamics to determine long-term dynamics of the instantaneous parameters.

11. The non-transitory computer-readable medium of claim 8 , wherein the processing the contextual parameters and the instantaneous parameters is conducted through features extraction schemes.

12. The non-transitory computer-readable medium of claim 11 , wherein the features extraction schemes include:

processing the instantaneous parameters and the contextual parameters in overlapping or non- overlapping time windows to compute varying degree of moments including at least one of: mean, standard deviation, skew, or kurtosis, and

measuring dispersion including at least one of: mean absolute difference, median absolute deviation, coefficient of variation, or entropy, to derive time domain features.

13. The non-transitory computer-readable medium of claim 11 , wherein the features extraction schemes include:

processing the instantaneous parameters and the contextual parameters in time windows to compute statistics of power density in at least one of different frequency sub-bands, cross spectrum, or coherence, to derive frequency domain features.

14. The non-transitory computer-readable medium of claim 11 , wherein the features extraction schemes include:

processing the instantaneous parameters and the contextual parameters in time windows to compute statistical information from short-term Fourier transform and wavelet transform to derive time-frequency domain features.

15. The non-transitory computer-readable medium of claim 11 , wherein the features extraction schemes include processing the instantaneous parameters and the contextual parameters to derive information including time domain features, frequency domain features, and time-frequency domain features.

16. The non-transitory computer-readable medium of claim 15 , wherein the derived information is appended into a universal set feature bank for further processing.

17. The non-transitory computer-readable medium of claim 16 , wherein the plurality of body temperature estimates determined by the plurality of estimators is based on a subset of the universal set feature bank.

18. The non-transitory computer-readable medium of claim 17 , wherein the plurality of body temperature estimates determined by the plurality of estimators is further based on prior training data that contains features of the universal set feature bank and a reference body temperature.

19. The non-transitory computer-readable medium of claim 18 , wherein within each of the plurality of estimators, multiple versions are trained based on availability of features within the universal set feature bank, and a selector is employed to pick a high performing estimator.

20. A wearable sensor including a memory storing instructions to determine patient body temperature, and that, when executed, cause one or more processors to perform operations comprising:

measuring at least one of ambient, instantaneous, or patient condition sensor data of a patient; extracting from the at least one of ambient, instantaneous, or patient condition sensor data of the patient, two classes of parameters including:

instantaneous parameters that are entities that are derived based on the sensor data of the patient, and

contextual personalized parameters that are time series of the instantaneous parameters determined by smoothing or filtering with different kernels including at least one of: low pass filtering, band pass filtering, high pass filtering, or wavelets to determine short-term dynamics of the instantaneous parameters, and wherein as increased sensor data is accumulated, the smoothing is performed over longer time frames than the smoothing for the short-term dynamics to determine long-term dynamics of the instantaneous parameters;

processing the contextual personalized parameters and instantaneous parameters by utilizing a plurality of trained models by a plurality of estimators to determine a plurality of body temperature estimates, respectively;

implementing bias and trend correction procedures to correct the plurality of body temperature estimates by improving trend matching and accuracy of the plurality of body temperature estimates;

determining a confidence level of each of the plurality of body temperature estimates based on meta-features of a feature set that each of the plurality of estimators uses; and

determining a final body temperature based on a combination of the plurality of body temperature estimates weighted based on the confidence level of each of the body temperature estimates.