Method and device for calibrating physiological readings, electronic device using method, and non-transitory storage medium
Inventors
Chen, Cheng • LIU, PING-HAO • XIANG, NENG-DE • HU, MING-SHUN
Assignees
Territory Health Innovation Medical Technology Chengdu Co Ltd
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-11793470-B2
Publication Date
2023-10-24
Expiration Date
Abstract
A method for calibrating measurements of human physiological state applied on an electronic device couples the device to at least one uncalibrated measurement device, worn or placed. Readings of a physiological state are collected by the uncalibrated measurement device and the readings are input into a model for calibration. A sub-module corresponding to the type of reading, for blood pressure or heart rate for example, is called up from the calibration model and a regression algorithm is applied to the calibration model to generate a calibrated reading, each sub-module in the model corresponding to one type of reading. The method also outputs calibrated physiological measurements by the called-up sub-module. A related electronic device and a non-transitory storage medium are also disclosed.
Core Innovation
The invention relates to calibrating measurements of physiology applied on an electronic device communicationally coupled to at least one uncalibrated measurement device. A physiological reading collected by the uncalibrated measurement device is obtained and input into a model for calibration, and the model calls a sub-module corresponding to the reading. Each sub-module corresponds to one physiological reading and calibrates the reading via a regression algorithm, outputting a calibrated physiological measurement.
For calibrating a same kind of reading of the physiological state, the invention selects at least two regression algorithms from preset sorting algorithms to train and create at least two modules. The at least two created modules are averaged to generate the sub-module in the model for calibration. Sample data comprising physiological measurements of a user and standard physiological information is obtained, divided into a training set and a validation set, and used to support selecting an appropriate trained analysis model.
An analysis model is established based on a regression algorithm and trained via the training set with physiological measurements as input and standard physiological information as output. The trained analysis model is validated via the validation set to generate predicted accuracy, and the trained analysis model is determined as being the module if the predicted accuracy is not less than a preset threshold. In some embodiments, calibrated physiological measurements are compared with a preset threshold range to generate a prompt, and trained analysis model selection can be iteratively resumed by adjusting the regression algorithm and/or the number of samples in the training set while re-validating until the predicted accuracy meets the preset threshold.
Claims Coverage
The independent claims cover a method and corresponding electronic device and non-transitory storage medium for calibrating physiological measurements using a calibration model with regression-based sub-modules, one per reading type. The inventive features include multi-algorithm training with averaging, training and validation using sample data paired with standard physiological information, and selecting modules based on a preset accuracy threshold.
Calibration model with per-reading regression sub-modules
Inputting a physiological reading collected by an uncalibrated measurement device into a model for calibration to generate a calibrated physiological measurement, where a sub-module corresponding to the reading is called from the model for calibration and the sub-module calibrates the reading via a regression algorithm, with each sub-module corresponding to one physiological reading.
Multi-algorithm sub-module creation and averaging for a same kind of reading
Selecting at least two regression algorithms from preset sorting algorithms to train and create at least two modules for a same kind of reading of the physiological state, and averaging the at least two created modules to generate the sub-module in the model for calibration.
Training and validation using physiological measurements and standard physiological information
Obtaining sample data comprising physiological measurements of a user and standard physiological information, dividing the sample data into a training set and a validation set, establishing an analysis model based on a regression algorithm, training the analysis model via the training set with physiological measurements as input and standard physiological information as output, and validating the trained analysis model via the validation set to generate predicted accuracy.
Module selection based on predicted accuracy and a preset threshold
Determining whether the predicted accuracy of the trained analysis model is less than a preset threshold, and determining the trained analysis model as being the module if the predicted accuracy is not less than the preset threshold.
Preset threshold range comparison to generate a prompt
Comparing a calibrated physiological measurement to a preset threshold range and generating a prompt when the calibrated physiological measurement is outside the preset threshold range.
Iterative resume-training driven by predicted accuracy threshold
Adjusting the regression algorithm of the analysis model and/or the number of samples in the training set to resume training, validating the resumed trained analysis model via the validation set to generate resumed predicted accuracy, and repeating until the predicted accuracy is not less than the preset threshold.
Combining multi-device sample sets using device codes
Obtaining sample data from at least two uncalibrated measurement devices and standard physiological information from at least two standard measurement devices, and combining the physiological measurements and the standard physiological information based on the codes of the uncalibrated measurement devices and the codes of the standard measurement devices.
Example to-be-calibrated measurement device categories
The to-be-calibrated measurement devices comprise one or more of a smart bracelet, smartwatch, portable oximeter, and portable scale.
Across the independent claims, the core coverage is calibrating uncalibrated physiological readings on an electronic device using a calibration model that calls regression-based sub-modules per reading type. The claims further cover creating calibration sub-modules by selecting at least two preset regression algorithms and averaging the created modules, and selecting the analysis model as a module based on predicted accuracy compared to a preset threshold after training and validation using physiological measurements paired with standard physiological information.
Stated Advantages
Documented Applications
No documented applications found
Interested in licensing this patent?