Method for quantifying the risk of falling of an elderly adult using an instrumented version of the FTSS test
Inventors
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-9877667-B2
Publication Date
2018-01-30
Expiration Date
Abstract
Methods and systems may provide for estimating falls risk based on inertial sensor data collected during a Five Times Sit-to-Stand (FTSS) test. In an embodiment, a classifier model may be trained with inertial sensor data collected from a sample of people performing the FTSS test and their self-reported falls history. In an embodiment, one or more features related to steadiness or smoothness of the person's movement may be calculated. In an embodiment, one or more features related to timing of the FTSS test, such as a total time taken to complete the FTSS test or to complete individual sit-stand-sit (SSS) phases of the test, may be calculated. In an embodiment, supervised pattern recognition techniques may train the classifier model to classify a person as being likely to fall or not being likely to fall based on FTSS-related feature values collected from that person.
Core Innovation
The invention relates to estimating falls risk using computer-implemented processing of acceleration data from tri-axial inertial sensors attached to a person during transitions between a standing state and a sitting state. A first inertial sensor is attached to the person's lower body and a second inertial sensor is attached to the person's upper body while the person transitions at least one times from standing to sitting or from sitting to standing. Falls history information of the person is received and used together with sensor-derived features to support risk estimation.
The method determines multiple feature values from the received acceleration data, including a value indicating a total time for completing the transitioning between the sitting state and the standing state. From the second acceleration data, it determines a value indicating steadiness of movement, where a jerk of the movement is calculated as a derivative of second acceleration data along a sensor axis to measure steadiness along the sensor axis, and it also determines feature values indicating mean, coefficient of variation, or root mean square of the second acceleration data and a spectral edge frequency of the second acceleration data.
After determining the feature values, the method generates a classifier model based on the determined time-related feature, steadiness-related feature, statistical and spectral features, and the falls history information. After generating the classifier model, it receives acceleration data from inertial sensors attached to another person and inputs a subset of the features as input features of the classifier model. It calculates a quantitative value for a probability of a risk of falling for the another person and outputs a classification of likely to fall or not being likely to fall based on the calculated probability.
In one implementation, the method is additionally defined to determine total time using a time at the start of a first transition and a time at the end of a fifth transition, and the transitioning occurs at least five times in either direction between the standing state and the sitting state. In further refinements, additional acceleration-derived values such as acceleration amplitude along a body axis and mean jerk along the body axis are used, and the classifier model is generated as a linear discriminant classifier model.
Claims Coverage
The independent claims covered are clm-00001 (method) and clm-00008 (non-transitory computer-readable medium). Across these independent claims, five main inventive features are used to compute a quantitative probability of fall risk and output likely-to-fall versus not-likely-to-fall classifications, using tri-axial lower-body and upper-body sensor data, falls history, timing/steadiness/statistical/spectral features, and a generated classifier model with a feature subset as input for another person.
Tri-axial lower-body and upper-body acceleration sensing during standing-sitting transitions
Measuring first acceleration data from a first inertial sensor and second acceleration data from a second inertial sensor, the first and second inertial sensors comprise tri-axial accelerometers and are attached to at least one person transitioning at least one times from a standing state to a sitting state or from a sitting state to a standing state, wherein the first inertial sensor is attached to the person's lower body and the second inertial sensor is attached to the person's upper body.
Falls-history-informed feature extraction for total time
Determining, from the first acceleration data and using the one or more processors, a first value of one or more features indicating a total time for the at least one person to complete the transitioning between the sitting state and the standing state.
Jerk-derived steadiness and acceleration statistics plus spectral edge frequency
Determining, from the second acceleration data and using the one or more processors, a second value of the one or more features indicating steadiness of movement of the at least one person, wherein a jerk of the at least one person's movement is calculated as a derivative of second acceleration data along a sensor axis to measure the steadiness of movement of the least one person along the sensor axis; and determining, from the second acceleration data and using the one or more processors, a third value of the one or more features indicating a mean, coefficient of variation, or root mean square of the second acceleration data and a fourth value of the one or more features indicating a spectral edge frequency of the second acceleration data.
Classifier model generation using feature subset and falls history
Generating, at the one or more processors, a classifier model based on the determined first value, the determined second value, the determined third value, the determined fourth value, and the falls history information of the at least one person; and after the generation of the classifier model, receiving acceleration data from inertial sensors attached to another person; inputting a subset of the one or more features as input features of the classifier model.
Quantitative fall-risk probability and likely-to-fall versus not-likely-to-fall output
Calculating, via the one or more processors using both the generated classifier model and the received acceleration data from the another person, a quantitative value for a probability of a risk of falling of the another person; and outputting a classification of likely to fall or not being likely to fall for the another person based on the calculated probability.
The independent claims consistently require tri-axial acceleration sensing from lower and upper body during standing-to-sitting transitions, extraction of a total time feature from lower-body data, computation of steadiness using jerk as a derivative from upper-body data together with mean, coefficient of variation, root mean square, and spectral edge frequency features, generating a classifier model using these features plus falls history, and using the trained classifier to compute a quantitative fall-risk probability and output likely-to-fall versus not-likely-to-fall for another person.
Stated Advantages
Higher specificity and improved ROC AUC compared to models using only age and total FTSS time.
Provides quantitative fall probability and a classification of likely to fall versus not likely to fall.
Documented Applications
Estimating elderly falls risk during Five Times Sit-to-Stand (FTSS) using instrumented FTSS (iFTSS) with tri-axial inertial sensors attached to the thigh and sternum.
Classifying another person as likely to fall or not likely to fall by calculating a probability of risk of falling based on received acceleration data and falls history information.
Interested in licensing this patent?