Method for detecting and responding to falls by residents within a facility

• US 10,226,204 B2
• Filed: 06/19/2017
• Issued: 03/12/2019
• Est. Priority Date: 06/17/2016
• Status: Active Grant

• Current Assignee: Philips North America LLC (Koninklijke Philips N.V.)
• Original Assignee: Philips North America LLC (Koninklijke Philips N.V.)
• Inventors: Heaton, Richard, Devdas, Vikram, Erichsen, Dan, Edwards, Joe
• Primary Examiner(s): Haile, Benyam
• Application Number:
  US15/627,186
• Publication Number:
  US 20180000385A1
• Time in Patent Office: 631 Days
• CPC Class Codes: A61B 5/0022 Monitoring a patient using a global network, e.g. telephone networks, internet
• CPC Class Codes: A61B 5/01 Measuring temperature of body parts ; Diagnostic temperature sensing, e.g. for malignant or inflamed tissue clinical contact thermometers G01K13/20
• CPC Class Codes: A61B 5/11 Measuring movement of the entire body or parts thereof, e.g. head or hand tremor, mobility of a limb for measuring pulse A61B5/02 ;A61B5/1038 takes precedence; motion detection to correct for motion artifacts in physiological signals A61B5/721
• CPC Class Codes: A61B 5/1117 Fall detection
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: G01P 15/00 Measuring acceleration; Measuring deceleration; Measuring shock, i.e. sudden change of acceleration
• CPC Class Codes: G01P 15/0891 with indication of predetermined acceleration values G01P15/135 takes precedence
• CPC Class Codes: G08B 21/043 detecting an emergency event, e.g. a fall
• CPC Class Codes: G08B 21/0446 worn on the body to detect changes of posture, e.g. a fall, inclination, acceleration, gait
• CPC Class Codes: G08B 25/001 Alarm cancelling procedures or alarm forwarding decisions, e.g. based on absence of alarm confirmation
• CPC Class Codes: G08B 25/016 Personal emergency signalling and security systems emergency non-personal manually actuated alarm activators G08B25/12
• CPC Class Codes: G08B 29/186 Fuzzy logic; neural networks
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16Z 99/00 Subject matter not provided for in other main groups of this subclass
• CPC Class Codes: H04M 3/5116 for emergency applications
• CPC Class Codes: H04W 4/023 using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds

Abstract

One variation of a method for detecting and responding to falls by residents within a facility includes: at a wearable device worn by a resident, writing sensor data from a sensor integrated into the wearable device to a buffer, inputting sensor data into a compressed fall detection model—defining a compressed form of a complete fall detection model and stored locally on the wearable device—to detect a fall event at a first time, and transmitting a corpus of sensor data from the buffer and a cue for confirmation of the fall event to a local wireless hub in response to detecting the fall event; and, remotely from the wearable device, inputting the corpus of sensor data into the complete fall detection model—stored remotely from the wearable device—to confirm the fall event and dispatching a care provider to assist the resident in response to confirming the fall event.

Claims

1. A method for detecting and responding to falls by residents within a facility, the method comprising:

at a wearable device worn by a resident;

writing sensor data from a sensor integrated into the wearable device to a buffer;

inputting sensor data read from the sensor into a compressed fall detection model, the compressed fall detection model comprising a first neural network, defining a first quantity of nodes, to calculate a first confidence score for occurrence of a fall event at the wearable device at a first time, characterized by a first memory footprint, stored locally on the wearable device, and defining a compressed form of a complete fall detection model;

in response to the first confidence score exceeding a local threshold score;

detecting a fall event at the wearable device;

wirelessly connecting to a local wireless hub;

transmitting a corpus of sensor data from the buffer and a cue for automated confirmation of the fall event to the local wireless hub, the corpus of sensor data recorded over a duration of time extending up to approximately the first time; and

disconnecting from the local wireless hub;

at a remote computer system;

inputting the corpus of sensor data into the complete fall detection model, the complete fall detection model comprising a second recurrent neural network, defining a second quantity of nodes greater than the first quantity of nodes, to calculate a second confidence score for occurrence of the fall event at the wearable device at the first time, defining a second memory footprint greater than the first memory footprint, and stored remotely from the wearable device; and

in response to the second confidence score exceeding a global threshold score;

confirming the fall event at the wearable device; and

dispatching a care provider to assist the resident.

2. The method of claim 1, further comprising, at the wearable device:

tracking a step rate of the resident based on motion data read from an inertial sensor integrated into the wearable device;

tracking a skin temperature of the resident based on temperature values read from a temperature sensor integrated into the wearable device; and

setting the local threshold score as an inverse function of step rate and deviation of skin temperature of the resident from a target skin temperature.

3. The method of claim 1, further comprising:

calculating a frequency of historical fall events, stored in a fall event database, across a group of residents as a function of time of day; and

setting the global threshold score as an inverse function of frequency of historical fall events at a current time of day.

4. The method of claim 1:

further comprising;

at the wearable device, broadcasting a beacon signal;

calculating a location of the resident based on a known location of the local wireless hub within the facility in response to receipt of the beacon signal from the wearable device at the local wireless hub;

wherein dispatching the care provider to assist the resident comprises;

generating a notification comprising a prompt to assist the resident;

inserting a pointer to the location into the notification; and

transmitting the notification to a computing device affiliated with the care provider.

5. The method of claim 4:

wherein writing sensor data to the buffer comprises writing motion data read from an accelerometer and a gyroscope, integrated into the wearable device, to the buffer defining a duration of multiple minutes; and

wherein broadcasting the beacon signal comprises broadcasting the beacon signal multiple times per minute.

6. The method of claim 4, further comprising:

at the wearable device, outputting an audible alarm at a second time in response to detecting the fall event;

at the remote computer system;

verifying the fall event at the wearable device in response to failure of the resident to clear the alarm within a threshold duration following the second time at the wearable device; and

deescalating a priority of the notification at the computing device in response to receipt of an input to clear the alarm within the threshold duration following the second time at the wearable device.

7. The method of claim 1, further comprising, at the local wireless hub:

attempting connection to a remote computer system via a computer network;

in response to establishing connection to the remote computer system, transmitting the corpus of sensor data to the remote computer system; and

in response to failing to establish connection to the remote computer system;

inputting the corpus of sensor data into an intermediate fall detection model to confirm the fall event, the intermediate fall detection model stored on the local wireless hub, defining a second abbreviated form of the complete fall detection model, and characterized by an intermediate memory footprint between the first memory footprint of the compressed fall detection model and the second memory footprint of the complete fall detection model; and

in response to confirming the fall event, transmitting a local prompt to assist the resident, through a local wireless network, to a computing device affiliated with the care provider.

8. The method of claim 1, further comprising:

at the remote computer system;

generating the complete fall detection model characterized by the second memory footprint;

generating an intermediate fall detection model representing an abbreviated form of the complete fall detection model and characterized by an intermediate memory footprint between the first memory footprint of the compressed fall detection model and the second memory footprint of the complete fall detection model;

generating the compressed fall detection model characterized by the first memory footprint less than the second memory footprint;

distributing the intermediate fall detection model to the local wireless hub via a computer network; and

distributing the compressed fall detection model to the wearable device via the computer network;

at the local wireless hub;

in response to receiving the corpus of sensor data from the wearable device, inputting the corpus of sensor data into the intermediate fall detection model to detect the fall event; and

in response to detecting the fall event, returning the corpus of sensor data to the remote computer system executing the complete fall detection model to confirm the fall event.

9. The method of claim 1:

wherein inputting sensor data into the compressed fall detection model comprises;

calculating the first confidence score for occurrence of the fall event at the wearable device at the first time based on;

a change in ambient pressure values, read from an ambient pressure sensor integrated into the wearable device over the duration of time; and

similarity of motion data, read from an accelerometer and a gyroscope integrated into the wearable device over the duration of time, to a motion pattern of a fall represented in the compressed fall detection model; and

wherein writing sensor data to the buffer comprises writing ambient pressure values read from the ambient pressure sensor and motion data read from the accelerometer and the gyroscope to the buffer; and

wherein transmitting the corpus of sensor data to the local wireless hub comprises transmitting ambient pressure values and motion data to the local wireless hub in response to the confidence score exceeding the local threshold score.

10. The method of claim 1:

wherein writing sensor data into the wearable device to the buffer comprises writing sensor data to the buffer stored locally on the wearable device worn on a wrist of the resident;

further comprising, in response to detecting the fall event at the wearable device, transmitting a prompt to a second wearable device arranged on the resident to transmit sensor data contained in a second buffer stored on -the second wearable device to the local wireless hub; and

wherein inputting the corpus of sensor data into the complete fall detection model remotely from the wearable device to confirm the fall event comprises confirming the fall event remotely from the wearable device and the second wearable device based on the corpus of sensor data received from the wearable device and sensor data received from the second wearable device.

11. The method of claim 1:

wherein dispatching the care provider to assist the resident comprises;

accessing a digital incident report;

inserting a name of the resident, affiliated with the wearable device, into the digital incident report;

inserting a prompt to verify the fall event into the digital incident report; and

transmitting the digital incident report to a computing device affiliated with the care provider; and

further comprising;

extracting verification of the fall event from the digital incident report completed by the care provider at the computing device;

labeling the corpus of sensor data with verified fall event labels;

appending a training set with the corpus of sensor data; and

retraining the complete fall detection model on the training set.

12. The method of claim 11, further comprising, in response to retraining the complete fall detection model:

retraining the compressed fall detection model on the training set, the compressed fall detection model exhibiting a compressed memory footprint smaller than a complete memory footprint of the complete fall detection model; and

pushing an update for the compressed fall detection model to the wearable device via a computer network.

13. The method of claim 1, further comprising:

at the wearable device, transmitting an alert and a second corpus of sensor data from the buffer to the local wireless hub in response to receiving selection of an input region on the wearable device at a second time, the second corpus of sensor data different from the first corpus of sensor data and comprising sensor data recorded to the buffer by the wearable device up to the second time;

remotely from the wearable device;

in response to receiving the alert, dispatching a second care provider to assist the resident;

accessing confirmation of a second fall event, involving the resident, submitted by the second care provider;

scanning the second corpus of sensor data for a segment of sensor data exhibiting correlation to a fall;

labeling the segment of sensor data with a fall event label;

appending a training set with the segment of sensor data; and

retraining the complete fall detection model according to the training set.

14. The method of claim 1:

wherein writing sensor data to the buffer comprises writing sensor data to the buffer defining a first duration;

further comprising;

at the wearable device, writing a second corpus of sensor data to a second buffer defining a second duration greater than the first duration;

remotely from the wearable device;

at a second time, receiving verification of a second fall event from a second care provider, the second fall event involving the resident and undetected by the wearable device;

querying the wearable device for contents of the second buffer;

scanning the second corpus of sensor data, received from the wearable device, for a segment of sensor data exhibiting correlation to a fall;

labeling the segment of sensor data with a fall event label;

appending a training set with the segment of sensor data; and

retraining the complete fall detection model according to the training set.

15. A method for detecting and responding to falls by residents within a facility, the method comprising:

at a wearable device worn by a resident;

writing sensor data from a sensor integrated into the wearable device to a buffer;

inputting sensor data read from the sensor into a compressed fall detection model to detect a fall event at the wearable device at a first time, the compressed fall detection model comprising a first neural network, defining a first quantity of nodes, to calculate a first confidence score for occurrence of a fall event at the wearable device at a first time, characterized by a first memory footprint, defining a compressed form of a complete fall detection model, and stored locally on the wearable device;

in response to detecting the fall event at the wearable device;

detecting a fall event at the wearable device;

wirelessly connecting to a local wireless hub;

transmitting a corpus of sensor data from the buffer and a cue for automated confirmation of the fall event to the local wireless hub, the corpus of sensor data recorded over a duration of time extending up to approximately the first time; and

disconnecting from the local wireless hub;

in response to the wearable device detecting the fall event, issuing a fall alarm for a care provider to assist the resident;

remotely from the wearable device;

inputting the corpus of sensor data into the complete fall detection model to verify the fall event at the wearable device at the first time, the complete fall detection model comprising a second recurrent neural network, defining a second quantity of nodes greater than the first quantity of nodes, to calculate a second confidence score for occurrence of the fall event at the wearable device at the first time, defining a second memory footprint greater than the first memory footprint and stored remotely from the wearable device; and

in response to failure to verify the fall event at the wearable device based on the corpus of sensor data and the complete fall detection model, clearing the fall alarm.

16. The method of claim 15, further comprising:

at the wearable device, transmitting an alert and a second corpus of sensor data from the buffer to the local wireless hub in response to receiving selection of an input region on the wearable device at a second time, the second corpus of sensor data different from the first corpus of sensor data and comprising sensor data recorded to the buffer by the wearable device up to the second time;

remotely from the wearable device;

in response to receiving the alert, dispatching a second care provider to assist the resident;

accessing confirmation of a second fall event, involving the resident, submitted by the second care provider;

scanning the second corpus of sensor data for a segment of sensor data exhibiting correlation to a fall;

labeling the segment of sensor data with a fall event label;

appending a training set with the segment of sensor data; and

retraining the complete fall detection model according to the training set.

17. The method of claim 15, further comprising, at the local wireless hub:

attempting connection to a remote computer system via a computer network

in response to establishing connection to the remote computer system, transmitting the corpus of sensor data to the remote computer system; and

in response to failing to establish connection to the remote computer system;

inputting the corpus of sensor data into an intermediate fall detection model to confirm the fall event, the intermediate fall detection model stored on the local wireless hub, defining a second abbreviated form of the complete fall detection model, and characterized by an intermediate memory footprint between the first memory footprint of the compressed fall detection model and the second memory footprint of the complete fall detection model; and

in response to confirming the fall event, transmitting a local prompt to assist the resident, through a local wireless network, to a computing device affiliated with the care provider.

18. A method for detecting and responding to falls by residents within a facility, the method comprising:

at a wearable device worn by a resident;

writing sensor data from a sensor integrated into the wearable device to a buffer on a first time interval;

transmitting contents of the buffer to a local wireless hub on a second time interval longer than the first time interval;

on a third time interval shorter than the second time interval, scanning sensor data read from the sensor for fall events based on a compressed fall detection model, the compressed fall detection model comprising a first neural network, defining a first quantity of nodes, to calculate a first confidence score for occurrence of a fall event at the wearable device at a first time, characterized by a first memory footprint, defining a compressed form of a complete fall detection model, and stored locally on the wearable device;

in response to detecting a first fall event at a first time;

wirelessly connecting to a local wireless hub;

transmitting a first corpus of sensor data currently stored on the buffer and a cue for automated confirmation of the first fall event to the local wireless hub at approximately the first time; and

disconnecting from the local wireless hub;

in response to the wearable device detecting the first fall event, issuing a fall alarm for a care provider to assist the resident at approximately the first time;

remotely from the wearable device;

scanning the first corpus of sensor data for verification of the first fall event based on the complete fall detection model, the complete fall detection model comprising a second neural network, defining a second quantity of nodes greater than the first quantity of nodes, to calculate a second confidence score for occurrence of the fall event at the wearable device at the first time, defining a second memory footprint greater than the first memory footprint, and stored remotely from the wearable device; and

in response to failure to verify the first fall event based on the first corpus of sensor data and the complete fall detection model, clearing the fall alarm at approximately the first time.

19. The method of claim 18, further comprising, remotely from the wearable device:

in response to receipt of a second corpus of sensor data from the buffer of the wearable device on the second time interval sans detection of a fall event by the wearable device at a second time, scanning the second corpus of sensor data for a second fall event based on the complete fall detection model; and

in response to detecting the second fall event, issuing a second fall alarm for a care provider to assist the resident at approximately the second time.