Methods and apparatus for detecting hand-to-mouth behavior

• US 10,335,061 B2
• Filed: 10/22/2016
• Issued: 07/02/2019
• Est. Priority Date: 11/07/2015
• Status: Active Grant

• Current Assignee: Massachusetts Institute of Technology
• Original Assignee: Massachusetts Institute of Technology
• Inventors: Azaria, Asaph, Mayton, Brian, Paradiso, Joseph
• Primary Examiner(s): Macchiarolo, Peter J
• Assistant Examiner(s): Jenkins, Jermaine L
• Application Number:
  US15/331,840
• Publication Number:
  US 20170127979A1
• Time in Patent Office: 983 Days
• Field of Search: None
• CPC Class Codes: A61B 2562/0209 Special features of electrodes classified in A61B5/24, A61B5/25, A61B5/283, A61B5/291, A61B5/296, A61B5/053
• CPC Class Codes: A61B 2562/0219 Inertial sensors, e.g. accelerometers, gyroscopes, tilt switches
• CPC Class Codes: A61B 2562/0223 Magnetic field sensors
• CPC Class Codes: A61B 2562/043 in a linear array
• CPC Class Codes: A61B 2562/06 Arrangements of multiple sensors of different types
• CPC Class Codes: A61B 5/053 Measuring electrical impedance or conductance of a portion of the body
• CPC Class Codes: A61B 5/1114 Tracking parts of the body
• CPC Class Codes: A61B 5/1123 Discriminating type of movement, e.g. walking or running A61B5/1116, A61B5/112 take precedence
• CPC Class Codes: A61B 5/1128 using image analysis A61B5/1127 takes precedence
• CPC Class Codes: A61B 5/16 Devices for psychotechnics using teaching or educational appliances G09B1/00 - G09B7/00; Testing reaction times ; Devices for evaluating the psychological state
• CPC Class Codes: A61B 5/6824 Arm or wrist
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A sensor system detects hand-to-mouth behavior. The system includes an electrical bio-impedance spectrometer and an inertial measurement unit. The sensor system may be worn on the forearm. The sensor system recognizes hand-to-mouth behavior in real-time, facilitating monitoring and immediate interventions An electrode positioning strategy optimizes the device'"'"'s sensitivity and accuracy. Machine learning algorithms are leveraged to infer the hand-to-mouth detection. A prototype of the sensor system achieves 92% detection accuracy for recurrent usage by a single user and 90% accuracy for users that have not been previously encountered.

Claims

1. A method comprising, in combination:

(a) an electrical bio-impedance (EBI) spectrometer taking a first set of measurements, which EBI spectrometer includes four crossed electrodes that are electrically connected to skin of a forearm of a subject, which four crossed electrodes comprise a first IC electrode, a second IC electrode, a first PU electrode and a second PU electrode, the first set of measurements being taken by

(i) driving an alternating electrical current such that the alternating current

(A) undergoes frequency sweeps, and

(B) passes through an electrical circuit that includes the forearm and the first and second IC electrodes, and

(ii) measuring, during the frequency sweeps, voltage difference between the first and second PU electrodes;

(b) a magnetometer worn on the forearm taking a second set of measurements of orientation of the forearm;

(c) a gyroscope worn on the forearm taking a third set of measurements of orientation of the forearm;

(d) an accelerometer worn on the forearm taking a fourth set of measurements of acceleration of the forearm; and

(e) one or more computers detecting, based on the first, second, third and fourth sets of measurements, hand-to-mouth behavior of the subject.

2. The method of claim 1, wherein the four crossed electrodes are positioned on the forearm in substantially a (6,5) pattern.

3. The method of claim 1, wherein:

(a) the four crossed electrodes are positioned on the forearm in substantially a (6,5) pattern; and

(b) the one or more computers perform a Random Forests classification algorithm to detect the hand-to-mouth behavior.

4. The method of claim 1, wherein the four crossed electrodes are positioned on the forearm in substantially a (10,1) pattern.

5. The method of claim 1, wherein:

(a) the four crossed electrodes are positioned on the forearm in substantially a (10,1) pattern; and

(b) the one or more computers perform a K-Nearest Neighbor classification algorithm to detect the hand-to-mouth behavior.

6. The method of claim 1, wherein the one or more computers perform a Random Forests classification algorithm in order to detect the hand-to-mouth behavior.

7. The method of claim 1, wherein, during a training step, the one or more computers:

(a) optimize parameters of a set of classification algorithms; and

(b) select the classification algorithm, out of the set of classification algorithms, that produces the highest mean accuracy of detection of hand-to-mouth behavior.

8. The method of claim 1, wherein:

(a) the one or more computers perform a classification algorithm to detect the hand-to-mouth behavior of the subject; and

(b) training of the classification algorithm is performed on one or more persons other than the subject and is not performed on the subject.

9. A method comprising, in combination:

(a) an electrical bio-impedance (EBI) spectrometer taking EBI measurements, which EBI spectrometer includes four electrodes that are electrically connected to skin of a forearm of a subject, which four electrodes comprise a first IC electrode, a second IC electrode, a first PU electrode and a second PU electrode, the EBI measurements being taken by

(i) driving an alternating electrical current such that the alternating current

(A) undergoes frequency sweeps, and

(B) passes through an electrical circuit that includes the forearm and the first and second IC electrodes, and

(ii) measuring, during the frequency sweeps, voltage difference between the first and second PU electrodes; and

(b) one or more computers detecting, based on the EBI measurements, hand-to-mouth behavior of the subject.

10. The method of claim 9, wherein the four electrodes are crossed.

11. The method of claim 9, wherein the four electrodes are positioned on the forearm in substantially a (6,5) pattern.

12. The method of claim 9, wherein:

(a) the four electrodes are positioned on the forearm in substantially a (6,5) pattern; and

(b) the one or more computers perform a Random Forests classification algorithm to detect the hand-to-mouth behavior.

13. The method of claim 9, wherein the four electrodes are positioned on the forearm in substantially a (10,1) pattern.

14. The method of claim 9, wherein:

(a) the four electrodes are positioned on the forearm in substantially a (10,1) pattern; and

(b) the one or more computers perform a K-Nearest Neighbor classification algorithm to detect the hand-to-mouth behavior.

15. The method of claim 9, wherein:

(a) the one or more computers perform a classification algorithm to detect the hand-to-mouth behavior of the subject; and

(b) training of the classification algorithm is performed on one or more persons other than the subject and is not performed on the subject.

16. The method of claim 9, wherein the method further comprises:

(a) an accelerometer, magnetometer or gyroscope taking additional measurements; and

(b) the one or more computers taking the additional measurements as in input, when detecting hand-to-mouth behavior of the subject.

17. A system comprising:

(a) an electrical bio-impedance (EBI) spectrometer for taking EBI measurements through four electrodes that are electrically attached to a forearm of a subject; and

(b) one or more computers that are programmed

(i) to detect, based on the EBI measurements, hand-to-mouth behavior of the subject, and

(ii) to perform a Random Forests classification algorithm in order to detect the hand-to-mouth behavior.

18. A system comprising:

(a) an electrical bio-impedance (EBI) spectrometer for taking EBI measurements through four electrodes that are electrically attached to a forearm of a subject, and

(b) one or more computers that are programmed

(i) to detect, based on the EBI measurements, hand-to-mouth behavior of the subject,

(ii) to optimize parameters of a set of classification algorithms, and

(iii) to select the classification algorithm, out of the set of classification algorithms, that produces the highest mean accuracy of detection of hand-to-mouth behavior.