Granted Patent B2

US 12,749,576 · App. 17/776,056 · Granted Sep 29, 2026

Inventors: Ke Wang (Charlottesville, VA); Leon S. Farhi (Charlottesville, VA); Boris P. Kovatchev (Charlottesville, VA)

Assignee: University of Virginia Patent Foundation

G16H40/67G16H10/60G16H20/17G16H50/30

Abstract

Claims (502)

1 . A computer-implemented method for compressing continuous glucose monitor (CGM) data of a subject, for use with an interventional device to improve the safety and/or efficacy of therapy for said subject, comprising:

receiving CGM data profiles of said subject;

extracting glycemic risk profiles from said CGM data profiles;

compressing said CGM data profiles into low-dimensional representations using a trained neural network encoder via a cost function weighted by said extracted glycemic risk profiles;

transmitting said low-dimensional encoded representations of CGM profiles to a secondary source;

reconstructing said transmitted low-dimensional encoded representations of CGM profiles to full-dimensional CGM profiles via a trained neural network decoder;

wherein said compressing step and said reconstructing step defines encryption and decryption, respectively;

wherein:

said transmitted reconstructed full-dimensional CGM profiles are configured to allow:

said interventional device to operationally take action on said subject in response to receiving said transmission to improve the safety and/or efficacy of therapy for said subject; and

wherein said improvement of the safety and/or efficacy of therapy for said subject may include one or more of the following:

preventing a hypoglycemic event(s) from occurring in said subject;

preventing a hyperglycemic event(s) from occurring in said subject;

reducing excessive glucose variability occurring in said subject;

reducing postprandial glucose excursions occurring in said subject;

reducing the risk for hypoglycemia;

reducing the risk for hyperglycemia;

optimizing delivery of antidiabetic drugs/compounds (including, insulin); or

lowering glycated hemoglobin (HbA1c).

2 . The method of claim 1 , wherein said interventional device includes one or more of anyone of the following:

insulin pump device;

decision support system;

low glucose suspend system;

connected insulin pens;

automated insulin delivery systems; or

intelligent patch or intelligent transplant.

3 . The method of claim 1 , wherein said secondary source includes one or more of anyone of the following:

local memory;

remote memory; or

display or graphical user interface.

4 . The method of claim 1 , wherein the neural network of the neural network encoder and the neural network of the neural network decoder includes one of the following:

artificial neural network (ANN);

convolutional neural network (CNN); or

recurrent neural networks (RNN).

5 . The method of claim 4 , wherein the CNN is an autoencoder.

6 . The method of claim 1 , wherein the cost function includes one or more of anyone of the following:

maximum likelihood cost function;

absolute deviation cost function; or

mean squared error cost function.

7 . The method of claim 6 , wherein said mean squared error cost function is represented by the following formula:

∑

i

=

1

n

w

i

[

0

,

1

]

(

Y

i

-

Y

^

i

)

2

wherein:

Y i =observed result, for any i=1, 2, . . . , n

Ŷ i =predicted result, for any i=1, 2, . . . , n

w i =the weight of the i th glycemic risk profile, such that 0≤w i ≤1, for any i=1, 2, . . . , n

n=number of profiles,

whereby:

the mean squared cost function is weighted by the corresponding glycemic risk profile or a function of the corresponding glycemic risk profile.

8 . The method of claim 7 , wherein the sum of weights equals 1, wherein

∑

i

=

1

n

⁢

w

i

=

1

.

9 . The method of claim 1 , further comprising, prior to the extraction, preprocessing the received CGM data profiles.

10 . The method of claim 9 , wherein the preprocessing comprises discarding a specified percentage of incomplete CGM data profiles.

11 . The method of claim 10 , wherein the specified percentage of incomplete CGM data profiles includes one of the following:

range of 0 percent and less than about 50 percent;

range of 0 percent and less than about 40 percent;

range of 0 percent and less than about 30 percent;

range of 0 percent and less than about 20 percent;

range of 0 percent and less than about 10 percent; or

about 10 percent.

12 . A system configured for compressing continuous glucose monitor (CGM) data of a subject, for use with an interventional device to improve the safety and/or efficacy of therapy for said subject, comprising:

a computer processor;

a memory configured to store instructions that are executable by the computer processor, wherein said processor is configured to execute the instructions to:

receive CGM data profiles of said subject;

extract glycemic risk profiles from said CGM data profiles;

compress said CGM data profiles into low-dimensional representations using a trained neural network encoder via a cost function weighted by said extracted corresponding glycemic risk profiles;

transmit said low-dimensional encoded representations of CGM profiles to a secondary source;

reconstruct said transmitted low-dimensional encoded representations of CGM profiles to full-dimensional CGM profiles via a trained neural network decoder;

wherein said compression and said reconstruction defines encryption and decryption, respectively;

wherein:

said transmitted reconstructed full-dimensional CGM profiles are configured to allow:

said interventional device to operationally take action on said subject in response to receiving said transmission to improve the safety and/or efficacy of therapy for said subject; and

wherein said improvement of the safety and/or efficacy of therapy for said subject may include one or more of the following:

preventing a hypoglycemic event(s) from occurring in said subject;

preventing a hyperglycemic event(s) from occurring in said subject;

reducing excessive glucose variability occurring in said subject;

reducing postprandial glucose excursions occurring in said subject;

reducing the risk for hypoglycemia;

reducing the risk for hyperglycemia;

optimizing delivery of antidiabetic drugs/compounds (including, insulin); or

lowering glycated hemoglobin (HbA1c).

13 . The system of claim 12 , wherein said interventional device includes one or more of anyone of the following:

insulin pump device;

decision support system;

low glucose suspend system;

connected insulin pens;

automated insulin delivery systems; or

intelligent patch or intelligent transplant.

14 . The system of claim 12 , wherein said secondary source includes one or more of anyone of the following:

local memory;

remote memory; or

display or graphical user interface.

15 . The system of claim 12 , wherein the neural network of the neural network encoder and the neural network of the neural network decoder includes one of the following:

artificial neural network (ANN);

convolutional neural network (CNN); or

recurrent neural networks (RNN).

16 . The system of claim 15 , wherein the CNN is an autoencoder.

17 . The system of claim 12 , wherein the cost function includes one or more of anyone of the following:

maximum likelihood cost function;

absolute deviation cost function; or

mean squared error cost function.

18 . The system of claim 17 , wherein said mean squared error cost function is represented by the following formula:

∑

i

=

1

n

w

i

[

0

,

1

]

(

Y

i

-

Y

^

i

)

2

wherein:

Y i =observed result, for any i=1, 2, . . . , n

Ŷ i =predicted result, for any i=1, 2, . . . , n

w i =the weight of the i th glycemic risk profile, such that 0≤w i ≤1, for any i=1, 2, . . . , n

n=number of profiles,

whereby:

the mean squared cost function is weighted by the corresponding glycemic risk profile or a function of the corresponding glycemic risk profile.

19 . The system of claim 18 , wherein the sum of weights equals 1, wherein

∑

i

=

1

n

⁢

w

i

=

1

.

20 . The system of claim 12 , further comprising, prior to the extraction, preprocessing the received CGM data profiles.

21 . The system of claim 20 , wherein the preprocessing comprises discarding a specified percentage of incomplete CGM data profiles.

22 . The system of claim 21 , wherein the specified percentage of incomplete CGM data profiles includes one of the following:

range of 0 percent and less than about 50 percent;

range of 0 percent and less than about 40 percent;

range of 0 percent and less than about 30 percent;

range of 0 percent and less than about 20 percent;

range of 0 percent and less than about 10 percent; or

about 10 percent.

23 . A computer program product, comprising a non-transitory computer-readable storage medium containing computer-executable instructions for compressing continuous glucose monitor (CGM) data of a subject, for use with an interventional device to improve the safety and/or efficacy of therapy for said subject, said instructions causing the computer to:

receive CGM data profiles of said subject;

extract glycemic risk profiles from said CGM data profiles;

compress said CGM data profiles into low-dimensional representations using a trained neural network encoder via a cost function weighted by said glycemic risk profiles;

transmit said low-dimensional encoded representations of CGM profiles to a secondary source;

reconstruct said transmitted low-dimensional encoded representations of CGM profiles to full-dimensional CGM profiles via a trained neural network decoder;

wherein said compression and said reconstruction defines encryption and decryption, respectively;

wherein:

said transmitted reconstructed full-dimensional CGM profiles are configured to allow:

said interventional device to operationally take action on said subject in response to receiving said transmission to improve the safety and/or efficacy of therapy for said subject; and

wherein said improvement of the safety and/or efficacy of therapy for said subject may include one or more of the following:

preventing a hypoglycemic event(s) from occurring in said subject;

preventing a hyperglycemic event(s) from occurring in said subject;

reducing excessive glucose variability occurring in said subject;

reducing postprandial glucose excursions occurring in said subject;

reducing the risk for hypoglycemia;

reducing the risk for hyperglycemia;

optimizing delivery of antidiabetic drugs/compounds (including, insulin); or

lowering glycated hemoglobin (HbA1c).

24 . The computer program product of claim 23 , wherein said interventional device includes one or more of anyone of the following:

insulin pump device;

decision support system;

low glucose suspend system;

connected insulin pens;

automated insulin delivery systems; or

intelligent patch or intelligent transplant.

25 . The computer program product of claim 23 , wherein said secondary source includes one or more of anyone of the following:

local memory;

remote memory; or

display or graphical user interface.

26 . The computer program product of claim 23 , wherein the neural network of the neural network encoder and the neural network of the neural network decoder includes one of the following:

artificial neural network (ANN);

convolutional neural network (CNN); or

recurrent neural networks (RNN).

27 . The computer program product of claim 26 , wherein the CNN is an autoencoder.

28 . The computer program of claim 23 , wherein the cost function includes one or more of anyone of the following:

maximum likelihood cost function;

absolute deviation cost function; or

mean squared error cost function.

29 . The computer program of claim 28 , wherein said mean squared error cost function is represented by the following formula:

∑

i

=

1

n

w

i

[

0

,

1

]

(

Y

i

-

Y

^

i

)

2

wherein:

Y i =observed result, for any i=1, 2, . . . , n

Ŷ i =predicted result, for any i=1, 2, . . . , n

w i =the weight of the i th glycemic risk profile, such that 0≤w i ≤1, for any i=1, 2, . . . , n

n=number of profiles,

whereby:

the mean squared cost function is weighted by the corresponding glycemic risk profile or a function of the corresponding glycemic risk profile.

30 . The computer program product of claim 29 , wherein the sum of weights equals 1, wherein

∑

i

=

1

n

⁢

w

i

=

1

.

31 . The computer program of claim 23 , further comprising, prior to the extraction, preprocessing the received CGM data profiles.

32 . The computer program product of claim 31 , wherein the preprocessing comprises discarding a specified percentage of incomplete CGM data profiles.

33 . The computer program product of claim 32 , wherein the specified percentage of incomplete CGM data profiles includes one of the following:

range of 0 percent and less than about 50 percent;

range of 0 percent and less than about 40 percent;

range of 0 percent and less than about 30 percent;

range of 0 percent and less than about 20 percent;

range of 0 percent and less than about 10 percent; or

about 10 percent.

34 . A computer-implemented method for compressing continuous glucose monitor (CGM) data of a subject, for use with an automated insulin delivery system to improve the safety and/or efficacy of therapy for said subject, comprising:

receiving CGM data profiles of said subject;

extracting glycemic risk profiles from said CGM data profiles;

compressing said CGM data profiles into low-dimensional representations using a trained neural network encoder via a cost function weighted by said extracted glycemic risk profiles;

transmitting said low-dimensional encoded representations of CGM profiles to a secondary source;

reconstructing said transmitted low-dimensional encoded representations of CGM profiles to full-dimensional CGM profiles via a trained neural network decoder;

wherein said compressing step and said reconstructing step defines encryption and decryption, respectively;

configuring said transmitted reconstructed full-dimensional CGM profiles to allow:

said automatic insulin delivery system, configured for delivering insulin to said subject, to operationally take action on said subject in response to receiving said transmission to improve the safety and/or efficacy of therapy for said subject;

wherein said improvement of the safety and/or efficacy of therapy for said subject may include one or more of the following:

predicting a hypoglycemic event for preventing said hypoglycemic event from occurring in said subject; or

predicting a hyperglycemic event for preventing said hyperglycemic event from occurring in said subject;

automatically adjusting said insulin delivery by said automated insulin delivery system in response to said predicted hypoglycemic event or hyperglycemic event to prevent said hypoglycemic event or hyperglycemic event from occurring in said subject; and

wherein said operational action taken includes delivering said insulin, by said automated insulin delivery system, to said subject according to said adjusted insulin delivery in response to said predicted hypoglycemic event or hyperglycemic event to prevent said hypoglycemic event or hyperglycemic event from occurring in said subject.

35 . The method of claim 34 , wherein said secondary source includes one or more of anyone of the following:

local memory;

remote memory; or

display or graphical user interface.

36 . The method of claim 34 , wherein the neural network of the neural network encoder and the neural network of the neural network decoder includes one of the following:

artificial neural network (ANN);

convolutional neural network (CNN); or

recurrent neural networks (RNN).

37 . The method of claim 36 , wherein the CNN is an autoencoder.

38 . The method of claim 34 , wherein the cost function includes one or more of anyone of the following:

maximum likelihood cost function;

absolute deviation cost function; or

mean squared error cost function.

39 . The method of claim 38 , wherein said mean squared error cost function is represented by the following formula:

∑

i

=

1

n

w

i

[

0

,

1

]

(

Y

i

-

Y

^

i

)

2

wherein:

Y i =observed result, for any i=1, 2, . . . , n

Ŷ=predicted result, for any i=1, 2, . . . , n

w i =the weight of the i th glycemic risk profile, such that 0≤w i ≤1, for any i=1, 2, . . . , n

n=number of profiles,

whereby:

the mean squared cost function is weighted by the corresponding glycemic risk profile or a function of the corresponding glycemic risk profile.

40 . The method of claim 39 , wherein the sum of weights equals 1, wherein

∑

i

=

1

n

⁢

w

i

=

1

.

41 . The method of claim 34 , further comprising, prior to the extraction, preprocessing said received CGM data profiles.

42 . The method of claim 41 , wherein said preprocessing comprises discarding a specified percentage of incomplete CGM data profiles.

43 . The method of claim 42 , wherein the specified percentage of incomplete CGM data profiles includes one of the following:

range of 0 percent and less than about 50 percent;

range of 0 percent and less than about 40 percent;

range of 0 percent and less than about 30 percent;

range of 0 percent and less than about 20 percent;

range of 0 percent and less than about 10 percent; or

about 10 percent.

44 . A system configured for compressing continuous glucose monitor (CGM) data of a subject, for use with an automated insulin delivery system to improve the safety and/or efficacy of therapy for said subject, comprising:

a computer processor;

a memory configured to store instructions that are executable by the computer processor, wherein said processor is configured to execute the instructions to:

receive CGM data profiles of said subject;

extract glycemic risk profiles from said CGM data profiles;

compress said CGM data profiles into low-dimensional representations using a trained neural network encoder via a cost function weighted by said extracted corresponding glycemic risk profiles;

transmit said low-dimensional encoded representations of CGM profiles to a secondary source;

reconstruct said transmitted low-dimensional encoded representations of CGM profiles to full-dimensional CGM profiles via a trained neural network decoder;

wherein said compression and said reconstruction defines encryption and decryption, respectively;

configure said transmitted reconstructed full-dimensional CGM profiles to allow:

said automatic insulin delivery system, configured for delivering insulin to said subject, to operationally take action on said subject in response to receiving said transmission to improve the safety and/or efficacy of therapy for said subject;

wherein said improvement of the safety and/or efficacy of therapy for said subject may include one or more of the following:

predicting a hypoglycemic event for preventing said hypoglycemic event from occurring in said subject; or

predicting a hyperglycemic event for preventing said hyperglycemic event from occurring in said subject;

automatically adjust said insulin delivery by said automated insulin delivery system in response to said predicted hypoglycemic event or hyperglycemic event to prevent said hypoglycemic event or hyperglycemic event from occurring in said subject; and

wherein said operational action taken includes delivery of said insulin, by said automated insulin delivery system, to said subject according to said adjusted insulin delivery in response to said predicted hypoglycemic event or hyperglycemic event to prevent said hypoglycemic event or hyperglycemic event from occurring in said subject.

45 . The system of claim 44 , wherein said secondary source includes one or more of anyone of the following:

local memory;

remote memory; or

display or graphical user interface.

46 . The system of claim 44 , wherein the neural network of the neural network encoder and the neural network of the neural network decoder includes one of the following:

artificial neural network (ANN);

convolutional neural network (CNN); or

recurrent neural networks (RNN).

47 . The system of claim 46 , wherein the CNN is an autoencoder.

48 . The system of claim 44 , wherein the cost function includes one or more of anyone of the following:

maximum likelihood cost function;

absolute deviation cost function; or

mean squared error cost function.

49 . The system of claim 48 , wherein said mean squared error cost function is represented by the following formula:

∑

i

=

1

n

w

i

[

0

,

1

]

(

Y

i

-

Y

^

i

)

2

wherein:

Y i =observed result, for any i=1, 2, . . . , n

Ŷ i =predicted result, for any i=1, 2, . . . , n

w i =the weight of the i th glycemic risk profile, such that 0≤w i ≤1, for any i=1, 2, . . . , n

n=number of profiles,

whereby:

the mean squared cost function is weighted by the corresponding glycemic risk profile or a function of the corresponding glycemic risk profile.

50 . The system of claim 49 , wherein the sum of weights equals 1, wherein

∑

i

=

1

n

⁢

w

i

=

1

.

51 . The system of claim 44 , further comprising, prior to the extraction, preprocessing the received CGM data profiles.

52 . The system of claim 51 , wherein the preprocessing comprises discarding a specified percentage of incomplete CGM data profiles.

53 . The system of claim 52 , wherein the specified percentage of incomplete CGM data profiles includes one of the following:

range of 0 percent and less than about 50 percent;

range of 0 percent and less than about 40 percent;

range of 0 percent and less than about 30 percent;

range of 0 percent and less than about 20 percent;

range of 0 percent and less than about 10 percent; or

about 10 percent.

54 . A computer program product, comprising a non-transitory computer-readable storage medium containing computer-executable instructions for compressing continuous glucose monitor (CGM) data of a subject, for use with an automated insulin delivery system to improve the safety and/or efficacy of therapy for said subject, said instructions causing the computer to:

receive CGM data profiles of said subject;

extract glycemic risk profiles from said CGM data profiles;

compress said CGM data profiles into low-dimensional representations using a trained neural network encoder via a cost function weighted by said glycemic risk profiles;

transmit said low-dimensional encoded representations of CGM profiles to a secondary source;

reconstruct said transmitted low-dimensional encoded representations of CGM profiles to full-dimensional CGM profiles via a trained neural network decoder;

wherein said compression and said reconstruction defines encryption and decryption, respectively;

configure said transmitted reconstructed full-dimensional CGM profiles to allow:

said automatic insulin delivery system, configured for delivering insulin to said subject, to operationally take action on said subject in response to receiving said transmission to improve the safety and/or efficacy of therapy for said subject;

wherein said improvement of the safety and/or efficacy of therapy for said subject may include one or more of the following:

predicting a hypoglycemic event for preventing said hypoglycemic event from occurring in said subject; or

predicting a hyperglycemic event for preventing said hyperglycemic event from occurring in said subject;

automatically adjust said insulin delivery by said automated insulin delivery system in response to said predicted hypoglycemic event or hyperglycemic event to prevent said hypoglycemic event or hyperglycemic event from occurring in said subject; and

said operational action taken includes delivery of said insulin, by said automated insulin delivery system, to said subject according to said adjusted insulin delivery in response to said predicted event to prevent said hypoglycemic event or hyperglycemic event from occurring in said subject.

55 . The computer program product of claim 54 , wherein said secondary source includes one or more of anyone of the following:

local memory;

remote memory; or

display or graphical user interface.

56 . The computer program product of claim 54 , wherein the neural network of the neural network encoder and the neural network of the neural network decoder includes one of the following:

artificial neural network (ANN);

convolutional neural network (CNN); or

recurrent neural networks (RNN).

57 . The computer program product of claim 56 , wherein the CNN is an autoencoder.

58 . The computer program of claim 54 , wherein the cost function includes one or more of anyone of the following:

maximum likelihood cost function;

absolute deviation cost function; or

mean squared error cost function.

59 . The computer program of claim 58 , wherein said mean squared error cost function is represented by the following formula:

∑

i

=

1

n

w

i

[

0

,

1

]

(

Y

i

-

Y

^

i

)

2

wherein:

Y i =observed result, for any i=1, 2, . . . , n

Ŷ i =predicted result, for any i=1, 2, . . . , n

w i =the weight of the i th glycemic risk profile, such that 0≤w i ≤1, for any i=1, 2, . . . , n

n=number of profiles,

whereby:

the mean squared cost function is weighted by the corresponding glycemic risk profile or a function of the corresponding glycemic risk profile.

60 . The computer program product of claim 59 , wherein the sum of weights equals 1, wherein

∑

i

=

1

n

⁢

w

i

=

1

.

61 . The computer program of claim 54 , further comprising, prior to the extraction, preprocessing the received CGM data profiles.

62 . The computer program product of claim 61 , wherein the preprocessing comprises discarding a specified percentage of incomplete CGM data profiles.

63 . The computer program product of claim 62 , wherein the specified percentage of incomplete CGM data profiles includes one of the following:

range of 0 percent and less than about 50 percent;

range of 0 percent and less than about 40 percent;

range of 0 percent and less than about 30 percent;

range of 0 percent and less than about 20 percent;

range of 0 percent and less than about 10 percent; or

about 10 percent.

Assignments (2)

Continuity (2)

References Cited (115)