METHOD OF EARLY DETECTION OF MULTIPLE SCLEROSIS

• US 20180092591A1
• Filed: 09/30/2016
• Published: 04/05/2018
• Est. Priority Date: 09/30/2016
• Status: Active Grant

• Current Assignee: Recursion Pharmaceuticals, Inc.
• Original Assignee: Vium, Inc. (Recursion Pharmaceuticals, Inc.)
• Inventors: Ford, Daniel J., Schaevitz, Laura
• Granted Patent: US 10,244,979 B2
• Time in Patent Office: Days
• CPC Class Codes: A61B 2503/40 Animals
• CPC Class Codes: A61B 5/1118 Determining activity level
• CPC Class Codes: A61B 5/4076 Diagnosing or monitoring particular conditions of the nervous system A61B5/4821, A61B5/4824 take precedence
• CPC Class Codes: A61B 5/7257 using Fourier transforms
• CPC Class Codes: A61B 5/7275 Determining trends in physiological measurement data; Predicting development of a medical condition based on physiological measurements, e.g. determining a risk factor
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/30 for calculating health indices; for individual health risk assessment

Abstract

A method of early detection of multiple sclerosis (MS) in an animal in a vivarium is described. Animal activity data is collected at multiple times during the night. Sequential time regions of the night are identified as high-activity, activity-drop, or low-activity regions. Embodiments are described to quantify a drop, during the night, of an animal'"'"'s activity level. These quantified activity-drop scalars for consecutive nights are accumulated in an animal health dataset. Then, a health detection function is applied to this dataset that, in response to the level of activity change and the speed of activity change, predicts or detects MS in the animal. One embodiment quantifies an activity-drop by fitting straight-line curves through the data in the three nightly regions. Another embodiment uses a Fourier transform on a circle and a linear combination. Another embodiment compares areas under data curves in the regions. Animals may be housed in cages with other animals.

Claims

1. A method of early detection of multiple sclerosis (MS) in an animal in a vivarium comprising the steps of:

(a) selecting a single “

activity drop metric”

with an associated scalar “

activity drop value”

;

(b) selecting an “

MS health detection function”

whose input comprises an “

animal health dataset” and

whose output comprises a likelihood scalar representing the likelihood that the animal has MS;

(c) collecting a nightly activity scalar of the animal repeatedly and continually for a night and placing the nightly activity scalar into a set of “

nightly activity data”

;

(d) identifying automatically three consecutive time regions in the nightly activity data;

a “

high activity region,”

an “

activity drop region,” and

a “

low activity region”

;

(e) applying the activity drop metric to the three consecutive regions, generating a nightly activity drop value;

(f) adding the nightly activity drop value into the animal health dataset, wherein the animal health dataset comprises the resulting nightly activity drop values;

(g) applying the MS health detection function to the animal health dataset, generating a likelihood scalar for each iteration;

(h) iterating steps (c) through (g) for sequential nights until a terminating condition is reached;

wherein the early detection of MS comprises the likelihood scalars from step (g).

2. The method claim 1 wherein:

the MS health detection function further comprises an output that is a confidence value;

wherein in step (g) the MS health detection function further generates a confidence value for the animal; and

wherein the early detection of MS in the animal further comprises the confidence values from steps (g).

3. The method claim 1 wherein comprising the additional steps:

(i) locating an intermediate point within the activity drop region in the night;

(j) computing a first best-fit straight line curve through a region surrounding the intermediate point;

(k) computing a second best-fit straight line curve for the high-activity region;

(l) computing a third best-fit straight-line curve for the low-activity region;

(m) computing a first point midway between (the intersection point of the first and second best-fit straight lines curves, a “

second point”

) and (the intersection point of the first and third best-fit straight lines curves, a “

third point”

);

wherein steps (i) through (m) are the activity drop metric;

wherein the high activity region is updated to the region from the start of the night to the second point; and

the low activity region is updated to begins at the third point; and

the activity drop region is updated to the region between the second and third points; and

wherein the activity value of the first point is the activity drop value for the night.

4. The method claim 3 wherein comprising the additional steps:

(n) iterating steps (i) through (m) until a terminating condition is reached;

wherein the new intermediate point in step (i) for each iteration is the activity drop value of the prior iteration; and

wherein the region surrounding the intermediate point is the activity drop region from the prior iteration; and

wherein the first, second and third best-fit straight line curves are recomputed each iteration;

wherein the first, second and third points are recomputed each iteration; and

wherein the activity value of the first point of the last iteration is the activity drop value for the night.

5. The method claim 1 wherein comprising the additional steps:

(o) averaging the nightly activity data for a first initially predetermined period beginning at the beginning of the night, the “

starting region;

”

(p) averaging the nightly activity data for a second initially predetermined period after the starting region, the “

ending region;

”

(q) computing a first best-fit straight line curve from end of the starting region to the start of the ending region;

(r) computing a second best-fit straight line curve through the starting region;

(s) computing a third best-fit straight line curve through the ending region;

(t) computing a first point midway between (the intersection point of the first and second best-fit straight lines curves, a “

second point”

) and (the intersection point of the first and third best-fit straight lines curves, a “

third point”

);

wherein steps (o) through (t) are the activity drop metric;

wherein the high activity region is updated to be the region from the beginning of the night to the second point;

the low activity region is updated to begin at the third point, and the activity drop region is updated to be the region between the second and third points; and

wherein the activity value of the first point is the activity drop value for the night.

6. The method claim 5 wherein comprising the additional steps:

(u) iterating steps (o) through (t) until a terminating condition is reached;

wherein the new starting region for each iteration is the high activity period of the prior iteration;

wherein the new ending region for each iteration is the low activity period of the prior iteration;

wherein the first, second and third best-fit straight line curves are recomputed each iteration;

wherein the first, second and third points are recomputed each iteration; and

wherein the activity value of the first point of the last iteration is the activity drop value for the night.

7. The method claim 1 wherein:

the MS health detection function is responsive to the activity drop value for at least one night.

8. The method claim 7 wherein:

the MS health detection function is computed, at least in part, using the additional method steps;

(v) executing at least a subset of method steps (a) through (h) repeatedly using different animals and observing the actual MS onset of the each different animal;

(w) modifying the MS health detection function using a best-fit method such that the MS health detection function best determines the observed actual MS onset of the each different animal.

9. The method claim 1 wherein:

the MS health detection function is binary valued.

10. A method of early detection of multiple sclerosis (MS) in an animal in a vivarium comprising the steps of:

(x) selecting a single “

linear combination coefficient set”

that generates a scalar linear combination value when applied responsively to a set of discreet transform values;

(y) selecting a “

MS health detection function”

whose input comprises an animal health dataset and whose output comprises a likelihood scalar representing the likelihood that the animal has MS;

(z) collecting an activity scalar of the animal repeatedly and continually for a night;

a set of “

nightly activity data”

;

(aa) computing a Fourier transform on a circle responsive to the nightly activity data, generating a set of discreet transform values;

(bb) applying the linear combination coefficient set responsively to the set of discreet transform values;

wherein the resulting scalar linear combination value is a nightly activity value;

(cc) adding the nightly activity value into the animal health dataset;

wherein the animal health dataset comprises the nightly activity values;

(dd) applying the MS health detection function to the animal health dataset, generating an animal likelihood scalar for the animal;

(ee) iterating steps (z) through (dd) for sequential nights until a terminating condition is reached;

wherein the early detection of MS in the animal comprises the animal likelihood scalars from steps (dd).

11. The method claim 10 wherein:

the MS health detection function further comprises an output that is a confidence value;

wherein in step (dd) the MS health detection function further generates a confidence value for the animal; and

wherein the early detection of MS in the animal further comprises the confidence values from steps (dd).

12. A method of early detection of multiple sclerosis in an animal in a vivarium comprising the steps of:

(ff) selecting a single “

LASSO metric”

with an associated scalar “

activity drop value”

;

(gg) selecting an “

MS health detection function”

whose input comprises an animal health dataset and whose output comprises a likelihood scalar representing the likelihood that the animal has MS;

(hh) collecting an activity scalar of the animal repeatedly and continually for a night;

a set of “

nightly activity data”

;

(ii) computing a LASSO best-fit of a first piece-wise linear function to the nightly activity data, generating a LASSO L0, L1 and L2;

(jj) applying the LASSO metric to the generated LASSO L0, L1 and L2, generating a nightly activity drop value;

(kk) adding the nightly activity drop value into the “

animal health dataset,”

wherein the animal health dataset comprises the resulting nightly activity drop values;

(ll) applying the MS health detection function to the animal health dataset;

(mm) iterating steps (hh) through (ll) for sequential nights until a terminating condition is reached;

wherein the early detection of MS comprises the likelihood scalars from step (ll).

13. The method of claim 12 wherein:

the first step-wise linear function comprises two linear pieces; and

wherein the nightly activity drop value comprises a linear combination of LASSO L0, L1 and L2.

14. The method of claim 12 wherein:

the first step-wise linear function comprises of three linear pieces; and

wherein the nightly activity drop value comprises a linear combination of LASSO L0, L1 and L2

15. A method of early detection of multiple sclerosis in an animal in a vivarium comprising the steps of:

(nn) selecting a “

MS health detection function”

whose input comprises an animal health dataset and whose output comprises a likelihood scalar representing the likelihood that the animal has MS;

(oo) collecting an activity scalar of the animal repeatedly and continually for a night;

generating a set of “

nightly activity data”

;

(pp) computing a baseline low-activity level responsive to the nightly activity data;

(qq) modifying the set of nightly activity data by subtracting the baseline low-activity level from the elements of the set;

(rr) identifying automatically two sequential regions for each night in the nightly activity data;

a “

high activity region,” and

a “

low activity region”

;

(ss) computing a high activity value for each night equal to the integration of values in the set of nightly activity data taken within the time interval of the high activity region;

(tt) adding the high activity value into the “

animal health dataset,”

wherein the animal health dataset comprises the resulting nightly high activity values;

(uu) applying the MS health detection function to the animal health dataset;

(vv) iterating steps (oo) through (uu) for sequential nights until a terminating condition is reached;

wherein the early detection of MS comprises the likelihood scalars from step (uu).

16. The method of claim 15 wherein:

(ww) computing a low activity value for each night equal to the integration of values in the set of nightly activity data within the time interval of the low activity region;

(xx) adding the low activity value into the “

animal health dataset,”

wherein the animal health dataset comprises the resulting nightly activity drop values;

wherein step (ww) is performed before or after step (ss);

wherein step (xx) is performed before or after step (tt); and

wherein early MS health detection function is responsive to both the high activity values and the low activity values in the animal health dataset.

17. The method of claim 15 wherein:

the MS health detection function is further responsive to an activity level of the animal prior to an induction date for the animal into a study.

18. The method of claim 17 wherein:

the MS health detection function is normalized responsive to the activity level of the animal prior to an induction date for the animal into the study.