Ex Parte El Solh

Patent Trial and Appeal Board

Jun 14, 2016

11840889 (P.T.A.B. Jun. 14, 2016)

UNITED STA TES p A TENT AND TRADEMARK OFFICE
APPLICATION NO. FILING DATE
111840,889 08/17/2007
26712 7590 06/16/2016
HODGSON RUSS LLP
THE GUARANTY BUILDING
140 PEARL STREET
SUITE 100
BUFFALO, NY 14202-4040
FIRST NAMED INVENTOR
AliElSolh
UNITED STATES DEPARTMENT OF COMMERCE
United States Patent and Trademark Office
Address: COMMISSIONER FOR PATENTS
P.O. Box 1450
Alexandria, Virginia 22313-1450
www .uspto.gov
ATTORNEY DOCKET NO. CONFIRMATION NO.
011520.00633 1123
EXAMINER
LIN, JERRY
ART UNIT PAPER NUMBER
1631
NOTIFICATION DATE DELIVERY MODE
06/16/2016 ELECTRONIC
Please find below and/or attached an Office communication concerning this application or proceeding.
The time period for reply, if any, is set in the attached communication.
Notice of the Office communication was sent electronically on above-indicated "Notification Date" to the
following e-mail address( es):
ipdocketing@hodgsonruss.com
PTOL-90A (Rev. 04/07)
UNITED STATES PATENT AND TRADEMARK OFFICE
BEFORE THE PATENT TRIAL AND APPEAL BOARD
Ex parte ALI EL-SOHL 1
Appeal2014-000169
Application 11/840,889
Technology Center 1600
Before DONALD E. ADAMS, JEFFREY N. FREDMAN, and
JOHN G. NEW, Administrative Patent Judges.
NEW, Administrative Patent Judge.
DECISION ON REQUEST FOR REHEARING
1 Appellant states the real party-in-interest is The Research Foundation of
State University of New York. App. Br. 2.
Appeal2014-000169
Application 11/840,889
Appellant filed a Request for Rehearing (hereinafter "Request") under
37 C.F.R. § 41.52(a)(l) for reconsideration of our Decision of March 30,
2016 (hereinafter "Decision").
The Decision affirmed the Examiner's Final Rejection of claims 1-20
as unpatentable under 35 U.S.C. § 103(a) as follows:
1. Claims 1-20 stand rejected as unpatentable under 35 U.S.C.
Decision 2.
§ 103(a) as being obvious over the combination of Michael J.
Banner et al., Power of Breathing Determined Noninvasively
with Use of an Artificial Neural Network in Patients with
Respiratory Failure, 34(4) CRIT. CARE MED. 1052-59 (2006)
("Banner"), Zoe Oliver & Victor Hoffstein, Predicting Effective
Continuous Positive Airway Pressure, 117(4) CHEST 1061---64
(2000) ("Oliver"), and Statnikov et al. (US 2011/0246403 Al,
October 6, 2011) ("Statnikov").
In their Request, Appellant seeks reconsideration of our Decision
affirming the Examiner's rejection of claims 1-20. Specifically, Appellant
argues the panel misapprehended or overlooked certain arguments presented
by Appellant. Request 1.
Claim 1 is representative of the claims on appeal and recites:
1. A method for evaluating artificial neural networks (ANNs) for
predicting a continuous positive airway pressure level,
compnsmg:
a) collecting information from human subjects to provide
a dataset that includes an entry for each human subject wherein
2
Appeal2014-000169
Application 11/840,889
each entry includes a neck circumference, a body mass index, an
apnea-hypopnea index, and an actual effective pressure value;
b) randomly separating the entries of the dataset into n
subsets;
c) creating n unique training sets, wherein n is an integer
greater than 1, and wherein each training set has n-1 of the n
subsets;
d) creating n ANNs using a processor, each of the ANNs
being created from a different one of the training sets and
configured to provide a predicted effective pressure value from
the neck circumference, body mass index and apnea-hypopnea
index entries of the dataset;
e) calculating a mean squared error for each of the n ANN s
by comparing the predicted effective pressure value to the actual
effective pressure value for each of the entries;
f) calculating an average by averaging the mean squared
errors;
g) determining which of the mean squared errors is closest
to the average; and,
h) selecting the ANN corresponding to the mean squared error
that is determined to be closest to the average.
See App. Br. 11.
ANALYSIS
Issue 1
Appellant contends that, in reaching its Decision, the Board
misapprehended and/or overlooked Banner's alleged teaching of but a single
3
Appeal2014-000169
Application 11/840,889
training set. Request 2. 2 Appellant notes the Board relied upon Banner's
teaching; "[V]alid data were divided into 2-min segments, with no more than
one 2-min segment per valid ventilator setting. Patients through February
2003 (115 patients, 206 valid 2-min segments) were used for training the
neural network." Id. (quoting Banner 1059). According to Appellant, the
quoted passage of Banner describes only a single training set of data, where
this single data set is made up of 206 valid 2-min segments from 115
patients. Id. Appellant further notes that Banner introduces the Appendix
by describing the three data sets of "training, cross-validation, and testing."
Id. (quoting Banner 1059, col. 2).
Appellant also notes the Board further cites Banner as teaching "data
sets from 200 patients, divided into 2-minute stable segments." Request 2
(citing Decision 9 (quoting Banner 1054)). Appellant argues this portion of
Banner describes a single training set which is made up of data from 200
patients-not 200 training sets. Id. Appellant contends this data from 200
patients is the same data that was described in Banner's Appendix (i.e., the
Board's earlier cite of p. 1059). In particular, Banner describes that data
from 200 patients were divided into a set of 115 patients ("used for training
the neural network") and a set of 85 patients ("used-for cross-validation").
Id. (citing Banner 1059, col. 3).
We do not agree. Appellant attempts to persuade us that the passage
of Banner reciting: "[V]alid data were divided into 2-min segments, with no
more than one 2-minute segment per valid ventilator setting. Patients
through February 2003 (115 patients, 206 valid 2-min segments) were used
2 Appellant's Request does not supply page numbers. We consequently
refer to the number of the page in the brief, with the cover page acting as
page "1 ".
4
Appeal2014-000169
Application 11/840,889
for training the neural network" can only be reasonably interpreted as
teaching a single training data set. However, Appellant's Specification
discloses:
A method for evaluating artificial neural networks
("ANN s") for predicting continuous positive airway pressure is
depicted generally by flowcharts 10, 11 and 13 of Figs. 1-3. The
method includes collecting information 12 from human subjects
to provide entries 14 for a dataset 16.
The dataset 16 may be randomly separated in process 18
into n subsets 20, where n is an integer. Then, n training sets 24
are created as is illustrated as process 22 in Fig. 1. Each training
set 24 may have n-1 of then subsets 20.
Spec. i-fi-f 17, 18. The Specification thus discloses assembling a large dataset
and subdividing it into subunits to be used as "training sets 24." Banner
similarly teaches a single large dataset, divided into 2-minute intervals, with
no more than one 2-min segment per valid ventilator setting. We conclude
that a person of ordinary skill would find it obvious, in light of the teachings
of Banner, that a single large set of collected data can be subdivided into a
number of subsets (i.e., 2-minute intervals) to be used as training sets for
ANNs.
Issue 2
Appellant further contends the Board misapprehended and/ or
overlooked the teachings of Banner in concluding that Banner's description
of a "wide variety of ANN s" teaches that n ANN s are created, each created
from a different one of the n training sets. Request 2. Appellant points to
the Board's quoting Banner's teaching that: "[a] wide variety of ANN
architectures were studied and compared on the basis of their complexity
5
Appeal2014-000169
Application 11/840,889
and performance." Id. (quoting Decision 9). Appellant contends that,
although no conclusion is expressly drawn based on the citation, the
implication is that this reference to multiple ANN s in Banner meets the
claim limitation described in Step B.1:
[C]reating n ANNs (where n is an integer greater than 1) using a
processor, each of the ANNs being created from a different one
of the training sets and configured to provide a predicted
effective pressure value from the neck circumference, body mass
index and apnea-hypopnea index entries of the dataset.
See App. Br. 7.
Appellant points out that the claim limitation requires more than the
use of multiple ANNs, it also requires that the number of ANNs (n) is the
same as the number of training sets (n ), and that each ANN is created from a
different one of the training sets (where n is greater than 1 from Step A).
Request 3. Appellant argues Banner does not disclose the use of n ANN s
corresponding ton training sets. Id. Rather, Appellant contends, Banner
cannot teach such a correspondence, because Banner teaches only a single
training set.
We are not persuaded. Banner explicitly teaches the use of multiple
architectures of ANN s on the basis of their complexity and performance,
using the mean squared error as the criterion used for training and selecting
the neural network. Banner 1059. Moreover, as we have related supra, we
do not agree with Appellant's contention that there is only a single data set.
We consequently reaffirm our conclusion that one of ordinary skill in the art
would realize that it would be obvious, in light of the teachings of Banner,
that that n ANN s are created, each created from a different one of the n
training sets.
6
Appeal2014-000169
Application 11/840,889
Consequently, we have granted Appellant's Request for Rehearing to
the extent that we have considered Appellant's arguments, but we decline to
provide Appellant with the requested relief.
DENIED
7