Automated Detection of Cigarette Smoking Puffs from Respiration Measurements Amin Ahsan Ali Syed Monowar Hossain Karen Hovsepian Md Mahbubur Rahman Kurt Plarre and Santosh ID: 586943
Download Presentation The PPT/PDF document "mPuff" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
mPuff: Automated Detection of Cigarette Smoking Puffsfrom Respiration Measurements
Amin Ahsan Ali, Syed Monowar Hossain, Karen Hovsepian, Md. Mahbubur Rahman, Kurt Plarre, and Santosh KumarUniversity of Memphis
Puff
detection
IPSN 2012, Beijing, ChinaSlide2
2
Acknowledgements
Saul
ShiffmanPsychologyUniversity of Pittsburgh
Mustafa al’AbsiBehavioral Science
University of MinnesotaEmre
ErtinElectrical & Computer EngineeringThe Ohio State UniversitySlide3
Causes cancer in different organs throughout the body leads to cardiovascular and respiratory diseasesharms reproduction3
Smoking KillsIn US alone
Tobacco is the cause for one of every five deaths
Smokers die 13-14 years younger Public health burden of
$193 billion annuallyRef: Mukdad A. H., et al., Actual Causes of Death in the United States, 2000,
JAMA 2004Slide4
The 6
month quit rate was 3-5% in 5 of the 6 studiesRef: Hughes J.R.,et. al., Shape of the relapse curve and long-term abstinence among untreated smokers, Hughes,, Addiction, 99(1),pp. 29—38, ’04.4
Self-Quitting is NOT an Option
survival curve
relapse curve self-quitters control groups in cessation programsSlide5
Eight (out of 27) divisions at NIH award research grants for smoking cessation programsNIH
alone awards $350+ million annually in smoking research1Still, smoking continues to be prevalentEach day about 3,000 people become new daily smokersSuccess rate of most smoking cessation programs is less than 10%The reason seems to beMost research are self report based that introduces biasNO reliable method to detect smoking and intervene at the right moment
5
Research in Smoking
[1] Estimates of Funding for Various Research, Condition, and Disease Categories (RCDC), NIH (http://report.nih.gov/categorical_spending.aspx)Slide6
6
Current State of the ArtDevices are available that measure and display/store CO levels in a single breath exhaled through a mouthpiece attached to them
piCO+/Micro+ are designed for use as motivational aid
CReSS is used
to
observe smoking patterns and the degree of tobacco intake
Requires compliance from the users
May cause embarrassment using in front of others
piCO
+ and Micro+
CReSS
Pocket
DOES NOT DETECT SMOKING & CANNOT PROVIDE REAL TIME INTERVENTIONSSlide7
AutoSense System for Data Capture in Field
Using AutoSense System , Plarre et. al., showed that stress can be detected reliably from respiration measurements. (IPSN’112)2Plarre et. al., "Continuous Inference of Psychological Stress from Sensory Measurements Collected in the Natural Environment”, IPSN, 2011
Chestband
sensors:
ECG, Respiration, GSR, Ambient & Skin Temp. , AccelerometerArmband sensors:
Alcohol (WrisTAS) , Temp., GSR, Accelerometer
Wireless
Android
G1 Smart
Phone
Ten wireless sensors in two wearable units
Long lifetime (10+ days)
Continuous
Assessment of
Physiology,
Stress,
and Addictive Behaviors in Field
Details about
AutoSense
is available in an ACM Sensys’11
1
Paper
1
Ertin
et. al., "
AutoSense
: Unobtrusively Wearable Sensor Suite for
Inferencing
of Onset, Causality, and Consequences of Stress in the Field,“
SenSys
, 2011.Slide8
8Memphis Study40 daily smokers and social drinkersOne week of
AutoSense wearing in the fieldStress, drinking, smoking, and craving for cigarettes are reportedNational Institute on Drug Abuse (NIDA) Study20 drug users undergoing treatmentTwo lab sessions and 4 weeks of wearing AutoSense in the fieldSmoking, craving, and stress events are marked in the labCraving, stress, and drug usage are reported in the fieldJohns Hopkins Study10 drug users in residential treatmentDrug self-administration sessions are marked in the lab
Ongoing Field Studies with
AutoSenseSlide9
9
ItemMemphis StudyNIDA Study# of participants completed
20
3# of person days worth of data
140 days34 days
Amount of good quality sensor data76,312 min
22,125 min# hours worth of data1,272 hours
369 hours
Avg data collection per day
9
hours/day
10.8
hours/day
# of EMA received
2145 (or 16/day)
253 (7.5/day)
% of EMA answered
94%
91.3%
# of smoking self-report
953 (or 6.8
/day)
116 (or 3.4
/day)
# of drinking self-report
101 (5.6/week)
---
# of craving self-report
---
10
# of drug
used self-report
---
6
Data Collection StatisticsSlide10
Inference of Conversation and Stress from respiration is shown to be possible1,2Most smokers smoke during conversations and with other smokersStress has been found to be a predictor of smoking
10Inferences from Respiration Real time detection of smoking opens up the opportunity for
Analysis of contexts of smoking Finding true predictors of smoking
[1] Md. Mahbubur
Rahman, Amin Ahsan Ali,et. al., "
mConverse: Inferring Conversation Episodes from Respiratory Measurements Collected in the Field," In Proceedings of ACM Wireless Health, San Diego, CA. 2011. [2] K. Plarre
, A. Raij, et. al., "Continuous Inference of Psychological Stress from Sensory Measurements Collected in the Natural Environment," In Proceedings of the 10th ACM/IEEE International Conference on Information Processing in Sensor Networks ( IPSN), Chicago, IL, 2011.Slide11
11
Respiration Chest Band
Concept of
mPuff
Smoking puff detection
Cell Phone Captures SignalSlide12
12Show
puff detection is possible from respiration Open up research on automated smoking detectionIdentify several new
respiration features
Key ContributionsSlide13
13
System Overview Features Computed from each cycle SVM Classifier Trained using carefully labeled data Semi-Supervised Classifier
Utilizing the data collected in natural environmentSlide14
14
Inhalation Duration
Exhalation Duration
Stretch
Features
IERatio
= Inhalation Duration / Exhalation Duration
Respiration Duration = Inhalation + Exhalation DurationSlide15
15
Respiration during Smoking
PUFFS
Compared to Non-puff cycles PUFFs have:
HIGH Stretch
Similar upper & lower parts Relative to neighboring cycles significant change in stretch and other features
Slide16
Feature Statistics16
New Features(1)
Stress
Speaking
Smoking
Unlike Conversation or
Stress, SMOKING PUFF Cycles have
LONGER Stretch
Symmetric across the mid-axis
Significant relative change in stretch and other features Slide17
New Features (2)
SmokingRunning
Unlike RunningSMOKING PUFFS Cycles
Significant relative change in stretch and Exhalation Duration Slide18
1st Difference of Inhalation, Exhalation durations, IE ratio, Stretch
18New FeaturesTOTAL OF 12 New Features
Ratio of Stretch
(Exhalation Duration) of a cycle, C to the avg.
across neighboring
cyclesWe considered a window of 5 cycles centered around C
Upper (Lower) Stretch value
– taking the difference of peak(valley) amplitude from the running mean of valley amplitudesSlide19
ClassificationSupport Vector Machines are used as classifier
Classify respiration cycles to Puffs and Non-puffsPuff data obtained from 10 participants13 sessions Non-puffs come from smoking sessionsStress dataConversation dataPhysical Activity dataArea Under ROC Curve (AUC) metric to assess the performancebecause we have highly imbalanced class sizes.19
ClassificationSlide20
20
TrainingSlide21
21
TestingSlide22
Unlabeled data collected from 4 participantsThey provide self-reports of smoking episodesMay not report at the beginning of smoking episodesDistance to self report is added as a featureAn S3VM is employed
Increases accuracy to 87%22Semi-Supervised ClassifierSlide23
Limitations & Future WorkDuration of smoking session:
6.62 ± 1.66 minutesPuff duration: 1.09 ± 0.53 secondsInter-puff interval: 28.38 ± 14.57 secondsNumber of puffs/smoking session: 12.38 ± 0.92
Consistent with previous lab and field based topology studies
23
Smoking TopologySlide24
Performance of classifier in different confounding events
24LimitationsSlide25
ConclusionNoveltyFirst system to show smoking puff detection is possible from respiration
which can enable scientific studies onHow stress level can predict smoking behaviorHow conversation is related to smoking behaviorNew predictors of smokingIdentifies several new discriminatory features from respiration which may help in detecting other states, e.g.,Eating, and drinking from respirationOngoing WorkUse correlated contexts and robust features to develop an automated smoking session detectorAddress the system reliability and efficiency issues25
ConclusionsSlide26
AutoSense Papers26
Further Reading[System] E. Ertin, N. Stohs, S. Kumar, A. Raij, M. al'Absi, T.Kwon
, S. Mitra, Siddharth Shah, and J. W. Jeong
, “AutoSense: Unobtrusively Wearable Sensor Suite for Inferencing of Onset, Causality, and Consequences of Stress in the Field,” ACM SenSys, 2011
.[Conversation] M. Rahman, A. Ahsan Ali, K. Plarre, M. al'Absi, E. Ertin, and S. Kumar, “mConverse
: Inferring Conversation Episodes from Respiratory Measurements Collected in the Field,” ACM Wireless Health, 2011.[Incentive] M. Mustang, A. Raij, D. Ganesan, S. Kumar and S. Shiffman, “
Exploring Micro-Incentive Strategies for Participant Compensation in High Burden Studies,” to appear in ACM UbiComp, 2011.[Stress] K. Plarre, A. Raij, M. Hossain, A. Ali, M. Nakajima, M. al'Absi, E. Ertin, T.
Kamarck, S. Kumar, M. Scott, D. Siewiorek, A.
Smailagic, and L. Wittmers, “Continuous Inference of Psychological Stress from Sensory Measurements Collected in the Natural Environment
,” ACM IPSN, 2011.
[Privacy]
A. Raij, A.
Ghosh
, S. Kumar and M. Srivastava, “Privacy Risks Emerging from the Adoption of
Inoccuous
Wearable Sensors in the Mobile Environment,” In ACM CHI, 2011
.