![Using SpiroSmart
]](https://image.slidesharecdn.com/phoneassensor2014-140131144943-phpapp02/85/Phone-As-A-Sensor-Technology-mHealth-and-Chronic-Disease-73-320.jpg)
![Using SpiroSmart
]](https://image.slidesharecdn.com/phoneassensor2014-140131144943-phpapp02/85/Phone-As-A-Sensor-Technology-mHealth-and-Chronic-Disease-74-320.jpg)
![Using SpiroSmart
]](https://image.slidesharecdn.com/phoneassensor2014-140131144943-phpapp02/85/Phone-As-A-Sensor-Technology-mHealth-and-Chronic-Disease-75-320.jpg)
Phone As A Sensor Technology: mHealth and Chronic Disease
- 1. phone-as-a-sensor technology: mhealth and chronic disease > slide to unlock eric c. larson | eclarson.com Assistant Professor Computer Science and Engineering
- 2. iPhone 9 M Now with iColon. > slide to unlock
- 4. arch algorithms OS databases & data mining software Comp Sci. & Engr. languages AI & robotics networking symbolic computing
- 5. arch algorithms OS databases & data mining software mobile Comp Sci. & Engr. languages AI & robotics networking symbolic computing
- 6. arch algorithms OS databases & data mining software mobile Comp Sci. & Engr. mhealth languages AI & robotics networking symbolic computing
- 7. arch algorithms OS databases & data mining software Comp Sci. & Engr. mhealth languages AI & robotics networking symbolic computing mobile
- 9. the promise of mhealth:
- 10. the promise of mhealth: revolutionize medicine
- 11. the promise of mhealth: revolutionize medicine eliminate doctor visits
- 12. the promise of mhealth: revolutionize medicine eliminate doctor visits remote / automatic diagnosis
- 13. the promise of mhealth: revolutionize medicine eliminate doctor visits remote / automatic diagnosis equalize developing countries
- 14. the promise of mhealth: revolutionize medicine eliminate doctor visits remote / automatic diagnosis equalize developing countries
- 15. the promise of mhealth: revolutionize medicine eliminate doctor visits remote / automatic diagnosis equalize developing countries
- 16. heart rate zombie run fitness trainer stress check current mhealth glucose buddy
- 17. heart rate zombie run stress check current mhealth 43,000 apps for health on the app store fitness trainer glucose buddy
- 18. heart rate zombie run stress check current mhealth 43,000 apps for health on the app store 96% are for calorie counting & exercise fitness trainer glucose buddy
- 19. heart rate zombie run stress check current mhealth 43,000 apps for health on the app store 96% are for calorie counting & exercise 4% are remote monitoring fitness trainer glucose buddy
- 21. consider physician’s needs connecting with patient
- 22. consider physician’s needs connecting with patient tracking baselines
- 23. consider physician’s needs connecting with patient tracking baselines personalized trending data
- 24. consider physician’s needs connecting with patient tracking baselines personalized trending data managing chronic disease
- 25. consider physician’s needs connecting with patient tracking baselines personalized trending data managing chronic disease 30% of all US healthcare spending is on chronic disease
- 26. mhealth and chronic disease management
- 27. mhealth and chronic disease management compliance? cost? doctor patient? data reliability?
- 29. compliance
- 30. compliance
- 32. phone as a sensor baseline sensor quantity
- 33. phone as a sensor baseline embedded sensors sensor quantity
- 34. phone as a sensor baseline embedded sensors sensor estimated quantity processing
- 36. accelerometer gyroscope magnetometer /compass dual camera / flash 1+ microphones proximity sensor capacitive sensor gps motorized actuator wireless antenna (s)
- 37. compliance++; cost--; dr_pat *= 10; accelerometer gyroscope magnetometer /compass dual camera / flash 1+ microphones proximity sensor capacitive sensor gps motorized actuator wireless antenna (s)
- 38. accelerometer gyroscope magnetometer /compass dual camera / flash 1+ microphones compliance++; proximity sensor cost--; capacitive sensor dr_pat *= 10; gps motorized actuator data reliability? wireless antenna (s)
- 39. what can the mobile phone sense with clinical accuracy?
- 44. spirometer device that measures amount of air inhaled and exhaled.
- 50. volume-time graph FVC volume FEV1 time FEV1: Forced Expiratory Volume in 1 second FVC: Forced Vital Capacity
- 51. volume-time graph FVC volume FEV1 1 sec. time FEV1: Forced Expiratory Volume in 1 second FVC: Forced Vital Capacity
- 52. volume-time graph FEV1% = FEV1/FVC FVC volume FEV1 1 sec. time FEV1: Forced Expiratory Volume in 1 second FVC: Forced Vital Capacity
- 53. FEV1% = FEV1/FVC FEV1: Forced Expiratory Volume in 1 second FVC: Forced Vital Capacity
- 54. FEV1% = FEV1/FVC > 80% healthy 60 - 79% mild 40 - 59% moderate < 40% severe FEV1: Forced Expiratory Volume in 1 second FVC: Forced Vital Capacity
- 57. flow-volume graph PEF flow 1 sec. FEV1 FVC volume PEF: Peak Expiratory Flow FEV1: Forced Expiratory Volume in 1 second FVC: Forced Vital Capacity
- 60. obstructive diseases ! resistance in air path leads to reduced air flow
- 61. obstructive diseases ! resistance in air path leads to reduced air flow
- 62. restrictive diseases ! lungs are unable to pump enough air and pressure
- 63. restrictive diseases ! lungs are unable to pump enough air and pressure
- 67. home spirometry
- 68. home spirometry ! faster detection rapid recovery trending
- 69. challenges with home spirometry high cost barrier patient compliance less coaching limited integration
- 70. SpiroSmart availability cost portability more effective coaching interface integrated uploading
- 71. Using SpiroSmart
- 72. Using SpiroSmart
- 76. Using SpiroSmart
- 77. Using SpiroSmart
- 81. audio
- 85. audio flow features measures regression curve regression Flow(L/s) lung function 10 10 5 5 0 0 1 0 40 Volume(L) Volume(L) FEV1 FVC PEF Flow(L/s) 15 15 2 Volume(L) 4 2 Volume(L) 1 3 3 3 4 2 1 3 0 20 1 2 4 6 time(s) 8 10 4
- 87. flow features estimation 0.5 0 −0.5 −1 0 1 2 3 4 5 6 time(s) source vocal tract 2500 frequency(Hz) amplitude 1 2000 1500 1000 500 0 output 7
- 88. flow features estimation lpc8raw amplitude 1 0.5 0.5 0 −0.5 0 flow estimation features amplitude −0.5 1 2 3 4 5 6 7 time(s) 0.5 0 −0.5 resonance tracking lpc8raw 1 2 3 4 5 6 time(s) source vocal tract 2500 2000 1500 1000 500 0 −1 1 0 7 amplitude −1 0 envelope detection −1 10 frequency(Hz) amplitude 1 output 1 2 3 4 5 6 7 time(s) 0.5 0 −0.5 −1 0 auto-regressive estimate 1 2 3 4 time(s) 5 6 7
- 89. measures regression 8 flow(L/s) 6 4 2 ground truth 0 −1 0 1 2 3 4 5 time(s) feature value 0.4 0.3 0.2 0.1 0 feature 1 0 1 2 4 5 time(s) 0.4 feature value 3 0.3 0.2 0.1 feature 2 0 −1 0 1 2 time(s) 3 4 5
- 90. 8 6 flow(L/s) measures regression 7.1 4 2 PEF features ground truth 0 −1 0 1 2 3 4 5 time(s) feature value 0.4 0.3 0.2 0.1 0 feature 1 0 1 2 4 5 time(s) 0.4 feature value 3 0.3 0.2 0.1 feature 2 0 −1 0 1 2 time(s) 3 4 5
- 91. 8 6 flow(L/s) measures regression 7.1 4 2 PEF features ground truth 0 −1 0 1 2 3 4 5 time(s) feature value 0.4 0.3 0.2 0.1 0 feature 1 0 1 2 4 5 time(s) 0.4 feature value 3 0.3 0.2 0.1 feature 2 0 −1 0 1 2 time(s) 3 4 5 0.35 0.33
- 92. 8 6 flow(L/s) measures regression 7.1 3.2 4 2 PEF features ground truth 0 −1 0 1 2 3 4 5 time(s) feature value 0.4 0.3 0.2 0.1 0 FEV1 features feature 1 0 1 2 3 4 5 time(s) 0.4 feature value 0.35 0.33 0.3 0.2 0.1 feature 2 0 −1 0 1 2 time(s) 3 4 5
- 93. 8 6 flow(L/s) measures regression 7.1 3.2 4 2 PEF features ground truth 0 −1 0 1 2 3 4 5 time(s) feature value 0.4 0.3 0.2 0.1 0 FEV1 features feature 1 0 1 2 3 4 0.12 0.17 5 time(s) 0.4 feature value 0.35 0.33 0.3 0.2 0.1 feature 2 0 −1 0 1 2 time(s) 3 4 5
- 94. measures regression FEV1 features PEF features
- 95. measures regression FEV1 features PEF features bagged decision tree bagged decision tree
- 96. measures regression FEV1 features PEF features bagged decision tree bagged decision tree output output
- 97. curve regression feature 1 0.3 0.2 0.1 0 1 3 4 5 time(s) 0.4 feature value 2 feature N 0.3 0.2 0.1 0 ... 0 −1 ... windowed machine learning regression feature value 0.4 0 1 2 time(s) 3 4 5
- 98. curve regression feature 1 0.3 0.2 0.1 0 1 3 4 5 time(s) 0.4 feature value 2 ... 0 −1 ... feature N 0.3 0.2 0.1 0 8 6 flow(L/s) windowed machine learning regression feature value 0.4 4 2 0 0 1 2 3 4 5 time(s) curve output
- 99. curve regression 0.4 feature value feature value 0.4 0.3 0.2 0.1 0 −1 0 1 2 3 time(s) bagged decision tree 4 0.3 0.2 0.1 0 5 0 1 2 time(s) 3 4 5
- 100. curve regression 0.4 feature value feature value 0.4 0.3 0.2 0.1 0 −1 0 1 2 3 time(s) bagged decision tree 4 0.3 0.2 0.1 0 5 0 1 2 3 4 time(s) CRF 5
- 101. curve regression 0.4 feature value feature value 0.4 0.3 0.2 0.1 0 −1 0 1 2 3 4 0.3 0.2 0.1 0 5 0 1 2 bagged decision tree ! 6 flow(L/s) 4 CRF 8 4 2 0 −1 3 time(s) time(s) 0 1 2 time(s) 3 4 5 5
- 102. study design x3 x3
- 103. study enrollment
- 104. study enrollment study a participants 52 18-75 years old, mostly healthy
- 105. study enrollment study a participants 52 18-75 years old, mostly healthy study b participants 10 12-17 years old, mixed healthy/abnormal
- 106. study enrollment study a participants 52 18-75 years old, mostly healthy study b 10 12-17 years old, mixed healthy/abnormal study c participants 56 10-69 years old, mostly abnormal enrolled by pulmonologists participants
- 113. can SpiroSmart curves be used for diagnosis?
- 114. survey
- 115. survey • normal/abnormal subjects curves
- 116. survey • normal/abnormal subjects curves • 5 pulmonologists
- 117. survey • normal/abnormal subjects curves • 5 pulmonologists • unaware if from SpiroSmart / spirometer
- 124. results restrictive inadequate normal minimal obstructive mild obstructive moderate obstructive severe obstructive
- 125. results restrictive inadequate normal minimal obstructive mild obstructive moderate obstructive severe obstructive identical 64%
- 126. results restrictive inadequate normal minimal obstructive mild obstructive moderate obstructive severe obstructive one off 10% identical 64%
- 127. results restrictive inadequate normal minimal obstructive mild obstructive moderate obstructive severe obstructive one off 10% identical 64%
- 128. results restrictive inadequate normal minimal obstructive mild obstructive moderate obstructive severe obstructive false positive 14% one off 10% identical 64%
- 129. results false negative 4% restrictive inadequate normal minimal obstructive mild obstructive moderate obstructive severe obstructive false positive 14% one off 10% identical 64%
- 130. results error 8% false negative 4% restrictive inadequate normal minimal obstructive mild obstructive moderate obstructive severe obstructive false positive 14% one off 10% identical 64%
- 131. results error 8% false negative 4% restrictive inadequate normal minimal obstructive mild obstructive moderate obstructive severe obstructive false positive 14% ! ! abnormal vs normal one off 96% identical 10% 64%
- 132. appropriate for trending and screening
- 134. global health non-profit patient and doctors
- 135. global health non-profit patient and doctors pharmaceutical drug trials
- 152. neonatal jaundice in the US kernicterus: 21 hazardous jaundice: 1158 extreme jaundice: 2,317 severe jaundice: 35,000 phototherapy: 290,000 visible jaundice: 3.5 million births/year: 4.1 million
- 153. Method Accuracy Disadvantages TSB Gold standard (r=1.0) Painful, costly, inconvenient, delayed Accurate (r= 0.75 -0.93) Meter = $7000 tips $5 unavailable in most physician offices TcB Visual assessment (provider or parent) not accurate (r= 0.36 - 0.7), underestimates severity No standardization lighting, pigmentation
- 154. bilirubin level in blood blood draw jaundice level
- 156. bilirubin level in blood blood draw estimated jaundice level yellowness camera processing
- 158. bilicam
- 160. study A participants 48 newborns 0-4 days old, collected in nursery 3 hospitals in Washington & Philadelphia
- 161. study A participants 48 newborns 0-4 days old, collected in nursery 3 hospitals in Washington & Philadelphia
- 162. bilicam Color Linearization • Camera Settings Adjustment • Light Source Estimation Image Segmentation • Quality Control for Distance, Lighting, and Shadow • Sternum, Forehead, Card Segmented Color Calibration • Dynamic Least Squares Regression • Automatic Feature Selection Neonatal Skin Response to Bilirubin • Skin Independent Color Transformations Applied • Multivariate Machine Learning Regression
- 163. bilicam initial results bilirubin level 20 15 10 5 0 r=0.91 mg/dl 20 5 10 15 bilicam estimation
- 164. bilicam initial results bilirubin level 20 15 10 5 0 r=0.91 non white mg/dl white 20 5 10 15 bilicam estimation
- 165. bilicam initial results bilirubin level 20 TcB = 0.85 15 10 5 0 r=0.91 non white mg/dl white 20 5 10 15 bilicam estimation
- 166. bilicam initial results bilirubin level 20 TcB = 0.85 BiliCam = 0.84 15 10 5 0 r=0.91 non white mg/dl white 20 5 10 15 bilicam estimation
- 167. bilicam future work
- 168. bilicam future work • near term: screening
- 169. bilicam future work • near term: screening • medium term: more data
- 170. bilicam future work • near term: screening • medium term: more data • long term: developing world
- 171. bilicam future work • near term: screening • medium term: more data • long term: developing world “in many resource poor nations, hyperbilirubinemia is the second or third leading cause of infant mortality and disability”
- 174. future research oxygen volume, VO2
- 175. cardiac output and blood pressure
- 178. eclarson.com eclarson@lyle.smu.edu > slide to unlock @ec_larson Thank You!
- 179. phone-as-a-sensor technology: mhealth and chronic disease eclarson.com eclarson@lyle.smu.edu > slide to unlock @ec_larson eric c. larson | eclarson.com Assistant Professor Computer Science and Engineering collaborators: ! Joseph Camp Shwetak Patel Jim Stout, MD Jim Taylor, MD Margaret Rosenfeld, MD Gaetano Boriello Mayank Goel Lilian DeGreef
- 180. phone-as-a-sensor technology: mhealth and chronic disease eclarson.com eclarson@lyle.smu.edu @ec_larson eric c. larson | eclarson.com Assistant Professor Computer Science and Engineering collaborators: ! Joseph Camp Shwetak Patel Jim Stout, MD Jim Taylor, MD Margaret Rosenfeld, MD Gaetano Boriello Mayank Goel Lilian DeGreef