Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and...
Transcript of Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and...
![Page 1: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/1.jpg)
Potential of In-Vehicle and Smartwatch Data Streams for
Improved Diabetes Management
MARTIN MARITSCH, SIMON FÖLL, FELIX WORTMANN
Bosch IoT Lab, a cooperation of ETH Zurich, University of St. Gallen and Bosch
![Page 2: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/2.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 2
Bosch IoT Lab
Current Project Landscape
A Connected Business B IoT Platform Economy C IoT Technology Exploitation
Blockchain-based
P2P Energy Markets
In-Vehicle Hypoglycemia
Detection and Warning
IoT Platform Business
Models
In-Vehicle Affect Reco-
gnition and Regulation
IoT Performance
Management
From Connectivity to
Margin
Data Strategy for the
IoT and AI
Wearable-supported
Diabetes Management
Equipment as a
Service
![Page 3: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/3.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 3
Headwind
Starting Point and Goal
Boris Vukcevic (midfielder at TSG 1899 Hoffenheim)
cause of accident: hypoglycemia (2012)
Source: welt.de, 2020; hackingdiabetes.org, 2020.
Goal: Design and Evaluation of a Vehicle Hypoglycemia Warning System in Diabetes
![Page 4: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/4.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 4
Headwind
Why not CGM?
Time
delay
CGM
rejection
No
reimbursement
High
financial burden
Source: Basu et al., 2013; Keenan et al., 2009; Rebrin et al., 2010.
![Page 5: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/5.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 5
Headwind
What about autonomous driving and vision zero?
Source: NY Times, 2019.
![Page 6: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/6.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 6
Headwind
Goals and partners
How must in-vehicle hypoglycemia warnings be
designed that they are
(a) perceived by drivers and
(b) that they lead to actual behavioral reactions?
RQ3
How does diagnostic accuracy compare to state-
of-the-art methods to detect hypoglycemia (e.g.
self-measurement of capillary blood glucose and
continuous glucose measurement)?
RQ2
To which degree of accuracy can hypoglycemia in
diabetic patients be detected from (a) today's and
(b) future (including physiological and video data)
real-time vehicle car sensor data streams?
RQ1
Objectives
SE
NS
ES
UP
PO
RT
Partner
![Page 7: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/7.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 7
HeadwindApproach
WP3: Driving in the FieldWP2: Driving in the Field
Basic Sensing Module
Basic Support Module
WP1: Driving Simulator
Enhanced Sensing Module
Enhanced Support Module
Integrated Sensing Module
Integrated Support Module
WP4: Project Management & Dissemination
Vehicle Hypo Warning System
triggersBuild & Evaluate
Build & Evaluate
Enhance & Evaluate
Enhance & Evaluate
Integrate & Evaluate
Proof of feasibility in pilot study
Demonstrate
WP0: Preparation
Q4 2017/Q1 2018
![Page 8: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/8.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 8
HeadwindWP1: Driving simulator
Highway
Town
Rural
![Page 9: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/9.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 9
HeadwindWP1: Data gathering
Recorded at 30 Hz
Driver behaviorThrottle, brake, steering wheel, …
Simulator valuesDistance to intersection, lateral position,
headway time, …
Logitech C920
Driver face recording
2x Full-HD 30fps
H.264 encoded stream
1. CAN 2. Video 3. Consumer Eye Tracker
Tobii 4C
90Hz
Gaze points
Head position/rotation
![Page 10: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/10.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 10
HeadwindWP1: Data gathering
4. Professional ECG 5. Consumer Smartwatch
Garmin vivoactive 3
Heart rate inter-beat-intervals
Sensor fusion with
accelerometer data*
Lifecard CF
3-lead ECG
Heart rate inter-beat-intervals
Countour XT,
Biosen C-Line, Dexcom G6
XT: venous blood glucose
C-Line: venous blood glucose
G6: sensor glucose
6. Glucose
![Page 11: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/11.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 11
HeadwindWP1: Clamp procedure
![Page 12: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/12.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 12
HeadwindFirst results from WP0
• Key variables (e.g., “velocity” and
“steering speed”) are significant at
the 1% level
Statistical significance
• Random forest
• ROC AUC: 0.72
• Balanced Accuracy: 0.62
• Deep neural networks
• ROC AUC: 0.74
• Balanced Accuracy: 0.66
Predictive models
• Driving behavior of 5 individuals (3 non-
diabetic and 2 with type 1 diabetes)
• Data for training and testing predictive
models are from disjoint groups of
subjects
• We run 1-fold cross-validation on
subject level, i.e. we train the model on
all subjects except for one, which is used
for testing and repeat this until every
subject has been in the testing set
Training of
predictive models
Source: Kraus et al., 2018.
Early prediction
model shows
between-subject
predictability
![Page 13: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/13.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 13
HeadwindOutlook WP2: Driving in the Field
Test track: tank training field of the Swiss Army
Measure CAN/Video/Audio Intervention interface
Medical equipment
!
Instructor pedals
![Page 14: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/14.jpg)
Confidential | Bosch IoT Lab | 11.02.20
© Bosch Software Innovations GmbH. All rights reserved, also regarding any disposal, exploitation, reproduction, editing, distribution as well as in the event of
applications for industrial property rights. 14
RADARWearable-Based Dysglycemia Detection and Warning
Hypoglycemia Normoglycemia
3.9 mmol/L 10 mmol/L
Smartwatch shows reasonable
classification performance
Classification model captures physio-
logical response during hypoglycemia
Your blood glucose
level is probably low
TIME OF DAY
TENDENCY
SITUATION
PHYSIOLOGY
ML-based classifier for
hypoglycemia1 Explainable AI to rely on sound
cause-effect relationships2 Explainable decision-making
for everyday life3
Model evaluation
Baseline*Empatica E4 +
fasting glucose
AUC 0.5 0.815
Accuracy 0% 88.1%
Sensitivity 0% 72.3%
Specificity 100% 90.6%
Source: Maritsch et al., 2020.
![Page 15: Potential of In-Vehicle and Smartwatch Data Streams for ... · Potential of In-Vehicle and Smartwatch Data Streams for Improved Diabetes Management MARTIN MARITSCH, SIMON FÖLL, FELIX](https://reader034.fdocuments.us/reader034/viewer/2022043022/5f3e36a455dde208d000e156/html5/thumbnails/15.jpg)
Potential of In-Vehicle and Smartwatch Data Streams for
Improved Diabetes Management
MARTIN MARITSCH, SIMON FÖLL, FELIX WORTMANN
Bosch IoT Lab, a cooperation of ETH Zurich, University of St. Gallen and Bosch