nano-tera.ch Body Powered SenSE
bodypoweredsense.ch
Zero Power Medical Devices
Annual Plenary Meeting Bern, May 4th, 2015
P.-A. Farine, J. Gutknecht, T. Gross, C. Hierold, D. Atienza, D. Briand, M. Knyazeva, G. Wohlrab Milad Ataei, PhD student, ESPLAB, EPFL
Introduction
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
• Patient’s physiological state requires long-term monitoring in order to precise a diagnosis or to evaluate the efficacy of an on-going treatment. • Today’s wearable solutions invade the user’s normal life as sensing platforms have batteries issues including size, weight, operating lifetime or convenience.
BodyPoweredSense:
L’EXPRESS - L’IMPARTIAL MERCREDI 17 JUILLET 2013
o Improve Health Care through Smart, Convenient Wearable Sensors o Advance Human Energy Harvesting towards a “fit and forget” and “no recharge” goal. o Target Energy Optimization and Usage o Apply the technology in Real and Demanding Clinical Use cases, ensuring fitness for purpose o Involve Industry in multi-disciplinary IT research
Convenient optimized system in Energy and Monitoring functionality from low levels to high levels
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
BodyPoweredSense –Partitioning and Challenges Amplified and digitized EEG and ECG signals
Energy Harvesters
Power Supply
rgy e En
DC/DC Converter
p m A
- Optimized output power - A lightweight, compact and comfortable wearable system
AC/DC Converter
ADC
- Modelling and predication of energy harvesting - Energy-aware runtime for smart medical sensors
Integrated Circuit
µC DSP
µC Low power
RAM
Personal Environmental monitoring powered by solar cells
- Optimize the methods of analysis for long-term lowdensity EEG
- EEG recording and data analysis for Epilepsy in children - Advice on hospital best practice
a gFrw sn tim p lfO Se
Apply the technology in Real and Demanding Clinical Use cases
l ru e gth axisn m
- Development and optimization of compliant nano-composite piezoelectric materials
at rD so n Se
rgyifo e En
B LA ESP
MNS LAB
Low Power Processor unit
Low Power Processor unit
/e rgyatio e n
Optimised Power Conversion from the Energy Harvesting (EH) sources to storage - Power converter efficiency - Limited sizes - Ultra-low power design
- Energyefficient and real-time multilead biosignal analysis
- Provide a complete medical Human Centered Design - Provide medical processing platform
Low Density EEG
- Support data capture trials Epilepsy in children (KISPI), Alzheimer’s disease in the elderly (CHUV)
PFIM
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
System Energy Budgets EEG Sensor
Estimated Power Consumption 1.5mW to 3mW, over 24 hours 44.5mWH for 64 electrodes Estimated Energy Harvesting levels: TEG 72 cm2 1.2mW, Solar 5 Cm2 75mWH, estimated total circa 100mWH
Available Power: 4mW Converted power: 2mW
ECG Sensor
Memory
Estimated Power Consumption 0.5mW to 1 mW , 24mWH Estimated Energy Harvesting levels: 9 cm2 TEG 0.2mW, Piezo 1000 Cm2 1mW, estimated total circa 30mWH
Electrodes uC Amplifier ADC Data flow Raw data size 24h uC frequency uC Consumption Sub total (uC+A+ADC) Energy per day Flash storage energy Total per day
# # uW uW kbps Gb MHz uW mW mWh mWh mWh
1 5 10 1.2 0.1
16 1 80 160 19.2 1.7 1 400 0.6 15.4 0.6 15.9
32 1 160 320 38.4 3.3 1 400 0.9 21.1 1.2 22.3
64 2 320 640 76.8 6.6 1 800 1.8 42.2 2.3 44.5
128 4 640 1280 153.6 13.3 1 1600 3.5 84.5 4.6 89.1
uC
ADC
Amp
MSP430FR5739 uC active power Flash memory Flash memory
uW/MHz
400
MB/J Gb/mWh
100 2.88
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
Thermoelectric energy harvesting of human body heat, MNS Local energy harvesting and conversion
Large-area harvesting (20 cm2) nnkk SSi i
e at pl Al
Heat sink
r Ai in Sk
Electronics TEG Heat transfer Electrode 12 mm
• Autonomous sensor nodes • Biopotential readout
Thielen, et al. ICT 2014
ΔT
Isolation
Electronics Reference electrode • Data processing • Wireless communication
Heat sinks Thielen, et al. NT Annual Meeting 2014
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
EEG electrode for application in hair with TEG, MNS Current status: - Prototyping - Testing of new TEGs
Heat sink Stacked TEG Thermal insulation Heat transfer structure AgCl pins
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
Compliant piezoelectric harvesting band, SAMLAB
Synthesis of homogeneous piezoelectric nanocomposites via stabilized nanoparticle suspensions. Utilization of large area fabrication methods: Inkjet printing for electrode deposition, bar casting for active layer deposition. Piezoelectric NPs / PDMS Piezoelectric Composite
Material optimization • • •
Ball milled BaTiO3 NPs Carbon filler (CNT, CB) PDMS, SU-8
Milled and filtered BaTiO3 after 24hours
Nanocomposite (PDMS/BaTiO3/CNT)
NP/CNT/PDMS mixture Au/Ti Nanocomposite (PDMS/BaTiO3/CNT)
Au/Ti
PDMS dielectric Kapton Measured Piezoelectric Constant: 5 pC/N McCall et al. (2014): > 100 pC/N
Kapton
0.1 µW / cm2 (x 1000 cm2) = 0.1 mW
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
Energy conversion implementation, ESPLAB
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
Energy conversion design and measurements, ESPLAB
Challenges:  Harvesters produce small power at low voltages, processing are power hungry: Power converter efficiency 
It will fit in size of an electrode: Integrated circuit with limited passive sizes To realize a high efficient system here, 4 criteria have to keep in mind: 1- Ultra low power blocks in circuit level 2- Minimizing loss 3- Limited form factor 4- Maximizing power transfer
Demonstrator, LST
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
• Will demonstrate • TEG and Solar energy harvesting • Modeling and Prediction of energy harvesting • EEG recording and seizure prediction using energy aware run time
ACTUAL PCBs of Processor Unit and Active electrode
Final version with ESPLAB’s chip
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
Real-world testing, MNS and LST
First feedback: - Improvements on wire connections needed for kids - Results vary strongly for different subjects ďƒ Statistics needed
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
Smart WBSN concept for ECG, ESL Smart Node
Displays the received data and relays to medical personnel
ECG
Noise filtering
ECG delineation
ECG Analysis (arrhythmia)
ECG-aware compression Energy-efficient and real-time multi-lead ECG biosignal analysis (1) (2) (3) (4)
Low-complexity filtering methods using integer computing and bio-signal operators Limited-memory methods to dynamically adapt ECG thresholds for each person Low-energy multi-lead ECG arrhythmia analysis in real-time Energy-efficient communication. storage exploiting biosignal-adapted compression
Automated ECG diagnosis
Detection of normal beats (N), premature ventricular contraction (V), left bundle branch block (L), right bundle branch block (R), and atrial premature beats (A)
Classification by majority voting
Battery of n(n-1)/2 one-vs-one classifiers Neuro fuzzy classifier accuracy: ~ 84% Support vector machine accuracy: ~ 91%
Filtering Lead 1 Delineation
Filtered ECG DWT Transform Fiducial Points
Random Projection Random Projection
Classification
Diagnosis (‘N’, ‘V’, ‘L’, ‘R’, ‘A’)
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
High- to low-density EEG monitoring Goal: Method: Steps:
To Optimize methods for monitoring of some pathological conditions with low density EEG. To use functional connectivity based on source analysis of EEG. To adjust these methods from high density to low density EEG. Step 1
Functional connectivity based on source analysis of high-densty EEG: To use an usual analysis of hdEEG.
High Density EEG
61 sensors
The differences of functional connectivity were observed in PNES patients vs. healthy controls [1].
2
The result of functional connectivity analysis, using cortical partial coherence, were different with different EEG densities.
110 sensors
1
Step 2
Functional connectivity based on cortical partial coherence analysis of high-densty and lowdensty EEG: To use a method designed to work with high and low density EEG.
18 sensors
[1] Barzegaran, E., Knyazeva, M., JNNP, accepted, 2015.
Step 3
validate cortical partial coherence method for ldEEG using simulated data: To simulate EEG data with different locations of active sources in the brain, and evaluate the method for estimation of sources
Low Density EEG
3
This Method performs accurately on simulated data when the location of sources are close to the electrodes.
4
This method does not perform accurately when the location of sources are not close to the electrodes.
EEG signal analysis
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
• Supported partners with EEG signal capture of BPS test subjects • Collaboration with ETH-NSG with Real recording experience and data analysis for the early detection of epilepsy • Provided ESPLAB with training data for good and bad electrodes analysis • Provided advice on hospital procedures and best practice • Provided signal to noise data to ESPLAB for circuit design considerations
1. Introduction | 2. Harvesters | 3. Conversion | 4. Processing | 5. Clinical | 6. Industrial partner
Industrial partners •
PFIM, Support to clinical approach and to data capture trials , providing children as test subjects and clinical attendance
•
ByElement, has provide a complete medical HCD Engineering consisting of several developer workshops with medical experts, applying HCD. Methodology, identification of critical acceptance elements, identification and description of medical usage scenarios, including device acceptance dimensions.
Children: Wearable Energy Harvesting Measurement and EEG capture
•
EXPLORIS, collaborating with ETH-NSG and with CHUV provide medical processing platform
Focus group meeting ETH Zurich
Human Centred Design in practice
END OF THE PRESENTATION
Thank you for your attention Acknowledgments
Christopher Borsa SAMLAB, EPFL
Elham Barzegaran CHUV
Francisco Rincon ESL, EPFL
Ivana Unkovic LST, ETHZ
Moritz Thielen MNS, ETHZ