Photoplethysmography-basedsystem for atrial fibrillationdetection during hemodialysisDainius Stankevicius, A. Petrenas, A. Solosenko, M. Grigutis, T.Januskevicius, L. Rimsevicius, and V. Marozasdainius.stankevicius@ktu.lt

Vilnius university hospitalSANTARIŠKIŲ KLINIKOS

Introduction (1)

Hemodialysis:– 350 000 people in the EU– 5400 dialysis centers in the EU– 60 % of chronic kidney disease patients– 3 times per week– 4 hours long

1/20

Introduction (2)

Atrial fibrillation (AF):– The most common arrhythmia– 33 million people around the world– Risk of stroke and heart failure

Connection:– Hemodialysis increases risk of developing AF– AF occurs during hemodialysis– Related to changes in fluid balance

2/20

Research question

Can we detect AF during hemodialysis in anunobtrusive and still reliable way?

3/20

Proposal

• Electrocardiography is encumbering• Photoplethysmography might be a solution

– Definitely less obtrusive!– Also less reliable...

• A system consisting of:– Specialized low power device– Embedded algorithm to run the detection

4/20

We have built a device

• ARM Cortex-M0, 16 MHz, 16 kB SRAM• PPG sampling at 250 Hz• 24-hour battery life

5/20

Methodology: Start

ppg(n)

6/20

Raw PPG

7/20

Methodology: Preprocessing

ppg(n) Low pass filterfc

LMSw1...w10m

+-x(n) = 1

ppgl(n)

ppga(n)

y(n)

8/20

Raw & Preprocessed PPGs

9/20

Methodology: SSF

SSFN

ppga(n) s(n)

10/20

Slope Sum Function (SSF):

PPG and SSF

11/20

Methodology: Thresholding

SSFN

ThresholdingT, Tk

max{s(n)}arg[max{s(n)}]

median{p(m),...,p(m-4)}

i(m)-i(m-1)

ppga(n) s(n)

T

p(m)

i(m)

∆i(m)

st(n)SSFN

ppga(n) s(n)

12/20

Thresholding and peak-to-peak

13/20

Methodology: AF detector

i(m)-i(m-1)

Low complexityAF detector

adt

i(m)

∆i(m) O(m)

14/20

Methodology: overall

Low pass filterfc

LMSw1...w10m

SSFN

ThresholdingT, Tk

max{s(n)}arg[max{s(n)}]

median{p(m),...,p(m-4)}

i(m)-i(m-1)

+-

Low complexityAF detector

adt

x(n) = 1

ppg(n) ppgl(n)

ppga(n)

s(n)

T

p(m)

i(m)

∆i(m) O(m)

st(n)

y(n)

15/20

Testing: hardware-in-the-loop

Real environment

Simulated environment

Pre-recorded PPG withAF

Raw data file saved inmicroSD card

Normal PPGsampling routine... Read PPG sample PPG

hardware

16/20

Results

0 50 100 150 200-1

0

1

2

0 50 100 150 200100

200

300

400

0 50 100 150 2000

1

2

0 50 100 150 2000

1Outa Outb Threshold

Time, s0 50 100 150 200

0

5000

17/20

Conclusion

The proposed system has a potential to beapplied for an unobtrusive real-time AFdetection using solely thephotoplethysmogram.

18/20

Future directions

• Introduce signal quality metrics• Introduce morphology-based features• Determine the optimal placement of the

device

19/20

Acknowledgment

• This work was supported by CARRE(No.611140) project, funded by theEuropean Commission FrameworkProgram 7.

• CARRE – CARdio REnal disease• http://www.carre-project.eu

20/20