=Paper= {{Paper |id=Vol-1337/paper1 |storemode=property |title=Software-based Prediction of Cannula Occlusion During Extracorporeal Blood Circulation Through Networked Medical Data |pdfUrl=https://ceur-ws.org/Vol-1337/paper1.pdf |volume=Vol-1337 }} ==Software-based Prediction of Cannula Occlusion During Extracorporeal Blood Circulation Through Networked Medical Data== https://ceur-ws.org/Vol-1337/paper1.pdf

Software-based Prediction of Cannula Occlusion during
Extracorporeal Blood Circulation through Networked
Medical Data

André Stollenwerk1,∗ , Jan Kühn1 , Marian Walter2 , Christian Brendle2 , Nabil Wardeh3 ,
Rolf Rossaint3 , Steffen Leonhardt2 , Stefan Kowalewski1 , Rüdger Kopp3
1
Informatik 11 - Embedded Software, 2 Philips Chair for Medical Information Technology,
both RWTH Aachen University, 52056 Aachen
3
Clinic for Anesthesiology, RWTH Aachen University Clinic, 52056 Aachen
∗
corresponding author: stollenwerk@embedded.rwth-aachen.de

Abstract

This paper presents a novel method to predict the occlusion of a with-
drawing cannula during extracorporeal circulation due to a networked
intensive care setup. During in-vivo experiments we were able to detect
the cannula suction up to 90 seconds prior to the collapse of extracor-
poreal blood flow. The elaborated metric is based on heart rate, ex-
tracorporeal blood flow and blood pressure at the withdrawing cannula
conjoined in a cyber-medical system.

1 Introduction
Intensive Care Units (ICU) develop towards an increasingly complex composition of medical devices. In order
to optimize the treatment of a life threatening condition a high number of different medical devices is used to
monitor and control the vital signs. However, still the devices are only sparsely interconnected to each other
[GBF12].
Thereby ICU treatments get more and more invasive and a higher need for functional safety arises. One
possible new dimension in overcoming this drawback is the joint analysis of connected sensors in order to derive
additional safety relevant information. By utilizing these cyber-medical systems new therapies are available for
critically ill patients.
In a variety of ICU-established therapies the patient’s blood is treated outside the human body. This form of
treatment is usually used to support the patient’s circulation (e.g. dialysis, extracorporeal membrane oxygenation
(ECMO), ventricular assist devices (VAD)). One of the possible hazards during such a treatment is the occlusion
of the sucking cannula by the withdrawing vessels wall [SAM+ 12, HRB+ 04].

1.1 Worked Example
This paper focuses on the detection of the collapse of a blood vessel during an extracorporeal circulation (EC).
The drawn blood vessel may collapse if blood is actively sucked from the human body. Subsequently the collapsing
vessels’ wall can obstruct successively the holes of the drawing cannula. This results in a rapid breakdown of the
extracorporeal circulation. In clinical applications the EC has currently to be stopped after a cannula occlusion.

1
3.5

Withdrawal Cannula Pressure [mmHg]
3

2.5
Blood Flow [l/min]

2 0

−50

−100

0
−200 −150 −100 −50 0
Time [s]

Figure 1: Blood flow and pressure prior to the occlusion of a vessel to the withdrawing cannula
Following the EC can be reestablished after a calming phase of some seconds (with no extracorporeal blood
flow). Hence the patient’s extracorporeal support can no longer be ensured.
Due to our introduced cyber-medical system architecture of sensors and actuators [SGW+ 11] we were enabled
to predict the occlusion of the blood withdrawing cannula.

2 Extracorporeal Circulation
Several therapies in intensive care medicine are based on EC. All these methods have in common that the
extracorporeal blood flow is preset [PCR+ 13]. The resulting extracorporeal blood flow (Qext ) and the blood
pressure at the point of withdrawing and reinsertion (pw and pr ) is usually measured. In an ICU these values are
not available on a focused device in general. Therefore we introduced an embedded safety network [SWW+ 09],
which interconnects all sensors and actuators in the medical setup. Thus, every node within this network to
know all measurement values and set points, respectively.
Cannula occlusion may be caused by e.g. dislocation of the dawning cannula within the vasculature or hypov-
olemia due to dehydration [SAM+ 12]. If a vessel’s wall occludes the withdrawing cannula and the extracorporeal
blood flow collapses, the EC has to be stopped, before it can be reestablished. Subsequently the vasculature
needs time to refill the vessel. Afterwards the EC can be restarted and slowly increased.
Based on the analysis of the data of cannula occlusion occurrences a model was establishes, which enables the
prediction of cannula occlusion within a horizon of 50 to 90 seconds. Figure 1 shows the blood flow Qext through
the EC and the pressure at the withdrawing cannula pw . Qext collapses at the timestamp 0 s and can only be
reestablished by stopping the pump. The first peaks in the signal can be observed about 90 seconds prior to
the collapse. For reasons of vividness the plotted blood flow signal is low-pass filtered. In the following we will
elaborate the measure used to predict the cannula occlusion and comment on our implementation.

3 Base Conditions for Cannula Occlusion Detection
The presented work bases on the setup of a veno-venous ECMO, which was applied to pigs (m ≈ 55 kg) in animal
experiments. Blood is taken from the vena cava inferior, by a DP2 blood pump (Medos, Stolberg, Germany),
passed through a Medos Hilite 7000 Oxygenator and returned into the vena jugularis externa (see Figure 2).

2
Oxygenator
Blood Pump
Figure 2: Cannulization and setup of ECMO therapy

The withdrawing branch of the cannulization is a twofold design in order to reduce the risk of cannula occlusion.
Nevertheless the previously described problem still exists. In this setup in general a slight pulsation of the tubing
can be observed prior to an occlusion of the cannula to the surrounding vessel.
The used rotary blood pump was driven by a self-developed console specific for this application. The original
manufacturer’s restrictions with respect to the pump interfaces enforced a proprietary development. The pump
can be operated as a revolution speed controlled device or as a blood flow controlled device, respectively. In the
latter case a blood flow sensor has to be introduced to the setup and connected to the network. For this purpose
a flow sensor model HT 110 with an HQ9XL sensor probe (Transonic Systems ultrasonic, Ithaca, NY, USA) was
used. For measuring the pressure at the withdrawing cannula a piezoelectric pressure sensor (Smiths Medical,
Ashford, Kent, UK) were introduced to the setup, which also feed in the embedded safety network. These two
sensor readings are analyzed.

3.1 Extracorporeal Blood Flow

The blood flow through the EC (Qext ) shows an oscillation before the blood flow collapses. This oscillation reflects
the pulsation of the tubing. In order to quantize this oscillation we calculated the Fourier transform of Qext .
Figure 3 shows the spectrum of Qext according to Figure 1 as well as a plot of the maximum of the logarithmic
power spectral density Emax,log of the signal. In the spectrum an increased energy can in general be observed at
the heart rate and the respective multiples. In this case the heart rate of the patient was approximately 70 bpm.
At the same time the constant component is noisy. Hence a filter which attenuates the constant component and
the pulse is applied.
The preprocessed signal can be used to calculate the energy of the corrected flow signal. In Figure 3 a
significant increase of spectral energy (Emax.log ) can be recognized. Therefore we conclude that this event can
be detected about 80 to 90 seconds before the EC has to be stopped (at t = 0 s).

3
200 200

100 100

0 0

−100 −100

−200 −200
Time [s]

−300 −300

−400 −400

−500 −500

−600 −600

−700 −700

−800 −800
1 2 3 4 Emax,log
Frequency [Hz]

Figure 3: Spectrum and maximum of power spectral density of the blood flow signal
3.2 Withdrawing Pressure
The blood pressure measured at the withdrawing cannula pw will lead to a comparable spectrum like presented
in the previous section. In general the pressure difference over a pipe and the resulting blood flow (if it is a
laminar flow) are linked via the Hagen-Poiseuille equation:

dV π · r4 · ∆p
= . (1)
dt 8·η·l
Where V is the flow, ∆p the resulting pressure difference over the length l of a pipe with the radius r and
η is the dynamic viscosity of the moved fluid. In this worked example the Hagen-Poiseuille equation cannot be
utilized due to underconstrained parameters. Blood is a shear thinning (non-Newtonian) fluid hence there is no
fix dynamic viscosity η [HB59]. In addition the tubing of the EC and above all the geometric properties of the
patients’ vasculature are not known in detail. Therefore no direct relation between the extracorporeal blood flow
and the withdrawing pressure can be given.
In the worked example the utilized pressure sensor has a less significant dynamic compared to the blood flow
sensor. Hence the spectrum of the blood flow signal offers information with a higher resolution. Nevertheless
the blood pressure signal offered valuable information and redundancy with respect to functional safety. If one
of the sensors fails in operation we still can analyse a reduced information set, that allows prediction of cannula
suction.
Considered from the physical way a blood vessel can only collapse if there is no positive pressure within the
vessel. In addition the likelihood of collapsing increases in proportion to the negative pw . Therefore the whole
model is only calculated if pw < 0. In addition the smaller the pw and the higher the variance of pw the higher
the probability of cannula occlusion is assumed.

4 Implementation
The introduced algorithm was implemented in C on the embedded safety network. Due to the existing soft-
ware architecture [SGW+ 11] the calculation can be embedded to any of the nodes or be moved between them
respectively.

4
4

3.5

Measure of Confidence
Blood Flow [l/min]

2.5

Limit
0
−200 −150 −100 −50 0
Time [s]

Figure 4: Extracorporeal blood flow and calculated measure of confidence for the occlusion of the withdrawing
cannula

The analysis as introduced in Section 3.1 needs information about the current pulse, which is feed into the
network by a patients monitor (datex ohmeda AS/3). Both presented measures of confidence from the previous
sections 3.1 and 3.2 are summed up. The measure of confidence for the introduced example is plotted in Figure
4. In this case the occlusion of the dawning cannula is predicted 87 seconds before the blood flow collapses.

5 Evaluation
The presented algorithm’s design is based on five events of cannula occlusion, which encountered during about
153 hours of animal experiments. These animal experiments were conducted in order to develop a safe closed
loop control for extracorporeal lung assist therapy [SWW+ 09]. During this research the clinically known problem
of cannula occlusion witnessed. Hence we investigated on a software-based prediction of this hazard.
The limit for the measure of confidence to predict the cannula occlusion was empirically determined in a
way that there are no false positives in the animal experiment data. Hence all events could be detected. A
retrospective analysis shows that this metric can be used as indicator for hypovolemia.
We also tried to introduce an animal model for the occlusion of a withdrawing cannula. Unfortunately it was
not possible to develop a model with stable and reproducible behavior.

6 Conclusion
This paper presents an approach to predict the occlusion of a withdrawing cannula by a blood vessel’s wall
during an extracorporeal blood circulation. The occlusion can be predicted up to 90 seconds in advance. This
does not only prevent the patient from possible harm, but also allows new therapies to confide on extracorporeal
circulation. The presented measure of confidence for the cannula occlusion is another step towards a safer
cyber intensive care unit. Further investigations will have to show the general robustness of this method und
applicability to different types of blood pumps.

5
7 Outlook
The presented work settles on a data base of five events. Therefore it should be refined with a larger base. This
could be achieved on the one hand by evaluating a lager EC data set or on the other hand by establishing a
reproducible and stable animal model for cannula occlusion.
The evaluation showed that the introduced metric can be used as an indicator for hypovolemia. The proceeding
investigations shall elaborate the according potential and limits.

Acknowledgements
This work was supported by the German Research Foundation DFG (DFG - Grant PAK 138/2). The authors
gratefully acknowledge this allowance.

References
[GBF12] David Gregorczyk, Timm Bubhaus, and Stefan Fischer. A proof of concept for medical device inte-
gration using web services. In 9th International Multi-Conference on Systems, Signals and Devices
(SSD), pages 1–6, 2012.
[HB59] Robert H. Haynes and Alan C. Burton. Role of the non-Newtonian behavior of blood in hemody-
namics. American Journal of Physiology – Legacy Content, 197(5):943–950, 1959.

[HRB+ 04] Mark R. Hemmila, Stephen A. Rowe, Tamer N. Boules, Judiann Miskulin, John W. McGillicuddy,
Douglas J. Schuerer, Jonathan W. Haft, Fresca Swaniker, Saman Arbabi, Ronald B. Hirschl, and
Robert H. Bartlett. Extracorporeal life support for severe acute respiratory distress syndrome in
adults. Annuals of Surgery, 240(4):595–605; discussion 605–7, Oct 2004.
[PCR+ 13] Matthew L. Paden, Steven A. Conrad, Peter T. Rycus, Ravi R. Thiagarajan, and ELSO Registry.
Extracorporeal life support organization registry report 2012. ASAIO J, 59(3):202–210, 2013.
[SAM+ 12] David Sidebotham, Sara J. Allen, Alastair McGeorge, Nathan Ibbott, and Timothy Willcox. Ven-
ovenous extracorporeal membrane oxygenation in adults: practical aspects of circuits, cannulae, and
procedures. Journal of cardiothoracic and vascular anesthesia, 26(5):893–909, 2012.

[SGW+ 11] André Stollenwerk, Florian Göbe, Marian Walter, Tobias Wartzek, Rüdger Kopp, Jutta Arens, and
Stefan Kowalewski. Smart data provisioning for model-based generated code in an intensive care
application. In High Confidence Medical Devices, Software, and Systems & Medical Device Plug-
and-Play Interoperability (HCMDSS/MDPnP 2011), Chicago, USA, 2011.
[SWW+ 09] André Stollenwerk, Marian Walter, Tobias Wartzek, Rüdger Kopp, Jutta Arens, and Stefan
Kowalewski. A safety and control concept for extracorporeal membrane oxygenation. The inter-
national Journal of Artificial Organs, 32(7):428, September 2009.