Diagnosis Using a Minimal Cardiac Model Including Reflex Actions
C.E.Hann1, J.G. Chase1, S. Andreassen2, B. W. Smith2, G. M. Shaw3
1Department
of Mechanical Engineering, University of Canterbury, Christchurch, New Zealand
for Model Based Medical Decision Support, Aalborg University, Aalborg, Denmark
3Department of Intensive Care Medicine, Christchurch Hospital, Christchurch, New Zealand
2Centre
Objectives



Simulate two common disease states from onset including reflex actions
Develop a linear and convex identification method that only requires data readily available in an
intensive care unit (ICU)
Accurately identify each time varying disease state as well as all other patient parameters in the
presence of 10% uniformly distributed noise.
Methodology
A minimal cardio-vascular system (CVS) model [1] is employed. Reflex actions include: vaso-constriction,
venous constriction, increased heart rate and increased ventricular contractility, as incorporated every
heartbeat. Integral-based parameter identification [2,3] is used to identify model parameters in each 8-beat
period. The pressure waveforms through the aorta and pulmonary artery, cardiac output and the maximum and
minimum ventricular volumes with 10% uniform noise are the assumed measurements.
Pulmonary Embolism
Pulmonary Artery Pressure (mmHg)
Introduction
Heart disease is difficult to diagnose due to often confusing clinical a data. A minimal cardiac model has been
developed that captures the major dynamics of the cardiovascular system (CVS). To assist medical staff in
diagnosis and treatment, a fast accurate patient-specific parameter identification method, which can account
for time varying disease state and the body’s natural reflex response, is required.
Rpul increases 20% every 8 beats to simulate pulmonary embolism
Mean arterial pressure: slightly down - Cardiac output: down Pulmonary vein pressure: substantially increased
The integral-based identification method identified each disease state within 9% in the presence of 10%
uniformly distributed noise, as shown in Tables 1 and 2. All other parameters, including blood inertances, are
identified with a total mean error of 7.6%. Without including inertances (brackets), total mean error was 4.3%.
Results
Pericardial Tamponade
Aortic Pressure (mmHg)
V0,pcd decreases 20 ml every 8 heart
beats for 40 beats simulating pericardial
tamponade
TABLE 1: PERICARDIAL TAMPONADE (DETERMINING V0,pcd (ml))
Change
True
value
Optimized
value (ml)
Error (%)
V0,pcd
Mean
(all parameters)
Standard
Deviation
First
180
176
2.22
5.60 (2.32)
8.89 (3.23)
Second
160
158
1.25
8.03 (4.45)
9.32 (4.41)
Third
140
138
1.43
6.84 (2.96)
9.29 (5.18)
Fourth
120
117
2.50
8.95 (5.13)
10.46 (4.15)
Fifth
100
100
0
8.58 (5.22)
10.17 (5.37)
TABLE 2: PULMONARY EMBOLISM (DETERMINING Rpul (mmHg s ml-1))
Mean arterial pressure: 100 to 88 mmHg - Cardiac output: 5.3 to 4.1 L/min –
Pulmonary vein pressure: 2 to 7.9 mmHg.
References
[1] B.W. Smith, J.G. Chase, R.I. Nokes, G.M. Shaw and G. Wake. Minimal haemodynamic system model including
ventricular interaction and valve dynamics. Med. Eng. Phys, 26(2):131-139,2004.
[2] C.E. Hann, J.G. Chase, G.M. Shaw, and B.W. Smith. Identification of patient specific parameters for a minimal cardiac
model. Proc 26th International Conf of IEEE Engineering in Med and Biology Society (EMBS 2004), San Francisco, CA,
Sept 1-5, pages 813-816,2004.
[3] C.E. Hann, J.G. Chase, J. Lin, T. Lotz, C.V. Doran, and G.M. Shaw. Integral-based parameter identification for longterm dynamic verification of a glucose-insulin system model. Computer Methods and Programs in Biomedicine,
77(3):259-270, 2005.
Change
True
value
Optimized
value
Error (%)
Rpul
Mean (all
parameters)
Standard
Deviation
First
0.1862
0.1907
2.41
7.93 (3.36)
10.01 (6.35)
Second
0.2173
0.2050
5.67
6.32 (2.92)
7.62 (3.94)
Third
0.2483
0.2694
8.50
7.26 (4.55)
9.50 (5.46)
Fourth
0.2794
0.2721
2.60
8.90 (6.71)
10.84 (5.37)
Fifth
0.3104
0.2962
4.59
7.64 (4.89)
9.61 (5.24)
Conclusion
The model accurately captured the physiological trends in two common disease states simulated from onset.
Integral based optimization successively identified each disease state in the presence of significant measurement
noise. These results demonstrate the potential of using this model in a clinical setting.