Transcript CVS Modeling - University of Canterbury
Diagnosing cardiac disease states using a minimal cardiovascular model
Christina Starfinger Wednesday, 21 March 2007 HRSC Scientific Meeting
Current Clinical Process
•
Introduction Model
Measure
Identification Prediction
limited clinical data
Results People
•
Diagnose
based on data trends observed and approximate mental models of physiological function and pathology – typically based on an “
average patient
” •
Treat
by selecting a therapy based on standard protocols or methods for diagnosed disease state Problems: – – No means of aggregating often conflicting data into a clear picture Many disease states show very similar measurements clinically confusing the diagnostic picture – – Reflex actions can mask disease states until they are acute Limited measurements are typically used (“task overload”)
Model-Based Solution
Introduction Model Identification Prediction Results People
1.
Identify
patient specific parameters from
measured
create patient specific CVS model clinical data to
2.
Diagnose
disease state from patient-specific parameter values
3.
Treat
using patient-specific model to predict results of different therapeutic strategies/interventions (
Therapy Decision Support
)
Outcomes:
Real-time, patient specific CVS model Assistance in diagnoses and therapy selection Uses data and catheters typically found in the ICU
Introduction Model
Model
Identification Prediction Results People
Left ventricle (LV) Right ventricle (RV) Vena Cava, Aorta, Pulmonary artery/vein Pulmonary circulation Systemic circulation
Introduction Model
Vena Cava
Model
Identification Prediction Results
Right ventricle
People
Pulmonary Artery Aorta Pulmonary Vein Left ventricle Smith, BW, Chase, JG, Shaw, GM and Nokes, RI (2006). “Simulating Transient Ventricular Interaction Using a Minimal Cardiovascular System Model ,” Physiological Measurement, IOP, Vol 27, pp. 165-179
Identification Problem
Results Introduction Model Identification Prediction People Measurements Required
: 1.
Max/Min Pressure in aorta (SAP, DAP) 2.
3.
4.
5.
Max/Min Pressure in pulmonary artery (SPAP, DPAP) Max/Min Volume in left ventricle (LVEDV,LVESV) Max/Min Volume in right ventricle (RVEDV,RVESV) Heart Rate
Model Parameters Identified:
Integral-based methods
Given or estimated
Lav,Lmt,Ltc,Lpv,Eeslvf,Eesrvf,Polvf,Porvf,Rav,Rmt,Rtc,Rpv, Eao,Epa,Evc,Epu,Rsys, Rpul
Find
Measurements required are a minimal set compared to the total number of parameters identified in the model
Hann, CE, Chase, JG and Shaw, GM (2006). “Integral-based Identification of Patient Specific Parameters for a Minimal Cardiac Model ,” Computer Methods and Programs in Biomedicine, Vol 81(2), pp. 181-192
Diagnosis from ID
Introduction Model Identification Prediction Results People Parameters:
Contractilities, increased during increased sympathetic activity (HR ↑) Lav,Lmt,Ltc,Lpv,Eeslvf,Eesrvf,Polvf,Porvf,Rav,Rmt,Rtc,Rpv, Eao,Epa,Evc,Epu,Rsys, Rpul Pulmonary vascular resistance and systemic vascular resistance, increased for example in PE, and PHT Resistances for 4 heart valves, increased for stenosis
Pulmonary Embolism
Introduction Model Identification Prediction Results
• • •
Experiment:
6 pigs, 32.75kg ± 1.83kg
pulmonary embolization induced with autologous blood clots clots were injected every two hours with decreasing concentrations
People
• •
Measurements:
aortic pressure (P ao ) and pulmonary artery pressure (P pa ) are measured using micromanometer-tipped catheters (Sentron pressure-measuring catheter;Cordis, Miami, FL) right and left ventricle pressures and volumes (V lv ,V rv ,P lv ,P rv ) are measured using 7F, 12 electrodes (8-mm interelectrode distance) conductance micromanometer tipped catheters (CD Leycom, Zoetermeer, The Netherlands) •
Goal:
Accurate identification at all stages of induced embolism with no false parameter value changes
Data from Ghuysen et al, Hemodynamics Laboratory, Univ of Liege
Results for PE (ID) - 1
Results Introduction Model Identification Prediction Left ventricle (30 mins) People Right ventricle (30 mins)
Errors ~5% and all peak and stroke values captured
Results for PE (ID) - 2
Results Introduction Model Identification Prediction People
Pulmonary resistance (Pig 2) Rpul – All pigs Right ventricle expansion index (RVEDV/LVEDV, Pig 2) Reflex actions (Pig 2)
PEEP Titrations
Introduction Model Identification Prediction Results
• • •
Experiment:
6 pigs (2 used so far for analysis: 21kg) 5 PEEP titrations at different blood volumes: – – – – – Baseline Removal of blood (Hypovolemia) Reinfusion of blood Infusion of saline Infusion of more saline Each PEEP titration included PEEP levels of 0,10 and 20 cmH 2 0
People
• •
Measurements:
Measurements recorded using PiCCO, Servo-i, Vigilance and SC9000 monitors V lv , V rv were estimated based on TBV and GEDV •
Goal:
Predict the effect of PEEP therapy on SV (etc) in presence of different blood volumes
Data from Smith et al MMDS, Aalborg University
Results for PEEP-Prediction
Introduction Model Identification Prediction Results People
Goals:
• Match peak pressures • Match stroke volumes • Volumes close • Pressures accurate
Results for PEEP-Prediction
Introduction Model Identification Prediction Results People
Errors ~10% + Trends Captured!
Conclusions
Identification Prediction Results Introduction Model People
• Model-based
Identification
,
Diagnosis
and
Therapy Decision Support
methods presented • Validated on two clinical data sets using porcine animal models – Pulmonary embolism – PEEP intervention at different blood volumes • Results show good accuracy (<10% error) for critical parameters – No false parameter value changes in identification implies a model of the proper level of complexity for these disease states • Future Work = Septic Shock (April 2007) and further disease states
Acknowledgement
Dr Geoff Chase Dr Geoff Shaw
Denmark (PEEP Data)
Dr Chris Hann
Results People Introduction Model Engineers and Docs Identification Prediction
Questions
???
Belgium (PE Data)
Prof Steen Andreassen Dr Bram Smith Dr Bram Smith Dr Thomas Desaive