Volume Quantification by Contrast-Enhanced Ultrasound
The realized in-vitro set-up (Figure 1) consisted of an open circuit with a roller pump, Cobe Stoeckert multiflow bloodpump (Stoeckert Instruments, Munich, Germany), a water-filled basin, a network of tubes with a variable volume simulating the pulmonary vessels, and a pressure stabilizer. The whole set-up was filled with tap water which was degassed by 24-hour rest. The temperature was maintained at 37°C with heating devices and thermostats at different positions in the set-up. The in- and outflow tubes of the network were submerged in a water-filled basin. The submerged segment of the tubes was made of a thin polyurethane layer (Ultracover®, Microtek™ Medical BV, Zutphen, the Netherlands) in order to limit interference with ultrasound measurements. In the water-filled basin, a transesophageal (TEE) probe (X7–2 t, Philips Healthcare, MA, USA) was directly submerged in water to optimize the acoustic impedance while insonifying the submerged tubes. Two 0.014" high-fidelity pressure wires (Radiwire, St Jude Medical Inc, St. Paul, MN, USA) were inserted in these tubes. These wires measure temperature at 0–25 Hertz (Hz) with an accuracy of 0.05°C within a temperature range of 15 to 42°C. Distal to the centrifugal pump, cold saline and UCA were injected into the inflow tube through an injection point consisting of a single lumen central venous line (Blue flextip catheter, Arrow®, Reading, PA, USA). Between the inflow and outflow tubes, outside the basin, the circuit expanded into a network of eight tubes and converged back into a single outflow tube. This network was made of tubing which is used for cardiopulmonary bypass, matching the roller pump. The tubing (Medtronic, Minneapolis, MN, USA) of the network had a diameter of ¼" with a wall size of /32". The afferent and efferent tubes had a diameter of ½". The length of tubing was adapted to create a physiologic range of volumes. The network could be clamped at different positions to create different volumes. The hydrodynamic circuit was open to avoid UCA recirculation and the hydrostatic pressure of the circuit was stabilized at the output. All tubes were isolated with polyethylene covers (Climaflex®, NMC, Eynatten, Belgium) to prevent temperature loss to the surroundings.
(Enlarge Image)
Figure 1.
The in-vitro set-up in a schematic overview. The variable network can be clamped at different points to create different volumes.
A commercially available scanner (iE33, Philips Healthcare, Andover, MA, USA) was used to obtain cross-sectional B-mode images of the inflow and outflow tubes. Harmonic imaging at 2.7 - 5.4 MHz was used in order to increase the signal-to-noise ratio (SNR) for low UCA concentration together with a low mechanical index (MI) of 0.2 to reduce bubble disruption. Frame rate was set at 27 Hz, the same time-gain and lateral-gain compensation were employed over all measurements, compression was set at 50 dB, general gain at 60%, and image depth was 8 cm with the focus being at the level of both tubes.
For direct application of the indicator dilution theory, a linear relationship between UCA concentration and detected acoustic intensity is necessary. Therefore, we measured the different acoustic intensities of different doses of UCA (SonoVue®, Bracco SpA, Geneva, Italy) diluted in saline at room temperature and at 4°C. This calibration was performed according to the protocol described by Mischi et al. It had a twofold objective: finding the range of UCA concentrations that show a linear relationship with the measured acoustic intensity, and investigating the effect of temperature on the UCA behavior. The relationship between SonoVue®-concentration and measured acoustic intensity was linear below 1.5 mg/L (Figure 2) at room temperature and 1 mg/L at 4°C. Above these concentrations shadowing was seen.
(Enlarge Image)
Figure 2.
Acoustic intensity calibration curves of SonoVue® at room temperature and at a temperature <4°C. Acoustic intensity is presented on the Y-axis at different temperatures and at different concentrations of SonoVue®. At room temperature (red circle) there is attenuation above 1.5 mg/L; at a temperature <4°C (light blue square) attenuation occurs at a concentration between 1 mg/L and 1.5 mg/L. A linear relationship between concentration and acoustic intensity is seen below 1 to 1.5 mg/L at both temperatures.
Thermodilution measurements were performed using the pressure wires as described above. These pressure wires have temperature sensing tips that were positioned in the polyurethane tubes and were intercepted by the ultrasound beam for contrast quantification. The temperature sensors of both pressure wires were connected to a Wheatstone bridge adjusted to half-bridge configuration in order to output measured IDCs from both sensors. The electrical circuit further comprised a feedback amplifier (INA 118, Burr-Brown Corporation, Tucson, AZ, USA), a power supply (Delta Elektronica, Zierikzee, the Netherlands), and a data acquisition board (NI USB-6341, National Instruments, Austin, TX, USA). The bridge was balanced by manual adjustment of the value of an embedded potentiometer. The output signal was amplified in such a way that the full range of the analog-to-digital converter of the data acquisition card (0–10 V) was exploited. High frequency noise suppression was achieved by placing an additional capacitance in parallel with the input impedance of the amplifier. All devices were shielded and grounded to minimize ambient disturbances. The thermodilution curves were acquired with LabVIEW (National Instruments, Austin, TX, USA) and processed in MATLAB® 2009b (The Mathworks, Natick, MA, USA). The full system was calibrated by mapping the measured voltage as a function of temperature in a water-filled basin measured by a digital thermometer (Keithley 871, Keithley Instruments, Cleveland, OH, USA). The calibration showed a linear relationship with a slope of 0.65 V/°C and r = 0.999. These results confirmed the system linearity for temperatures in a range 24°C – 40°C.
Different flows were generated by adjusting the rounds per minute (rpm) of the centrifugal pump. Six flows were used for the measurements that varied between 1 and 4 liters per minute in increments of 0.5 liter per minute. Flow was measured using a flow sensor (Flow controller ARS 260, Biotech, Vilshofen, Germany), at the end of the circuit. By clamping different bifurcations of the variable network, four different volumes were generated, namely 890 milliliter (mL), 718 mL, 530 mL, and 356 mL (Figure 1). These volumes have been chosen to cover a range that is slightly broader than the pulmonary blood volumes reported in patients, which range from 271 mL/m (~500 mL) to 421 mL/m (~800 mL) in heart failure patients. Every measurement was repeated three times, at six flows and four volumes. With every measurement a bolus of 0.2 mL SonoVue® diluted in 20 mL cold saline (4°C) was injected. The change in acoustic intensity on B-mode ultrasound was stored in an uncompressed format for subsequent analysis with commercially available software (QLAB 8, Philips Healthcare, Andover, MA, USA). This software allows drawing of multiple regions of interest (ROIs) to obtain acoustic IDCs. Two ROIs were drawn within the thin polyurethane layer of the inflow and outflow tube in the water-filled basin. An additional movie file shows this in more detail (see Additional file 1). The IDCs were processed and fitted by the local density random walk (LDRW) model using MATLAB® 2009b. The LDRW model was employed since it provides both the best least square error fit to the IDC and a physical description of the dilution process. The MTT of the contrast bolus between the injection and the detection sites was directly derived from the parameters of the fitted model. Volumes were then calculated as the product between the measured flow and the difference in MTT between the two curves.
The MTT can be derived using two different methods. First, the MTT of each IDC can be estimated as the first order statistical moment of the fitted model, using the double fit method (Figure 3). Second, the indicator dilution system can also be interpreted as a linear system; therefore, the impulse response approach can be employed. The impulse response of the system between the two indicator detection sites was estimated by means of a parametric deconvolution technique, using the system input and output signals represented by the measured IDCs. The estimated impulse response is represented by the LDRW model, which allows blood volume assessments (Figure 3). The advantage of using a deconvolution technique over a double IDC fitting consists of the independency of the resulting impulse response from the injection function.
(Enlarge Image)
Figure 3.
Indicator dilution curves (IDCs) fitted by LDRW double fit method and impulse response method. These IDCs were constructed from post-processing imaging analysis at the level of inflow and outflow tubes at a flow of 2 L/min and true volume of 718 mL. The dotted black line is the IDC at the level of the inflow tube and outflow tube. In the upper picture, the red and blue lines depict the IDCs by the double fit method according to the LDRW model of the inflow and outflow tube. The vertical dashed lines represent the MTTs of each curve. In the bottom picture, the green line depicts the IDC according to the LDRW model impulse response method.
All data were reported as mean values ± standard deviation (SD) or as median ± interquartile range (IQR) depending on the distribution of the variables of three consecutive measurements. The first goal was to investigate the agreement between measured volumes by both techniques and the true set-up volumes. Statistical significance was considered as a two-sided P<0.05. Bland-Altman analysis was used to determine the agreement between measured volumes and the true volumes. The effect of the different flows on the volume measurement was also investigated and reported in dedicated plots. Reproducibility was assessed by the intraclass-correlation coefficient (ICC). ICC consists of a basic calculation as repeated-measures analysis of variance (ANOVA) and the intraobserver reliability (ICC (1,1)). ICC assesses the agreement of quantitative variables on its reliability and consistency. The second goal was to analyze the correlation between the CEUS volumes and thermodilution volumes, assuming thermodilution as the gold standard. Correlation coefficients were assessed using the Pearson correlation coefficient R or the Spearman correlation coefficient rs depending on normal distribution or non-normal distribution of variables, respectively. Statistical analysis was performed using GraphPad Prism version 5.03 (GraphPad Software, San Diego, CA, USA) except for the intraclass-correlation, which was analyzed by Unistat® Statistical Package for Windows™ version 6.0 (Unistat House, London, England). Statistical analysis was performed by using all data (n=72 measurements) to exclude a bias.
Methods
In-vitro Set-up
The realized in-vitro set-up (Figure 1) consisted of an open circuit with a roller pump, Cobe Stoeckert multiflow bloodpump (Stoeckert Instruments, Munich, Germany), a water-filled basin, a network of tubes with a variable volume simulating the pulmonary vessels, and a pressure stabilizer. The whole set-up was filled with tap water which was degassed by 24-hour rest. The temperature was maintained at 37°C with heating devices and thermostats at different positions in the set-up. The in- and outflow tubes of the network were submerged in a water-filled basin. The submerged segment of the tubes was made of a thin polyurethane layer (Ultracover®, Microtek™ Medical BV, Zutphen, the Netherlands) in order to limit interference with ultrasound measurements. In the water-filled basin, a transesophageal (TEE) probe (X7–2 t, Philips Healthcare, MA, USA) was directly submerged in water to optimize the acoustic impedance while insonifying the submerged tubes. Two 0.014" high-fidelity pressure wires (Radiwire, St Jude Medical Inc, St. Paul, MN, USA) were inserted in these tubes. These wires measure temperature at 0–25 Hertz (Hz) with an accuracy of 0.05°C within a temperature range of 15 to 42°C. Distal to the centrifugal pump, cold saline and UCA were injected into the inflow tube through an injection point consisting of a single lumen central venous line (Blue flextip catheter, Arrow®, Reading, PA, USA). Between the inflow and outflow tubes, outside the basin, the circuit expanded into a network of eight tubes and converged back into a single outflow tube. This network was made of tubing which is used for cardiopulmonary bypass, matching the roller pump. The tubing (Medtronic, Minneapolis, MN, USA) of the network had a diameter of ¼" with a wall size of /32". The afferent and efferent tubes had a diameter of ½". The length of tubing was adapted to create a physiologic range of volumes. The network could be clamped at different positions to create different volumes. The hydrodynamic circuit was open to avoid UCA recirculation and the hydrostatic pressure of the circuit was stabilized at the output. All tubes were isolated with polyethylene covers (Climaflex®, NMC, Eynatten, Belgium) to prevent temperature loss to the surroundings.
(Enlarge Image)
Figure 1.
The in-vitro set-up in a schematic overview. The variable network can be clamped at different points to create different volumes.
Ultrasound System and Settings
A commercially available scanner (iE33, Philips Healthcare, Andover, MA, USA) was used to obtain cross-sectional B-mode images of the inflow and outflow tubes. Harmonic imaging at 2.7 - 5.4 MHz was used in order to increase the signal-to-noise ratio (SNR) for low UCA concentration together with a low mechanical index (MI) of 0.2 to reduce bubble disruption. Frame rate was set at 27 Hz, the same time-gain and lateral-gain compensation were employed over all measurements, compression was set at 50 dB, general gain at 60%, and image depth was 8 cm with the focus being at the level of both tubes.
Calibration
For direct application of the indicator dilution theory, a linear relationship between UCA concentration and detected acoustic intensity is necessary. Therefore, we measured the different acoustic intensities of different doses of UCA (SonoVue®, Bracco SpA, Geneva, Italy) diluted in saline at room temperature and at 4°C. This calibration was performed according to the protocol described by Mischi et al. It had a twofold objective: finding the range of UCA concentrations that show a linear relationship with the measured acoustic intensity, and investigating the effect of temperature on the UCA behavior. The relationship between SonoVue®-concentration and measured acoustic intensity was linear below 1.5 mg/L (Figure 2) at room temperature and 1 mg/L at 4°C. Above these concentrations shadowing was seen.
(Enlarge Image)
Figure 2.
Acoustic intensity calibration curves of SonoVue® at room temperature and at a temperature <4°C. Acoustic intensity is presented on the Y-axis at different temperatures and at different concentrations of SonoVue®. At room temperature (red circle) there is attenuation above 1.5 mg/L; at a temperature <4°C (light blue square) attenuation occurs at a concentration between 1 mg/L and 1.5 mg/L. A linear relationship between concentration and acoustic intensity is seen below 1 to 1.5 mg/L at both temperatures.
Thermodilution Measurement
Thermodilution measurements were performed using the pressure wires as described above. These pressure wires have temperature sensing tips that were positioned in the polyurethane tubes and were intercepted by the ultrasound beam for contrast quantification. The temperature sensors of both pressure wires were connected to a Wheatstone bridge adjusted to half-bridge configuration in order to output measured IDCs from both sensors. The electrical circuit further comprised a feedback amplifier (INA 118, Burr-Brown Corporation, Tucson, AZ, USA), a power supply (Delta Elektronica, Zierikzee, the Netherlands), and a data acquisition board (NI USB-6341, National Instruments, Austin, TX, USA). The bridge was balanced by manual adjustment of the value of an embedded potentiometer. The output signal was amplified in such a way that the full range of the analog-to-digital converter of the data acquisition card (0–10 V) was exploited. High frequency noise suppression was achieved by placing an additional capacitance in parallel with the input impedance of the amplifier. All devices were shielded and grounded to minimize ambient disturbances. The thermodilution curves were acquired with LabVIEW (National Instruments, Austin, TX, USA) and processed in MATLAB® 2009b (The Mathworks, Natick, MA, USA). The full system was calibrated by mapping the measured voltage as a function of temperature in a water-filled basin measured by a digital thermometer (Keithley 871, Keithley Instruments, Cleveland, OH, USA). The calibration showed a linear relationship with a slope of 0.65 V/°C and r = 0.999. These results confirmed the system linearity for temperatures in a range 24°C – 40°C.
Ultrasound Contrast Measurement
Different flows were generated by adjusting the rounds per minute (rpm) of the centrifugal pump. Six flows were used for the measurements that varied between 1 and 4 liters per minute in increments of 0.5 liter per minute. Flow was measured using a flow sensor (Flow controller ARS 260, Biotech, Vilshofen, Germany), at the end of the circuit. By clamping different bifurcations of the variable network, four different volumes were generated, namely 890 milliliter (mL), 718 mL, 530 mL, and 356 mL (Figure 1). These volumes have been chosen to cover a range that is slightly broader than the pulmonary blood volumes reported in patients, which range from 271 mL/m (~500 mL) to 421 mL/m (~800 mL) in heart failure patients. Every measurement was repeated three times, at six flows and four volumes. With every measurement a bolus of 0.2 mL SonoVue® diluted in 20 mL cold saline (4°C) was injected. The change in acoustic intensity on B-mode ultrasound was stored in an uncompressed format for subsequent analysis with commercially available software (QLAB 8, Philips Healthcare, Andover, MA, USA). This software allows drawing of multiple regions of interest (ROIs) to obtain acoustic IDCs. Two ROIs were drawn within the thin polyurethane layer of the inflow and outflow tube in the water-filled basin. An additional movie file shows this in more detail (see Additional file 1). The IDCs were processed and fitted by the local density random walk (LDRW) model using MATLAB® 2009b. The LDRW model was employed since it provides both the best least square error fit to the IDC and a physical description of the dilution process. The MTT of the contrast bolus between the injection and the detection sites was directly derived from the parameters of the fitted model. Volumes were then calculated as the product between the measured flow and the difference in MTT between the two curves.
The MTT can be derived using two different methods. First, the MTT of each IDC can be estimated as the first order statistical moment of the fitted model, using the double fit method (Figure 3). Second, the indicator dilution system can also be interpreted as a linear system; therefore, the impulse response approach can be employed. The impulse response of the system between the two indicator detection sites was estimated by means of a parametric deconvolution technique, using the system input and output signals represented by the measured IDCs. The estimated impulse response is represented by the LDRW model, which allows blood volume assessments (Figure 3). The advantage of using a deconvolution technique over a double IDC fitting consists of the independency of the resulting impulse response from the injection function.
(Enlarge Image)
Figure 3.
Indicator dilution curves (IDCs) fitted by LDRW double fit method and impulse response method. These IDCs were constructed from post-processing imaging analysis at the level of inflow and outflow tubes at a flow of 2 L/min and true volume of 718 mL. The dotted black line is the IDC at the level of the inflow tube and outflow tube. In the upper picture, the red and blue lines depict the IDCs by the double fit method according to the LDRW model of the inflow and outflow tube. The vertical dashed lines represent the MTTs of each curve. In the bottom picture, the green line depicts the IDC according to the LDRW model impulse response method.
Statistics
All data were reported as mean values ± standard deviation (SD) or as median ± interquartile range (IQR) depending on the distribution of the variables of three consecutive measurements. The first goal was to investigate the agreement between measured volumes by both techniques and the true set-up volumes. Statistical significance was considered as a two-sided P<0.05. Bland-Altman analysis was used to determine the agreement between measured volumes and the true volumes. The effect of the different flows on the volume measurement was also investigated and reported in dedicated plots. Reproducibility was assessed by the intraclass-correlation coefficient (ICC). ICC consists of a basic calculation as repeated-measures analysis of variance (ANOVA) and the intraobserver reliability (ICC (1,1)). ICC assesses the agreement of quantitative variables on its reliability and consistency. The second goal was to analyze the correlation between the CEUS volumes and thermodilution volumes, assuming thermodilution as the gold standard. Correlation coefficients were assessed using the Pearson correlation coefficient R or the Spearman correlation coefficient rs depending on normal distribution or non-normal distribution of variables, respectively. Statistical analysis was performed using GraphPad Prism version 5.03 (GraphPad Software, San Diego, CA, USA) except for the intraclass-correlation, which was analyzed by Unistat® Statistical Package for Windows™ version 6.0 (Unistat House, London, England). Statistical analysis was performed by using all data (n=72 measurements) to exclude a bias.
SHARE
