2.1.

Image Acquisition and Phase Average Filtering

High-speed bright field imaging (2000 frames per second) was used to capture cardiac contraction over a 2-s window (approximately four cardiac cycles) in zebrafish ranging in age from 3 days post fertilization (dpf) to 6-dpf ($n=4$ at each stage). Acquisition rates at this speed ensure that the desired pixel displacement is met for PIV analysis ($1/4$ of the interrogation window) at the high spatial resolution used in this study (0.28 μm pixel size). At this rate of acquisition, we obtained $\sim 800$ instantaneous velocity measurement time points per cycle (depending on fish age).

The temporal oversampling due to the high-speed acquisition of data can be used as an opportunity to improve the velocity measurements by binning the measurements into 100 sections in each cardiac cycle and averaging them to reduce erroneous vectors. Mathematical analysis of blood flow has successfully modeled cardiac output with a finite Fourier model having seven harmonics;³⁴ as a result, a minimum of 14 bins is required to accurately model cardiac function. Measurement-based studies have used larger numbers of bins to ensure the accuracy of results, for example, 42 bins per cardiac cycle, as used by Leo et al.³⁵ Cardiac defects may result in changes to the pattern of blood flow as well as the magnitude, resulting in a more complex model; therefore, we have utilized 100 bins in our analysis to ensure the accurate measurement of irregular flow patterns. Additionally, as each imaging sequence contains multiple cardiac cycles, it is possible to phase average the PIV measurements. These steps require first dividing each cycle into bins and then aligning the periods so that they are in phase with each other. By initially performing PIV on a small section of the overall image, the onset of contraction could be determined and used to align the cardiac cycles (Fig. 1). The binning of images from each cardiac cycle also removes the effect of variance in the length of each cardiac cycle (a variance of 10 to 15 ms being common). Each frame within a cycle is allocated a relative bin within the cardiac cycle. This provides the ability to compare frames from different periods in this quasi-periodic flow, enhancing the accuracy of the velocity measurements acquired by reducing erroneous vectors. Because of the high temporal resolution of the acquisitions ($\sim 800$ frames for one cardiac cycle of a 4-dpf embryo), each of the 100 bins contains approximately eight averaged velocity measurements from each cycle.

Fig. 1

Flow data is interrogated by performing PIV on a small section of the image to identify the start of each cardiac cycle. Each cardiac cycle is then divided into 100 bins. The velocity measurements from each bin in multiple cardiac cycles are combined to create a temporally averaged velocity measurement. For reference each cardiac cycle is approximately 800 frames in length.

To remove the contribution of the static heart structures from the PIV analysis of blood flow, which causes underestimation of velocity, averaged images are created for each of the 100 bins within a cycle. Each average image is subtracted from the individual frames within the bin, resulting in the filtering of the static heart structures, and those moving less than $0.07\mathrm{mm}/\mathrm{s}$ (which would appear stationary at these frame rates) are also removed from the imaging sequence as illustrated in Fig. 2 (Video 1). Following filtering, PIV analysis, using the full temporal resolution available, is carried out on the unbinned filtered images.

Fig. 2

Bright field microscopy images of a 4-dpf zebrafish heart. (a) Raw image of the zebrafish heart, (b) average image formed by the binning process described in Fig. 1, and (c) final image produced by subtracting the average image (b) from the raw image (a) (Video 1, QuickTime, 9.5 MB) [URL: http://dx.doi.org/10.1117/1.JBO.17.3.036007.1].

2.2.

Particle Image Velocimetry Analysis

The computational expense of PIV can be reduced by defining the region of interest within the images for analysis. We carried out imaging on zebrafish expressing green fluorescent protein under the control of the cardiac myosin light chain 2 promoter. For each bright field acquisition, we captured a corresponding sequence of fluorescent images at a rate of 100 frames per second immediately following bright field capture. These fluorescent images were manually phase matched to the bright field images in order to provide a synchronized data set (Fig. 3) and thresholded to create a mask for the PIV analysis. The fluorescence image clearly identifies the area of interest for velocity calculation and is ideal for masking of the bright field images. The mask images were used to restrict the area of the image analyzed for PIV, reducing the computational expense of the technique. Additionally, by utilizing a mask we immediately combine heart wall information obtained from fluorescence imaging with the velocity measurements obtained from bright field imaging.

Fig. 3

Synchronization of bright field and fluorescent images for PIV mask creation. (a) Bright field microscopy image, (b) fluorescence image used for mask creation, and (c) combination of the two images following temporal alignment. The fluorescence image is seen to clearly identify the area of interest for velocity calculation.

PIV analysis was carried out on each of the masked data sets to provide detailed velocity measurements over the period of zebrafish development from 3 to 6-dpf. Figure 4 (Video 2) shows the peak flow from the atrium to the ventricle and clearly illustrates the complexities of the flow being investigated. Video 2 illustrates the temporal resolution of the averaging technique, showing all 100 time-points (10 bins per second). These data show that for large parts of the cardiac cycle, the flow within the ventricle is negligible, with short periods of high inflow or outflow. Flow into the ventricle displays a vortical nature brought about by the enlargement of the chamber, while flow through the ventricular-bulbar valve is relatively parallel. Immediately after the valve the flow once again shows vortical nature owing to the expansion of the bulbus arteriosus.

Fig. 4

Blood flow velocity measurements in a 6-dpf zebrafish during peak flow from the atrium into the ventricle. Vectors show direction and magnitude of velocity (standard color rainbow, red is high). Vectors are evaluated at 4.5-µm spacing with an interrogation window size of $18\times 18\mathrm{\mu m}$. For clarity only every second vector is shown. Video 2 provides all 100 time-point measurements (QuickTime, 11.2 MB) [URL: http://dx.doi.org/10.1117/1.JBO.17.3.036007.2].

To determine the improvement in PIV analysis following image filtering, we compared the cross correlation peaks obtained with and without cardiac-phase filtering to an idealized result generated with synthetic data (Fig. 5). Stationary structures within the image pairs will be included in the cross correlation, resulting in a shift of the apparent maximum of cross correlation towards the centroid of these structures and, in this instance, reducing the measured velocity. The noise resulting from these structures and the distortion of the cross correlation peak are clearly evident in the unfiltered data [Fig. 5(b)] but removed as a result of cardiac-phase filtering [Fig. 5(c)]. To examine the effect of cardiac-phase filtering on the velocity measurements, we compared the unfiltered data to the filtered data [Fig. 6(a)]. This comparison identified more than a 3-fold difference in average velocity during peak flow ($1.1\mathrm{mm}/\mathrm{s}$ in filtered data compared to $0.35\mathrm{mm}/\mathrm{s}$ in the unfiltered data sets).

Fig. 5

Cross correlation images used in PIV analysis. (a) Cross correlation peak for synthetic, ideal data. (b) Cross correlation from 6-dpf unfiltered images. A clear ridge is evident distorting the peak, indicative of low-frequency noise such as stationary structures. (c) Filtering of the data analyzed in (b) demonstrating successful removal of noise from the analysis and improvement in the cross correlation, as evidenced by the similar appearance to the ideal case in (a).

Fig. 6

Average velocity measurements through the ventricular-bulbar valve. (a) Comparison of velocity measurements before (blue squares) and after (red circles) cardiac-phase average filtering in a 6-dpf zebrafish. (b) Symbols provide velocity measurements for individual fish (red circles, green squares, blue triangles, and gray diamonds) at 5-dpf, and the solid line shows the average of these. Video 3 provides the bright field imaging of the fish displaying retrograde flow through the ventricular-bulbar valve (green squares, QuickTime, 10.2 MB). (c) Average velocity measurements for each age group: 3-dpf (gray diamonds), 4-dpf (blue triangles), 5-dpf (green squares) and 6-dpf (red circles). Retrograde flow is evident in 3-dpf fish prior to systole between 0.8 and 1 period. (Video 3, MOV, 9.7 MB) [URL: http://dx.doi.org/10.1117/1.JBO.17.3.036007.3]

Within each age group filtered velocity measurements were highly reproducible and were used to form an average for each time point [Fig. 6(b)]. Figure 6(b) also shows the sensitivity of our approach. Note that Fig. 6(b) shows one fish exhibiting retrograde motion through the ventricular-bulbar valve in the latter part of the cycle. Remarkably, this retrograde flow, clearly evident in the quantitative analysis of this fish, can be seen to correspond to the leaking of one red blood cell (RBC) at a time back through the valve (Video 3).

Analysis of blood flow between 3 and 6-dpf identified a reduction in both peak velocity and the contractile period of the cardiac cycle as the fish increase in age [Fig. 6(c)]. The contractile period is seen to reduce approximately 10% each day between 3 and 6-dpf. Interestingly, this reduction in contractile period is not accounted for by a substantial change in heart rate (average 158 beats per minute). A steady decrease in the velocity is also witnessed during this age range [Fig. 7(a)]. Figure 7(b) provides the maximum velocity measurement recorded through the valve, rather than the spatial average used in Fig. 6, which nevertheless shows a similar trend to the spatially averaged results. This gives an indication of the maximum velocities measured in the center of the valve. We observe a high variance in maximum velocity between individuals, particularly in the younger age groups. Even within a single batch of fertilized embryos raised in the same conditions, there is subtle variation in the rate of development between individuals. We believe the high variance is due to these subtle variations in developmental rate as well as variation between individuals. The difference in development become less significant with time, and therefore there is decreasing variation with increasing age.

Fig. 7

Maximum average velocity (a) and maximum velocity (b) measurements acquired through the ventricular-bulbar valve. Maximum velocity and maximum average velocities are measured before and after spatial averaging is performed, respectively. Linear regression lines are provided for clarity.

The measurements provided in Figs. 6 and 7 should not be confused with flow measurements or with oxygen transportation capacity. They represent the velocity at which the present RBCs are traveling. Two possible explanations for this reduction in both contractile time and velocity are the increase in size of the ventricular-bulbar valve and the increased hematocrit brought about by the ongoing production of RBCs during the early development of the embryonic zebrafish. These factors, individually or combined, may account for the reduction in velocity, while maintaining or increasing the fish’s ability to circulate oxygen.

It is important that automated analyses of flow are able to detect subtle changes in heart function. One area of interest is the reverse flow through the valve. The motion witnessed in Fig. 6(c) immediately after systole in the 4-, 5-, and 6-dpf fish could easily be mistaken for retrograde flow through the valve; however, this motion is due to the closure of the valve pushing several RBCs back into the ventricle. While this trait is not present in the 3-dpf fish, reverse velocity measurements are evident immediately before systole. This motion is true retrograde flow caused by the force applied to the ventricular-bulbar valve as the ventricle expands. This force opens the underdeveloped valve, causing RBCs to move back into the ventricle through the ventricular-bulbar valve from the difference in pressure. These results demonstrate the capability of this approach to detect subtle developmental changes in the ventricular-bulbar valve.

4.1.

Microparticle Image Velocimetry

An in-house code was used for performing the image processing and velocity measurements in this study.³⁶ This code has been developed over a number of years and rigorously validated by Fouras et al.³⁷ and Nesbitt et al.³⁸ Whole field velocity measurements were performed on the ventricle of each subject, which were masked by fluorescence image data. An interrogation window size of $18\times 18\mathrm{\mu m}$ was used with a spacing between measurements of 4.5 µm.

Velocity measurements though the ventricular-bulbar valve were calculated on a subregion of the full images $18\times 18\mathrm{\mu m}$ in size ($64\times 64$ pixels). An initial large subregion of $45\times 45\mathrm{\mu m}$ ($160\times 160$ pixels) was chosen so that the ventricular-bulbar valve was centered within the region. An interrogation window size of $18\times 18\mathrm{\mu m}$ was used with a spacing between measurements of 2.25 μm. The larger initial subregion ensures that RBCs in the desired subregion do not leave the image in the subsequent frame. The measurements outside the desired subregion are later discarded, and a spatial average of the remaining vectors is used to obtain the average velocity through the valve. Because of the small region size, no masking was used for these measurements.

4.2.

Preparation of Zebrafish Samples

Homozygous transgenic cardiac myosin light chain2-GFP zebrafish embryos³⁹ were collected and raised in E3 embryo medium (50 mm NaCl, 1.67 Mm KCl, 3.3 mM CaCl, 3.3 mM ${\mathrm{MgSO}}_4$, in deionized water) at 28 °C. Embryos were anesthetized with 0.016% tricaine (3-amino benzoic acid ethyl ester) and mounted in 0.7% low-melting-point agarose dissolved in E3 embryo medium together with 0.016% tricaine, prior to imaging. The embryos were mounted in Fluorodish imaging chambers (World precision instruments).

4.3.

Imaging Setup

An inverted microscope (Leica DM ILM) with a $20\times$  objective lens (Nikon bright field CFI Plan Apo $20\times /0.75$) was coupled via an optical spacer (Leica $2.5\times$) to a high-speed camera (IDT Y4) to obtain an effective pixel size of 0.28 µm. A three-axis robotic sample stage (Aerotech) was used for high-resolution ($x$, $y$ resolution, 1 μm; $z$ resolution, 0.1 μm) specimen positioning. Zebrafish with fluorescently labeled cardiac muscle cells were mounted such that viewing the heart was possible from below, enabling the main chambers of the heart to be in focus. Specimens were aligned to minimize flow in the out-of-plane direction through the ventricular-bulbar valve. Consistent orientation of the embryos was achieved by using the anatomical features surrounding the heart. Bright field imaging was used for PIV measurements, while fluorescence imaging was used to acquire heart wall information. The fluorescence images were acquired immediately after the bright field acquisition, with the only change to the optical path between the specimen and the camera being the inclusion of a dichroic filter. Temperature during image acquisition was $22.5\pm 0.5°\mathrm{C}$.