Ultrafast two-photon fluorescence imaging of cerebral blood circulation in the mouse brain in vivo

Significance Characterizing blood flow by tracking individual red blood cells as they move through vessels is essential for understanding vascular function. With high spatial resolution, two-photon fluorescence microscopy is the method of choice for imaging blood flow at the cellular level. However, its application is limited to a low flow speed regimen in anesthetized animals by its slow focus scanning mechanism. Using an ultrafast scanning module, we demonstrated two-photon fluorescence imaging of blood flow at 1,000 two-dimensional frames and 1,000,000 one-dimensional line scans per second in the brains of awake mice. These ultrafast measurements enabled us to study hemodynamic and fluid mechanical regimens beyond the reach of conventional methods.


Figures S1 to S5
Table S1 Legends for Movies S1 to S11 SI References Other supplementary materials for this manuscript include the following: We found that at high flow velocity, the PIV method was superior to an iterative Radon Transform method employed by us in a previous publication(1).To understand the origin behind this difference, we generated artificial blood flow data and tested the performance of these two methods against the ground truth (SI Appendix, Fig. S4).
We first generated the ground-truth temporally fluctuating positive velocity profiles by summing a constant baseline velocity with multiple sinusoidal functions with manually selected distinct amplitudes and frequencies.Next, we generated videos with 1,000 2D frames at 1 kHz containing a single blood vessel in which blood cells traveled at the ground-truth velocity.The positive flow direction was defined to be from left to right.Each frame was 50 µm × 50 µm, with a pixel size of 0.5 µm.For the first frame, pixels inside the vessel were initially set to the same user-chosen value while pixels outside the vessel was given the value 0.Then, we added 5 RBCs to the vessels that were almost equally distributed along the vessel but for random shifts of 1-5 pixels.The pixel values within the RBCs were modeled as a negative 2D Gaussian function with a standard deviation of 6 µm and a user-chosen amplitude.Gaussian noise was then added to the entire frame using MATLAB 'imnoise' function with a user-chosen SNR.Subsequent frames were generated the same way, except the locations of blood cells were calculated from those in the preceding frame using the ground-truth velocity profile.Once a blood cell exited the image from the right, a new blood cell was introduced from the left.Once the simulation video was generated, a kymograph was extracted along the centerline of the blood vessel.Radon transform and PIV analyses were then performed, and the extracted velocities compared with the ground truth (Fig. S4, images simulated with 5 sinusoidal functions for velocity profile, 128 for initial pixel value in vessel, amplitude of 60 for RBC image, 1dB for SNR).Radon transform mistook shadows of distinct RBCs as the streak left by a single RBC (Fig. S4 D, F), generating erroneous velocity measurements increasingly frequently at high flow speeds (Fig. S4 I, J), while PIV method reliably reproduced the ground truth up to 25 mm/s for 1 kHz 2D imaging data.

Figure S1 .
Figure S1.Cell morphology and flow reversion directly visualized by kHz full-frame imaging.(A) Superficial blood vessels with reversed flow.Top left: time-averaged image.Bottom left: example single-frame image.Top right: 4-frame average; White arrow: cell aggregate.Bottom right: 4frame average: white arrowheads: crescent-shape cells.(B) (Top) Time-averaged and (Bottom) single-frame images of a deep capillary.(C, D) kymograph and velocity plot along the green dashed line in (A).(E, F) kymograph and velocity plot along the green dashed line in (B).Orange rectangles, periods containing reversed flows.

Figure S2 .
Figure S2.Image registration can be essential for accurate velocity measurement.(A) Unregistered (top) and registered (bottom) time-averaged images.(B) Kymographs extracted along the green dashed lines in unregistered and registered images, respectively.Note the missing section in the unregistered data due to vessel movement.(C) Flow velocity calculated from kymographs in (B).Note the velocity deviations between the unregistered and registered data, caused by motion artifacts.

Figure S4 .
Figure S4.Cross-correlation-based PIV method outperforms iterative Radon transform in simulated blood flow images.(A) Individual frames of simulated 1kHz FACED images.Red arrowheads: a simulated blood cell traveling at an average speed of 4.0 mm/s.(B-E) Kymographs measured along the centerline of simulated vessel with average flow speed of 0.5, 4.0, 20.0, and 25.0 mm/s, respectively.(F) Zoomed-in view of the blue box in E. Red arrowheads: a fast-moving RBC.Yellow arrowheads: multiple distinct RBCs.(G-J) Extracted velocity profiles with PIV and Radon transform, compared with the ground truth.

Figure S5 .
Figure S5.Comparison of results from SIFT flow and 1D PIV.(A) Example temporally varying flow profiles from SIFT-flow (black) and 1D PIV (red).(B) Correlation between the mean velocity from 61 line ROIs in 12 blood vessels with multi-file flows (gray dots) with the two methods.Black line has a slope of 1.