High-speed two-photon microscopy with adaptive line-excitation

Abstract

We present a two-photon fluorescence microscope designed for high-speed imaging of neural activity at cellular resolution. Our microscope uses an adaptive sampling scheme with line illumination. Instead of building images pixel by pixel via scanning a diffraction-limited spot across the sample, our scheme only illuminates the regions of interest (i.e., neuronal cell bodies) and samples a large area of them in a single measurement. Such a scheme significantly increases the imaging speed and reduces the overall laser power on the brain tissue. Using this approach, we performed high-speed imaging of the neuronal activity in mouse cortex in vivo. Our method provides a sampling strategy in laser-scanning two-photon microscopy and will be powerful for high-throughput imaging of neural activity.

Fig. 1.

Principle, optical setup, and point-spread function (PSF) of the two-photon scanning microscope with adaptive line-excitation. (a)–(c) Working principles of adaptive line-excitation. (a) A high-resolution video of the neuronal activity is acquired in the calibration step through an equivalent point scanning strategy. (b) The ROIs (i.e., neuronal cell bodies) could then be segmented. They are then binarized into a mask and loaded to the DMD. The laser (920 nm femtosecond laser) light is first shaped to a short line and incident to the DMD, which is located at the conjugate plane of the sample plane and works as an intensity spatial modulator. (c) The beam diffracted from the DMD carries the information of the ROI, and illuminates the corresponding part of the sample ROI. Only the ROIs but not the background region are imaged. Top, illustration of the excitation scheme on the sample, with the arrow showing the beam scanning direction. Bottom, (a)(c) zoom-in view of a sub-region of the image recorded by the photomultiplier tube (PMT). (b) Binary mask loaded on the DMD. The four ROIs are labeled in different numbers. (d) Schematic of the two-photon microscope setup with adaptive line-excitation scheme. The laser beam is first shaped into a short line, which is scanned by a resonant scanner and a galvanometer mirror onto the DMD. Light diffracted from the DMD is relayed to the sample plane through a relay system and the tube lens and objective lens. The fluorescence is detected by the PMT. M, mirror. (e) Measured PSF in the lateral direction ( $xy$ ) for line-excitation. (f) Measured axial PSF using 5 µm fluorescent beads. Scale bar in (a) and (c) is 10 µm.

Fig. 2.

Simulation of the adaptive line-excitation sampling in calcium imaging. (a) Time-series standard-derivation projection frame of the simulated high-resolution calcium imaging video. (b) Mask for (a) showing individual ROIs, constructed from the ground truth in simulation. Some of the neurons have spatial overlaps with others. (c) Time-series average projection frame of the video constructed by the adaptive line-excitation sampling on the original high-resolution video, using the mask shown in (b). The frame was resized to the same dimension as the original video. The spatial footprints of the extracted ROIs through CalmAn are outlined in red. (d) Extracted normalized temporal activity traces of the representative neurons in (c), which were indicated by the yellow arrows in (c). Black, ground truth with noise and background included; green, extracted traces using CalmAn on the original high-resolution video; red, extracted traces using CalmAn on the adaptive line-excitation sampled video. (e) Intersection over union (IoU) between the CalmAn-extracted spatial footprint and the ground truth of individual neurons in the high-resolution video versus that calculated for the adaptive line-excitation sampled video. The green lines are indications of $\mathrm{I}\mathrm{o}\mathrm{U}=0.33$ . The color of each dot shows the signal-to-noise ratio (SNR) of individual neurons in the original high-resolution video. The SNR is defined as the maximum of the denoised temporal signal (with signal baseline included) over the standard deviation of the noise. The ROIs with spatial overlap with others are indicated by diamond shape symbols, while the isolated ROIs are plotted in circular dots. (f) Same as (e), but for the Pearson correlation coefficient (PCC) between the CalmAn-extracted traces and the noise-free ground truth traces. The green lines are indications of $\mathrm{P}\mathrm{C}\mathrm{C}=0.9$ . (g) Three pairs of comparisons illustrating that the neuronal signals from the neurons with spatial overlaps could be demixed in the adaptive line-excitation sampled video. In each pair group, the top-left and top-right figures show the spatial footprint of the neurons in the high-resolution ground truth and the time-series average projection frame of adaptive line-excitation sampled video, respectively. The middle panel shows the temporal traces of the ROI contoured in red: gray, ground truth without noise and background; black, ground truth with noise and background included; red, CalmAn-extracted from the adaptive line-excitation sampled video; magenta, CalmAn-extracted from the high-resolution video. Similarly, the bottom panel shows the temporal traces of ROI contoured in green: gray, ground truth without noise and background; black, ground truth with noise and background included; green, CalmAn-extracted from the adaptive line-excitation sampled video; cyan, CalmAn-extracted from the high-resolution video. All the traces are normalized to [0,1]. Scale bar in (a) and (b) is 50 µm.

Fig. 3.

Validation of the adaptive sampling scheme through phantom samples. (a),(b) Projection of the DMD mask onto a uniform fluorescent slab through adaptive short-line excitation. (a) DMD mask ( $1280\times 800\mathrm{p}\mathrm{i}\mathrm{x}\mathrm{e}\mathrm{l}\mathrm{s}$ ) with characters “UC DAVIS.” (b) Top: raw image ( $256\times 64\mathrm{p}\mathrm{i}\mathrm{x}\mathrm{e}\mathrm{l}\mathrm{s}$ ) acquired from PMT through adaptive line-excitation with bi-directional scanning; bottom: interpolated and resized image ( $500\times 695\mathrm{p}\mathrm{i}\mathrm{x}\mathrm{e}\mathrm{l}\mathrm{s}$ ) with square pixels. (c)–(f) Imaging of a phantom sample with randomly distributed fluorescent beads, in 12 µm diameters. (c) A high-resolution image of the sample was acquired through the equivalent point scanning approach. (d) Binary mask on the DMD. The two corner regions of the mask were outside the DMD active regions and were not displayed. (e) Single frame of the recording from the sample acquired at 198 Hz, using the adaptive line-excitation with bi-directional scanning. (f) Simulated high-speed image based on the binary mask and the bi-directional scanning trajectory of the resonant scanner. Scale bar in (b),(c),(e),(f) is 50 µm.

Fig. 4.

In vivo calcium imaging of mouse V1 at 150 µm depth using the adaptive line-excitation two-photon microscope. (a) Time-series average projection frame of the high-resolution image from the effective point scanning method, which includes the fine contours of the neurons and the background. The image has $256\times 320\mathrm{p}\mathrm{i}\mathrm{x}\mathrm{e}\mathrm{l}\mathrm{s}$ over a $500\text{µ}\mathrm{m}\times 695\text{µ}\mathrm{m}$ FOV. The periodic lines were artifacts and formed due to the tilted scanning trajectory and corresponding shifted DMD pixels (Supplement 1, Note 3, Supplement 1, Fig. S6). They did not impact the segmentation results. (b) Time-series average projection frame of the line-scanning sampled video ( $256\times 64\mathrm{p}\mathrm{i}\mathrm{x}\mathrm{e}\mathrm{l}\mathrm{s}$ ) for the same FOV in (a) without adaptive sampling, where all the pixels in the DMD were turned on. (c) Corresponding binary mask for (a), constructed from the ROI segmentation algorithm of SUNS. (d) Time-series average projection frame of the line-scanning sampled video ( $256\times 64\mathrm{p}\mathrm{i}\mathrm{x}\mathrm{e}\mathrm{l}\mathrm{s}$ ) with adaptive sampling, by applying the mask on DMD. Only the ROIs defined in (c) were illuminated and sampled. The spatial footprints of the extracted ROIs through CalmAn are outlined in red. (e) Temporal activity traces of the representative neurons recorded at 198 Hz, which were indicated by the yellow arrows in (d). The temporal activity traces were extracted by CalmAn. Scale bar in (a),(b),(d) is 50 µm.