Fig. 1. Experimental scheme for low exposure confocal microscopy based on photonic timestamped reconstruction. (a) The optical path: The excitation laser is expanded into parallel light and filtered before entering an X–Y scanning galvo system. The galvo mirrors coordinate with a bi-telecentric optical system containing a scanning lens and a tube lens to scan linearly on the focal plane. The emission light of the sample is filtered and collected by the lens and pinhole in the reflected path of BS. SPAD detects the signal. The electrical part: A controller syncs the positions of galvo mirrors, the laser pulses, and the high-time-resolution TCSPC module. Central: pulse generation principle, F: filter, BS: 50:50 beam splitter, SL: scanning lens, TL: tube lens, M: mirror, O: objective, S: sample, AL: achromatic lens, FC: fiber coupler, SPAD: single-photon avalanche diode, TCSPC: time-correlated single-photon counting module, GVS: dual-axis galvo scanning system. (b) The time sequences of the controller and processor: X, Y control the position of mirrors in order to make the illuminated laser scan in a raster way (shown in a schematic diagram on the left). In every position, a series of pulses are generated to excite the fluorophores on the focal plane, and the TCSPC module records the location signal from the controller and the timestamps of excited photons from the optical path. P: the laser pulse controller sequence, R: the emission photon received in the fiber coupler.

Fig. 2. Overview of the reconstruction strategy. 2D results of mouse kidney tissue are shown: (a), (d), (g) are reconstructed from the first 10-photon timestamped data; (b), (e), (h) first 20-photon; (c), (f), (i) first 30-photon. (a)–(c) are achieved by maximum-likelihood estimation. Then discrete wavelet transformation is conducted in (d)–(f). A deep-learning algorithm is exploited in (g)–(i). The mouse kidney tissue section is stained with Alexa Fluor 488 wheat germ agglutinin (green). Scale bars in (a)–(i) are 5µm. (j) shows the SSIM (structural similarity) difference among all the groups (n=6). Centerline, medians; limits, 75% and 25%; p*<0.05, p**<0.01, p***<0.001; ns: not significant (t test). (k), (l) are the enlarged mouse Bowman’s capsules in the white dashed boxes of (a), (g). The scale bars in (k), (l) are 2µm. (m) shows the plot profiles from the red dash lines of (k), (l). Max-L: maximum likelihood estimation, DWT: discrete wavelet transform, DL: deep learning.

Fig. 3. Multi-channel results of PT-Confocal. (a) A photon-count accumulation control of the two-channel result without timestamped information of the first 10 photons. (b), (c) The two-color results, which analyze the same number of photons every pixel in (a), illustrate the capacity of the system in low-flux biomedical fluorescent microscopy. The mouse kidney tissue section is stained with Alexa Fluor 488 wheat germ agglutinin (green) and DAPI DNA (blue). The scale bar is 5µm.

Fig. 4. 3D reconstruction of PT-Confocal. (a) shows the 3D spatial structure of mouse glomerulus and renal tubules in a spectrum-colored 2D map. The full image pixel size is 400×400. (c), (e), (g) The details of the mouse glomerulus in the relative z of 2.13µm from the white dashed box 1 in (a). (b), (d), (f) The details of the mouse Bowman’s capsule in the relative z of 3.20µm from the white dashed box 2 in (a). The mouse kidney tissue section is stained with Alexa Fluor 488 wheat germ agglutinin; scale bars are 5µm. C: control, Max-L: maximum likelihood estimation, DL: deep learning.