Fig. 1. (a) Schematic showing the principle of super-resolution optical fluctuation imaging. The image shows an example of two adjacent emitters, which could not be resolved due to the optical diffraction limit. By computing the autocorrelation (AC) result of intensity for each pixel with time-lags τ, two fluorophores could be distinguished successfully. (b) The diagram of our experimental setup: M1-M2, mirrors; D1-D3, dichroic mirrors; L1-L4, lenses; LBC, laser beam coupler; SMF, single-mode fiber.

Fig. 2. Workflow of TRUS framework. (a) Schematic showing the TRUS training procedure. In the forward propagating process, a sparse SOFI image (reconstructed from 20 frames) and an additional widefield image are first fed into the network to generate an initial output image (TRUS), and then the weights of the network are updated according to the partial derivative of the objective function, and an iteration is complete. This iteration continues until the minimum value of the objective functions is obtained. (b) In the inference procedure, the optimal model weights are loaded into the network in advance. The SOFI image (reconstructed from 20 frames) and corresponding widefield image are acquired and fed into the network for TRUS reconstruction.

Fig. 3. Reconstruction comparison of different test targets with the second-order SOFI, SACD, and TRUS framework. (a) Simulated widefield image and (b) corresponding ground-truth image. (c) Results reconstructed by second-order SOFI, SACD, and TRUS at 0.5 on-time ratio (defined as the proportion of time fluorophores remain in the active state). (d) Results reconstructed by second-order SOFI, SACD, and TRUS at 0.067 on-time ratio (defined as the proportion of time fluorophores remain in the active state). (e) The values of the structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) of TRUS (yellow line), SACD (green line), and second-order SOFI (purple line) versus on-time ratio.

Fig. 4. Comparison of (a) SACD reconstructed from 20 frames, (b) SACD reconstructed from 3000 frames, (c) second-order SOFI reconstructed from 3000 frames, (d) fourth-order SOFI reconstructed from 3000 frames, and (e) TRUS, with (f) HIFI-SIM as a reference. Scale bar: 5 µm.

Fig. 5. TRUS framework universally extends the spatial resolution of SIM under different biological samples. (a)–(c) Representative widefield images and corresponding results reconstructed by conventional second-order SOFI and TRUS of outer mitochondrial membrane, F-actin, and microtubule samples. Magnified views of the regions marked by the white boxes are shown on the right side. The line profiles of intensity across the white line are shown. (d) Statistical resolution comparisons of second-order SOFI and TRUS in the cases of outer mitochondrial membrane, F-actin, and microtubule samples. Resolution enhancements were measured by the decorrelation analysis (n=5, right). Scale bars: 6 µm (left), 1.5 µm [right, (a)–(c)].

Fig. 6. TRUS application in dual-color imaging and three-dimensional imaging. (a) Representative widefield images and results reconstructed by second-order SOFI (reconstructed from 3000 frames), sparse SOFI (reconstructed from 20 frames), and TRUS of microtubule (red) labeled with Alexa Fluor 647 and outer mitochondrial membrane (green fire blue) labeled with Alexa Fluor 488. (b) Subcellular structures in the regions marked by the white boxes in (a). (c) Intensity profiles along the corresponding yellow line shown in (b). (d) Color-coded, 3D distributions of microtubule in COS-7 cells labeled with Alexa Fluor 647 obtained with second-order SOFI (left) and TRUS (right). (e) Horizontal section views of widefield, sparse SOFI (reconstructed from 20 frames), second-order SOFI (reconstructed from 1000 frames), and TRUS from (d). (f) Intensity profiles along the corresponding white line shown in (e). Scale bars: 10 µm in (a), 2 µm in (b), and 6 µm in (c), (d).

Fig. 7. High-throughput super-resolution (SR) imaging with TRUS. (a) Application of TRUS (with 20 frames and an additional widefield) to high-throughput SR imaging of a 1.0mm2 area. Microtubules are identified in COS-7 cells labeled with Alexa Fluor 647. (b) Enlarged views of the white boxed regions in (a) and line profiles of intensity across the white line. (c) Decorrelation resolution enhancement factor (compared with widefield) maps over the entire FoV of the second-order SOFI image (top) and TRUS image (bottom). Scale bars: 100 µm in (a) and 5 µm in (b).