October 10, 2018
by Abhijit Marar
We started studying the signal-to-noise ratio of the system, by initially running simulations. Different simulations were performed to show the effect of read noise on the performance of the system. Figure 1. shows the degradation in the quality of the reconstruction as the read noise increases. The simulations assumed that the light source in the center of the frame was emitting 5000 photons. (a), (b) and (c) are the one of the raw holograms with read noise of Δ= 0.1, Δ = 0.3 and Δ = 1.0 respectively and (d), (e) and (f) are the reconstructions of the raw holograms. It can be seen that for a constant number of photons, as the read noise increases the raw data recorded by the sensor deteriorates, thus degrading the quality of the reconstruction.
Fig.2, confirms the argument made above. For a constant number of photons (5000) emitted but the source, increasing the read noise, reduces the signal to noise ratio. The EMCCD camera being currently used in the lab right now measures a read noise of approximately 1.0 at an exposure of 50ms with an intensity of 50 mW being output from the laser. The plot in Fig2. shows that we will require more than 5000 photons to be able to successfully reconstruct the image of a bead or single molecule at that exposure time and laser power.
The plot in the figure below (Fig.3) makes the same point. It shows that for a constant read noise of 0.5 electrons, higher photons are needed to achieve high signal-to-noise-ratios.
We continued to perform STORM imaging with our collaborator. After multiple runs of imaging microtubules, our collaborator was satisfied with imaging parameters and wanted to try imaging neurons. The images below show the results from our first attempt at imaging the neurons from a fruit fly (Drosophila). The image shows the filopodia of the neuron. The image on the left is the wide-field image that is diffraction limited and the image on the right is the STORM image that has been reconstructed from 52000 frames.