Sparsity-based background removal for STORM super-resolution images

TL;DR

The paper addresses background fluorescence in STORM by introducing a sparsity-based background removal method using SLNet to compute a low-rank background and sparse signal. It trains SLNet in an unsupervised manner to estimate the low-rank component and derives sparse frames $S = (M - L)_{\ge 0}$, yielding cleaner inputs for STORM reconstruction. On two datasets (glial cells and microtubules), SLNet achieves higher localization precision, lower emitter FWHM, and better agreement with wide-field references as measured by NanoJ-SQUIRREL, with training times under a few minutes. The approach is lightweight, dataset-agnostic, and publicly available, offering a practical pre-processing step for improved STORM results.

Abstract

Single-molecule localization microscopy techniques, like stochastic optical reconstruction microscopy (STORM), visualize biological specimens by stochastically exciting sparse blinking emitters. The raw images suffer from unwanted background fluorescence, which must be removed to achieve super-resolution. We introduce a sparsity-based background removal method by adapting a neural network (SLNet) from a different microscopy domain. The SLNet computes a low-rank representation of the images, and then, by subtracting it from the raw images, the sparse component is computed, representing the frames without the background. We compared our approach with widely used background removal methods, such as the median background removal or the rolling ball algorithm, on two commonly used STORM datasets, one glial cell, and one microtubule dataset. The SLNet delivers STORM frames with less background, leading to higher emitters' localization precision and higher-resolution reconstructed images than commonly used methods. Notably, the SLNet is lightweight and easily trainable (<5 min). Since it is trained in an unsupervised manner, no prior information is required and can be applied to any STORM dataset. We uploaded a pre-trained SLNet to the Bioimage model zoo, easily accessible through ImageJ. Our results show that our sparse decomposition method could be an essential and efficient STORM pre-processing tool.

Figures (6)

• Figure 1: Left: The three main steps conducted in STORM, where the background removal methods are applied in the pre-processing stage. Right: The sparse decomposition process and the architecture of the neural network SLNet. The network approximates the low-rank representation stack from the input stack, and the sparse images are calculated as a subtraction of the two. The parameter k controls the size of the input stack (here, $k=3$).
• Figure 2: Super-resolution reconstruction of glial cells using ThunderSTORM and three background removal methods: median, "rolling ball" with radius 3 pixels and our SLNet background removal method. (1), (2), (3) Magnified views of three selected regions with scale bar $1\mu m$.
• Figure 3: Super-resolution reconstruction of microtubules using ThunderSTORM and three background removal methods: median, "rolling ball" with radius 10 pixels and our SLNet background removal method. (1), (2), (3) Magnified views of three selected regions with scale bar $1\mu m$.
• Figure 4: The measurements were taken for the raw frames and the three methods (SLNet, Median Background Removal, and "Rolling Ball" algorithm) on the glial cells and microtubules dataset Leterrier:34Kumar:35. Left: Box plot presenting quantitative results of around 45 measured emitters' FWHM on the STORM emitter images. Right: Box plot presenting quantitative results of around 20 measured tubules' FWHM of the reconstructed super-resolution images.
• Figure 5: Wide-field images of both datasets and SQUIRREL analysis comparing the four respective STORM reconstructions. SLNet delivers better linearity in reconstruction concerning the wide-field image of datasets.