Scientists have used various approaches to improve spatial resolution in fluorescence microscopy. Deconvolution, in which images are numerically deblurred based on a knowledge of the microscope’s point spread function, is one approach. Deconvolution, however, can lead to noise-amplification artifacts between the sample and the image, decreasing the image resolution. To prevent artifacts, researchers at Boston University developed an image deblurring algorithm that sharpens images via pixel reassignment. The new method, called deblurring by pixel reassignment (DPR), produces a deblurring effect like deconvolution, but without the drawbacks associated with conventional deconvolution algorithms. Deblurring by pixel reassignment remaps raw fluorescent microscopy images to sharpen images. Courtesy of Zhao and Mertz, doi: 10.1117/1.AP.5.6.066004. Deblurring by postprocessing can also lead to negativities and failure to conserve local linearity between the sample and the image. Because DPR relies solely on pixel reassignment, no negativities are possible in the final reconstruction of the image. Intensity levels are rigorously conserved by DPR, eliminating the need for additional procedures to ensure local linearity. DPR’s ability to sharpen images without introducing noise artifacts or negativities could improve fluorescence microscopy for a range of scientific applications. To ensure consistent results, raw fluorescence images are preconditioned before DPR is applied. The purpose of preconditioning is to standardize the raw images before the algorithm is used. The actual sharpening of the image is then performed by pixel reassignment. Intensities (i.e., pixel values) at each grid location (i.e., pixel) are reassigned to neighboring locations according to the direction and magnitude of the locally normalized image gradient, scaled by a gain parameter. Because pixels are generally reassigned to off-grid locations, their pixel values are distributed to the nearest on-grid reassigned locations as weighted by their proximity. Unlike Wiener deconvolution, which is performed in Fourier space using a division operation, DPR operates entirely in real space with no division operation that can significantly amplify noise. Unlike Richardson-Lucy deconvolution, DPR is noniterative and can be performed in a single pass. Conventionally, the resolution of a microscope is defined by the minimum separation distance required for two points to be resolved based on a predefined standard. DPR can reduce this separation distance. The DPR algorithm helps distinguish nearby fluorophores, even when they are separated by distances smaller than the conventional resolution limit, helping to facilitate, for example, the application of single-molecule localization microscopy in dense samples. Resolution enhancement with DPR. (a) DPR with gain 1 and 2 applied to simulated images of two closely spaced point objects. Left column: The two point objects are separated by 1.68 σ. Right column: The two point objects are separated by 1.41 σ. The scale bar represents 2 σ. (b) The simulated results of the intensity dips of two closely spaced point objects with varying separation distances in raw, DPR gain 1, and DPR gain 2 images. (c) DPR gain 1 and 2 applied to the confocal images of the BPAE cells. The scale bar represents 600 nm. Courtesy of Zhao and Mertz, doi: 10.1117/1.AP.5.6.066004. To demonstrate the effectiveness of DPR, the researchers applied it to single-molecule localization, structural imaging of engineered cardiac tissue, and volumetric zebrafish imaging. These diverse, real-world applications showed DPR’s potential to improve the clarity of microscopic images. DPR provides the ability to sharpen images while preserving larger structures. It could be used in scenarios where samples contain both small and large structures, adding to its versatility as a research tool. The DPR algorithm can be applied to general microscope modalities and fluorophore types and requires minimal assumptions about the emission point spread function to use. It can be used on both a sequence of raw images, a few images, and even a single image to enable temporal analysis of fluctuating fluorophore statistics. The researchers have made the DPR algorithm available as a MATLAB function compatible with Windows or macOS. The DPR algorithm could provide a user-friendly solution to enhancing the spatial resolution of microscopy images, by improving image clarity while avoiding common noise-related issues. “Because of its ease of use, speed, and versatility, we believe DPR can be of general utility to the bioimaging community,” said Jerome Mertz, professor of biomedical engineering at Boston University. The research was published in Advanced Photonics (www.doi.org/10.1117/1.AP.5.6.066004).