Benchmarking Quickpalm and Other Molecule Localization Software for Super-Resolution Microscopy

Abstract

Image processing software is used to estimate the positions of fluorescent molecules in super-resolution microscopy images. The rejection methods (for distinguishing single-molecule and multi-molecule images) and fitting algorithms (for estimating molecular positions) used determine the quality and reliability of the reconstructed images. We benchmarked and analyzed the ImageJ plug-in called QuickPALM. To do this, we used simulations of two-molecule images (with realistic point spread functions and Poisson noise model) as input. These images were generated with known molecule positions, to test the predictions of QuickPALM against the actual images. Our analysis showed that the accuracy dropped significantly when molecules were less than one wavelength apart, as there was rejection of multiple molecules in close proximity. With our results, we were also able to show that localization bias occurred towards the center of each pixel, distorting the overall image. The result was confirmed by simulating a PALM experiment on the US Air Force Test Pattern. We also wrote our own molecule localization plug-in (for ImageJ) utilizing second-moments (as a rejection criterion) and a fast maximum likelihood estimator (for estimating molecular positions), to compare with QuickPALM. We were able to show that the new plug-in had improved precision and accuracy when using the same single-molecule and two-molecule images. We also showed that the maximum likelihood estimator had no bias towards the center of each pixel when processing the US Air Force Test Pattern.