A gradient-based, GPU-accelerated, high-precision contour-segmentation algorithm with application to cell membrane fluctuation spectroscopy.

Michael Mell,Francisco Monroy,Jorge Bernardino De La Serna

doi:10.1371/journal.pone.0207376

Abstract

We present a novel intensity-gradient based algorithm specifically designed for nanometer-segmentation of cell membrane contours obtained with high-resolution optical microscopy combined with high-velocity digital imaging. The algorithm relies on the image oversampling performance and computational power of graphical processing units (GPUs). Both, synthetic and experimental data are used to quantify the sub-pixel precision of the algorithm, whose analytic performance results comparatively higher than in previous methods. Results from the synthetic data indicate that the spatial precision of the presented algorithm is only limited by the signal-to-noise ratio (SNR) of the contour image. We emphasize on the application of the new algorithm to membrane fluctuations (flickering) in eukaryotic cells, bacteria and giant vesicle models. The method shows promising applicability in several fields of cellular biology and medical imaging for nanometer-precise boundary-determination and mechanical fingerprinting of cellular membranes in optical microscopy images. Our implementation of this high-precision flicker spectroscopy contour tracking algorithm (HiPFSTA) is provided as open-source at www.github.com/michaelmell/hipfsta.

Highlights

Quantitative imaging is progressively empowering the analytic toolbox of cellular biology with high-performance observational facilities accessing new biophysical markers resolved in space and time [1,2]
The contour-segmentation algorithm we present here harnesses the computational power of general purpose graphics processing units (GPGPU) having recently become available to significantly improve on current segmentation methods in the sub-pixel performance of nanometer resolution
We have developed a high-precision flicker spectroscopy contour tracking algorithm based on an intensity-gradient segmentation schema (HiPFSTA, available open-source at www.github. com/michaelmell/hipfsta), which is designed to run on graphical processing units and makes extensive use of image-oversampling using the processing power of these processors

Summary

Introduction

Quantitative imaging is progressively empowering the analytic toolbox of cellular biology with high-performance observational facilities accessing new biophysical markers resolved in space and time [1,2]. The positions for performing the linear fits become independent of the pixelgrid by interpolating pixel-intensities, which overcomes a subtle, but significant issue with Pecreaux’ method that is related with the fixed positions of the centers of the pixels These mutual improvements have allowed us to generate a new hybrid algorithm, more robust and precise than the previous ones, which enhances the accuracy of the segmentation method down to 2nm spatial precision in determining changes in the position of the contour halo. All these improvements empower the proposed method with a significantly higher performance than previous ones, in terms of enhanced spatiotemporal accuracy that allows resolving the membrane fluctuations with a higher precision in amplitude and in spatial and time resolution, both in real and reciprocal domains Such a breakthrough in segmentation performance should be crucial to approach new problems of cell mechanics where membrane fluctuations can be exploited as a relevant observable. Imaging membrane fluctuations could become an excellent observational method for non-invasive and non-stressing probing the mechanical phenotypic traits of cellular membranes where the intrinsic rigidity of the cytoskeleton necessarily imposes low amplitudes and short correlation times and distances of the membrane fluctuations

Method

Findings

Conclusion