Skip to main content

Resultados relacionados

Resultados totales (Incluyendo duplicados): 2
Encontrada(s) 1 página(s)

DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

Optical flow driven interpolation for isotropic FIB-SEM reconstructions

  • González-Ruiz, Vicente
  • García-Ortiz, Juan Pablo
  • Fernández-Fernández, María Rosario
  • Fernández, José Jesús
[Background and Objective]: Focused Ion Beam - Scanning Electron Microscopy (FIB-SEM) allows three-dimensional ultrastructural analysis of cells and tissues at the nanoscale. The technique iteratively removes a section of the sample with a FIB and takes an SEM image from the exposed surface. The section thickness is usually higher than the image pixel size to reduce acquisition time, thus resulting in anisotropic resolution. In this work, we explore novel interpolation methods along the sectioning direction to produce isotropic resolution and facilitate proper interpretation of the FIB-SEM 3D volumes., [Methods]: Classical interpolation methods are usually applied in this context under the assumption that the changes through successive images are relatively smooth. However, the actual 3D arrangement of the structures in the sample may cause significant changes in the biological features between consecutive images of the FIB-SEM stacks. We have developed a novel interpolation strategy that accounts for this variation by using the Optical Flow (OF) to estimate it. As an intermediate stage, OF-compensated images are produced by aligning the spatial regions of the biological structures. Interpolated images are then generated from these OF-compensated images. The final isotropic stack is assembled by interleaving the interpolated images with the original images of the anisotropic stack., [Results]: OF-driven and classical interpolation methods were compared using an objective assessment based on Pearson Correlation Coefficient (PCC) and a qualitative evaluation based on visual results, using public datasets and representative anisotropy conditions. The objective assessment demonstrated that the OF-driven interpolation always yields higher PCC values, with interpolated images closer to the ground truth. The qualitative evaluation corroborated those results and confirmed that classical interpolation may blur areas with substantial changes between consecutive images whereas OF-driven interpolation provides sharpness., [Conclusions]: We have developed an OF-driven interpolation approach to generating FIB-SEM stacks with isotropic resolution from experimental anisotropic data. It adapts to the rapid variation of the biological structures observed through the images of the FIB-SEM stack. Our approach outperforms classical interpolation and manages to produce sharp interpolated views in cases where there are significant changes between consecutive experimental images., This work has been partially supported by the Spanish Ministry of Science and Innovation/AEI (grants SAF2017-84565-R, RTI2018-095993-B-I00, PID2021-123278OB-I00, PID2021-123424OB-I00, TED2021-132020B-I00), by the Spanish National Research Council (CSIC, grant 2020AEP029), by the European initiative Corbel (project id 2301), by the Junta de Andalucía (grants UAL18-TIC-A020020-B and P18-RT-1193) and by FEDER., Peer reviewed
DIGITAL.CSIC. Repositorio Institucional del CSIC
Publication Artículo científico (JournalArticle)

HPC enables efficient 3D membrane segmentation in electron tomography

  • Moreno, J. J.
  • Garzón, E. M.
  • Fernández, José Jesús
  • Martínez-Sánchez, Antonio
Electron Tomography (ET) is a powerful three-dimensional (3D) imaging technique used in structural biology and biomedicine to allow the visualization of the interior of cells at close-to-molecular resolution. Interpretation of the 3D volumes in ET is usually challenging due to the complexity of the cellular environment, noise conditions and other factors. Automated segmentation methods focused on membranes of the cells and organelles greatly facilitate visualization and interpretation of the 3D volumes. However, they are typically computationally expensive and spend significant processing time on standard computers. In this work, we introduce efficient implementations of one of the methods most commonly used in the ET field for membrane segmentation. They were developed by using High Performance Computing (HPC) techniques to make the most of modern CPU-based and GPU-based computing platforms. A thorough evaluation of the performance on state-of-the-art machines was carried out. The HPC implementations succeed in achieving remarkable speedups, which are around 100× on GPUs, and making it possible to process large 3D volumes in the order of seconds or a few minutes., This work has been supported by the projects and contracts: RTI2018-095993-B-I00, SAF2017-84565-R, PID2021-123278OB, PID2021-123424OB, TED2021-132020B (funded by MCIN/AEI/10.13039/501100011033/FEDER “A way to make Europe”); UAL18-TIC-A020020-B and P18-RT-1193 (both funded by Junta de Andalucía and FEDER); PAPI-21-GR-2011-0048 (funded by the University of Oviedo) and FUO-200-21 (funded by Thermo Fisher Scientific); and a FPU fellowship (FPU16/05946 funded by MCIN/AEI/10.13039/501100011033/ “El FSE invierte en tu futuro”) awarded to J.J. Moreno., Peer reviewed