Three-dimensional GPU-accelerated active contours for automated localization of cells in large images - PubMed (original) (raw)

Three-dimensional GPU-accelerated active contours for automated localization of cells in large images

Mahsa Lotfollahi et al. PLoS One. 2019.

Abstract

Cell segmentation in microscopy is a challenging problem, since cells are often asymmetric and densely packed. Successful cell segmentation algorithms rely identifying seed points, and are highly sensitive to variablility in cell size. In this paper, we present an efficient and highly parallel formulation for symmetric three-dimensional contour evolution that extends previous work on fast two-dimensional snakes. We provide a formulation for optimization on 3D images, as well as a strategy for accelerating computation on consumer graphics hardware. The proposed software takes advantage of Monte-Carlo sampling schemes in order to speed up convergence and reduce thread divergence. Experimental results show that this method provides superior performance for large 2D and 3D cell localization tasks when compared to existing methods on large 3D brain images.

PubMed Disclaimer

Conflict of interest statement

The authors have declared that no competing interests exist.

Figures

Fig 1

Fig 1

(a) A snakuscule is defined by two points p and q. (b) Initial configuration of multitude snakuscules congregated together at a distance 1.5R.

Fig 2

Fig 2

(a) A DAPI stained brain tissue slice. (b) The initial configuration and (c) final configuration of snakuscules on a zoomed region.

Fig 3

Fig 3

(a) A 3D snakuscule defined by two points p and q; to simplify computations 3D snakusules identifier points are considered in the same y and z levels. (b) The initial configuration consists of every two neighboring 3D contours located at a distance 1.5R apart. (c) The middle section of the 3D snakuscule with four distinct regions are shown in different colors. (d) The weight function assigns a weight to any portion of the 3D snakuscule shown in (c).

Fig 4

Fig 4. Energy changes of the 3D snake with and without normalization term w.r.t its radius.

Energy function with normalization term has a single local minimum when the snake fit the blob.

Fig 5

Fig 5. Execution time for the method implemented on CPU and GPU.

Time axis is plotted on a logarithmic scale.

Fig 6

Fig 6. Execution time of snakuscules (2D) on GPU with and without Monte-Carlo sampling.

Fig 7

Fig 7. Execution time of 3D snakuscules on GPU with and without Monte Carlo sampling.

Fig 8

Fig 8. Parallel implementation on GPU.

(a) Assignment of each contour to a thread block. (b) The algorithm implemented on GPU.

Fig 9

Fig 9

(a) A section of DAPI stained mouse hippocampus 3D image and (b) the final configuration of 3D snakuscules on the section.

Fig 10

Fig 10. Evaluation of different algorithms on the same DAPI stained image.

Our proposed method (3D snakuscule) provides results which matches the ground truth better than others.

Fig 11

Fig 11. Representing one section of different 3D datasets used for cell detection.

(a) Fluo-N3DH_CE (b) Fluo-N3DH-SIM+ (c) Mouse-Brain.

Fig 12

Fig 12. Quantification of the effect of Gaussian-distributed noise on localization accuracy.

Reducing SNR to 0.2 results in a drop in F-measure from 0.96 to 0.75.

Fig 13

Fig 13

(a) Theoretical occupancy, green, and achieved occupancy using different size images (different number of initial 3D snakuscules) with and without Monte Carlo integration, blue and red respectively. Parallel 3D snakuscules provide an achieved occupancy (purple) comparable to the theoretical limit. (b) The diagram shows execution time of the method implemented on GPU using MC integration before and after further parallelization for different number of initiated 3D snakuscules.

Similar articles

Cited by

References

    1. Su H, Yin Z, Huh S, Kanade T, Zhu J. Interactive Cell Segmentation Based on Active and Semi-Supervised Learning. IEEE Transactions on Medical Imaging. 2016;35(3):762–777. 10.1109/TMI.2015.2494582 - DOI - PubMed
    1. Irshad H, Veillard A, Roux L, Racoceanu D. Methods for nuclei detection, segmentation, and classification in digital histopathology: a review—current status and future potential. IEEE reviews in biomedical engineering. 2014;7:97–114. - PubMed
    1. Lin G, Adiga U, Olson K, Guzowski JF, Barnes CA, Roysam B. A hybrid 3D watershed algorithm incorporating gradient cues and object models for automatic segmentation of nuclei in confocal image stacks. Cytometry. 2003;56A(1):23–36. 10.1002/cyto.a.10079 - DOI - PubMed
    1. Xing F, Xie Y, Yang L. An Automatic Learning-Based Framework for Robust Nucleus Segmentation. IEEE Transactions on Medical Imaging. 2016;35(2):550–566. 10.1109/TMI.2015.2481436 - DOI - PubMed
    1. Wahlby C, Sintorn IM, Erlandsson F, Borgefors G, Bengtsson E. Combining intensity, edge and shape information for 2D and 3D segmentation of cell nuclei in tissue sections. Journal of Microscopy. 2004;215(1):67–76. 10.1111/j.0022-2720.2004.01338.x - DOI - PubMed

Publication types

MeSH terms

LinkOut - more resources