Figure 1.
Application of 3D BLE to synthetic phantoms corrupted with Gaussian and impulse noise.
Performance of the 3D BLE, 3D recursive and 3D Canny filters was assessed using a volume of 3D synthetic phantoms contaminated with increasing levels of Gaussian and impulse noise. (A1–A5) 2D sections taken from synthetic volumes contaminated with increasing levels of Gaussian and impulse noise. (B1–B5) 3D surface rendering of results (B1–B5) obtained from the 3D BLE filter. (C1–C5) Surface rendering of the 3D recursive-filtered synthetic dataset. (D1–D5) Surface rendering of the 3D Canny-filtered synthetic dataset.
Table 1.
Statistical evaluation of filter performance using synthetic volumes contaminated with different levels of Gaussian and impulse noise shown in Figure 1.a
Figure 2.
Application of 3D BLE to synthetic phantoms corrupted with simulated cytosolic noise.
Performance of the 3D BLE, 3D recursive and 3D Canny filters was assessed using the same volume of 3D synthetic phantoms shown in Figure 1, but contaminated with different levels of simulated experimental noise. (A1–A5) A graphical representation of the SNR present in the five representative cases shown. Coloring in A1–A5 is as follows: green dotted line shows the contrast and intensity of the original signal; red line shows the contrast and intensity of the noise; green solid line shows the scaling and shifting of signal profile towards noise profile. Overall, the graph shows the probability density function (G(I)) of the normal distribution (B1–B5) 2D sections taken from synthetic volumes contaminated with experimental noise. (C1–C5) 3D surface rendering of results obtained following application of the 3D BLE filter to the synthetic dataset. (D1–D5) Surface rendering of 3D recursive-filtered test dataset. (E1–E5) Surface rendering of 3D Canny-filtered dataset.
Table 2.
Statistical evaluation of filter performance using synthetic volumes contaminated with different levels of simulated experimental noise shown in Figure 2.a
Figure 3.
Detection of molecular volumes using 3D BLE.
The ability of the3D BLE, 3D recursive and 3D Canny filters to resolve molecular contours was assessed using a test volume populated with 3D GroEL molecules. A representative region of the test volume showing 9 molecules is shown. (A1–A5) SNR illustrated as for Figure 2. (B1–B5) 2D sections taken from synthetic volumes contaminated with experimental noise. (C1–C5) Surface rendering of results following application of the 3D BLE filter applied to the test volume contaminated with experimental noise. (D1–D5) Surface rendering of 3D recursive-filtered test volume. (E1–E5) Surface rendering of 3D Canny-filtered test volume.
Figure 4.
Extraction of molecular contours from an electron tomogram subvolume.
Application of the 3D BLE, 3D recursive and 3D Canny filters to a subvolume of an experimentally-recorded tomogram of a resin-embedded C. reinhardtii cell. (A) Unprocessed, central 2D cross-section of the subvolume extracted from the 3D tomogram showing a region of the chloroplast heavily populated with putative macromolecular assemblies (dark objects). The inset in (A) highlights a randomly chosen single particle, represented as an isosurface rendering and shown at a selected number of orientations around the y-axis. (B) 3D surface rendering of results obtained from application of the 3D BLE filter. (C) Surface rendering of the 3D recursive-filtered subvolume. (D) Surface rendering of the 3D Canny-filtered subvolume.
Figure 5.
Segmentation of the Golgi region of an insulin-secreting pancreatic beta cell line HIT-T15.
(A) A tomographic slice (slice 33) extracted from the reconstructed volume reported in [23]. The region demarcated by a red box is shown in (B). (B) Objects were segmented by manually drawing colored lines (contours) using IMOD. (C) Surface-rendered 3D model of the Golgi region analysed in (B) by manual segmentation. (D) 3D BLE-filtered tomogram. (E) Contours detected automatically by the 3D BLE were then manually colored for comparison to the manually segmented volume shown in (B). (F) Surface-rendered 3D model generated by automatic segmentation of the same region shown in B. Coloring in (C–D) and (E–F) is as follows: the seven cisternae that comprise the Golgi in the region - C1, light blue; C2, pink; C3, cherry red; C4, green; C5, dark blue; C6, gold; C7, bright red. ER, yellow; membrane-bound ribosomes, blue; free ribosomes, orange; mitochondria, bright green; dense core vesicles, bright blue; clathrin-negative vesicles, white; clathrin-positive compartments and vesicles, bright red; clathrin-negative compartments and vesicles, purple; mitochondria, dark green. (G) A tomographic slice revealing the outer and inner membrane architecture of a mitochondrion in the Golgi region. (H) Surface rendering shows that automated 3D segmentation facilitated by the application of 3D BLE detects the mitochondrial membranes.
Table 3.
Comparison of processing resources consumed by each of the three filters evaluated using a synthetic volume (385×512×128 voxels) contaminated with different levels of Gaussian and impulse noise.