Figure 5. The 3D mode of ILEE renders... | Rockefeller University Press

$The 3D mode of ILEE renders cytoskeleton indices at much higher accuracy than the 2D mode, due to advanced data structure. We utilized image samples (n = 31, 25*800*800, Fig. 2) to train an actin image simulation model that generates 31 artificial 3D actin images with ground truth. By comparing the performance of 3D and 2D ILEE over the artificial images, it was demonstrated that the 3D mode displays high fidelity to the ground truth while the 2D mode suffered from loss of information and systemic bias. (a) The general experimental scheme. The raw 3D images are analyzed to train an actin image simulation model which uses statistical approaches to mimic the genuine actin filament, diffraction noise, and ground noise. Then 31 artificial actin images are regenerated with specific ground truth in terms of both actin segmentation and cytoskeleton indices. The 3D and 2D ILEE were applied to the artificial images, whose results were then compared to their corresponding ground truth. Scale bars are marked as red/yellow solid line. See Fig. S16 and Materials and methods for a detailed description of the algorithm. (b) A table comparing the absolute accuracy of ILEE by the 3D and 2D mode. It was demonstrated that 2D mode losses ∼90% of pixels (voxels) and ∼50% of the total length of actin samples and is less accurate for most of the indices. Results are presented as mean ± standard deviation. Exclamation mark (!): contrast difference due to different definitions of dimensional space in their units; not rigorously comparable. (c) The linear correlation levels between ILEE results and the ground truth for all indices, by 2D and 3D mode. The absolute values of each index of the ground truth, 3D ILEE, and 2D ILEE are normalized by making their relative folds to the minimum data point of each group. A higher correlation coefficient r indicates a better capability to differentiate indices at different levels relatively, regardless of their absolute fidelity. The 3D mode has a generally better capability of differentiating high and low values. “r” represents Pearson correlation coefficient; “p” represents possibility to reject the null-hypothesis of no existent correlation by two-sided t test. Data distribution was assumed to be normal but not formally tested.$

Figure 5.

The 3D mode of ILEE renders cytoskeleton indices at much higher accuracy than the 2D mode, due to advanced data structure. We utilized image samples (n = 31, 25*800*800, Fig. 2) to train an actin image simulation model that generates 31 artificial 3D actin images with ground truth. By comparing the performance of 3D and 2D ILEE over the artificial images, it was demonstrated that the 3D mode displays high fidelity to the ground truth while the 2D mode suffered from loss of information and systemic bias. (a) The general experimental scheme. The raw 3D images are analyzed to train an actin image simulation model which uses statistical approaches to mimic the genuine actin filament, diffraction noise, and ground noise. Then 31 artificial actin images are regenerated with specific ground truth in terms of both actin segmentation and cytoskeleton indices. The 3D and 2D ILEE were applied to the artificial images, whose results were then compared to their corresponding ground truth. Scale bars are marked as red/yellow solid line. See Fig. S16 and Materials and methods for a detailed description of the algorithm. (b) A table comparing the absolute accuracy of ILEE by the 3D and 2D mode. It was demonstrated that 2D mode losses ∼90% of pixels (voxels) and ∼50% of the total length of actin samples and is less accurate for most of the indices. Results are presented as mean ± standard deviation. Exclamation mark (!): contrast difference due to different definitions of dimensional space in their units; not rigorously comparable. (c) The linear correlation levels between ILEE results and the ground truth for all indices, by 2D and 3D mode. The absolute values of each index of the ground truth, 3D ILEE, and 2D ILEE are normalized by making their relative folds to the minimum data point of each group. A higher correlation coefficient r indicates a better capability to differentiate indices at different levels relatively, regardless of their absolute fidelity. The 3D mode has a generally better capability of differentiating high and low values. “r” represents Pearson correlation coefficient; “p” represents possibility to reject the null-hypothesis of no existent correlation by two-sided t test. Data distribution was assumed to be normal but not formally tested.

This Feature Is Available To Subscribers Only