Joined in 2023 
82 
63 About this deal
Plus, VOLOOM has protective ceramic coated plates, as well as ionic technology that help to seal the cuticle and protect from damage. All of these features protect the hair. Reconstructions using ( a) LS based on landmarks by observer 1, ( b) OPT, ( c) SIFT, ( d) HSR, ( e) RVSS, ( f) ESA, ( g) MIM and ( h) Voloom. Optimized parameters and the most suitable resolution were used for each method. The locations of the four landmark points on each section are indicated with dots, shown together with lines of best fit to each of the four series of points. Note that the scale of the vertical axis is different from the horizontal axes in the visualization. Viewing the high-resolution color version of the Figure online is recommended. (Color version of this figure is available at Bioinformatics online.) 4 Discussion
LS: Least-squares fitting of an affine transformation to the landmarks was implemented in MATLAB R2016b. The result is in principle unaffected by error accumulation ( Xu et al., 2015). First, we analyzed whether our metrics depend on image resolution (see Supplementary Results). TRE, ATRE, Jaccard and ΔA-% are essentially invariant to image resolution. They can be compared across different datasets and resolutions, as long as the accumulation of interpolation errors is avoided. RMSE, NCC, MI, NMI, f 2 and f 3 depend both on resolution and image content, and these metrics should thus only be compared within the same dataset and resolution. In all following analyses, we used images subsampled to pixel sizes of 7.36 and 3.68 µm, referred to as low and high resolution, respectively. The pixel sizes are close to the 5 µm section spacing and metrics computed from these images are not distorted by interpolation errors. Furthermore, we will only present RMSE as a measure of pixelwise similarity and f 2 as a measure of reconstruction smoothness due to their strong correlations with NCC, MI, NMI and f 3 (see Supplementary Table S1 for details). 3.2 Automated parameter tuning All methods benefited from parameter tuning on both image resolutions based on most of the metrics, using either set of landmarks for evaluation (see Table 1 and Supplementary Results). Of the top three methods, MIM and RVSS obtained better accuracy using high resolution images and ESA worked better on the low resolution images. ESA and MIM reached similar mean TRE values, slightly better than RVSS and approaching or exceeding the accuracy of LS. In terms of maximum TRE and ATRE, the three methods were comparable, but RVSS reached slightly lower ATRE than ESA or MIM. Among all tools, ESA and MIM also obtained the highest Jaccard index values. The RMSE and f 2 metrics do not allow comparison across different image resolutions and one should note that MIM’s output was always stored at the lower resolution for technical reasons. Considering these limitations, we can observe that ESA performed best in terms of these metrics on both image resolutions ahead of RVSS. Changes in tissue area introduced by ESA, MIM and RVSS were moderate. Behind the top three, most other tools reached accuracy comparable to each other. The worst results were obtained using default parameters and for some methods, most notably ESA and RVSS, they were even comparable to the unregistered original images.Of the evaluated methods, LS, HSR and Voloom do not have tunable parameters. For OPT, SIFT, RVSS, ESA and MIM, we tuned the parameters automatically, minimizing the mean TRE computed for the prostate dataset. Parameter optimization took approximately 1500 hours in total to compute, producing 23 terabytes of data.
The two samples selected for this study are markedly different in their histological composition. The fact that the top methods performed well on both the prostate and the liver dataset without any retuning of parameters indicates that these methods are not overly sensitive to tissue appearance, and that the results obtained in this study are not specific to a single dataset. However, some variation in the relative performance of the algorithms on the two datasets was still observed. Thus, collecting and annotating additional datasets representing diverse tissue types and other histological stainings, such as immunohistochemistry, remains an important goal for future studies. In the case of MIM, which had to be operated interactively, we evaluated each combination of tunable values by a parameter sweep. Tunable parameters of the other methods were optimized via Bayesian optimization ( Shahriari et al., 2016; Snoek et al., 2012), which is well-suited for such problems, where the objective function is computationally expensive to evaluate, nonconvex, multimodal, and typically has low to moderate dimensionality. Bayesian optimization has been shown to perform favorably in comparison to other global optimization algorithms on benchmarking functions ( Jones, 2001) as well as on real WSI data ( Teodoro et al., 2017). We used MATLAB’s bayesopt implementation ( https://www.mathworks.com/help/stats/bayesian-optimization-algorithm.html) with mean pairwise TRE as the objective function. We utilized a Gaussian process model of the objective function and an automatic relevance determination (ARD) Matérn 5/2 kernel ( Snoek et al., 2012) with ‘expected-improvement-plus’ as the acquisition function ( Bull, 2011). Reconstructions with output image dimensions over fivefold compared to the input due to extreme error accumulation were considered failures. The number of variables to optimize was 2 (OPT), 4 (SIFT), 7 (RVSS) or 15 (ESA). We first optimized SIFT alone and used the optimal values for the SIFT step of RVSS and ESA. See Supplementary Table S1 for descriptions of the parameters. The number of seed points was set to twice the number of variables. We ran 30 iterations for OPT due to its simple objective function ( Kartasalo et al., 2016) and 100 iterations for the other tools. We used the prostate images subsampled by factors of 8 and 16, except for ESA, for which optimization was only feasible using the factor 16. Parameters optimized for ESA using the lower resolution were scaled to be used with the high resolution images. Computations were run on a workstation with Intel Xeon E5-1660 v3 3 GHz and 64 GB of RAM (low resolution) and a cluster node with Intel Xeon E5-2680 v3 2.5 GHz and 128 GB of RAM (high resolution). 3 Results 3.1 Effect of image resolution on evaluation metrics First, make sure your hair is dry and styled as you like. (VOLOOM can be used on hair that is freshly styled or on 2nd or 3rd day hair). Part your hair normally. Repeat this process as you move VOLOOM down the hair shaft, two to three times, stopping at about eye or cheekbone level. You can experiment with more or less, depending on the length of your hair. Digitalization of pathology has been accelerated by improvements in technology allowing acquisition of whole slide images (WSI) ( Ghaznavi et al., 2013; Griffin and Treanor, 2017). Besides computer-aided facilitation of pathologists’ tasks, digital pathology can enable new approaches like 3D histology, where three-dimensional reconstructions of samples are formed in silico based on serial sections ( Magee et al., 2015; Roberts et al., 2012). While other techniques allow imaging directly in 3D, they are currently incapable of matching the subcellular resolution and throughput of whole slide imaging. Examples of potential applications include construction of data-driven computer models and improved diagnostics of diseases associated with changes in the 3D microarchitecture of tissue. Moreover, 3D histology is compatible with established histopathological interpretation techniques and biochemical assays such as immunohistochemistry or in situ hybridization. This raises interesting prospects in view of recent advances in spatially resolved omics ( Mignardi et al., 2017; Ståhl et al., 2016). Pairing imaging with genomic, epigenomic, transcriptomic and proteomic data in the spatial context of tissue holds great promise for pathology and other fields ( Koos et al., 2015). Taking a step further, this could be performed in 3D to truly probe the relationships between structural and functional features as well as the heterogeneity and interplay between different cell types in tumors, and significant projects are now pursuing these goals ( Ledford, 2017; Rusk, 2016). These kind of approaches have already led to the creation of brain atlases ( Amunts et al., 2013; Johnson et al., 2010; Lein et al., 2007). Such high-dimensional data also represent an exciting challenge for new ways of scientific visualization based e.g. on virtual reality techniques ( Calì et al., 2016; Ledford, 2017; Theart et al., 2017).
MIM: Medical Image Manager, trial v. 0.94, was applied using images subsampled by a factor of 4 (magnification of 5×) as input. Sections 130 and 24 were used as references for the prostate and liver, respectively. We varied the initial magnification (0.3125×, 0.625×, 1.25× or 2.5×) and the number of non-rigid levels (1, 2, 3 or 4), thus modifying the image resolution used. The optimization mostly converged close to the final solution in a handful of iterations (see Supplementary Results). By inspecting the variation in mean TRE values obtained during the process it is possible to reach a semi-quantitative view of the sensitivity of each method towards parameter adjustments. OPT and SIFT produced similar results for most parameter combinations while ESA, MIM and especially RVSS exhibited more sensitivity to parameter tuning. VOLOOM has been designed to help you achieve maximum results while minimizing the potential for hair damage. It is to be used only on the hair near the scalp and a few inches down the hair shaft. This hair is rich in natural protective oils – your own natural heat protection. Unlike other hot tools, it is never used on the ends of hair, most prone to damage. Based on this study, methods utilizing locally varying transformations (ESA, MIM, RVSS, Voloom) were superior to those constrained to global affine models (OPT, SIFT, HSR). ESA was the only method to consistently outperform or match the other approaches on two datasets based on the majority of metrics. In the case of the higher quality prostate dataset, differences in accuracy between the tools were rather subtle. All three top-performing methods on this dataset incorporate an elastic transformation model: MIM and RVSS use a B-spline grid and ESA is based on a piecewise linear mesh. While methods relying on a global transformation model also performed reasonably well, the additional accuracy offered by elastic transformations could be crucial when microstructure at the cellular scale is of interest. In the case of the liver sample, more profound differences between the methods were observed, likely due to the more challenging tissue content and the presence of deformations, which cannot be compensated for using a global model. ESA, MIM and Voloom stood out from the other methods. While Voloom appeared to be less accurate on average compared to ESA and MIM based on mean TRE, it demonstrated the lowest maximum and accumulated errors of all automated methods, indicating capability to avoid propagation of errors even in the presence of considerable deformations. The ability of the algorithms to tolerate such deformations is a significant benefit. Due to the mostly manual nature of histological sectioning and brittleness of the thin tissue sections, deformations in the form of folds and tears often occur. This challenge is especially encountered in 3D histology, when uninterrupted sequences of sections are desired.
Great Deal 
New Comment