A BaSiC tool for background and shading correction of optical microscopy images.
Peng T, Thorn K, Schroeder T, Schroeder T, Wang L, Theis FJ, Marr C, Navab N. · 2017
Read A BaSiC tool for background and shading correction of optical microscopy images. by Peng T, Thorn K, Schroeder T, Schroeder T, Wang L, Theis FJ, Marr C, Navab N. (2017) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc5472168.txt.
Introduction
Peng and co-authors introduce BaSiC, a software tool for fixing microscopy images, in Nature Communications. Uneven light can shade the edges of an image and skew measured values. Time-lapse films also drift as background fades. Such flaws can distort counts of cells and tissue traits. The team built a method that needs no manual settings.
BaSiC splits each image set into a smooth background part and a sparse detail part. From that split it estimates the shading pattern and the camera offset. It then corrects both, plus drift over time. The authors report high accuracy with fewer input images than rival tools. BaSiC is free as a Fiji/ImageJ plugin.
Catalog record
| CREATOR | Peng T, Thorn K, Schroeder T, Schroeder T, Wang L, Theis FJ, Marr C, Navab N. |
|---|---|
| YEAR | 2017 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A BaSiC tool for background and shading correction of optical microscopy images By Tingying Peng, Kurt Thorn, Timm Schroeder, Lichao Wang, Fabian J Theis, Carsten Marr, Nassir Navab Published in: Nature Communications DOI: 10.1038/ncomms14836 ABSTRACT Abstract Quantitative analysis of bioimaging data is often skewed by both shading in space and background variation in time. We introduce BaSiC, an image correction method based on low-rank and sparse decomposition which solves both issues. In comparison to existing shading correction tools, BaSiC achieves high-accuracy with significantly fewer input images, works for diverse imaging conditions and is robust against artefacts. Moreover, it can correct temporal drift in time-lapse microscopy data and thus improve continuous single-cell quantification. BaSiC requires no manual parameter setting and is available as a Fiji/ImageJ plugin. Optical imaging is an indispensable tool in biomedical research. All modern optical imaging (whether whole slide imaging, high-content screening or high-throughput time-lapse microscopy) relies on image processing and quantification methods to analyse and interpret the acquired data. However, optical microscopy data, and especially fluorescence imaging, is often severely affected by shading or vignetting1, typically reflected as an attenuation of the brightness intensity from the centre of the optical axis to the edges. This not only degrades the visual quality of an image (for example, by causing discontinuities in whole slide images (WSIs)), but more critically compromises the downstream analysis of, for example, tissue composition or single-cell properties. Besides spatial shading effects, time-lapse movies often exhibit a temporal baseline drift due to background bleaching, which further skews the quantification of the dynamic behaviour of cellular and molecular properties2. The physical process of image formation can be approximated as a linear function3 that relates a measured image, Imeas(x) at location x, to its uncorrupted true correspondence, Itrue(x), as where the multiplicative term S(x) represents the change in effective illumination across an image (known as flat-field); the additive term D(x), known as dark-field, is dominated by camera offset and thermal noise, which are present even if no light is incident on the sensor. Existing shading correction methods can be generally divided into two groups: ‘prospective' approaches that determine S(x) and D(x) from extra reference images4 (Supplementary Note 1) and ‘retrospective' approaches, which rely on the actual image data itself and hence avoid collection of extra reference images (Supplementary Note 2). A number of multi-image based approaches have been recently published, for example, Smith et al.5 (Fiji Plugin ‘CIDRE'), Coster et al.6 and Singh et al.7 (the default module in CellProfiler). These approaches take advantage of shared S(x) among an image sequence and are usually more reliable than single-image based corrections. Yet they require large numbers of images to reach a stable performance and a manual fine-tuning of internal parameters, and their robustness to common bioimage artefacts (such as dust and fluorescence dye particles) has not been tested. Moreover, none of the existing methods is able to model and correct temporal drift (e.g. caused by photobleaching) for time-lapse movies. We propose BaSiC, a retrospective method for background and shading correction of image sequences, based on a sparse and low-rank decomposition. In comparison to existing shading correction tools, BaSiC requires fewer input images, works for diverse imaging conditions and is robust against typical image artefacts. Moreover, it can correct temporal drift for time-lapse microscopy data, and hence improve single-cell quantification. BaSiC is available as an easy-to-use Fiji/ImageJ plugin as usually requires no manual parameter tuning. RESULTS BaSiC workflow Inspired by Smith et al.5, we build our method on the shading model (equation (1)), which accounts for the effect of both S(x) and D(x). Such a full model is superior as compared to a partial model that considers S(x) only5. As shown in the schematic plot of Fig. 1, BaSiC first constructs a measurement matrix I (step I), which is then decomposed into a low-rank matrix IB and a sparse residual matrix IR (step II). The low-rank matrix has a maximum rank of two as each column is the sum of a scaled version of S(x) (with a scaling factor Bi) and D(x), which are all initialized with zeros (step III) and optimized by promoting the sparsity of the residual matrix with a reweighted L1-norm (step IV). In addition, smooth constraints are imposed on both S(x) and D(x) by regulating their sparsity in Fourier domain (Supplementary Fig. 1). The optimization is solved in an iterative fashion using the linearized augmented Lagrangian method8, which is widely used in sparse matrix decomposition like Robust PCA (ref. 9) and RASL (ref. 10) (for a detailed description of the mathematical derivation and matrix updating see Methods and Supplementary Note 3). An automatic parameter setting strategy determines the smooth regularization parameters for S(x) and D(x), adaptive to different image contents, so that tedious manual parameter tuning is avoided (Supplementary Note 4). We provide a statistical interpretation of BaSiC in Supplementary Note 5. With the estimation of S(x) and D(x), we can correct the intensity profile of each image tile of a WSI by reversing the image formation process, equation (1), leading to a homogenous appearance and correct stitching (Fig. 1d versus a). [Figure: BaSiC is an automatic correction method for both static multi-image microscopy data and dynamic time-lapse data.(a) A mosaic whole slide image (WSI) of a mouse brain slice showing intensity discontinuities and resulting stitching errors (indicated by black arrows). (b) A time-lapse movie corrupted by both shading in space and photobleaching in time. (c) The BaSiC workflow, (see text for detailed explanation). (d) The corrected WSI shows improved stitching with no discontinuities. (e) BaSiC corrects both spatial shading and background bleaching over time.] BaSiC requires few images for shading correction We first evaluate BaSiC using synthetic images, where the ground-truth of the shading-free image Itrue, the flat-field S(x) and the dark-field D(x) are known (Supplementary Note 6). We first compare BaSiC with CIDRE (ref. 5), the only other method that is able to simultaneously estimate S and D. BaSiC requires far fewer images to achieve the same accuracy as CIDRE (for example, 10 versus 100 images to reach an estimation score Γ(Sest)≤0.1 at three levels of cell density (cell density, Supplementary Fig. 2). We subsequently evaluate BaSiC using a comprehensive microscope image collection provided by Smith et al.5, which includes 10 real microscopy data sets, one photography data set and one synthetic microscopy data set (refer to Smith et al.5 for data set details). We assess the correction quality by the correction score, Γ′(Icorr), which is the mean absolute difference of pixel pairs in overlapping image regions for each data set after correction, normalized by the difference of the uncorrected pairs5. A Γ′(Icorr)<1 suggests reduced shading while Γ′(Icorr)>1 implies increased shading compared to the uncorrected images. Besides CIDRE, we compare BaSiC also with the methods used in Coster6 and Cellprofiler7 (refer to Methods for technical details), as well as two prospective methods (Calib-zero and Empty-zero)5. We find that BaSiC improves image quality with as few as five images (Fig. 2a). At around 100 images it matches prospective methods on average, although the performance varies for different data sets (Supplementary Fig. 3). Among retrospective methods, BaSiC significantly outperforms existing approaches, when fewer than 500 images are used (Fig. 2a). This is practically relevant since WSI