to BioTechniques free email alert service to receive content updates.
Visualization approaches for multidimensional biological image data
 
Curtis T. Rueden and Kevin W. Eliceiri
Laboratory for Optical and Computational Instrumentation, University of Wisconsin at Madison, Madison, WI, USA
BioTechniques, Vol. 43, No. S1, July 2007, pp. S31–S36
Full Text (PDF)

Introduction

Over the last 20 years, there have been great advances in light microscopy with the development of techniques such as confocal laser-scanning microscopy (CLSM) (1,2) and multiphoton laser-scanning microscopy (MPLSM) (3) and fluorescence-based techniques such as fluorescence resonance energy transfer (FRET) (4) and fluorescence lifetime imaging (FLIM) (5). These techniques have greatly improved the depth penetration, viability, and discrimination capabilities that can be obtained with light-based approaches. Optical sectioning allows for three-dimensional (3-D) time-lapse observation of dynamic events, and fluorescence-based techniques can assay not only the intensity of a fluorophore but other key attributes of fluorescence, such as spectral and lifetime information. Multidimensional light microscopy techniques encompass all these spatial, temporal, spectral, and lifetime dimensions and can also include additional dimensionality parameters such as the polarization state of a fluorophore. These techniques have revolutionized modern microscopy and yet have posed significant data management challenges in how to effectively store, analyze, and disseminate these complex data sets. Even with the more typical 3-D time-lapse (four-dimensional; 4-D) data sets, there have been significant challenges in how to analyze and disseminate this data. In this review we present some of the challenges in multidimensional approaches with an emphasis on 4-D data and present some of the approaches of visualizing such data.

The great advances in computing in recent years have allowed for the increased development of advanced multidimensional microscopy approaches by allowing for improved acquisition capabilities, including automated collection, larger data set collection, faster collection, and hardware-based image processing at acquisition. These computational advances have had a profound impact on how one can analyze data as well. With the advent of digital data collection and readily available personal computer workstations has come the widespread use of sophisticated computational analysis by the microscopist. However despite these advances, surprisingly there lacks a common infrastructure or framework in microscopy to help microscopists analyze and share data. Much of the analysis in microscopy is being done by commercial programs that use proprietary file formats and in many cases algorithms for their analysis. These proprietary programs, while very powerful, often are not transparent in their approach to the analysis. Below we detail some of the common approaches one can use in two-dimensional (2-D) and 3-D analysis and discuss some of the approaches in visualization and storage that may be utilized to deal with files of increased dimensionality.

Common Visualization Approaches

Visualization in 2-D

The most fundamental data visualization method for multidimensional microscopy data is one plane at a time, with controls for roaming between image slices across all dimensions. This technique is the most straightforward way to work within the limitations of a 2-D screen, allowing the user to verify and study what was actually collected by the instrument, without worry that analysis techniques have somehow introduced errors or otherwise altered the data. The software should allow the user to jump between image planes quickly; in particular, fast animation should be feasible for reproduction of the specimen's progress over time. This requirement is more challenging, but even more important, when the data set's size exceeds the computer's available memory. Lastly, it should be possible to probe individual pixels for their numerical values, since the images are inherently vague without them.

4-D data sets lend themselves well to plane-by-plane visualization, since the two dimensions, focal plane and time, can be construed as vertical and horizontal axes, respectively. This paradigm can be expressed in the user interface in a variety of ways, such as with horizontal and vertical scroll bars bordering the image that control the dimensional position, or through the use of arrow keys for navigation between image planes. The software could even display all planes at once, tiling them at reduced resolution in a 2-D matrix, for efficient comparison. These techniques become more cumbersome for dimensions beyond space and time, as additional axes must be introduced, resulting in image planes being distributed across three or more dimensions.

Once the data's spatial representation has been chosen, the other key component in its display is how to color the pixels. Intensity data sets have a single value at each pixel, which can be mapped along a continuum from black to white (grayscale). For data in full color, there are three-color components—often red, green, and blue—which can be mapped directly to red, green, and blue on the screen, respectively. It is often useful to manipulate the data's colorization, however. At minimum, brightness and contrast controls can be provided for affecting the overall lightness and variance between lowest and highest color values, respectively.

  1    2    3    4