Scientific data visualization, analysis and mining are among the key processes used in life sciences. Together with a vast array of novel imaging techniques, they give the researchers detailed insights into the complex relationships that exist within living cells. With an ever-increasing amount of multi-dimensional image data, the complex task of extracting and understanding multifaceted biological information should be aided by advanced image analysis software tools. Such tools must be capable of guiding and intuitively aiding their users throughout the image analysis process. Visualization and understanding of complex images of single molecules and organelles as well as whole cells, tissues and organs is a challenging tour de force that can benefit from advanced software systems tailored to provide the best qualitative and quantitative representation of image data.

Visualization of Multi-modal Images

Data visualization and the software tools used in this process have evolved dramatically in recent years making it possible to perform complex analyses of images of different specimens on relatively mid-range computers [1]. This gradual improvement in image processing, visualization and analysis led to a widespread popularity of standard PC-based software systems capable of dealing with multi-image series ranging from nanostructures acquired by electron microscopes (EM) up to whole brain sections now successfully obtained with single plane illumination microscopy (SPIM) [2] and similar techniques. It is increasingly common practice to combine several imaging methods, for example EM and fluorescence which produce multi-dimensional correlative microscopy image data [3, 4]. These images need significant amounts of time and computational power to complete analysis of both EM and fluorescence signals which can often span across several thousands of EM and optical sections.

It is because of this multi-modal image data stream that the software tools needed for the data analysis face several challenges. One such challenge is directly related to the multi-scale nature of images where the user's software is faced with detail levels spanning from nanometers of single molecular motors or transport assemblies in EM images, to several micrometers associated with entire cells or their elongated processes particularly common in neuroscience-related image data.

It is desirable that the software tool of choice can smoothly handle changes of scale when the user explores thousands of image planes and regions of interest within the same data set.

Image Data Size and Scale Challenges

Another important factor relevant for software tool functionality is its integration into one package. Given the abundance of both image data inputs and use of different software tools, the users may have a need to transfer vast amounts of multi-modal data from one tool to another. This can present significant problems with data sets exceeding hundreds of gigabytes which may have to be transferred due to incomplete functionality within one tool. Therefore integration of various analytical modules in one package is an advantage making the complex analysis of large data smoother. Imaris software suite is an example of tight integration of several image analysis tools organized in modules which can be combined by the user according to their application needs.

With image data coming in large volumes from various streams of imaging modalities, the users of visualization and analysis software will often have to work with different analytical routines involving numerous steps from pre-processing, image restoration and segmentation to the final image analysis steps. Some of these steps will require operator's input at certain key stages of analysis which may mean that a considerable amount of time needs to be dedicated throughout the process. In case of high volumes of analyses this could have a negative impact on the entire process when consistent judgment is required between analyses of consecutive data sets. It would be advantageous to enable the software tool to have a large degree of automation for some of the analytical steps so that the human input is required predominantly at the key steps of the process.

Usability of Image Analysis Tools

Such interactions between the user and the software tools need to be optimized to reduce the time it takes for users to become sufficiently acquainted with the software . This goal can be achieved in several ways and these will differ significantly depending on what level of usability users expect. In most cases streamlining the image analysis process in the form of a wizard with a pre-conceived series of steps from image acquisition through segmentation to final analysis makes it easier for users to learn and later to master the software tool [1, 5]. This model works well for multi-module suites where users can dynamically decide what level of analysis they require or to what extent they want to follow and supervise the individual steps of image analysis. Such modular setup also allows for a gradual development of users' skills scaling together with the increasing complexity of their image data sets making it possible to tailor the application to the exact analytical needs for a given data set.

Data Mining and Advanced Visualization

During analyses of multi-modal images the end result may often be a multi-dimensional output that features a vast array of segmented objects belonging to several functional categories that need to be individually classified and presented [6]. For example a user has acquired a 50 GB, 4D (3D+t) confocal image series that represents a cluster of differentiated PC12 cells forming neurite-like structures and the final visualization includes a map of connections between neurites, shapes and volumes of their neuronal sprouts. In addition, the visualization may feature three-dimensional positions and distribution of different populations of intracellular structures including large, dense-core vesicles and smaller vesicles. The final visualization is completed by the renders of cytoskeletal scaffolds and nuclear staining of DNA in the cell. The main challenge at this stage will be the representation of all those individual objects in a way that will be most informative and which will offer most insight into this relatively complex network of interrelations between cellular components. Innovative representation of such data can have a significant impact on data mining and understanding in case certain interdependencies within the analyzed image stack are not obviously discernible.

With similar tools that visualize image data in multiple dimensions it is easier to extract information from often intrinsically complex image data sets. This could be the key functionality of software analysis tools that should deliver the pivotal function of re-organization of multi-dimensional, multi-class/object images. Enhanced by the creation of series of fully customizable plots, these tools will help to understand often hidden relationships and associations between hundreds or thousands of visualized objects.

It is likely that the software tools currently available will keep evolving together with constantly changing imaging applications. With the constant increase in the number of combinations of multi-modal, multi-dimensional image data types, users will keep facing some of the challenges outlined above including usability, flexibility and modularity of integrated software tools. Finally the ever-increasing complexity of images will call for intuitive and insightful ways of representing the analyzed data in multiple dimensions to reflect some of the complexity of interplay between proteins, organelles, cells and organs on multiple levels.

References
[1] O'Donoghue S. I. et al.: Nature Methods 7, S2-S4, (2010)
[2] Huisken J. et al.: Science, Vol. 305, (2004)
[3] Oberti D. et al.: Front Neuroanat. 4: 24, 9, (2010)
[4] Spiegelhalter C. et al.: PLoS ONE, 5(2): e9014, (2010).
[5] Walter T. et al.: Nature Methods 7, S26, (2010)
[6] Berlanga M. L. et al.: Front Neuroanat.; 5: 17, (2011)