First, in terms of pure visualization, as mentioned above most computer displays only have 2 to 8 million pixels of display real estate. When a TB image is loaded, it is almost impossible to physically display every pixel from the dataset on the screen at one time. Therefore, a system that spends minutes or hours loading the entire image in memory to only display a small fraction of these pixels at any one time is generally wasteful. Instead, arivis built a system that only loads the pixels that we need, at the resolution we need, as and when we need them.

Typically this is done by creating a so-called pyramidal file structure, where the image is stored with multiple levels of redundancy with regard to zoom levels, so that when you load the image at the lowest zoom level you only load the lowest resolution layer, and as the user zooms in we remove that from the memory and load the next layer with a higher resolution. This is generally very efficient with regards to memory use, but highly inefficient with regards to data storage as the pyramidal file can end up being 1.5x time larger than the raw data just to give this ability to load the resolution as needed. Instead, arivis uses a redundancy-free file structure where the resolution is calculated as needed, meaning that even without compression the converted SIS file will typically be as little as a few 100MB larger than the raw image data. On top of this, we can also apply lossless compression, of the same type as is used in standard compressed file formats like ZIP, to get between 20-80% compression ratios depending on the image data without any loss of information that is critical for scientific image analysis. 

We cover 3D visualization of large datasets in more details here, but in short, here is a summary of the visualization concerns in 3D.

When rendering a dataset in 3D, similar constraints exist with regards to the number of pixels that can be displayed on the screen, but also the rendering capabilities of the system are also going to be limited by the graphics capabilities of the GPU and the available video memory. So, likewise, when rendering in 3D arivis will first load in memory only as much of the image data as can be readily handled by the GPU to enable a smooth user experience. This leads to a small, but usually acceptable, loss in resolution for the purpose of interactive visualization and navigation. arivis Vision4D also gives the possibility to render at the highest level of resolution as and when needed and when user interactivity is no longer required.