Fast Cross-sectional Display of Large Data Sets

One of the steps to reconstruct the 3-D geometry of biological objects from a stack of 2-D microscopic images, is to align the individual slices with respect to each other. Due to complex internal structures and imaging artifacts, automatic methods do not always lead to reliable results. Visual feedback consisting of sagittal and coronal cross-sections is very useful to validate the alignment result. For data sets which are too large to fit into main memory, this is an expensive operation as each slice in the volume needs to be referenced, resulting in a lot of disk I/O. Array data (including images) is conventionally stored in rowor column-major order. This is however not always optimal. In our application, we chose to store the data in Z-order (Morton order), resulting in a significantly improved performance.