This package contains both a rich library of functions for computational pathology and a collection of applications dealing with various problems, including spatial transcriptomics. This broad coverage comes with the price of installing many dependencies.

To avoid having to deal with various image (and data) formats (and their evolving specifications), we store the data as arrays in HDF5 files, with images saved in pyramidal structures. An image pyramid is stored as a group with datasets (named "data") representing its levels:

pyramid_<k>.h5
   /--- 0 <- highest resolution (largest image)
   /--- 1
   ...
   /--- n-1 <- lowest resolution 

To access a layer's data, one can use the path "//data" in HDF5 functions.

In addition, "pyramid_" has a few attributed (meta-data) guaranteed:

• max_level: number of levels in the pyramid
• channel_names: normally R, G, B, but others may be possible
• dimension_names: normally y, x, c
• mpp_x and mpp_y: the original resolution (microns-per-pixel) for x and y dimensions
• mag_step: magnification step (usually 2.0): scaling factor between pyramid levels
• objective_power: native objective used to acquire the image (e.g. 20.0 or 40.0)
• extent: a 2 x max_level array with extent[0,i] and extent[1,i] indicating the width and height of level i, respectively

A tissue section may have several pyramids (pyramid_0, ..., pyramid_n) as long as they refer to the exact same specimen, and they are aligned (registered). They may represent different modalities (e.g., H&E and IHC), or different versions of the same image (e.g., gray-scale or stain-normalized image). pyramid_0 is taken as the reference, and it defines the space in which the annotations are produced/saved, and all the analyses carried out.

1. Openslide
2. PyVIPS - make sure it is able to use openslide for reading whole-slide images.