Figure 1 shows a diagram of the data processing pipeline used for this research, described in more detail in the following section.
Methodological pipeline for the prediction of cognitive outcome. (1) We separate the different datasets (NOMIS and ADNI) into the necessary subsets (test, cross-validation and space set). (2) Different UMAP models are trained on one of the two space sets which results in the an embedding referred to as “UMAP embedding”. This embedding is linearly transformed so its axes are align with its principal axes. This transformation in turns results in a new embedding referred to as “t-UMAP embedding”. The latter embedding is fed to the probability atlas algorithm. The figures displayed under “UMAP embedding” and “t-UMAP embedding” were made using the NOMIS dataset with n_neighbors = 20 as the UMAP parameter. (3) The probability atlas…
