Following such an ordering, a fraction α of the weakest classes is discarded, generating a set of (1− α) C classes (the ones representing the strongest signals).
Five smaller classes gave suboptimal reconstructions, and the particles from these classes were discarded.
Pixels falling across two polygons, thus mixing two different classes, were discarded to limit using pixels with mixed spectral signature.
From the separation of the classes we discarded the set of points, L = [ L m 1, …, L m W ], from X that could not be discriminated in the LDA, such as X ′ = X - L = [ X m 1 ′, …, X m w ′ ′ ], where W' corresponds to the number of classes of W that are represented by at least one point of X after discrimination.
After using empirical Bayes to assign gene families to combinations of categories and rate classes, we discarded combinations that contained very few gene families, as distance estimates from these combinations are likely to have a very high variance.
Voxels assigned to the background class were discarded, and the remaining volume, i.e. tumour core and region of partial volume averaging, was kept as the final FLAB VOI in accordance with a previous description of this algorithm [9].
The samples within categories were selected randomly, but the samples were screened by listening, and the samples having a significant amount of content from other categories than their class were discarded.
20% of the worst particles in each class were discarded in each round of iteration.
At every refinement step, the 30% worst members of each class were discarded.
Loci that deviate from the above model (i.e. they show a continuous distribution, a multimodal distribution, or a trimodal distribution with a major intermediate class) are discarded from further analysis because they may represent multiple loci.
