- The SC that isn't in any ML is discarded.
- The graph is created and all the isolated sub-graph detected.
- The SC in the sub-graphs is labeled as one.
- If a sub-graph has two non-spurious classes and one is a merge including the SC and the other non - spurious, the SC and the spurious are merged together
- If a sub-graph has more than one non-spurious class, all the SC in that graph are removed.
- If only a non-spurious class is on the sub-graph the SC gets that label.
The changes in the classes will change the composition of the mergers, now some of them will have fewer members. All the merged classes with the same ML are equivalent and will be fused together. The other particular case isthe merge with just one related class, that mergers will be labeled as track of the same class. In the general case, this again creates the possibility of new equivalent mergers and it will need consecutive fusions until no more equivalent mergers are present. An example of the tracking that needs a class removal and fusion is shown in Figure \ref{748245}A. Finally, the remaining mergers are labeled correspondingly as a mixture of classes, to allow the track the activity of the class even when the quality of isolation is low.