. 2013 Jun 14;2:883–893. doi: 10.1016/j.nicl.2013.06.004

Table S1.

ECOC approach.

Class	Binary classifier
Class	PD vs MSA	PD vs PSP	PD vs CBD	MSA vs PSP	MSA vs CBD	PSP vs CBD
PD	1	1	1	0.5	0.5	0.5
MSA	0	0.5	0.5	1	1	0.5
PSP	0.5	0	0.5	0	0.5	1
CBD	0.5	0.5	0	0.5	0	0
Test scan: PD	0.5	0.7	0.8	0.4	0.6	0.5

Let's consider a test scan obtaining the probabilities shown in table S1, last line, for each binary classifier. When using the RVM probabilities, the distance measure used to assign the final class is the Manhattan distance, which would give for the four classes:

\begin{array}{l} L^{PD} = |1 - 0.5 |+| 1 - 0.7 |+| 1 - 0.8 |+| 0.5 - 0.4 |+| 0.5 - 0.6 |+| 0.5 - 0.5| = 1.2 \\ L^{MSA} = |0 - 0.5 |+| 0.5 - 0.7 |+| 0.5 - 0.8 |+| 1 - 0.4 |+| 1 - 0.6 |+| 0.5 - 0.5| = 2 \\ L^{PSP} = |0.5 - 0.5 |+| 0 - 0.7 |+| 0.5 - 0.8 |+| 0 - 0.4 |+| 0.5 - 0.6 |+| 1 - 0.5| = 2 \\ L^{CBD} = |0.5 - 0.5 |+| 0.5 - 0.7 |+| 0 - 0.8 |+| 0.5 - 0.4 |+| 0 - 0.6 |+| 0 - 0.5| = 2.2 \end{array}

where L represents the final score of each class. In the present case, the class ‘PD’ is assigned to the test point since it shows the smallest final score L, which leads to a correct classification.