ALGORITHM 1.

The pseudo-code of the proposed method.

Input: MAD/NC, PAD/NC, BAD/NC, MMCI, PMCI, BMCI

1 α = LASSO(train = MAD/NC).coefficients;

2 MAD/NC = MAD/NC[:, α! = 0], MMCI = MMCI[:, α! = 0];

3 scoreMRI = ELM(train = MAD/NC).outputScore(MMCI);

scorePET = ELM(train = PAD/NC).outputScore(PMCI);

scoreBio = ELM(train = BAD/NC).outputScore(BMCI);

4 scores = [scoreMRI, scorePET, scoreBio]; ## scores∈RN×3

Classification and Validation:

5 for n from 1 to 100:

6   scores = scores[random_permute,:];

Ten folds cross-validation:

7   separate scores into ten folds along first dimension;

8   for i from 1 to 10:

testSet = scores[foldth = = i,:];

trainSet = scores[others,:];

record predict = ELM(train = trainSet).classify(testSet);

end for

end for

9 statistics of 100 runs