NAGFS.m

%% Main function of NAG-FS framework for a fast and accurate classification.
% Details can be found in the original paper: https://www.scieC2edirect.com/scieC2e/article/abs/pii/S1361841519301367
% Islem Mhiri and Islem Rekik. "Joint input_ brain network atlas estimation and feature selection for neurological disorder diagnosis
%with application to autism",Medical Image Analysis, 2019, p. 101596.


%   ---------------------------------------------------------------------

%     This file contains the implementation of three key steps of our NAG-FS framework:
%     (1) Estimation of a centered and representative input_ network atlas,
%     (2) Discriminative connectional biomarker identification and  (3) disease classification:
%
%                        [AC1,AC2,ind] = NAG_FS(train_data,train_Labels,Nf,displays)
%
%                 Inputs:
%
%                          train_data: ((n-1) × m × m) tensor stacking the symmetric matrices of the training subjects
%                                      n the total number of subjects
%                                      m the number of nodes
%
%                          train_Labels: ((n-1) × 1) vector of training labels (e.g., -1, 1)
%
%                          Nf: Number of selected features
%
%                          displays: Boolean variables [0, 1].
%                                    if displays = 1 ==> display(Atlas of group 1, Atlas of group 2, top features matrix and the circular graph)
%                                    if displays = 0 ==> no display
%                 Outputs:
%                         AC1: (m × m) matrix storing the atlas of group 1
%
%                         AC2: (m × m) matrix storing the atlas of group 2
%
%                         ind: (Nf × 1) vector storing the indices of the top disciminative features
%
%
%     To evaluate our framework we used Leave-One-Out cross validation strategy.



%To test NAG-FS on random data, we defined the function 'simulateData' where the size of the dataset is chosen by the user.
% ---------------------------------------------------------------------
%     Copyright 2019 Islem Mhiri, Sousse University.
%     Please cite the above paper if you use this code.
%     All rights reserved.
%     """

%%------------------------------------------------------------------------------


 
 

  
 function [AC1,AC2,ind] = NAG_FS(train_data,train_Labels,Nf,displays)
  
 %% Step 1: Estimation of a centered and representative input_ network atlas
 % Extract C1 group and C2 group 
 s1 = 0;
 s2 = 0;
 [sz1,sz2,sz3] = size(train_data);
 for h = 1: length(train_Labels)
     
     if (train_Labels(h) == 1)
         s1 = s1+1;
         XC1(s1,:,:) = squeeze(train_data(h,:,:));
     else
         s2 = s2+1;
         XC2(s2,:,:) = squeeze(train_data(h,:,:));
     end
     
 end
 
 
 %Disentangling the heterogeneous distribution of the input_ networks using SIMLR clustering method
 
 % C1 group
 k = [];
 
 for l = 1: s1
     k1 = squeeze((XC1(l,:,:)));
     k2 = k1(:); %vectorize the matrix
     k = [k;k2'];
 end
 
 [t, S1, F1, ydata1,alpha1] = SIMLR(k,3,2);
 
 %C2 group
 kk = [];
 
 for l = 1: s2
     kk1 = squeeze(abs(XC2(l,:,:)));
     kk2 = kk1(:); %vectorize the matrix
     kk = [kk;kk2'];
 end
 [t1, S2, F2, ydata2,alpha2] = SIMLR(kk,3,2);
 
 % After using SIMLR, we extract each cluster independently for both classes
 
 % C1 group
 qC11 = 1;
 qC12 = 1;
 qC13 = 1;
 
 for qC1 = 1: s1
     
     if t(qC1) == 1
         Ca1(qC11,:,:) = abs(XC1(qC1,:,:));
         qC11 = qC11+1;
     elseif t(qC1) == 2
         Ca2(qC12,:,:) = abs(XC1(qC1,:,:));
         qC12 = qC12+1;
     elseif t(qC1) == 3
         Ca3(qC13,:,:) = abs(XC1(qC1,:,:));
         qC13 = qC13+1;
     end
     
 end
 
 %C2 group
 qC21 = 1;
 qC22 = 1;
 qC23 = 1;
 
 for qC2 = 1: s2
     
     if t1(qC2) == 1
         Cn1(qC21,:,:) = (XC2(qC2,:,:));
         qC21 = qC21+1;
     elseif t1(qC2) == 2
         Cn2(qC22,:,:) = (XC2(qC2,:,:));
         qC22 = qC22+1;
     elseif t1(qC2) == 3
         Cn3(qC23,:,:) = (XC2(qC2,:,:));
         qC23 = qC23+1;
     end
     
 end
 
 % For each cluster, we non-linearly diffuse and fuse all networks into a local centered network atlas using SNF
 
 % Setting all the parameters.
 K = 20;%number of neighbors, usually (10~30)
 alpha = 0.5; %hyperparameter, usually (0.3~0.8)
 T = 20; %Number of Iterations, usually (10~20)
 
 % C1 group
 
 for l = 1: (qC11-1)
     
     ll = num2str(l);
     Datap1.(['datap',ll,'']) = Standard_Normalization(squeeze(Ca1(l,:,:)));
     Distp1.(['distp',ll,'']) = dist2( Datap1.(['datap',ll,'']), Datap1.(['datap',ll,'']));
     Wp1.(['Wp1',ll,'']) = affinityMatrix(Distp1.(['distp',ll,'']), K, alpha);
 end
 
 Wall1 = struct2cell(Wp1);
 
 if qC11>2
     AC11 = SNF(Wall1,K,T) ;% First local network atlas for C1 group
 else
     AC11 = squeeze(Ca1(1,:,:));
 end
 
 for l = 1: (qC12-1)
     ll = num2str(l);
     Datap2.(['datap',ll,'']) = Standard_Normalization(squeeze(Ca2(l,:,:)));
     Distp2.(['distp',ll,'']) = dist2( Datap2.(['datap',ll,'']), Datap2.(['datap',ll,'']));
     Wp2.(['Wp1',ll,'']) = affinityMatrix(Distp2.(['distp',ll,'']), K, alpha);
 end
 
 Wall2 = struct2cell(Wp2);
 
 if qC12>2
     AC12 = SNF(Wall2,K,T); % Second local network atlas for C1 group
 else
     AC12 = squeeze(Ca2(1,:,:));
 end
 
 for l = 1: (qC13-1)
     ll = num2str(l);
     Datap3.(['datap',ll,'']) = Standard_Normalization(squeeze(Ca3(l,:,:)));
     Distp3.(['distp',ll,'']) = dist2( Datap3.(['datap',ll,'']), Datap3.(['datap',ll,'']));
     Wp3.(['Wp1',ll,'']) = affinityMatrix(Distp3.(['distp',ll,'']), K, alpha);
 end
 
 Wall3 =struct2cell(Wp3);
 if qC13>2
     AC13 = SNF(Wall3,K,T); % Third local network atlas for C1 group
 else
     AC13 = squeeze(Ca3(1,:,:));
 end
 % C2 group
 for l = 1: (qC21-1)
     ll = num2str(l);
     Datan1.(['datan',ll,'']) = Standard_Normalization(squeeze(Cn1(l,:,:)));
     Distn1.(['distn',ll,'']) = dist2( Datan1.(['datan',ll,'']), Datan1.(['datan',ll,'']));
     Wn1.(['Wn1',ll,'']) = affinityMatrix(Distn1.(['distn',ll,'']), K, alpha);
 end
 
 Walln1 = struct2cell(Wn1);
 if qC21>2
     AC21 = SNF(Walln1,K,T); % First local network atlas for C2 group
 else
     AC21 = squeeze(Cn1(1,:,:));
 end
 
 for l = 1: (qC22-1)
     ll = num2str(l);
     Datan2.(['datan',ll,'']) = Standard_Normalization(squeeze(Cn2(l,:,:)));
     Distn2.(['distn',ll,'']) = dist2( Datan2.(['datan',ll,'']), Datan2.(['datan',ll,'']));
     Wn2.(['Wn1',ll,'']) = affinityMatrix(Distn2.(['distn',ll,'']), K, alpha);
 end
 
 Walln2 = struct2cell(Wn2);
 if qC22>2
     AC22 = SNF(Walln2,K,T); % Second local network atlas for C2 group
 else
     AC22 = squeeze(Cn2(1,:,:));
 end
 
 for l = 1: (qC23-1)
     ll = num2str(l);
     Datan3.(['datan',ll,'']) = Standard_Normalization(squeeze(Cn3(l,:,:)));
     Distn3.(['distn',ll,'']) = dist2( Datan3.(['datan',ll,'']), Datan3.(['datan',ll,'']));
     Wn3.(['Wn1',ll,'']) = affinityMatrix(Distn3.(['distn',ll,'']), K, alpha);
 end
 
 Walln3 = struct2cell(Wn3);
 if qC23>2
     AC23 = SNF(Walln3,K,T); % Third local network atlas for C2 group
 else
     AC23 = squeeze(Cn3(1,:,:));
 end
 
 %% SNF-SNF
 
 AC1 = SNF({AC11,AC12,AC13},K,T); % Global network atlas for C1 group
 
 AC2 = SNF({AC21,AC22,AC23},K,T);% Global network atlas for C2 group
 
 D = abs(AC1-AC2);% Difference between matrices
 
 %% Step 2:  Discriminative connectional biomarker identification % % % % %
 D = triu(D,1); %Upper triangular part of matrix
 D1 = D(find(D));
 D1 = D1.';
 D2 = sort(D1,'descend')% Ranking features
 Dif = D2(1:Nf); % Extract the top Nf ranked features
 ind = [];
 
 for i = 1: Nf
     [a,b,pos] = intersect(Dif(i),D1); % Select the indices of top Nf ranked features
     ind = [ind;pos];
 end
 qq = [];
 qq1 = [];
 
 for h = 1 : Nf
     X = ind(h);
     [q,q1] = map_index_to_position_in_matrix(X,sz3);
     qq = [qq;q];
     qq1 = [qq1;q1];
 end
 
 topFeatures = zeros(sz3);
 
 for i = 1: Nf
     topFeatures(qq(i),qq1(i)) = D2(i);
     topFeatures(qq1(i),qq(i)) = D2(i);
 end
 
 %% Display Atlas 1, Atlas 2 and the circular graph of the top discriminative features
 
 
 if (displays)
     
     figure
     imagesc(AC1),title('Atlas of group 1 ','Color','b') % Display Atlas 1 of each LOO-CV iteration
     
     pause(2)
     figure
     imagesc(AC2) ,title('Atlas of group 2 ','Color','b') % Display Atlas 2 of each LOO-CV iteration.
     
     pause(2)
     figure
     imagesc(topFeatures) ,title('Top discriminative features ','Color','b') % Display Atlas 2 of each LOO-CV iteration.
     
     pause(2)
     
     close()
     figure
     J = circularGraph(topFeatures)
     
     pause(2)
     
 end
 end