Classification-driven region-growing for voxel classification.

I am not professional in Matlab Programming.I face a problem in re-implementing an algorithm.

It seems it is in an infinite loop that after long time I stopped it but there is nothing inside the LV matrix. the code is:

% neigb[]: 26 Neighbor locations (footprint)
th=.98; % threshold for controlling classifier since the number object voxels is much much lesser that background 
maxDist = Inf;
while ~isempty(VQ)
  idx=VQ(1);% removing first element from queue
  VQ(1)=[]; % deleting the first place in the queue
  [xi,yi,zi]=ind2sub([X2,Y2,Z2],idx); %converting index to position 
  LV(idx,1)=1; % marking the place as visited
  [~,ScoreF]=predict(Mdl_classifier1,Features_Test(idx,Feat));
  [~,ScoreT]=predict(Mdl_classifier2,Features_Test(idx,TFeat));
  if (ScoreF(2)>th && ScoreF(2)>ScoreT(2) )
      LV(idx,2)=.5;  % Point is classified as first object  
      % Finding 26 neighbors
      for j=1:26
          % Calculate the neighbour coordinate
          xn = xi+neigb(j,1); yn = yi+neigb(j,2); zn=zi+neigb(j,3);
          % Check if neighbour is inside or outside the image
          ins=(xn>=1)&&(yn>=1)&&(xn<=nRow)&&(yn<=nCol)&&(zn>=1)&&(zn<=nSli)...
             && sqrt((xn-initPos(1))^2+(yn-initPos(2))^2+(zn-initPos(3))^2)<maxDist;
          if (ins)
               idx=sub2ind([X2,Y2,Z2],xn,yn,zn);
              if (~LV(idx,1))% Check whether the point has already been visited or not
                VQ(end+1)=idx; % if the point has not been visited, add the neighbor point to the queue 
              end
          end
      end
  elseif (ScoreT(2)>th && ScoreT(2)>ScoreF(2)) % If the point is classified as object 2
      LV(idx,2)=1;    % Point is classified as object 2
      for j=1:26
          % Calculate the neighbour coordinate
          xn = xi+neigb(j,1); yn = yi+neigb(j,2); zn=zi+neigb(j,3);
          % Check if neighbour is inside or outside the image
          ins=(xn>=1)&&(yn>=1)&&(xn<=nRow)&&(yn<=nCol)&&(zn>=1)&&(zn<=nSli)...
              && sqrt((xn-initPos(1))^2+(yn-initPos(2))^2+(zn-initPos(3))^2)<maxDist;
          if (ins)
              idx = sub2ind([X2,Y2,Z2],xn,yn,zn);
             if (~LV(idx,1))
                VQ(end+1)=idx;
              end
          end
      end  % end for
  else
      LV(idx,2)=0;    % Point is classified background    
  end % end if        
end % end whiled

Answer 1

[Image Analyst]

Why don't you step through it with the debugger to figure out why no values in the LV matrix are getting changed? That's what everyone else would do in your situation. I don't see any reason why you can't do it too.