Add exclude_zero_mask_values (#66)

Adds option to exclude 0 values from mapset

Author: nharp189

CanlabCore/@image_vector/image_similarity_plot.m

@@ -304,7 +304,7 @@
 bicolor = false;
 treat_zero_as_data = 0; % Default of 0 means treat zero value as missing data.
 Error_STD=0;
-
+exclude_zero_mask_values = 0; % default so that 0 mask/pattern expression values are NOT excluded
 
 % optional inputs with default values
 % -----------------------------------
@@ -368,6 +368,10 @@
 
             case 'Error_STD'
                 Error_STD = 1;
+
+            case 'exclude_zero_mask_values'
+                exclude_zero_mask_values = 1;
+
             otherwise, warning(['Unknown input string option:' varargin{i}]);
         end
     end
@@ -489,7 +493,9 @@
 
             % r(im, :) = corr_matrix(double(mask.dat(:,im)), double(obj.dat));
 
-            if treat_zero_as_data==1
+            if treat_zero_as_data==1 && exclude_zero_mask_values == 1
+                r(im, :) = canlab_pattern_similarity(obj.dat, mask.dat(:, im), 'correlation','treat_zero_as_data', 'exclude_zero_mask_values'); 
+            elseif treat_zero_as_data==1 && exclude_zero_mask_values==0
                 r(im, :) = canlab_pattern_similarity(obj.dat, mask.dat(:, im), 'correlation','treat_zero_as_data'); % mask contains the pattern image (signature or binary mask) to apply.
             else
                 r(im, :) = canlab_pattern_similarity(obj.dat, mask.dat(:, im), 'correlation');

CanlabCore/Statistics_tools/canlab_pattern_similarity.m

@@ -62,6 +62,8 @@
 %       rather than being a missing value) or binary map. In this case, you
 %       can use 'treat_zero_as_data', 1 to treat the zeros as data values.
 %
+%   **exclude_zero_mask_values**
+%
 % :Outputs:
 %   **similarity_output**
 %       Matrix of similarity measures.
@@ -138,7 +140,6 @@
 %   than missing data
 %
 
-
 % ---------------------------------
 % Defaults and optional inputs
 % ---------------------------------
@@ -148,6 +149,7 @@
 doignoremissing = false; % ignore warnings for missing voxels
 doprintwarnings = true;  % print warnings regarding missing voxels, etc.
 treat_zero_as_data = false; % Treat zero value as missing data.
+exclude_zero_mask_values = false;
 
 if any(strcmp(varargin, 'ignore_missing'))
     doignoremissing = true;
@@ -181,6 +183,10 @@
     treat_zero_as_data = true;
 end
 
+if any(strcmp(varargin, 'exclude_zero_mask_values'))
+    exclude_zero_mask_values = true;
+end
+
 % if docosine && docorr, error('Choose either cosine_similarity or correlation, or no optional inputs for dot product'); end
 
 % ---------------------------------
@@ -210,7 +216,15 @@
 
 end
 
+if exclude_zero_mask_values
+    
+    badvals_mask = pattern_weights == 0 | isnan(pattern_weights);
+    
+else
+
+    badvals_mask = isnan(pattern_weights);
 
+end
 
 
 % ---------------------------------
@@ -232,7 +246,7 @@
         switch sim_metric
             case 'corr'
 
-                    similarity_output(:, i) = image_correlation(dat, pattern_weights(:, i), badvals);
+                similarity_output(:, i) = image_correlation(dat, pattern_weights(:, i), badvals, badvals_mask);
 
             case 'cosine'
 
@@ -377,12 +391,12 @@
 end % function
 
 
-function r = image_correlation(dat, pattern_weights, badvals, varargin)
+function r = image_correlation(dat, pattern_weights, badvals, badvals_mask)
 
 
 for i = 1:size(dat, 2)    % Loop because we may have different voxel exclusions in each image
 
-    inmask = ~badvals(:, i);
+    inmask = ~badvals(:, i) & ~badvals_mask(:, i);
 
     r(i, 1) = corr(pattern_weights(inmask), dat(inmask, i));  % Correlation, excluding out-of-image pattern_weights image-wise