Style, docs, misc minor fixes

vazirilab · Jul 19, 2018 · 91c1efd · 91c1efd
1 parent 0a68165
commit 91c1efd
Show file tree

Hide file tree

Showing 8 changed files with 442 additions and 170 deletions.
diff --git a/main_nnmf_SID.m b/main_nnmf_SID.m
@@ -269,9 +269,6 @@
     SID_output.S = SID_output.S';
 end
 
-%% Crop sensor movie
-sensor_movie = sensor_movie(SID_output.idx,:);
-
 %% Plot NNMF results
 close all;
 timestr = datestr(now, 'YYmmddTHHMM');
@@ -347,7 +344,7 @@
     colorbar;
     print(fullfile(Input.outdir, [timestr '_nnmf_component_recon_' num2str(i, '%03d') '.png']), '-dpng', '-r600');
 end
-pause(10);
+pause(2);
 close all;
 
 %% Save checkpoint
@@ -367,7 +364,7 @@
 
 if Input.filter
     disp('Filtering reconstructed spatial filters');
-    SID_output.segmm = filter_recon(SID_output.recon,opts);
+    SID_output.segmm = filter_recon(SID_output.recon, opts);
 else
     SID_output.segmm = SID_output.recon;
 end
@@ -381,24 +378,29 @@
 for ii=1:size(SID_output.segmm,u)
     SID_output.neuron_centers_per_component{ii} = segment_component(SID_output.segmm{ii},Input.segmentation.threshold);
     num(ii) = size(SID_output.neuron_centers_per_component{ii},1); %#ok<AGROW>
-    disp(num(ii));
+    disp(['Component ' num2str(ii) ': Found ' num2str(num(ii)) ' neuron candidates']);
 end
 
 ids = isoutlier(num, 'ThresholdFactor', 10);
 ids = (num > mean(num)) .* ids;
+outlier_ixs = find(ids);
 
-for ii=find(ids)
+for i = 1 : numel(outlier_ixs)
+    ii = outlier_ixs(i);
     threshold = 0.1;
     SID_output.neuron_centers_per_component{ii} = segment_component(SID_output.segmm{ii}, threshold);
     num(ii) = size(SID_output.neuron_centers_per_component{ii},1);
-    disp(num(ii));
+    disp(['Re-segmenting component with overly many neurons with higher threshold: Component ' num2str(ii) ': Found ' num2str(num(ii)) ' neuron candidates']);
 end
 
-[SID_output.neuron_centers_ini,SID_output.neur_id]=iterate_cluster(SID_output.neuron_centers_per_component,Input.cluster_iter,Input.neur_rad,dim);
+% Merge closely spaced neuron candidates from different NNMF components by finding clusters of candidates that have an extent smaller than Input.neuron_rad
+[SID_output.neuron_centers_ini, SID_output.neur_id] = iterate_cluster(SID_output.neuron_centers_per_component, Input.cluster_iter, Input.neur_rad, dim);
 
-figure;plot(hist(SID_output.neuron_centers_ini(:,3),size(SID_output.recon{1},3)));
+figure; histogram(SID_output.neuron_centers_ini(:,3), -0.5 : 1 : size(SID_output.recon{1},3) + 0.5);
 xlabel('Z plane index');
 ylabel('Neuron frequency');
+print(fullfile(Input.outdir, [timestr '_segmm_z-hist.png']), '-dpng', '-r300');
+
 if ~isfield(Input.segmentation,'top_cutoff')
     disp('Check the axial distribution and remove top/bottom artefacts');
     Input.segmentation.top_cutoff = input('Input top cutoff \n');
@@ -439,6 +441,9 @@
 %%
 clearvars -except sensor_movie Input SID_output mean_signal psf_ballistic Hsize m sensor_movie_max sensor_movie_min dim;
 
+%% Crop sensor movie
+sensor_movie = sensor_movie(SID_output.idx,:);
+
 %% Initiate forward_model
 %TODO: check performance of generate_forward_model() with matfile
 %psf_ballistic = matfile(Input.psffile);
@@ -455,13 +460,14 @@
     opts.image_size = SID_output.movie_size(1:2);
     opts.axial = Input.axial;
     opts.neur_rad = Input.neur_rad;
-    SID_output.forward_model_ini=generate_LFM_library_GPU(SID_output.recon,SID_output.neuron_centers_ini,round(SID_output.neur_id),psf_ballistic,opts);
+    SID_output.forward_model_ini = generate_LFM_library_GPU(SID_output.recon, SID_output.neuron_centers_ini, ...
+                                                            round(SID_output.neur_id), psf_ballistic, opts);
 end
 
 %% Generate template
-SID_output.template=generate_template(SID_output.neuron_centers_ini,psf_ballistic.H,SID_output.std_image,Input.template_threshold);
+SID_output.template = generate_template(SID_output.neuron_centers_ini, psf_ballistic.H, SID_output.std_image, Input.template_threshold);
 
-%% Crop model
+%% Remove neuron templates that don't have positive weights inside of to the overall crop region determined further up (based on crop_mask and/or crop_params)
 neur = find(squeeze(max(SID_output.forward_model_ini(:, SID_output.idx), [], 2) > 0));
 SID_output.forward_model_iterated = SID_output.forward_model_ini(neur, SID_output.idx);
 SID_output.neuron_centers_iterated = SID_output.neuron_centers_ini(neur, :);
@@ -499,15 +505,15 @@
     opts_temp.gpu_id = Input.gpu_ids(1);
 end
 
-if isfield(Input, 'bg_sub') && Input.bg_sub% && ~Input.use_std
+if isfield(Input, 'bg_sub') && Input.bg_sub % && ~Input.use_std
     SID_output.forward_model_iterated(end+1,:) = SID_output.bg_spatial(SID_output.idx);
-    SID_output.indices_in_orig=[SID_output.indices_in_orig' length(SID_output.indices_in_orig)+1];
+    SID_output.indices_in_orig = [SID_output.indices_in_orig' length(SID_output.indices_in_orig) + 1];
 end
 
-sensor_movie =double(sensor_movie);
+sensor_movie = double(sensor_movie);
 disp([datestr(now, 'YYYY-mm-dd HH:MM:SS') ': ' 'Starting temporal update']);
-SID_output.forward_model_iterated=(1./(sqrt(sum(SID_output.forward_model_iterated.^2....
-    ,2)))).*SID_output.forward_model_iterated;
+SID_output.forward_model_iterated = (1 ./ sqrt(sum(SID_output.forward_model_iterated .^ 2, 2))) ...
+                                    .* SID_output.forward_model_iterated;
 SID_output.timeseries_ini = LS_nnls(SID_output.forward_model_iterated(:,SID_output.microlenses(SID_output.idx)>0)', double(sensor_movie(SID_output.microlenses(SID_output.idx)>0,:)), opts_temp);
 disp([datestr(now, 'YYYY-mm-dd HH:MM:SS') ': ' 'Temporal update completed']);
 
@@ -541,22 +547,23 @@
         end
     end
 
-    [SID_output.forward_model_iterated,SID_output.timeseries_iterated,template_...
-        ,SID_output.indices_in_orig] = temporal_SID_update(...
-        sensor_movie,SID_output.forward_model_iterated,SID_output.timeseries_iterated...
-        ,template_,SID_output.indices_in_orig,opts_temp);
-
-
-    [SID_output.forward_model_iterated,SID_output.timeseries_iterated,template_...
-        ,SID_output.indices_in_orig] = merge_filters(...
-        SID_output.forward_model_iterated,SID_output.timeseries_iterated...
-        ,template_,SID_output.indices_in_orig,opts_temp);
-    disp([datestr(now, 'YYYY-mm-dd HH:MM:SS') ': ' 'Iteration ' num2str(iter) ' of ' num2str(num_iter) ' completed']);
-end
-SID_output.neuron_centers_iterated=SID_output.neuron_centers_ini(SID_output.indices_in_orig(1:end-1),:);
-
-SID_output.template_iterated=template_;
-opts_temp.warm_start=[];
+    [SID_output.forward_model_iterated, ...
+     SID_output.timeseries_iterated, template_, ...
+     SID_output.indices_in_orig] = temporal_SID_update(sensor_movie, ...
+                                                       SID_output.forward_model_iterated, ...
+                                                       SID_output.timeseries_iterated, ...
+                                                       template_, SID_output.indices_in_orig, opts_temp);
+
+    [SID_output.forward_model_iterated, ...
+     SID_output.timeseries_iterated, template_, ...
+     SID_output.indices_in_orig] = merge_filters(SID_output.forward_model_iterated, ...
+                                                 SID_output.timeseries_iterated, ...
+                                                 template_, SID_output.indices_in_orig, opts_temp);
+    disp([datestr(now, 'YYYY-mm-dd HH:MM:SS') ': ' 'Iteration ' num2str(iter) ' of ' num2str(Input.num_iter) ' completed']);
+end
+SID_output.neuron_centers_iterated = SID_output.neuron_centers_ini(SID_output.indices_in_orig(1:end-1), :);
+SID_output.template_iterated = template_;
+opts_temp.warm_start = [];
 clear sensor_movie;
 disp([datestr(now, 'YYYY-mm-dd HH:MM:SS') ': ' 'SID model optimization completed']);
 
@@ -614,6 +621,7 @@
 hold on;
 imagesc(squeeze(max(nmf_mip, [], 3)));
 scatter(SID_output.neuron_centers_iterated(:,2), SID_output.neuron_centers_iterated(:,1), 'r.');
+%scatter(SID_output.neuron_centers_ini(:,2), SID_output.neuron_centers_ini(:,1), 'r.');
 hold off;
 axis image;
 axis ij;
@@ -720,7 +728,7 @@
 axis image;
 colorbar();
 subplot(3,1,3);
-plot((1:size(SID_output.timeseries_iterated,2))/15, SID_output.timeseries_iterated(ix,:));
+plot((1:size(SID_output.timeseries_iterated,2)), SID_output.timeseries_iterated(ix,:));
 title(['Neuron candidate ' num2str(ix)]);
 
 %% Delete cached psf file
@@ -731,4 +739,6 @@
 
 %%
 disp([datestr(now, 'YYYY-mm-dd HH:MM:SS') ': ' 'main_nnmf_SID() returning'])
-%end
+
+%%
+end
diff --git a/src/generate_template.m b/src/generate_template.m
@@ -1,21 +1,21 @@
 function template=generate_template(centers,H,std_image,thres,shape)
 % GENERATE_TEMPLATE computes for each center of a putative neuron its 
-% template, a binary image fixing the area, where the LFM signature of the 
-%putative neuron is to be expected.
+% template, a binary image that defines the area where the LFM signature of the 
+% putative neuron is to be expected.
 %
 % Input: 
-% centers...        centroid of the putative neurons
-% H...              point spread function of the LFM
-% std_image...      standard deviation image of the LFM_movie
-% thres...          threshold for the estimation of the radius in which an
-%                   LFM signature is to be expected. The smaller 'thres'
-%                   the greater the radius.
-% shape...          boolean; if true the shape of the area, where the LFM 
-%                   signature of the putative neuron is to be expected, has
-%                   the shape of a disk, otherwise the shape of a square.
+% centers           Centroids of the putative neurons
+% H                 Point spread function of the LFM
+% std_image         Standard deviation image of the LFM_movie
+% thres             Threshold for the estimation of the radius in which an
+%                       LFM signature is to be expected. The smaller 'thres'
+%                       the greater the radius.
+% shape             boolean; if true the shape of the area, where the LFM 
+%                       signature of the putative neuron is to be expected, has
+%                       the shape of a disk, otherwise the shape of a square.
 %
 % Output:
-% template...       Library of templates of the putative neurons
+% template          Library of templates of the putative neurons
 
 if nargin<4
     thres=0.03;

diff --git a/src/iterate_cluster.m b/src/iterate_cluster.m
@@ -1,28 +1,29 @@
-function [centroid, ID, centers]=iterate_cluster(centers_cell,N,radius,dim)
-% ITERATE_CLUSTER: Algorithm clusters the elements of the cell array
-% centers_cell, such that the clusters have a maximal size of
-% radius and each of their elements is from a different cell of
-% centers_cell.
+function [centroid, ID, centers] = iterate_cluster(centers_cell, N, radius, dim)
+% ITERATE_CLUSTER clusters the elements of the cell array centers_cell, 
+% such that the clusters have a maximal radius according to input argument "radius"
+% and each of their elements is from a different cell of centers_cell.
 %
 % Input:
-% centers_cell...   cell array of array of size [m,3]
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+% centers_cell         cell array of array of size [m,3]
+% N                    number of iterations to perform
+% radius               max radius of cluster
+% dim                  vector of relative voxel edge lengths along the three dimensions
+
 centroid =[];
 for jj=1:size(centers_cell,2)
     centroid=[centroid' centers_cell{jj}']';
 end
 
-
 for nn=1:N
     centers=[];
     centroid_=[];
     centers_per_component_=centers_cell;
     for ii=1:size(centroid,1)
-        centers{ii}=[];
+        centers{ii}=[]; %#ok<AGROW>
         for jj=1:size(centers_per_component_,2)
             if ~isempty(centers_per_component_{jj})
                 rnd_ind=randperm(size(centers_cell{jj},1));
-                [a,n]=min(sum((dim.*(centers_per_component_{jj}(rnd_ind,:)-[centroid(ii,:)])).^2,2));
+                [a, n] = min(sum((dim .* (centers_per_component_{jj}(rnd_ind,:) - centroid(ii,:))) .^2, 2));
                 if a<radius^2
                     centers{ii}=[centers{ii}' centers_per_component_{jj}(rnd_ind(n),:)']';
                     centers_per_component_{jj}(rnd_ind(n),:)=inf*[1 1 1];
@@ -33,8 +34,8 @@
     end
     idx=randperm(size(centroid_,1));
     centroid=centroid_(idx,:);
-    size(centroid)
-    disp(nn);
+    %disp(size(centroid));
+    %disp(nn);
 end
 
 [~,n] = sort(sum(centroid,2));
@@ -44,7 +45,7 @@
     for ii=1:size(centroid,1)
         for jj=1:size(centers_per_component_,2)
             if ~isempty(centers_per_component_{jj})
-                a=min(sum((dim.*(centers_cell{jj}-[centroid(ii,:)])).^2,2));
+                a = min(sum((dim .* (centers_cell{jj} - centroid(ii,:))) .^2, 2));
                 if a<radius^2
                     ID(ii,jj)=true;
                 end
@@ -54,7 +55,5 @@
     ID=ID(n,:);
 end
 
-
 centroid = centroid(n,:);
-
-end
+end
diff --git a/src/merge_filters.m b/src/merge_filters.m
@@ -1,23 +1,20 @@
-function [forward_model,timeseries,template,indices_in_orig]...
-    =merge_filters(forward_model,timeseries...
-        ,template,indices_in_orig,opts)
-% MERGE_FILTERS merges components of the SID-nnmf, that overlap over 30% in
-% their templates and have a correlation of more than opts.limit (default:
-% 0.9).
-%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
-
- if nargin<5
-     opts=struct;
- end
-
- if ~isfield(opts,'limit')
-     opts.limit=0.90;
- end
-
-  if ~isfield(opts,'bg_sub')
-     opts.bg_sub=true;
- end
-
+function [forward_model, timeseries, template, indices_in_orig] = ...
+    merge_filters(forward_model, timeseries, template, indices_in_orig, opts)
+% MERGE_FILTERS merges components of the SID-nnmf, that overlap by more than 30% in
+% their templates and have a correlation of more than opts.limit in their temporal signals (default: 0.9)
+
+if nargin<5
+    opts=struct;
+end
+
+if ~isfield(opts,'limit')
+    opts.limit=0.90;
+end
+
+if ~isfield(opts,'bg_sub')
+    opts.bg_sub=true;
+end
+
 corr_m = corrcoef(timeseries')-eye(size(timeseries,1));
 corr_m = corr_m(1:end-opts.bg_sub,1:end-opts.bg_sub);
 

diff --git a/src/segment_component.m b/src/segment_component.m
@@ -1,18 +1,18 @@
 function centers = segment_component(component, threshold)
 
-    component=component/max(component(:));
-    component(isnan(component))=0;
-    component=component-threshold;
-    component(component<0)=0;
-    centers=[];
-    
-    B=reshape(component,[],1);
-    beads=bwconncomp((component));
-    for k=1:beads.NumObjects
-        qu=B(beads.PixelIdxList{1,k});
-        q=sum(B(beads.PixelIdxList{1,k}));
-        [a,b,c]=ind2sub(size(component),beads.PixelIdxList{1,k});
-        centers(k,:)=([a,b,c]'*qu/q)';
-    end
-    
+component=component/max(component(:));
+component(isnan(component))=0;
+component=component-threshold;
+component(component<0)=0;
+centers=[];
+
+B=reshape(component,[],1);
+beads=bwconncomp((component));
+for k=1:beads.NumObjects
+    qu=B(beads.PixelIdxList{1,k});
+    q=sum(B(beads.PixelIdxList{1,k}));
+    [a,b,c]=ind2sub(size(component),beads.PixelIdxList{1,k});
+    centers(k,:)=([a,b,c]'*qu/q)'; %#ok<AGROW>
+end
+
 end