|
@@ -10,12 +10,18 @@
|
|
|
|
|
|
function [D_full, w_fin, Geometry, optGoal] = NLP_optimizer_v2(varargin)
|
|
|
% This function performs the beamlet optimization
|
|
|
-% Inputs: (Pat_path, path2goal) or none
|
|
|
-% If paths are used they should be passed as strings.
|
|
|
-% Outputs: full dose image dose: [D_full, w_fin, Geometry, optGoal]
|
|
|
-%
|
|
|
% [D_full, w_fin, Geometry, optGoal] = NLP_beamlet_optimizer;
|
|
|
%
|
|
|
+% Inputs:
|
|
|
+% () OR
|
|
|
+% (Pat_path, path2goal) OR
|
|
|
+% (Pat_path, path2goal, beamlet_weights)
|
|
|
+% Pat_path, path2goal = strings to patient folder and optimal goals
|
|
|
+% beamlet_weights = initial beamlet weights
|
|
|
+%
|
|
|
+% Outputs:
|
|
|
+% full dose image dose: [D_full, w_fin, Geometry, optGoal]
|
|
|
+%
|
|
|
% Made by Peter Ferjancic 1. May 2018
|
|
|
% Last updated: 1. April 2019
|
|
|
|
|
@@ -32,8 +38,25 @@ else
|
|
|
path2goal = varargin{2};
|
|
|
end
|
|
|
|
|
|
-N_fcallback1 = 10000;
|
|
|
-N_fcallback2 = 200000;
|
|
|
+str = inputdlg({'N of iterations for initial calc', 'N of iterations for full calc', ...
|
|
|
+ 'Use pre-existing NLP_result to initiate? (y/n)'}, 'input', [1,35], {'10000', '500000', 'n'});
|
|
|
+N_fcallback1 = str2double(str{1}); % 10000 is a good guesstimate
|
|
|
+N_fcallback2 = str2double(str{2}); % 500000 is a good guesstimate
|
|
|
+pre_beamWeights = str{3};
|
|
|
+
|
|
|
+switch pre_beamWeights
|
|
|
+ case 'y'
|
|
|
+ [NLP_file,NLP_path,indx] = uigetfile([Pat_path '\matlab_files\*.mat'], 'Select NLP_result file');
|
|
|
+ load([NLP_path, NLP_file])
|
|
|
+ w_beamlets = NLP_result.weights;
|
|
|
+
|
|
|
+ load([Pat_path, '\all_beams.mat'])
|
|
|
+ if numel(all_beams) ~= numel(w_beamlets)
|
|
|
+ error('Provided weight number does not match beamlet number!')
|
|
|
+ end
|
|
|
+ case 'n'
|
|
|
+ disp('Initial beam weights will be calculated.')
|
|
|
+end
|
|
|
|
|
|
%% PROGRAM STARTS HERE
|
|
|
% - no tocar lo que hay debajo -
|
|
@@ -44,30 +67,70 @@ load(path2geometry)
|
|
|
load(path2goal)
|
|
|
[beamlets, beamlets_joined, numBeamlet, numBeam, beam_i_list] = get_beam_lets(Geometry, Pat_path);
|
|
|
|
|
|
+
|
|
|
+
|
|
|
%% -- OPTIMIZATION TARGETS --
|
|
|
% -- make the optimization optGoal structure --
|
|
|
|
|
|
for i_goal = 1:size(OptGoals.goals,1)
|
|
|
- optGoal{i_goal}=OptGoals.data{i_goal};
|
|
|
- optGoal{i_goal}.beamlets_pruned = sparse(beamlets(optGoal{i_goal}.ROI_idx, :));
|
|
|
- optGoal_beam{i_goal}=OptGoals.data{i_goal};
|
|
|
- optGoal_beam{i_goal}.beamlets_pruned = sparse(beamlets_joined(optGoal{i_goal}.ROI_idx, :));
|
|
|
+ answer = inputdlg(['# of supervoxels for "' OptGoals.data{i_goal}.name '" with ' num2str(numel(OptGoals.data{i_goal}.ROI_idx)) ' vox: ("0" to skip)'])
|
|
|
+
|
|
|
+ switch answer{1}
|
|
|
+ case '0'
|
|
|
+ % if not supervoxel, just select provided ROI_idx
|
|
|
+ optGoal{i_goal} = OptGoals.data{i_goal};
|
|
|
+ optGoal{i_goal}.beamlets_pruned = sparse(beamlets(optGoal{i_goal}.ROI_idx, :));
|
|
|
+ optGoal_beam{i_goal} = OptGoals.data{i_goal};
|
|
|
+ optGoal_beam{i_goal}.beamlets_pruned = sparse(beamlets_joined(optGoal{i_goal}.ROI_idx, :));
|
|
|
+
|
|
|
+ otherwise
|
|
|
+ % -- if supervoxel, merge given columns
|
|
|
+ % - make supervoxel map
|
|
|
+ mask = zeros(OptGoals.data{i_goal}.imgDim);
|
|
|
+ mask(OptGoals.data{i_goal}.ROI_idx) = 1;
|
|
|
+
|
|
|
+ superMask = superpix_group(mask, str2double(answer{1}));
|
|
|
+
|
|
|
+ superVoxList = unique(superMask);
|
|
|
+ superVoxList = superVoxList(superVoxList>0);
|
|
|
+
|
|
|
+ optGoal{i_goal} = OptGoals.data{i_goal};
|
|
|
+ optGoal{i_goal}.ROI_idx_old = optGoal{i_goal}.ROI_idx; % copy old index data
|
|
|
+ optGoal{i_goal}.ROI_idx = zeros(numel(superVoxList), 1);
|
|
|
+ optGoal{i_goal}.opt_weight = optGoal{i_goal}.opt_weight * numel(optGoal{i_goal}.ROI_idx_old)/numel(optGoal{i_goal}.ROI_idx);
|
|
|
+
|
|
|
+ optGoal_beam{i_goal} = OptGoals.data{i_goal};
|
|
|
+ optGoal_beam{i_goal}.ROI_idx_old = optGoal_beam{i_goal}.ROI_idx;
|
|
|
+ optGoal_beam{i_goal}.ROI_idx = zeros(numel(superVoxList), 1);
|
|
|
+ optGoal_beam{i_goal}.opt_weight = optGoal_beam{i_goal}.opt_weight * numel(optGoal_beam{i_goal}.ROI_idx_old)/numel(optGoal_beam{i_goal}.ROI_idx);
|
|
|
+
|
|
|
+ h_w1 = waitbar(0, 'merging superboxels');
|
|
|
+ for i_supVox = 1:numel(superVoxList)
|
|
|
+ supVox_idx = superVoxList(i_supVox);
|
|
|
+
|
|
|
+ waitbar(i_supVox/numel(superVoxList), h_w1)
|
|
|
+
|
|
|
+ idxList = find(superMask == supVox_idx);
|
|
|
+
|
|
|
+ optGoal{i_goal}.beamlets_pruned(i_supVox,:) = sparse(mean(beamlets(idxList, :),1));
|
|
|
+ optGoal_beam{i_goal}.beamlets_pruned(i_supVox,:) = sparse(mean(beamlets_joined(idxList, :),1));
|
|
|
+ % -- make new indeces
|
|
|
+ optGoal{i_goal}.ROI_idx(i_supVox) = idxList(1);
|
|
|
+ optGoal_beam{i_goal}.ROI_idx(i_supVox) = idxList(1);
|
|
|
+ end
|
|
|
+ close(h_w1)
|
|
|
+ end
|
|
|
+
|
|
|
+
|
|
|
end
|
|
|
|
|
|
-% optGoal_idx = ROI_goals.optGoal_idx;
|
|
|
-% targetMinMax_idx = ROI_goals.targetMinMax_idx;
|
|
|
+
|
|
|
|
|
|
% -- make them robust --
|
|
|
RO_params=0;
|
|
|
optGoal_beam = make_robust_optGoal(optGoal_beam, RO_params, beamlets_joined);
|
|
|
optGoal = make_robust_optGoal(optGoal, RO_params, beamlets);
|
|
|
|
|
|
-%% -- INITIALIZE BEAMLET WEIGHTS --
|
|
|
-w0_beams = ones(numBeam,1);
|
|
|
-w0_beams = mean(optGoal_beam{1}.D_final(optGoal{1}.ROI_idx) ./ (optGoal_beam{1}.beamlets_pruned*w0_beams+0.1)) * w0_beams;
|
|
|
-% w0_beams = w0_beams + (2*rand(size(w0_beams))-1) *0.1 .*w0_beams; % random perturbation
|
|
|
-w0_beams = double(w0_beams);
|
|
|
-
|
|
|
% -- CALLBACK OPTIMIZATION FUNCTION --
|
|
|
fun1 = @(x) get_penalty(x, optGoal_beam);
|
|
|
fun2 = @(x) get_penalty(x, optGoal);
|
|
@@ -88,31 +151,43 @@ options = optimoptions('fmincon');
|
|
|
options.MaxFunctionEvaluations = N_fcallback1;
|
|
|
options.Display = 'iter';
|
|
|
options.PlotFcn = 'optimplotfval';
|
|
|
-options.UseParallel = true;
|
|
|
+% options.UseParallel = true;
|
|
|
+options.UseParallel = false;
|
|
|
% options.OptimalityTolerance = 1e-9;
|
|
|
|
|
|
-fprintf('\n running initial optimizer:')
|
|
|
-
|
|
|
-%% Run the optimization
|
|
|
-% -- GET FULL BEAM WEIGHTS --
|
|
|
-tic
|
|
|
-w_beam = fmincon(fun1,w0_beams,A,b,Aeq,beq,lb_beam,ub,nonlcon,options);
|
|
|
-fprintf(' done!:')
|
|
|
-t=toc;
|
|
|
-disp(['Optimization time for beams = ',num2str(t)]);
|
|
|
-
|
|
|
-w_beamlets = ones(numBeamlet,1);
|
|
|
-numBeam=numel(unique(beam_i_list));
|
|
|
-for beam_i = 1:numBeam % assign weights to beamlets
|
|
|
- % beamlets from same beam get same initial weights
|
|
|
- w_beamlets(beam_i_list == beam_i) = w_beam(beam_i);
|
|
|
+%% -- INITIALIZE BEAMLET WEIGHTS --
|
|
|
+switch pre_beamWeights
|
|
|
+ case 'y'
|
|
|
+ % should have been assigned previously.
|
|
|
+ disp('Provided beamlet weights used for initial comparison')
|
|
|
+ case 'n'
|
|
|
+ % if initial beamlet weights are not provided, get quick estimate
|
|
|
+ fprintf('\n running initial optimizer:')
|
|
|
+ % initialize beamlet weights, OR
|
|
|
+ w0_beams = ones(numBeam,1);
|
|
|
+ w0_beams = mean(optGoal_beam{1}.D_final(optGoal{1}.ROI_idx) ./ (optGoal_beam{1}.beamlets_pruned*w0_beams+0.1)) * w0_beams;
|
|
|
+ w0_beams = double(w0_beams);
|
|
|
+
|
|
|
+ % -- GET BEAM WEIGHTS --
|
|
|
+ tic
|
|
|
+ w_beam = fmincon(fun1,w0_beams,A,b,Aeq,beq,lb_beam,ub,nonlcon,options);
|
|
|
+ fprintf(' done!:')
|
|
|
+ t=toc;
|
|
|
+ disp(['Optimization time for beams = ',num2str(t)]);
|
|
|
+
|
|
|
+ w_beamlets = ones(numBeamlet,1);
|
|
|
+ numBeam=numel(unique(beam_i_list));
|
|
|
+ for beam_i = 1:numBeam % assign weights to beamlets
|
|
|
+ % beamlets from same beam get same initial weights
|
|
|
+ w_beamlets(beam_i_list == beam_i) = w_beam(beam_i);
|
|
|
+ end
|
|
|
end
|
|
|
-w_beamlets = w_beamlets + (2*rand(size(w_beamlets))-1) *0.1 .*w_beamlets; % small random perturbation
|
|
|
-w_beamlets = 1.1* w_beamlets; % this just kicks the beamlets up a bit to make it easier for the optimizer to start
|
|
|
+
|
|
|
+%% FULL OPTIMIZATION
|
|
|
|
|
|
% -- GET FULL BEAMLET WEIGHTS --
|
|
|
options.MaxFunctionEvaluations = N_fcallback2;
|
|
|
-% tic
|
|
|
+tic
|
|
|
fprintf('\n running full optimizer:')
|
|
|
w_fin = fmincon(fun2,w_beamlets,A,b,Aeq,beq,lb,ub,nonlcon,options);
|
|
|
fprintf(' done!:')
|
|
@@ -128,10 +203,11 @@ NLP_result.dose = D_full;
|
|
|
NLP_result.weights = w_fin;
|
|
|
save([Pat_path, '\matlab_files\NLP_result.mat'], 'NLP_result');
|
|
|
|
|
|
-warning('this part needs modification')
|
|
|
-plot_DVH(D_full, optGoal)
|
|
|
-colorwash(Geometry.data, D_full, [-500, 500], [0, 60]);
|
|
|
-% plot_DVH_robust(D_full, optGoal, optGoal_idx)
|
|
|
+
|
|
|
+plot_DVH(Geometry, D_full)
|
|
|
+colorwash(Geometry.data, D_full, [-500, 500], [0, 35]);
|
|
|
+
|
|
|
+
|
|
|
end
|
|
|
|
|
|
%% support functions
|
|
@@ -173,27 +249,27 @@ function penalty = eval_f(x, optGoal, nrs_i, sss_i, rgs_i)
|
|
|
(optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.beamlets_pruned * x)-...
|
|
|
(optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.target)));
|
|
|
case 'min_sq'
|
|
|
- % penalize if achieved dose is higher than target dose
|
|
|
+ % penalize if achieved dose is lower than target dose
|
|
|
temp1=min(0, (optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.beamlets_pruned * x)-...
|
|
|
(optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.target));
|
|
|
- d_penalty = 1.0e0 * sum(temp1.^2);
|
|
|
+ d_penalty = 1.0e0 * sum(temp1.*temp1);
|
|
|
case 'max_sq'
|
|
|
% penalize if achieved dose is higher than target dose
|
|
|
temp1=max(0, (optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.beamlets_pruned * x)-...
|
|
|
(optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.target));
|
|
|
- d_penalty = 1.0e0 * sum(temp1.^2);
|
|
|
+ d_penalty = 1.0e0 * sum(temp1.*temp1);
|
|
|
case 'LeastSquare'
|
|
|
% penalize with sum of squares any deviation from target
|
|
|
% dose
|
|
|
- d_penalty = 1.0e-1* sum(((...
|
|
|
- optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.beamlets_pruned * x) - ...
|
|
|
- optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.target).^2);
|
|
|
+ temp1 = (optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.beamlets_pruned * x) - ...
|
|
|
+ optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.target;
|
|
|
+ d_penalty = 1.0e0* sum(temp1.^2);
|
|
|
case 'min_perc_Volume'
|
|
|
% penalize by amount of volume under threshold
|
|
|
perc_vox = numel(find((optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.target) -...
|
|
|
(optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.beamlets_pruned * x) > 0)) ...
|
|
|
/ numel(optGoal{goal_i}.nrs{nrs_i}.sss{sss_i}.rgs{rgs_i}.target);
|
|
|
- d_penalty = 3.0e5 * min(perc_vox-0.05, 0)
|
|
|
+ d_penalty = 3.0e4 * min(perc_vox-0.05, 0)
|
|
|
|
|
|
case 'max_perc_Volume'
|
|
|
% penalize by amount of volume under threshold
|
|
@@ -221,7 +297,7 @@ function optGoal = make_robust_optGoal(optGoal, RO_params, beamlets);
|
|
|
% Y - Y>0 moves image down
|
|
|
% Z - in/out.
|
|
|
|
|
|
- shift_mag = 2; % vox of shift
|
|
|
+ shift_mag = 1; % vox of shift
|
|
|
nrs_scene_list={[0,0,0]};
|
|
|
|
|
|
|