-
Notifications
You must be signed in to change notification settings - Fork 11
/
example.m
220 lines (183 loc) · 9.75 KB
/
example.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
%% example code
% tested with MATLAB R2017a
clear all; clc; close all;
% add paths
addpath('./extern');
addpath('./extern/NIfTI_20140122');
addpath('./extern/harris');
addpath('./include');
dataset_names={{'volunteerA','9'},{'volunteerA','26'},{'volunteerB','8'},...
{'volunteerB','26'},{'volunteerC','8'},{'volunteerC','26'},{'volunteerD','7'},...
{'volunteerD','27'}};
%% set dataset paths
patString = dataset_names{1}{1};
sliceString = dataset_names{1}{2};
staticImageFilename = ['./Data/',patString,'/bhs/sl_ss',sliceString,'_bh.nii.gz'];
dynamicImagePath = ['./Data/',patString,'/dyn/images/'];
fieldsPath = ['./Data/',patString,'/dyn/deeds_bh/'];
lmksPath = ['./Data/',patString,'/dyn/lmks/'];
%% load static image
staticImage = load_nii(staticImageFilename);
spacing = staticImage.hdr.dime.pixdim(2);
mask = logical(ones(size(staticImage.img))-(regiongrowing(staticImage.img,1,1,3000)+...
regiongrowing(staticImage.img,1,size(staticImage.img,2),3000)));
staticImageLM = fliplr(csvread([lmksPath,'sl_ss',sliceString,'_bh_lmks.txt']))+1;
%% find feature points
% harrisParams = [sigma, thresh, radius for NMS, display];
harrisParams = [2, 1000000, 8, 1];
% fpoints: (numOfFeaturePoints x 2)
% fpointsImage: binary image with size (imh x imw)
[fpoints, fpointsImage] = selectFPs(staticImage.img.*mask,harrisParams);
fpointsImageMask = fpointsImage(mask);
fpointsFieldMask = zeros(numel(fpointsImageMask)*2,1);
fpointsFieldMask(1:2:end) = fpointsImageMask;
fpointsFieldMask(2:2:end) = fpointsImageMask;
fpointsFieldMask = logical(fpointsFieldMask);
%% load images, fields and build model
numOfTrainingFields = 20; numOfsamples = 40;
numOfTestdata = 20; variability = 0.98;
testImages = zeros([size(staticImage.img),numOfTestdata]);
for frame=1:numOfTestdata
imageFilename = [dynamicImagePath,'sl_ss',sliceString,'_tt',num2str(frame+numOfTrainingFields),'.nii.gz'];
image = load_nii(imageFilename);
testImages(:,:,frame) = image.img;
end
fields = loadDeedsFieldsMasked(1:numOfsamples,sliceString,fieldsPath,mask,1);
[muStandard, UStandard, DStandard] = buildMotionModel( fields(:,1:numOfTrainingFields), variability);
%% load test images and perform block matching
BMBlockSize = 5; BMSearchWindow = 15;
Nfpoints = numel(fpoints(:,1));
blockSize = 2*BMBlockSize+1; searchSize = BMSearchWindow*2+1;
pad = BMSearchWindow+BMBlockSize;
[xs,ys] = meshgrid(-BMSearchWindow:BMSearchWindow,-BMSearchWindow:BMSearchWindow);
% compute SSC descriptor for reference image
SSCsigma = 1.2; SSCdelta = 3;
refImageSSC = SSC_descriptor2D(staticImage.img,SSCsigma,SSCdelta);
refImageSSC = padarray(refImageSSC,[pad,pad,0]);
sscDim = size(refImageSSC,3);
%---- prepare for the Block-Matching using GPU
BMIndex = [];
refBlock = zeros(blockSize,blockSize,sscDim,Nfpoints,'single');
filt = ones(blockSize,blockSize,sscDim,Nfpoints,'single');
gfilt = gpuArray(filt);
for j=1:Nfpoints
refBlock(:,:,:,j) = refImageSSC(fpoints(j,1)+pad-BMBlockSize:fpoints(j,1)+pad+BMBlockSize,...
fpoints(j,2)+pad-BMBlockSize:fpoints(j,2)+pad+BMBlockSize,:);
end
gRefBlock = gpuArray(reshape(refBlock,blockSize,blockSize,[]));
gRefSum2 = vl_nnconv(gRefBlock.^2,gfilt,[]);
grefSum2 = repmat(gRefSum2,[searchSize,searchSize,1])./(blockSize*blockSize*sscDim);
grefSum2 = reshape(grefSum2,searchSize,searchSize,Nfpoints);
fprintf('variables for GPU block-matching are prepared...\n');
% sparse model
muSparse=muStandard(fpointsFieldMask);
USparse=UStandard(fpointsFieldMask,:);
%% compute block-matching, coupled convex optimization with and without temporal term
TREidx = []; TREBM = []; TREBMCoupled = []; TREBMCoupledTemporal = [];
tstart = tic;
for frame=1:numOfTestdata
filenameLMs = [lmksPath,'sl_ss',sliceString,'_tt',num2str(frame+numOfTrainingFields),'_lmks.txt'];
oldField = fields(:,1);
%% Block-Matching: results in distance Map (distSSC)
% disp(['BM Frame: ',num2str(frame)]);
testImageSSC = SSC_descriptor2D(squeeze(testImages(:,:,frame)),SSCsigma,SSCdelta);
distSSC = conv_blockmatching(grefSum2,testImageSSC,fpoints,...
gRefBlock,gfilt,blockSize,searchSize,sscDim);
distSSC = reshape(permute(distSSC,[3,1,2]),Nfpoints,[]);
[~,bestind] = min(distSSC,[],2);
% compute sparse displacement field
xDisp = xs(bestind(:)); yDisp = ys(bestind(:));
%% Block-Matching field reconstruction
bmField = zeros(size(fields(:,1),1),1);
bmDisp = zeros(size(xDisp,1)*2,1);
bmDisp(1:2:end) = xDisp; bmDisp(2:2:end) = yDisp;
bmField(fpointsFieldMask) = bmDisp;
% reconstruction of dense field from block-matching result
approxWeights = USparse\(bmDisp-muSparse);
stddevLimits = 3*sqrt(DStandard(1:size(UStandard,2)));
stddevLimitsIndicator = abs(approxWeights)>stddevLimits;
approxWeights(stddevLimitsIndicator) = sign(approxWeights(stddevLimitsIndicator)).*...
stddevLimits(stddevLimitsIndicator);
approxFieldBM = UStandard*approxWeights+muStandard;
%% coupled convex optimization
lambdas=logspace(-1.5,0,4);
maxComp = [1,1,1,1,1,1,1,1,1,1,1,1,1,1].*size(UStandard,2);
coupledDisp = zeros(size(xDisp,1)*2,1);
[coupledDisp(1:2:end),coupledDisp(2:2:end),bestindCS] = coupledSearch(distSSC,...
USparse,lambdas,BMSearchWindow,muSparse,DStandard,maxComp);
% field reconstruction from coupled convex optimization result
approxWeights = regularizedRegression(USparse*diag(sqrt(DStandard(1:size(UStandard,2)))),...
coupledDisp-muSparse,1);
stddevLimits = 3.*ones(size(approxWeights));
stddevLimitsIndicator = abs(approxWeights)>stddevLimits;
approxWeights(stddevLimitsIndicator) = sign(approxWeights(stddevLimitsIndicator)).*...
stddevLimits(stddevLimitsIndicator);
approxFieldBMCoupled = (UStandard*diag(sqrt(DStandard(1:size(UStandard,2))))*...
approxWeights)+muStandard;
%% coupled convex optimization (temporal)
beta = [0.5,0.25,0,0,0,0];
coupledDispTemporal = zeros(size(xDisp,1)*2,1);
[coupledDispTemporal(1:2:end),coupledDispTemporal(2:2:end),bestindCS] = ...
coupledSearchTemporal(distSSC,USparse,lambdas,BMSearchWindow,muSparse,...
DStandard,maxComp,beta,oldField(fpointsFieldMask));
% field reconstruction from coupled convex optimization (temporal) result
approxWeights = regularizedRegression(USparse*diag(sqrt(DStandard(1:size(UStandard,2)))),...
coupledDispTemporal-muSparse,1);
stddevLimits = 3.*ones(size(approxWeights));
stddevLimitsIndicator = abs(approxWeights)>stddevLimits;
approxWeights(stddevLimitsIndicator) = sign(approxWeights(stddevLimitsIndicator)).*...
stddevLimits(stddevLimitsIndicator);
approxFieldBMCoupledTemporal = (UStandard*diag(sqrt(DStandard(1:size(UStandard,2))))*...
approxWeights)+muStandard;
%% compute TRE
if exist(filenameLMs, 'file') == 2
currentLMs = fliplr(csvread(filenameLMs))+1;
TREidx = [TREidx frame+numOfTestdata];
% load deeds dense field
fullField = loadDeedsField(frame,sliceString,fieldsPath,1);
xFullField = fullField(:,:,1);
yFullField = fullField(:,:,2);
% TRE Block-Matching
xFullField(mask) = approxFieldBM(1:2:end);
yFullField(mask) = approxFieldBM(2:2:end);
fullField(:,:,1) = xFullField;
fullField(:,:,2) = yFullField;
griddedInterpolantX = griddedInterpolant(fullField(:,:,1));
griddedInterpolantY = griddedInterpolant(fullField(:,:,2));
displ(:,1) = griddedInterpolantX(staticImageLM(:,2),staticImageLM(:,1));
displ(:,2) = griddedInterpolantY(staticImageLM(:,2),staticImageLM(:,1));
warpedLMs = displ +staticImageLM;
tre = mean(sqrt(sum((warpedLMs-currentLMs).^2,2))*spacing);
TREBM = [TREBM tre];
% TRE coupled convex optimization
xFullField(mask) = approxFieldBMCoupled(1:2:end);
yFullField(mask) = approxFieldBMCoupled(2:2:end);
fullField(:,:,1) = xFullField;
fullField(:,:,2) = yFullField;
griddedInterpolantX = griddedInterpolant(fullField(:,:,1));
griddedInterpolantY = griddedInterpolant(fullField(:,:,2));
displ(:,1) = griddedInterpolantX(staticImageLM(:,2),staticImageLM(:,1));
displ(:,2) = griddedInterpolantY(staticImageLM(:,2),staticImageLM(:,1));
warpedLMs = displ + staticImageLM;
tre = mean(sqrt(sum((warpedLMs-currentLMs).^2,2))*spacing);
TREBMCoupled = [TREBMCoupled tre];
% TRE coupled convex optimization (temporal)
xFullField(mask) = approxFieldBMCoupledTemporal(1:2:end);
yFullField(mask) = approxFieldBMCoupledTemporal(2:2:end);
fullField(:,:,1) = xFullField;
fullField(:,:,2) = yFullField;
griddedInterpolantX = griddedInterpolant(fullField(:,:,1));
griddedInterpolantY = griddedInterpolant(fullField(:,:,2));
displ(:,1) = griddedInterpolantX(staticImageLM(:,2),staticImageLM(:,1));
displ(:,2) = griddedInterpolantY(staticImageLM(:,2),staticImageLM(:,1));
warpedLMs = displ + staticImageLM;
tre = mean(sqrt(sum((warpedLMs-currentLMs).^2,2))*spacing);
TREBMCoupledTemporal = [TREBMCoupledTemporal tre];
end
end
tend = toc(tstart);
fprintf('\nmean TRE: \n');
fprintf('block-matching \t\t = %.2f mm\n',mean(TREBM));
fprintf('coupled convex \t\t = %.2f mm\n',mean(TREBMCoupled));
fprintf('coupled convex temporal = %.2f mm\n',mean(TREBMCoupledTemporal));
fprintf('computation time \t = %.2f s (%.2f s/frame)\n',tend,tend/numOfTestdata);