-
Notifications
You must be signed in to change notification settings - Fork 0
/
StackofSpiralsB0.m
228 lines (207 loc) · 10.8 KB
/
StackofSpiralsB0.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
221
222
223
224
225
classdef StackofSpiralsB0<StackofSpirals
%[obj] = StackofSpiralsB0(k,w,imSize,sens,B0map,adcTime,varargin)
%Extension of StackofSpirals class for Bo correction
%INPUTS:
% k - kspace trajectory,scaled -0.5 to 0.5 [Axis x #points x NInterleaves x NPartitions]
% w - density compensation function [ #points x NInterleaves x NPartitions]
% imSize - [NColxNLin] or [NColxNLinxNPar]
% sens - Coil sensitvities(can be empty) [ColxLinxParxCha] or []
% fm -Field map in rad/s [ColxLin] or [ColxLinxPar]
% adcTime : time in seconds of one interleaves [single column]
%
% varargin : Name-Value pair
% 'Method' : Interpolation method ({'MFI','MTI'})
% 'max_mem_GB' : max memory for MFI weight calculation (deafult 4 GB);
% 'CompMode' : Computation mode {'GPU3D','CPU3D','CPU2DHybrid'} (default GPU3D)
% 'precision' : {'single','double'} (default single)
% 'KernelWidth' : scalar for gridding kernel size/#neighbours(default 5)
% 'osf' : oversampling factor (default 1.5(GPU),2(CPU))
%ONLY relavent for GPU
% 'SectorWidth' : default(12) (recommended: 8 for 3D,16 for 2D)
% 'atomic' : true/false for atomic operation (default true)
% 'use_textures' : true/false for using textures on gpu (default true)
% 'balance_workload' : true/false for balanced operation (default true)
%
%
%Example USAGE:
%B0OP=StackofSpiralsB0(kxyz,(DCF3d),[FTOP.imSize 20],permute(csm,[2 3 4 1]),-1*fm,adcTime(:)*1e-6,...
% 'CompMode','CPU2DHybrid','precision','single',...
% 'Method','MFI');
%im3d=B0OP'*sig3d;%[ColxLinxParxCha] %Adjoint
%sigr=B0OP*im3d; %[ColxLinxPar] %forward
%im=spiralCGSENSE(B0OP,sig3d,'maxit',10,'reg','none'); %cgsense
%
%
%
%Dependencies:
%CPU: NUFFT functions from https://github.com/JeffFessler/mirt
%GPU: gpuNUFFT https://github.com/andyschwarzl/gpuNUFFT
%
%Author: praveen.ivp@gmail.com
properties
B0map
B0Para
tk
Interp_weights
mask
end
methods
function obj=StackofSpiralsB0(k,w,imSize,sens,B0map,adcTime,varargin)
obj@StackofSpirals(k,w,imSize,sens,varargin{:});
% obj.B0Para=struct('Method','MTI','Levels',[],'nLevels',0,'max_mem_GB',4,'mode','LeastSquares','precision','single');
obj=obj.ParseInputPara(varargin{:});
if(isempty(obj.op.sens)||isscalar(isempty(obj.op.sens)))
obj.mask=ones(size(B0map),'logical');
else
switch(obj.Mode)
case 'GPU3D'
obj.mask = reshape(obj.op.sens(1,:,1),[imSize(1) imSize(2) max(1,imSize(3))])>0;
otherwise
obj.mask=abs(obj.op.sens(:,:,:,1))>0;
end
end
if(isscalar(B0map)||isempty(B0map))
%if field map is scalar or empty, it becomes normal NUFFT operator
obj.tk=0;
obj.B0Para.nLevels=1;
obj.B0Para.Levels=0;
obj.Interp_weights=1;
obj.B0map=0;
else
if(strcmpi(obj.precision,'single'))
obj.tk=single(adcTime(:));
obj.B0map=single(B0map);
% obj.B0map=single(obj.B0map(obj.mask)); %(squeeze(abs(obj.CoilSens(1,:,:,:)))>0)
obj.B0Para.nLevels=single(ceil((2*max(abs(obj.B0map(:)))*max(obj.tk)/pi)));
% calculate weights
switch(obj.B0Para.Method)
case 'MFI'
obj.B0Para.Levels=single(linspace(min(obj.B0map(obj.mask)),1*max(obj.B0map(obj.mask)),obj.B0Para.nLevels));
obj=obj.CalcWeightsMFI();
case 'MTI'
obj.B0Para.Levels=single(linspace(0, max(obj.tk),obj.B0Para.nLevels));
obj=obj.CalcWeightsMTI();
end
else
obj.tk=double(adcTime(:));
obj.B0map=double(B0map);
% obj.B0map=single(obj.B0map(obj.mask)); %(squeeze(abs(obj.CoilSens(1,:,:,:)))>0)
obj.B0Para.nLevels=double(ceil((1*max(abs(obj.B0map(:)))*max(obj.tk)/pi)));
% calculate weights
switch(obj.B0Para.Method)
case 'MFI'
obj.B0Para.Levels=double(linspace(min(obj.B0map(obj.mask)),1*max(obj.B0map(obj.mask)),obj.B0Para.nLevels));
obj=obj.CalcWeightsMFI();
case 'MTI'
obj.B0Para.Levels=double(linspace(0, max(obj.tk),obj.B0Para.nLevels));
obj=obj.CalcWeightsMTI();
end
end
end
end
function obj=ParseInputPara(obj,varargin)
p=inputParser;
p.KeepUnmatched=1;
addParameter(p,'Method','MTI',@(x) any(validatestring(x,{'none','MFI','MTI'})));
addParameter(p,'max_mem_GB',4,@(x) isscalar(x));
addParameter(p,'fitMode','LeastSquares',@(x) any(validatestring(x,{'LeastSquares','LeastSquares_HighMemory','NearestNeighbour'})));
addParameter(p,'nLevels',0,@(x)isscalar(x));
addParameter(p,'Levels',[],@(x) isvector(x));
parse(p,varargin{:});
obj.B0Para=p.Results;
end
function [out]=mtimes(obj,bb)
% bb=[nFE,nIntlv,nPar,nCha]% adj case
% bb=[ImX,Imy,Imz,nCha] % forward case
if(obj.adjoint==1) % Reverse operator: kspace data to image
all_images=zeros([obj.imSize(1:2),max(1,obj.imSize(3)),max(1,size(bb,4)-obj.op.sensChn),length(obj.B0Para.Levels)],obj.precision);
switch(obj.B0Para.Method)
case 'MFI'
b0term=exp(1i.*obj.tk(:)*obj.B0Para.Levels(:).');
for idx_freq=1:length(obj.B0Para.Levels)
all_images(:,:,:,:,idx_freq)= mtimes@StackofSpirals(obj,bsxfun(@times,bb,b0term(:,idx_freq)));
end
out=sum(bsxfun(@times,permute((obj.Interp_weights),[2 3 4 5 1]),all_images),5);
case 'MTI'
for idx_tau=1:obj.B0Para.nLevels
sig_MTI=bsxfun(@times,obj.Interp_weights(idx_tau,:).',bb);
all_images(:,:,:,:,idx_tau)= mtimes@StackofSpirals(obj,sig_MTI);
all_images(:,:,:,:,idx_tau)=bsxfun(@times,all_images(:,:,:,:,idx_tau),exp(1i*obj.B0map*obj.B0Para.Levels(idx_tau)));
end
out=sum(all_images,5);
end
else % forward operation image to kspace
out=zeros([obj.dataSize max(size(bb,4),obj.op.sensChn)]);
switch(obj.B0Para.Method)
case 'MFI'
b0term=exp(-1i.*obj.tk(:)*obj.B0Para.Levels(:).');
for idx_freq=1:obj.B0Para.nLevels
temp1=bsxfun(@times,squeeze(conj(obj.Interp_weights(idx_freq,:,:,:))),bb);
temp2= mtimes@StackofSpirals(obj,temp1);
temp2=bsxfun(@times,b0term(:,idx_freq),temp2);
out=out+temp2;
end
case 'MTI'
for idx_tau=1:obj.B0Para.nLevels
temp1=bsxfun(@times,bb,exp(-1i*obj.B0map*obj.B0Para.Levels(idx_tau)));
temp2= mtimes@StackofSpirals(obj,temp1);
temp2=bsxfun(@times,temp2,conj(obj.Interp_weights(idx_tau,:).'));
out=out+temp2;
end
end
end
end
function obj=ctranspose(obj)
if(obj.adjoint==0)
obj.adjoint=1;
end
end
function obj=CalcWeightsMFI(obj)
obj.Interp_weights=zeros([size(obj.B0map) obj.B0Para.nLevels]);
% calculate weights for all values in b0maps
if(strcmp(obj.B0Para.fitMode,'LeastSquares'))
obj.mask=(obj.B0map~=0);
idx=find(obj.mask);
n_split=ceil(8*length(obj.tk)*numel(obj.B0map(obj.mask))/1e9/obj.B0Para.max_mem_GB);
warning('Too large matrix to invert: spliting into %d parts',n_split);
obj.Interp_weights=zeros(numel(obj.B0Para.Levels),numel(obj.B0map),obj.precision);
witk=single(exp(1i.*obj.tk*obj.B0Para.Levels));
for i=1:n_split
B=exp(1i*obj.tk*obj.B0map(idx(i:n_split:end)')); %this is a HUGE matrix limited by max_size
obj.Interp_weights(:,idx(i:n_split:end))=mldivide((witk'*witk),(witk'*B));
end
obj.Interp_weights=reshape(obj.Interp_weights,[numel(obj.B0Para.Levels),size(obj.B0map)]);
elseif(strcmp(obj.B0Para.fitMode,'LeastSquares_HighMemory'))
witk=exp(1i.*obj.tk*obj.B0Para.Levels);
%memory hog way but faster
B=exp(1i*obj.B0map(:)*obj.tk.').';
obj.Interp_weights=mldivide((witk'*witk),(witk'*B));
obj.Interp_weights=permute(reshape(obj.Interp_weights,[],size(obj.B0map,1),size(obj.B0map,2)),[2 3 1]);
elseif(strcmp(obj.B0Para.fitMode,'NearestNeighbour') )
for i=1:size(obj.Interp_weights,1)
for j=1:size(obj.Interp_weights,2)
[~,I]=min(abs(obj.B0Para.Levels-obj.B0map(i,j)));
obj.Interp_weights(i,j,I)=1;
end
end
end
end
function obj=CalcWeightsMTI(obj)
if(sum(obj.B0map,'all')==0 ||isempty(obj.B0map))
obj.Interp_weights=1;
else
% h=histogram(obj.B0map,"NumBins",prod(0.5*obj.NUFFTOP.imSize));
% bins=h.BinEdges(h.BinCounts>0)+0.5*h.BinWidth;
[Ncount,edges] = histcounts(obj.B0map(obj.mask),prod(0.5*obj.imSize(1:2))*max(1,obj.imSize(3)));
bins=edges(Ncount>0)+0.5*diff(edges(1:2));
wn_tau_l=exp(1i*(bins(:)*(obj.B0Para.Levels(:).')));
if(strcmp(obj.B0Para.fitMode,'LeastSquares'))
actual=exp(1i*col(bins)*obj.tk');
obj.Interp_weights=mldivide(wn_tau_l'*wn_tau_l,wn_tau_l'*actual);
elseif(strcmp(obj.B0Para.fitMode,'NearestNeighbour') )
error('Not implemented')
end
end
end
end
end