DissectionPhotoVolumes / data /code /Stats /UWphoto_script_GLMmodeling.m
introvoyz041's picture
Migrated from GitHub
72417f2 verified
Raw
History Blame Contribute Delete
13.6 kB
% Script to generate GLMs with the matched case info and volume statistics
% from both MRI and photo segmentations
%% load data
PHOTO_RECON_HOME='/home/acasamitjana/Results';%getenv('PHOTO_RECON_HOME');
figuresDir = fullfile(PHOTO_RECON_HOME,'figures');
% load(fullfile(figuresDir,'AdjustedCaseStats.mat'),'segVolumeInfo',...
% 'matchedInfo')
savedir = fullfile(figuresDir,'GLM');
if ~exist(savedir,'dir')
mkdir(savedir)
end
%% sort out cases to remove
% removalLevel 1 - missing volumes and Hemmorhage removed
% removalLevel 2 - poorly segmented ventricles
% removalLevel 3 - poor segmentations and defformation
% removalLevel 4 - biasing ventricle volumes
removalLevel = 1;
removalFlag = (0<[segVolumeInfo.removaltype]) &...
([segVolumeInfo.removaltype] <=removalLevel);
segVolumeInfo_kept=segVolumeInfo(~removalFlag);
savepath = fullfile(savedir,['GLMresults_removalLevel_',...
num2str(removalLevel)]);
%% get target demographics
flag_mrs = strcmp({segVolumeInfo_kept.segtype},'MRI');
flag_hrd = strcmp({segVolumeInfo_kept.segtype},'Hard');
flag_sft = strcmp({segVolumeInfo_kept.segtype},'Soft');
flag_female_disease = strcmp(matchedInfo.sex,'Female') & ...
strcmp(matchedInfo.CognitiveStatus,'Dementia');
fd_demographics = matchedInfo(flag_female_disease,{'caseID','age'});
flag_male_disease = strcmp(matchedInfo.sex,'Male') & ...
strcmp(matchedInfo.CognitiveStatus,'Dementia');
md_demographics = matchedInfo(flag_male_disease,{'caseID','age'});
flag_male_control = strcmp(matchedInfo.sex,'Male') & ...
strcmp(matchedInfo.CognitiveStatus,'No dementia');
mc_demographics = matchedInfo(flag_male_control,{'caseID','age'});
%% make info structures
% these structures hold the label volumes, age and caseIDs for the selected
% cases that have not been excluded. They are split into separate
% structures for MRI, Hard reconstructions, Hard reconstructions with
% volume correction and Soft reconstruction.
%
% They are also split by cohort fd - female disease, md - male disease,
% mc - male control.
mri_strct_fd = struct();
mri_strct_fd(size(fd_demographics,1)) = struct();
mri_strct_md = struct();
mri_strct_md(size(md_demographics,1)) = struct();
mri_strct_mc = struct();
mri_strct_mc(size(mc_demographics,1)) = struct();
hrd_strct_fd = struct();
hrd_strct_fd(size(fd_demographics,1)) = struct();
hrd_strct_md = struct();
hrd_strct_md(size(md_demographics,1)) = struct();
hrd_strct_mc = struct();
hrd_strct_mc(size(mc_demographics,1)) = struct();
crct_strct_fd = struct();
crct_strct_fd(size(fd_demographics,1)) = struct();
crct_strct_md = struct();
crct_strct_md(size(md_demographics,1)) = struct();
crct_strct_mc = struct();
crct_strct_mc(size(mc_demographics,1)) = struct();
sft_strct_fd = struct();
sft_strct_fd(size(fd_demographics,1)) = struct();
sft_strct_md = struct();
sft_strct_md(size(md_demographics,1)) = struct();
sft_strct_mc = struct();
sft_strct_mc(size(mc_demographics,1)) = struct();
strfields = fieldnames(segVolumeInfo_kept);
%% get info fd
keep_flag = true(size(fd_demographics,1),1);
for il = 1:size(fd_demographics,1)
selection_flag = strcmp({segVolumeInfo_kept.caseID},...
fd_demographics.caseID(il));
if any(selection_flag & flag_mrs)
mri_strct_fd(il).age=fd_demographics.age(il);
hrd_strct_fd(il).age=fd_demographics.age(il);
crct_strct_fd(il).age=fd_demographics.age(il);
sft_strct_fd(il).age=fd_demographics.age(il);
for jl=1:length(strfields)
mri_strct_fd(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_mrs).(strfields{jl});
hrd_strct_fd(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl});
if jl<5
crct_strct_fd(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl});
else
crct_strct_fd(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl})...
.*segVolumeInfo_kept(selection_flag &...
flag_hrd).volumeAdjustmentFactor;
end
sft_strct_fd(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_sft).(strfields{jl});
end
else
keep_flag(il)=false;
end
end
mri_strct_fd=mri_strct_fd(keep_flag);
hrd_strct_fd=hrd_strct_fd(keep_flag);
crct_strct_fd=crct_strct_fd(keep_flag);
sft_strct_fd=sft_strct_fd(keep_flag);
%% get info md
keep_flag = true(size(md_demographics,1),1);
for il = 1:size(md_demographics,1)
selection_flag = strcmp({segVolumeInfo_kept.caseID},...
md_demographics.caseID(il));
if any(selection_flag & flag_mrs)
mri_strct_md(il).age=md_demographics.age(il);
hrd_strct_md(il).age=md_demographics.age(il);
sft_strct_md(il).age=md_demographics.age(il);
for jl=1:length(strfields)
mri_strct_md(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_mrs).(strfields{jl});
hrd_strct_md(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl});
if jl<5
crct_strct_md(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl});
else
crct_strct_md(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl})...
.*segVolumeInfo_kept(selection_flag &...
flag_hrd).volumeAdjustmentFactor;
end
sft_strct_md(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_sft).(strfields{jl});
end
else
keep_flag(il)=false;
end
end
mri_strct_md=mri_strct_md(keep_flag);
hrd_strct_md=hrd_strct_md(keep_flag);
crct_strct_md=crct_strct_md(keep_flag);
sft_strct_md=sft_strct_md(keep_flag);
%% get info mc
keep_flag = true(size(mc_demographics,1),1);
for il = 1:size(mc_demographics,1)
selection_flag = strcmp({segVolumeInfo_kept.caseID},...
mc_demographics.caseID(il));
if any(selection_flag & flag_mrs)
mri_strct_mc(il).age=mc_demographics.age(il);
hrd_strct_mc(il).age=mc_demographics.age(il);
sft_strct_mc(il).age=mc_demographics.age(il);
for jl=1:length(strfields)
mri_strct_mc(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_mrs).(strfields{jl});
hrd_strct_mc(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl});
if jl<5
crct_strct_mc(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl});
else
crct_strct_mc(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_hrd).(strfields{jl})...
.*segVolumeInfo_kept(selection_flag &...
flag_hrd).volumeAdjustmentFactor;
end
sft_strct_mc(il).(strfields{jl})=segVolumeInfo_kept(...
selection_flag & flag_sft).(strfields{jl});
end
else
keep_flag(il)=false;
end
end
mri_strct_mc=mri_strct_mc(keep_flag);
hrd_strct_mc=hrd_strct_mc(keep_flag);
crct_strct_mc=crct_strct_mc(keep_flag);
sft_strct_mc=sft_strct_mc(keep_flag);
%% build GLM design matrix
% columns correspond to age, gender and disease respectively. For gender 1
% corresponds to female 0 to male. For disease 1 is dementia 0 is control.
% There is also a column of ones to account for the constant.
% numbers in each cohort
% N_fd - female diseased
N_fd = length(mri_strct_fd);
% N_md - male diseased
N_md = length(mri_strct_md);
% N_mc - male control
N_mc = length(mri_strct_mc);
N_total = N_fd+N_md+N_mc;
% predictors will be age, gender and disease, with a constant term
Design_mat = zeros(N_total,4);
Design_mat(:,4)=1;
% females with disease
for il= 1:N_fd
Design_mat(il,1) = mri_strct_fd(il).age;
Design_mat(il,2) = 1;
Design_mat(il,3) = 1;
end
% males with disease
for il= 1:N_md
Ientry = il+N_fd;
Design_mat(Ientry,1) = mri_strct_md(il).age;
Design_mat(Ientry,2) = 0;
Design_mat(Ientry,3) = 1;
end
% males without disease
for il= 1:N_mc
Ientry = il+N_fd+N_md;
Design_mat(Ientry,1) = mri_strct_mc(il).age;
Design_mat(Ientry,2) = 0;
Design_mat(Ientry,3) = 0;
end
% remove column for gender due to lack of female controls
Design_mat_maleOnly=Design_mat(N_fd+1:end,[1,3,4]);
% Orthogonal designmat
Design_mat = gsog(Design_mat);
% Orthogonal designmat_male
Design_mat_maleOnly = gsog(Design_mat_maleOnly);
%% run model
% not going to use volume corrections at the moment because I'm not sure
% they are working well.
volume_fields = fieldnames(mri_strct_fd);
volume_flag = ~ismember(volume_fields,{'age','caseID','segtype',...
'volumeAdjustmentFactor','removaltype','removalnotes'}');
volume_fields = volume_fields(volume_flag);
MRI_GLM_resultsStruct.N_femaleDisease = N_fd;
MRI_GLM_resultsStruct.N_maleDisease = N_md;
MRI_GLM_resultsStruct.N_maleControl = N_mc;
MRI_GLM_resultsStruct.N_total = N_total;
Hrd_GLM_resultsStruct.N_femaleDisease = N_fd;
Hrd_GLM_resultsStruct.N_maleDisease = N_md;
Hrd_GLM_resultsStruct.N_maleControl = N_mc;
Hrd_GLM_resultsStruct.N_total = N_total;
Crct_GLM_resultsStruct.N_femaleDisease = N_fd;
Crct_GLM_resultsStruct.N_maleDisease = N_md;
Crct_GLM_resultsStruct.N_maleControl = N_mc;
Crct_GLM_resultsStruct.N_total = N_total;
Sft_GLM_resultsStruct.N_femaleDisease = N_fd;
Sft_GLM_resultsStruct.N_maleDisease = N_md;
Sft_GLM_resultsStruct.N_maleControl = N_mc;
Sft_GLM_resultsStruct.N_total = N_total;
for il=1:length(volume_fields)
%% MRI GLM
Observations = [mri_strct_fd(:).(volume_fields{il}),...
mri_strct_md(:).(volume_fields{il}),...
mri_strct_mc(:).(volume_fields{il})]';
[b,dev,stats] = glmfit(Design_mat,Observations,'normal','constant',...
'off');
MRI_GLM_resultsStruct.full.(volume_fields{il}).b = b;
MRI_GLM_resultsStruct.full.(volume_fields{il}).dev = dev;
MRI_GLM_resultsStruct.full.(volume_fields{il}).stats = stats;
[b,dev,stats] = glmfit(Design_mat_maleOnly,Observations(N_fd+1:end),...
'normal','constant','off');
MRI_GLM_resultsStruct.male.(volume_fields{il}).b = b;
MRI_GLM_resultsStruct.male.(volume_fields{il}).dev = dev;
MRI_GLM_resultsStruct.male.(volume_fields{il}).stats = stats;
%% Hard GLM
Observations = [hrd_strct_fd(:).(volume_fields{il}),...
hrd_strct_md(:).(volume_fields{il}),...
hrd_strct_mc(:).(volume_fields{il})]';
[b,dev,stats] = glmfit(Design_mat,Observations,'normal','constant',...
'off');
Hrd_GLM_resultsStruct.full.(volume_fields{il}).b = b;
Hrd_GLM_resultsStruct.full.(volume_fields{il}).dev = dev;
Hrd_GLM_resultsStruct.full.(volume_fields{il}).stats = stats;
[b,dev,stats] = glmfit(Design_mat_maleOnly,Observations(N_fd+1:end),...
'normal','constant','off');
Hrd_GLM_resultsStruct.male.(volume_fields{il}).b = b;
Hrd_GLM_resultsStruct.male.(volume_fields{il}).dev = dev;
Hrd_GLM_resultsStruct.male.(volume_fields{il}).stats = stats;
%% volume corrected Hard GLM
Observations = [crct_strct_fd(:).(volume_fields{il}),...
crct_strct_md(:).(volume_fields{il}),...
crct_strct_mc(:).(volume_fields{il})]';
[b,dev,stats] = glmfit(Design_mat,Observations,'normal','constant',...
'off');
Crct_GLM_resultsStruct.full.(volume_fields{il}).b = b;
Crct_GLM_resultsStruct.full.(volume_fields{il}).dev = dev;
Crct_GLM_resultsStruct.full.(volume_fields{il}).stats = stats;
[b,dev,stats] = glmfit(Design_mat_maleOnly,Observations(N_fd+1:end),...
'normal','constant','off');
Crct_GLM_resultsStruct.male.(volume_fields{il}).b = b;
Crct_GLM_resultsStruct.male.(volume_fields{il}).dev = dev;
Crct_GLM_resultsStruct.male.(volume_fields{il}).stats = stats;
%% Soft GLM
Observations = [sft_strct_fd(:).(volume_fields{il}),...
sft_strct_md(:).(volume_fields{il}),...
sft_strct_mc(:).(volume_fields{il})]';
[b,dev,stats] = glmfit(Design_mat,Observations,'normal','constant',...
'off');
Sft_GLM_resultsStruct.full.(volume_fields{il}).b = b;
Sft_GLM_resultsStruct.full.(volume_fields{il}).dev = dev;
Sft_GLM_resultsStruct.full.(volume_fields{il}).stats = stats;
[b,dev,stats] = glmfit(Design_mat_maleOnly,Observations(N_fd+1:end),...
'normal','constant','off');
Sft_GLM_resultsStruct.male.(volume_fields{il}).b = b;
Sft_GLM_resultsStruct.male.(volume_fields{il}).dev = dev;
Sft_GLM_resultsStruct.male.(volume_fields{il}).stats = stats;
end
%% Save results
save(savepath,'segVolumeInfo_kept','*_strct_*','Design_mat*','*resultsStruct')