diff --git a/cat_main_registration.m b/cat_main_registration.m index 4fcf8d00..c63193d2 100644 --- a/cat_main_registration.m +++ b/cat_main_registration.m @@ -420,7 +420,7 @@ % Adaption to avoid boundary effects by correcting the voxel close % to the image boundary that are effected by the interpolation of % the field. - msk = cat_vol_morph( isnan( yi2(:,:,:,1)) ,'d',3); w( msk ) = NaN; + msk = cat_vol_morph( isnan( yi2(:,:,:,1)) ,'d',3); w( msk(:) ) = NaN; wa = cat_vol_approx(w,'nn',1,4); bg = cat_stat_nanmean( wa(msk(:)) ); msk = cat_vol_smooth3X(msk,2); w( isnan(w) ) = wa( isnan(w) ); w = wa .* msk + (1-msk) .* w; clear msk wa; @@ -497,7 +497,7 @@ % Adaption to avoid boundary effects by correcting the voxel close % to the image boundary that are effected by the interpolation of % the field. - msk = cat_vol_morph( isnan( yi2(:,:,:,1)) ,'d',3); w( msk ) = NaN; + msk = cat_vol_morph( isnan( yi2(:,:,:,1)) ,'d',3); w( msk(:) ) = NaN; wa = cat_vol_approx(w,'nn',1,4); bg = cat_stat_nanmean( wa(msk(:)) ); msk = cat_vol_smooth3X(msk,2); w( isnan(w) ) = wa( isnan(w) ); w = wa .* msk + (1-msk) .* w; clear msk wa; @@ -765,8 +765,8 @@ if job.extopts.verb && ti==1 %cat_io_cprintf([0 0 1], ' == Very large ventricles detected. Use seperate class and adapt schedule. == \n'); cat_io_addwarning([mfilename ':Hydrocephalus'],sprintf( ... - 'Very large ventricles detected (%0.2f%%%% of TIV). Use seperate class and adapt schedule. \\\\n', ... - sum(Yven(:))/sum(Yb(:))>0.1),1,[0 1]); + 'Very large ventricles detected (%0.2f%%%%%%%% of TIV). \\\\nUse seperate class and adapt schedule.', ... + 100 * sum(Yven(:))/sum(Yb(:)>.1)),1,[0 1]); end @@ -1046,32 +1046,37 @@ % thus we use the def2det function of the inverted deformations to obtain the old and % in my view a more appropriate jacobian determinant % The 2nd reason to use the old modulation is compatibility with cat_vol_defs.m - if 1 - yi = spm_diffeo('invdef', y , idim, Maffinerigid * inv(M1r\M1), inv(M1r\M1)); %#ok - yi2 = spm_diffeo('invdef', yi , odim, eye(4), eye(4)); - else - yi = spm_diffeo('invdef', y , idim, Maffinerigid * inv(M1r(1)\M1(1)), inv(M1r\M1) ); %#ok - yi2 = spm_diffeo('invdef', yi , odim, eye(4), eye(4)); - end - w = max( eps , abs(spm_diffeo('def2det', yi2 ) ) ); + + % RD20240924: first only for the jacobian without affine parameters / TIV as covariate + yi = spm_diffeo('invdef', y , idim, Maffinerigid * inv(M1r\M1), inv(M1r\M1)); + yi2 = spm_diffeo('invdef', yi , odim, eye(4), eye(4) / Maffinerigid); + w2 = max( eps , abs(spm_diffeo('def2det', yi2 ) ) ); + + % now for the final transformations + yi = spm_diffeo('invdef', y , idim, Maffinerigid * inv(M1r\M1), inv(M1r\M1)); %#ok + yi2 = spm_diffeo('invdef', yi , odim, eye(4), eye(4)); + % RD20240924: This would be the real jacobian that considers the also the affine registration + % in the "correct" but unexpected way and that would finally need TIV as covariate. + w = max( eps , abs(spm_diffeo('def2det', yi2 ) ) ); + % avoid boundary effects that are not good for the global measurements % use half registration resolution to define the amout of smoothing to reduce registration artefacts if 1 + %% vxd = M1(1)./M1r(1); - %msk = cat_vol_morph( isnan( yi2(:,:,:,1)) ,'d',2/vxd); w( msk ) = NaN; - msk = true(size(w)); msk(3:end-2,3:end-2,3:end-2) = false; w( msk ) = NaN; - wa = cat_vol_approx(w,'nn',1,4); bg = cat_stat_nanmean( wa(msk(:)) ); + msk = w<=eps; w( msk(:) ) = NaN; + wa = cat_vol_approx(w,'rec'); bg = cat_stat_nanmean( wa(msk(:)) ); msk = cat_vol_smooth3X(msk,1/vxd); w( isnan(w) ) = wa( isnan(w) ); w = wa .* msk + (1-msk) .* w; clear msk wa; fs = newres / 2; w = w - bg; spm_smooth(w,w,fs); w = bg + w; % spm smoothing boudary problem - end - - % yi = spm_diffeo('invdef', y , idim, Maffinerigid * inv(M1r\M1t), inv(M1r\M1)); %#ok - % yi2 = spm_diffeo('invdef', yi , odim, eye(4), eye(4)); - - - + + msk = w2<=eps; w2( msk(:) ) = NaN; + wa = cat_vol_approx(w2,'rec'); bg = cat_stat_nanmean( wa(msk(:)) ); + msk = cat_vol_smooth3X(msk,1/vxd); w2( isnan(w2) ) = wa( isnan(w2) ); + w2 = wa .* msk + (1-msk) .* w2; clear msk wa; + fs = newres / 2; w2 = w2 - bg; spm_smooth(w2,w2,fs); w2 = bg + w2; % spm smoothing boudary problem + end % yi2 for fast high quality output trans.warped = struct('y',yi,'yi',yi2,'w',w,'odim',odim,'M0',M0,'M1',M1,'M2',inv(Maffinerigid),'dartel',2,'fs',fs); @@ -1085,7 +1090,7 @@ end end uo(~isfinite(uo)) = eps; - trans.jc = struct('u',uo*0,'odim',odim,'dt2',w); + trans.jc = struct('u',uo,'odim',odim,'dt2',w2); end