Susceptibility values

[bhav...@gmail.com]

Hi Pascal/Alexey,

We ran a few patients with Parkinson's disease (MEDI+0) and have ROI values. We see mean susceptibility values both in positive and negative ranges for the same ROI. My understanding is that these numbers are relative to the CSF, and are not absolute. 

For example, with Putamen, we generally anticipate that these numbers would be positive for most of the patients with PD. We see a range of values, are concerned and not sure how to interpret these values. 

1) Do you suggest - that we repeat anything with regard to QSM processing, in patients with negative mean values?

2) Also, in the ROI analysis, do you include negative values for calculation of the mean?

Thanks,

--Bhavani

[drluk...@gmail.com]

Dear MEDI_Users,

I've run into the same issue. I get a lot of negative values in the putamen in all of my QSM images.

I'm wondering if it's to do with the CSF masks that are generated off of my R2* maps. I've included the CSF masks from a participant with small ventricles and one with large ones. There are quite a few gaps in the mask where high susceptibility values sit.

I've run the standard pipeline both with and without the CSF mask one 10 test-retest scans that I have and I get a very good ICC of 0.95 for the striatum if I don't include the CSF masks when generating the QSM image and a poor ICC of 0.6 if I include them.

I'm wondering if there is anything I'm doing wrong and if not what is the best way to apply referencing after the image generation, or if it's possible to just use the absolute values.

Below is the code I'm using the generate the QSM images. I just remove the R2* and CSF steps then remove lambda_CSF param from MEDI_L1 when I want to make the.

Many thanks for your advice,

Luke Vano

King's College London

 % Use accelerated (for Siemens and GE only) reading of DICOMs
        [iField,voxel_size,matrix_size,CF,delta_TE,TE,B0_dir,files]=Read_DICOM(qsm_dir);
       
        % Only use first 6 echos only
        iField = iField(:,:,:,1:6);
        TE = TE(1:6);
        iField1 = iField(:,:,:,1);
        iField6 = iField(:,:,:,6);
       
        % Estimate the frequency offset in each of the voxel using a complex
        % fitting (even echo spacing)
        [iFreq_raw, N_std] = Fit_ppm_complex(iField);
        %[iFreq_raw N_std] = Fit_ppm_complex_TE(iField,TE);

        % Compute magnitude images for the first and last echo
        iMag = sqrt(sum(abs(iField1).^2,4));
        iMag6 = sqrt(sum(abs(iField6).^2,4));

        % Spatial phase unwrapping (region-growing)- Use the 6th mag for the unwrap
        iFreq = unwrapPhase(iMag6, iFreq_raw, matrix_size);

        % Use FSL BET to extract brain mask- Use the 6th mag for generating the mask
        Mask = BET(iMag6,matrix_size,voxel_size);

        % R2* map needed for ventricular CSF mask
        R2s = arlo(TE, abs(iField));

        % Ventricular CSF mask for zero referencing
        %    Requirement:
        %        R2s:    R2* map
        Mask_CSF = extract_CSF(R2s, Mask, voxel_size);
       
        % Background field removasl using Projection onto Dipole Fields
        RDF = PDF(iFreq, N_std, Mask,matrix_size,voxel_size, B0_dir);
       
        cd(working_dir)

        save RDF.mat RDF iFreq iFreq_raw iMag N_std Mask matrix_size...
             voxel_size delta_TE CF B0_dir Mask_CSF;

        % Morphology enabled dipole inversion with zero reference using CSF (MEDI+0)
        QSM = MEDI_L1('lambda',1000,'lambda_CSF',100,'merit','smv',5);

[drluk...@gmail.com]

Dear MEDI_users,

I'm just responding again to see if anyone has been able to look into this.

Best wishes,

Luke