misc. to-dos

[arkiev]

• update doc with naming conventions for radiology community
• in group analysis html, plot mean + 95%CI. need to compute this in the T1/T2/ADC mapping part
• fix mri viewer in T1 and T2 mapping
• revert to png for contrast plots of individual ir and te plots
• currently only shows mean (SD). Go beyond mean values:
      • Report median (preferred for robustness to outliers).
      • Include variability metrics such as SD or MAD (recommended as a robust alternative). Include functionality to toggle between (and on/off) both Mean and Median Absolute Deviation
• Add SNR and CNR measurements alongside vial means and error bars.
      • create noise vial (ROI of same dimensions as vial, on same plane, outside phantom
• Get vial volumes to be equal (tall/short vials)
• tidy up outputs to be a single xls file per contrast containing mean, median, min, max and std
• verify T1 and T2 calculations
• currently toggling between mean and median doesn't update on measured vs reference ADC. fix this, check if it applies to T1/T2
• tidy up output DWI processing directory (remove directories only store intermediate files)

[BPaton]

• update doc with naming conventions for radiology community[ ] in group analysis html, plot mean + 95%CI. need to compute this in the T1/T2/ADC mapping part[x] fix mri viewer in T1 and T2 mapping[x] revert to png for contrast plots of individual ir and te plots[ ] currently only shows mean (SD). Go beyond mean values:
  • Report median (preferred for robustness to outliers).
  • Include variability metrics such as SD or MAD (recommended as a robust alternative). [ ] Add SNR and CNR measurements alongside vial means and error bars.[ ] Get vial volumes to be equal (tall/short vials)

For the SNR and CNR what methods and what voxels will you choose for different methods? This is important given ghosting and distortion present in some of the image contrasts/modalities?

[BPaton]

• update doc with naming conventions for radiology community[ ] in group analysis html, plot mean + 95%CI. need to compute this in the T1/T2/ADC mapping part[x] fix mri viewer in T1 and T2 mapping[x] revert to png for contrast plots of individual ir and te plots[ ] currently only shows mean (SD). Go beyond mean values:
  • Report median (preferred for robustness to outliers).
  • Include variability metrics such as SD or MAD (recommended as a robust alternative). [ ] Add SNR and CNR measurements alongside vial means and error bars.[ ] Get vial volumes to be equal (tall/short vials)

For the SNR and CNR what methods and what voxels will you choose for different methods? This is important given ghosting and distortion present in some of the image contrasts/modalities?

And for the errors bars +/- 2 * SE of the Mean would be ideal.

[arkiev]

For the SNR and CNR what methods and what voxels will you choose for different methods?

not sure just yet...was hoping to review this during the Thursday meeting :)

[arkiev]

@BPaton is this what you had in mind for MAD calculations within each vial?

for M in mean median; do
    for vial in A B C D...; do

        # Compute the mean or median intensity within the vial mask, store as scalar
        metric=$(mrstats input.mif -mask ${vial}.nii.gz -output $M)

        # Subtract that scalar from every voxel in the image, take absolute value
        mrcalc input.mif $metric -subtract -abs abs_dev.mif -force

        # Average the absolute deviations within the vial mask → scalar MAD
        mad=$(mrstats abs_dev.mif -mask ${vial}.nii.gz -output mean)

        # Print results for this vial and central tendency combination
        echo "Vial ${vial} | centre=${M} | MAD=${mad}"

    done  # end vial loop
done      # end M loop

[BPaton]

Won't this line:

# Average the absolute deviations within the vial mask → scalar MAD
mad=$(mrstats abs_dev.mif -mask ${vial}.nii.gz -output mean)</code>

Always result in the Mean Absolute Deviation? Rather than the Median Absolute Deviation?

[BPaton]

The Median Absolute Deviation (MAD) is defined for a dataset of X₁, X₂, X₃, X₄, ..., X_n and Median X is:

MAD = Median(|X_i - X|)

[BPaton]

The Median Absolute Deviation (MAD) is defined for a dataset of X1, X2, X3, X4, ..., Xn and Median X is:

MAD = (|Xi - X|)

So in other words, take all of the values, calculate the median for those values, subtract those individual values from the median (taking their absolute value). Then take the median of those deviation scores.

[arkiev]

So in other words, take all of the values, calculate the median for those values, subtract those individual values from the median (taking their absolute value). Then take the median of those deviation scores.

Nice catch, this has been fixed

[BPaton]

And if you are doing the mean absolute deviation then just substitute median for mean in those steps. The MAD is a slightly different measure of dispersion compared to the SD or even Variance as there is no normalisation or scaling applied based on the sample size.