Quality measure relResA in Osprey

Hi all,
We are running a data set through Osprey obtained with the PRESS sequence. We are considering reporting the (amongst other) relative residuals (relResA variable) as a quality measure of the data. However, we are struggling a bit with its interpretation. The definition given by Osprey is ‘Fit quality number for spectrum A relative amplitude of the residual compared to the standard deviation of the noise’.

But I was wondering what spectrum A refers to as we used an unedited sequence, does anyone know this? Also, I was wondering whether there was a cut-off value for relResA of ‘good’ or ‘bad’ data (or any direct interpretation of the value).

Thanks in advance!

Hi @Ilke,

Apologies for the unclear explanation. This value is calculated by comparing the relative amplitude of the residual to the standard deviation of the noise. The exact procedure is performed in fit/code/osp_fit_Quality.m.

Regarding the label “Spectrum A”: these labels are used in osprey to differentiate between the individual (and combinations of) sub-spectra in editing experiments. In your case, this is not needed and it simply refers to your non-edited acquisition.

Regarding your second question, I’m not aware of a firm ‘good/bad’ cutoff value, but in the original description of the method they do provide some assistance in interpretation:

This measure is close to 1 for a close to perfect fit. Values of Q significantly larger than 1 indicate significant differences between the fit and the experimental data which could indicate an inappropriate data model or a local minimum in the sum of least squares. On the other hand, Q values smaller than 1 strongly indicate overfitting of the model to the data.


1 Like

Thank you Chris for your quick and clear response!

1 Like