Hi all,
First of all, best wishes for the new year.
I understand that dwiintensitynorm scales all b0s and dwi’s of a certain dataset using the same scaling factor. Would it be acceptable to use in some cases a separate factor for b0s and dwi’s? I give an example:
For a certain study I have data from two (similar) protocols. They both acquire 10 x b0, 25 x b700, 40 x b1000 and 75 x b2800. The difference is the way the data is bundled in scan sequences. For the first half of the study each b-shell was acquired in a separate sequence, each time with one b0. A separate sequence with 7 b0 was added. After that, the scanner underwent an update (hate it when that happens during a study). During the second half of the study, the data was acquired in two sequences. The first had 5 b0s, 25 x b700 and 40 x b1000. The other one had 5 b0s and 75 x b2800.
Before I start any analysis I want to make sure the (slight) difference in acquisition has no influence on outcome parameters. Now I noticed that median white matter FA values from the ‘new’ protocol were significantly lower compared to results from the ‘old’ protocol (0.35 compared to 0.40). As the effects I’m looking for are assumed small, such bias is too large, so something needs to be done.
I also noticed that the ratio of b0 intensity over dwi intensity of each b-value differs between protocols. To give an example: for the b2800 data the b0/dwi ratio in the ‘old’ protocol was about 40, for the ‘new’ protocol it was about 55.
Initial attempts to work away the bias without altering the b0/dwi ratio (see also this topic), failed, so I think this ratio is the cause of the bias.
Therefore, in short my approach now is to calibrate b0 intensities (similar to dwiintensitynorm) to a common value, but instead of preserving the b0/dwi ratio, I make sure each dataset obtains the same b-value specific ratio.
I derived the ‘ground truth’ b0 intensity and b0/dwi ratio from the new protocol data alone: I computed median b0 intensity in white matter and the average across subjects. This is the target b0 value. Still from the new protocol data alone I computed the b0/dwi ratio of each b-value. These are the target ratios.
Then, all datasets (new and old protocol data) underwent scaling: each b0 was scaled (individually) towards the target b0 value. The dwi’s were scaled (combined per dataset, not individually) to meet the target b0/dwi ratios.
Using this approach, the median white matter FA values of new and old protocol data did not differ significantly anymore, and so the effect of the protocol was worked away. However, I’ld like to have a second opinion, on whether this approach is acceptable.
Looking forward to any opinions on this.
Cheers,
Thibo