Explanations on the functioning and issues regarding the dwi-dti pipeline

[MatthieuJoulot]

This discussion's purpose is to clarify the dwi-dti pipeline and the problems revolving around it.

1. We use MRtrix interfaces to generate the dti model and FA, AD, RD, MD and DECFA images. For this to work we start by converting the bval and bvec from fsl format to MRtrix using the FSL2MRtrix command. We then produce the tensors using DWI2Tensor and, based on the tensor we can produce the different images by using TensorMetrics.
2. The images in step "1." are then registered to MNI except the DECFA using ANTs's RegistrationSynQuick and ApplyTransform
3. The registered images in step "2." are thresholded and outputed using FSL's Threshold
4. The thresholded images in step "3." are then used to produce statistics which are outputed.
5. The images produced by the step "1." are thresholded with FSL's Threshold and outputed

You can see the graph of the pipeline below.

A few potential problems come from this process.

1. The DECFA outputed by the fifth step is asymetric, whereas the DECFA from first step is not. The threshold seems to be responsible for this. It may have had a wrong effect on the other images as well. Also, since the thresholding from the third and the fifth step are the same, the numerical values produced may be wrong.

This is the DECFA image after the thresholding.

This is the DECFA image before the thresholding.

2. The reasons for these thresholdings is that MRtrix's DWI2Tensor can produce negative values. This issue is detailed here and here.

3. We can question the general compatibility of MRtrix and FSL as it is discussed here and here

These issues make it difficult to trust the result outputed by the pipeline.

[MatthieuJoulot]

The simplest solution seems to be change and to use only FSL's dtifit instead of the combination of MRtrix commands we currently have, which apparently doesn't generate negative values.

That way, we can discard different steps since everything is handled by dtifit. Plus the question of the compatibility between MRtrix and FSL won't be anymore.

We can also go the other way around, and change FSL's threshold for MRtrix's. That way we also discard the compatibility problem. However, the compatibility problem is only partially discarded since the inputs are FSL compatible. Plus the problem of the negative values is not handled if we use this solution.

Hence the most likely evolution right now is that the dwi-dti pipeline will soon be changed to use FSL only instead of a combination of. MRtrix and FSL.