Strategy to synthesize UNIT1 contrast from T1map images

[Nilser3]

Description

The nih-ms-mp2rage dataset contains UNIT1-denoised data and T1-maps (see #280 (comment) ), however for MS lesion segmentation UNIT1 non-denoised is classically used (and it is the contrast that we have for the other MS MP2RAGE datasets: marseille-3T-mp2rage and basel-mp2rage),
so a strategy is required to be able to synthesize UNIT1 non-denoised images from the T1-maps.

UNI-den

T1-map

Strategy:

• The synthesis of UNI images from T1maps was reported in the work of Massire et al., 2021
• For this synthesis, some MP2RAGE acquisition parameters are required such as : nbefore, nafter, alpha1, alpha2, TR, MP2RAGE-TR, TI1 and TI2. Parameters not available for the nih-ms-mp2rage data,
  but if I take the parameters reported by Demortière et al., 2020
  I obtain the following results:

Synthetic UNI from non-den T1-map

I believe that we would have more precise results if we have the MP2RAGE parameters from nih-ms-mp2rage data.

Related issues:

#280

[jcohenadad]

I'm not a fan of this two-step strategy-- what if the synthesis misses minor lesions? Why not train a model with den-UNI1 and non-denUNI1 first and see what you get?