Quantitative MRI mOOCT1 Mapping · Exercise 6
A collaborator’s strange T1 values
You receive an email from a collaborator in your city asking for help with a T1 dataset they acquired on their new 3T scanner. They tried fitting the data with qMRLab already, but are getting T1 values in the brain that are beyond the expected values (eg, WM: 5000 ms, GM: 5000 ms). They send you their acquired data.
- a. Fit a T1 map using their data and plot a few voxel signal curves.
- b. Luckily, you own a standardized quantitative MRI NIST phantom with known T1 values, and are able to go to your collaborators scanner to acquire a dataset using their protocol. Here is data you acquired for a region of the phantom with a known T1 value (987 ms)* close to the expected WM values. Plot that data, and simulate the expected signal curve for this T1 value
- c. Comparing the human and phantom data and the simulated curves, what do you notice? What do you think may be the problem?
- d. Test your proposed problem with some of your simulation code.
- e. What would you propose to your collaborator? Is a new data acquisition protocol required, or can you provide some post-processing corrections prior to fitting the data?
* Data: ISMRM 2020 T1 mapping challenge, one brain and one NIST phantom scanned on the same GE 3T (TR 2550 ms), resampled to 10 TIs from 50 to 2500 ms. Fits run live in this page with qmrust.
a. Fit their data
a. Fit a T1 map using their data.
Every voxel comes out at 5000 ms, the upper limit of the fit, WM and GM alike. Expected at 3T: WM ≈ 0.9 s, GM ≈ 1.5 s (Figure 2.2). The fit has failed everywhere.
a. Voxel signal curves
a. … and plot a few voxel signal curves.
The fits miss the data: the signal falls from the shortest TI, dips, then rises again. No inversion recovery curve does that. To find out why, we need data whose T1 we know.
b. The NIST phantom, same scanner and protocol
b. Plot the data from a phantom region with a known T1, and simulate the expected signal curve for this T1.
c. What do you notice?
c. Comparing the human and phantom data and the simulated curves, what do you notice? What do you think may be the problem?
The data look like the simulation read backwards: the images may be stored in the reverse order of the TIs in the header.
d. Test the hypothesis
d. Test your proposed problem with some of your simulation code.
e. What to tell the collaborator
e. Is a new data acquisition protocol required, or can you provide some post-processing corrections prior to fitting the data?
No new acquisition. Reorder the images by the TI stored in each image’s own header, then refit. Check the result on the phantom.
Exercise 6: answers
| Part | Answer |
|---|---|
| a | Every voxel fits to 5000 ms (the fit's upper bound); the fitted curves miss the data, which fall, dip and rise again |
| b | The phantom sphere (987 ms) does not follow the simulated curve for its known T1 |
| c | The data look like the simulation mirrored left to right: images in reverse TI order |
| d | Reversing the simulated order matches the data, and refitting the reordered phantom recovers the NIST values |
| e | No new acquisition: reorder the images by TI and refit (WM ≈ 0.85–0.92 s, GM ≈ 1.25–1.39 s) |
Data: ISMRM 2020 T1 mapping challenge (osf.io/ywc9g), USC, GE 3T, resampled to 10 TIs. Fits: qmrust, running in the browser.