mrirecon / bloch-moba Goto Github PK

Nice work and nice presentation at Sedona!

I was wondering if convergence is a concern for this QMRI method?

My elementary understanding of the optimization problem in Eq. (19) (pdf version v1: https://arxiv.org/abs/2209.08027v1)
makes me think the stationary solution to it may not be unique.

If so, would you comment on what tricks you'd use to avoid getting unrealistic results from solving the optimization?

(Sorry posting this as an issue, feel free to close it at any time. This is probably more of a discussion, but the discussion section is not set up for this repo...)

To hopefully rationalize my concern, I thought that quantitative imaging in general has a focus on achieving a numerical accuracy.
Conventional MRF pretty much does a thorough search to pin down the values, while iterative solutions of non-convex problems may be caught into wrong local minima.