DOI: 10.1002/mrm.70443 ISSN: 0740-3194

T1 ‐weighting in Steady‐State FLASH MRI –Diffusion Is Not Only Supportive but Mandatory for the Contrast

Simon Weinmüller, Deepak Charles Chellapandian, Jonathan Endres, Jochen Leupold, Florian Gritsch, Fabian Wagner, Rainer Schneider, Moritz Zaiss

ABSTRACT

Purpose

FLASH imaging is widely assumed to produce a T 1 ‐weighted steady‐state contrast using RF‐ and gradient‐spoiling. We observed substantial overestimation of CSF signals in simulations, when diffusion was neglected and realistic proton density was applied. This work investigates the role of diffusion in steady‐state FLASH contrast formation and its implications for simulation‐based modeling and measurement.

Methods

FLASH sequences were simulated using phase graph simulations using a synthetic brain phantom with and without diffusion and realistic PD values to show the contrast change. The impact of neglecting diffusion in synthetic training data was evaluated using a segmentation network trained on simulated data and tested on in vivo measurement. Experimental validation of the contrast change was performed using a 3D‐printed brain phantom using silicone oil as a low‐diffusivity compartment.

Results

Without diffusion, simulations showed CSF signal intensities higher than WM, resulting in a contrast change. Diffusion suppresses higher‐order echoes in long T 2 tissues and is essential for achieving the T 1 ‐weighted steady‐state contrast. A NN trained without diffusion fails to generalize to in vivo data and measurements with silicone oil compartments confirm the contrast changes in low‐diffusivity media.

Conclusion

Diffusion is essential for realistic FLASH simulations of long T 2 tissues such as CSF. Steady‐state FLASH contrast arises from the interplay of RF‐, gradient‐spoiling, and “multi‐TR‐relaxation‐spoiling” governed by T 2 ‐decay and diffusion effects. For many quadratic phase cycling schemes, diffusion is required to obtain realistic T 1 ‐weighted contrast in MR simulations and should not be neglected in simulations or simulation‐based deep learning applications.

More from our Archive