Multi‐parametric artificial neural network fitting of phase‐cycled balanced steady‐state free precession data

Abstract

Purpose Standard relaxation time quantification using phase‐cycled balanced steady‐state free precession (bSSFP), eg, motion‐insensitive rapid configuration relaxometry (MIRACLE), is subject to a considerable underestimation of tissue T1 and T2 due to asymmetric intra‐voxel frequency distributions. In this work, an artificial neural network (ANN) fitting approach is proposed to simultaneously extract accurate reference relaxation times (T1, T2) and robust field map estimates ( B1+, ΔB0) from the bSSFP profile. Methods Whole‐brain bSSFP data acquired at 3T were used for the training of a feedforward ANN with N = 12, 6, and 4 phase‐cycles. The magnitude and phase of the Fourier transformed complex bSSFP frequency response served as input and the multi‐parametric reference set [T1, T2, B1+, ∆B0] as target. The ANN predicted relaxation times were validated against the target and MIRACLE. Results The ANN prediction of T1 and T2 for trained and untrained data agreed well with the reference, even for only four acquired phase‐cycles. In contrast, relaxometry based on 4‐point MIRACLE was prone to severe off‐resonance‐related artifacts. ANN predicted B1+ and ∆B0 maps showed the expected spatial inhomogeneity patterns in high agreement with the reference measurements for 12‐point, 6‐point, and 4‐point bSSFP phase‐cycling schemes. Conclusion ANNs show promise to provide accurate brain tissue T1 and T2 values as well as reliable field map estimates. Moreover, the bSSFP acquisition can be accelerated by reducing the number of phase‐cycles while still delivering robust T1, T2, B1+, and ∆B0 estimates.

Pure balanced steady‐state free precession imaging (pure bSSFP)

“…This effect was attributed to tissue microstructure leading to asymmetric frequency distributions which in turn lead to an asymmetric frequency response function 3 . A possible approach to remove this bias for a specific measurement setup was recently suggested using neural networks 18 . From a methods point‐of‐view, however, scanning is preferentially being performed in a regime where theory and measurements match and voxels forget about their intrinsic spectral composition rather than learning a specific bias away.…”

Constrained Ellipse Fitting for Efficient Parameter Mapping With Phase-Cycled bSSFP MRI

“…Finally, we explored the feasability of a learning-based correction for CELF parameter estimates. Artificial neural networks were recently demonstrated to reproduce non-bSSFP relaxometry maps from phase-cycled bSSFP signals [30]. Inspired by this recent method, we reasoned that neural network models can be trained to predict reference parameter maps given CELF-derived parameter estimates as input, and that the trained models can further improve accuracy as a postcorrection step to CELF.…”

“…CELF can also be leveraged to produce synthetic bSSFP images at varying flip angles, to improve tissue delineation at multiple distinct contrasts [15]. Lastly, the intrinsic microstructural sensitivity in CELF can be viewed as an additional contrast mechanism, and concurrent analysis of spin-echo and bSSFP-based parameter maps can allow estimation of tissue microstructural properties [30].…”

“…A promising approach based on standard bSSFP sequences is artificial neural networks for purely data-driven estimation of T 1 and T 2 maps. This learning-based method requires training data from a lengthy protocol with bSSFP and gold-standard mapping sequences; and it might require retraining for different protocols or pathologies [30]. In contrast, the model-based trueCISS method uses N = 16 acquisitions to compare the measured signal profile across phase cycles against a simulated dictionary of signal profiles [15].…”

Constrained Ellipse Fitting for Efficient Parameter Mapping with Phase-cycled bSSFP MRI

“…Finally, we explored the feasability of a learning-based correction for CELF parameter estimates. Artificial neural networks were recently demonstrated to reproduce non-bSSFP relaxometry maps from phase-cycled bSSFP signals [30]. Inspired by this recent method, we reasoned that neural network models can be trained to predict reference parameter maps given CELF-derived parameter estimates as input, and that the trained models can further improve accuracy as a post-correction step to CELF.…”