To demonstrate robust myelin water fraction (MWF) mapping using an artificial neural network (ANN) with multi‐echo gradient‐echo (GRE) signal.
Methods
Multi‐echo gradient‐echo signals simulated with a three‐pool exponential model were used to generate the training data set for the ANN, which was designed to yield the MWF. We investigated the performance of our proposed ANN for various conditions using both numerical simulations and in vivo data. Simulations were conducted with various SNRs to investigate the performance of the ANN. In vivo data with high spatial resolutions were applied in the analyses, and results were compared with MWFs derived by the nonlinear least‐squares algorithm using a complex three‐pool exponential model.
Results
The network results for the simulations show high accuracies against noise compared with nonlinear least‐squares MWFs: RMS‐error value of 5.46 for the nonlinear least‐squares MWF and 3.56 for the ANN MWF at an SNR of 150 (relative gain = 34.80%). These effects were also found in the in vivo data, with reduced SDs in the region‐of‐interest analyses. These effects of the ANN demonstrate the feasibility of acquiring high‐resolution myelin water images.
Conclusion
The simulation results and in vivo data suggest that the ANN facilitates more robust MWF mapping in multi‐echo gradient‐echo sequences compared with the conventional nonlinear least‐squares method.
I had a question about Figure 2. I noticed that in the case with no averaging, there is elevated MWF on the left and right of the brain (sorry, I don’t know the correct anatomical name for that region) for both NNLS and ANN. In the case with 4x averaging, the ANN approach still shows the elevated MWF values, but they are actually diminished in the NNLS approach. Can you comment on why that might be?
It almost seems like the elevated MWF values are an artifact of noise that the ANN approach is learning, so I wonder if there is some way to avoid learning that effect. Or maybe there’s another explanation I’m not thinking of.
Dear Steven,
Thank you for your interest in our research.
As for Fig.2, your observation is indeed correct! We think that this (ie your observations of elevated MWF, which is probably the superior longitudinal fasciculus) is a limitation of our proposed ANN, which use the magnitude value of the signal. This is probably due to the fiber orientation with respect to B0. Magnitude valued analysis can have bias in MWF estimation as a function of fiber orientation. On the other hand, using complex value can be more robust with respect to orientation (ref. Nam NeuroImage 116 (2015) 214–221). The NNLS that we used in Fig. 2 actually uses complex-valued signal for analysis, and therefore shows more homogeneous MWF values.
So, as a result, it would seem that training the NN with complex-valued signal would be beneficial! This is what we have been working on since afterwards.
Hope this helped!
Dong-Hyun Kim.
Thank you for your response. I’m still a little confused, though, because even the NNLS results had elevated MWF in the case with no averaging, so to me that indicates a problem with SNR, not fiber orientation. Is there something I’m missing, or is there a reason you think noise is not the main issue?
Thanks for sharing your thoughts. We believe it is more of an issue of orientation. The NNLS itself is less sensitive to orientation but still has limitations. If it were an SNR, we think is should not necessarily follow specific anatomy. Of course, the SNR of that specific anatomy might be bad and hence result in increased or decreased MWF. If it were this case, your observation could be correct!