Thank you for the response, I certainly value the input.
I would agree that it is a much better idea to correct the FLAIR using the
T2. However, after multiplying the FLAIR and the T1 together, hyperintense
clusters are represented on both inputs and so are overlapping. Of course,
this means that other properties of the FLAIR, including the bias field, are
now represented in both inputs to mfast. And I would guess that multiplying
two images would invalidate certain assumptions about certain properties of
the images, like possibly the noise distribution for instance. And these
were the kinds of concerns that I was thinking would leave this approach
fundamentally flawed.
But, on the other hand, our only reason for correcting the FLAIR image is
for the identification of white matter hyperintensities. And the
hyperintensity segmentation results are nearly identical when I use the
FLAIR + T2 or FLAIR + FLAIR*T1 correction approaches. So I guess I was
hoping to use this approach simply on the justification that it worked.
However, if it is genuinely a bad idea then we will need to consider using
the T2.
|