Hello,
I am using SPM5 to generate Grey, White & CSF segmented volumes. The problem
is that in all of these files there are a few pixels which have a value of NAN.
The fundamental problem here is that some of the float values at the end of
these volumes are not valid numeric representations for a 32-bit float
value, which Windows does not like at all. When I examined the data as
floats, the following occurs towards the end of the volume.
15476220 0.000000e+00 0.000000e+00 nan nan
15476240 nan nan nan nan
*
15502760 nan nan 0.000000e+00 0.000000e+00
15503000 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
*
15503120 nan nan nan nan
*
15503440 nan 0.000000e+00 0.000000e+00 0.000000e+00
15503460 0.000000e+00 0.000000e+00 0.000000e+00 0.000000e+00
*
15503760 0.000000e+00 0.000000e+00 nan nan
15504000 nan nan nan nan
*
The nan result comes from 32-bit words that are not a valid numeric
representation - in fact they have the specific bit pattern for NaN - Not a
Number. By the IEEE 754 floating point standard, NaNs are represented with
the exponent field filled with ones and bit 22 set in the mantissa:
NaN bit pattern (without sign bit) = 1111 1111 axxxxxxxxxxxxxxxxxxxxxx
Having said all this is there a way in SPM to eliminate, filter or restrict
the volumes created so that the NAN values do not end up in the volumes but
are replaces by some arbitrary or fixed numeric value.
Thanks,
BA
|