SingleIntensityAccumulationError: about medpy HOT 1 CLOSED

Dear potis,

the intensity standardization algorithm works by mapping the provided images to a common scale according to anchor points in their respective histograms. You provide a number of very small images (each on 12 values) with a rather hight number of landmark points (the anchor points).

The border anchor points are determined by your cutoff percentiles

numpy.array(numpy.percentile(good_trainingset[0], (1, 99)))

resulting in the anchor points

[ 0. , 8.89]

In between these two, nine further anchor points are set

numpy.array(numpy.percentile(good_trainingset[0], [10, 20, 30, 40, 50, 60, 70, 80, 90]))

which leads to

array([ 0. , 0.4, 2.3, 3.4, 4. , 4.6, 6.4, 7.8, 8. ])

As you can readily see, the left-most anchor point 0. and the subsequent one 0. fall onto the same value since good_trainingset[0] contains three zeros. Defining a linear connection between these would result in a slope of either zero or infinity, both of which are not a good idea. Therefore the error.

Using nearly as many anchor points as image pixels makes anyway not much sense. If you really must use such small images, look at the second part of error message (Another possibility would be to reduce the number of landmark percentiles landmarkp or to change their distribution.). This

irs = IntensityRangeStandardization(cutoffp=(1, 99), landmarkp=[25, 50, 75], stdrange='auto')
irs.train_transform(good_trainingset,surpress_mapping_check=True)

works, for example.

Regarding the background part of the message: MedPy is developed to work with medical images. These scans often have a large, mostly zero-valued background denoting air surrounding the body area scanned. This should be removed before the intensity range standardization. See here for an example.

I hope this helps you.

Best,
loli

from medpy.