Enforce constant number of bins(?)

[Daniel Gutmann]

Hey All,

I have previously been used to texture extraction tools that allow to specify the number of bins of the co-occurrence matrix, eg = 128.

In my current scenario Im work on a large MRI tissue dataset where I would like to normalize my data using pyradiomics imageoperations,

however how can I ensure I get the exact same number of bins for each image, if after intensity normalisation the min,max values 

between images vary widely, while the bin_width stays constant? Would a non-constant number of bins not affect the comparability of features

between images? Seems I am missing something!

What am I missing here? 

Daniel

[Joost van Griethuysen]

Hi Daniel,

PyRadiomics does not have the option for setting a fixed bin count, as a fixed bin count makes the values less comparable, instead of more. This is because a fixed bin count means that the “meaning” of difference between gray values is dependent on the range of gray values in the ROI. Take for example 2 images with 2 ROIs, with the range of gray values in the first being {0-100} and in the second {0-10}. If you use a fixed bin count, the “meaning” of 1 gray value difference is different (in the first it means 10 gray values different, in the second just 1). This means you are looking at texture based on very different contrasts.

Therefore, PyRadiomics uses a fixed bin width (parameter “binWidth”), which ensures texture feature values are calculated using the same “contrast” between gray values.[1] There are currently no specific guidelines as to what constitutes an optimal bin width. We try to choose a bin width in such a way, that the resulting amount of bins is somewhere between 30 and 130 bins. This allows for differing ranges of intensity in ROIs, while still keeping the texture features informative (and comparable inter lesion!).

 

[1] Leijenaar RTH, Nalbantov G, Carvalho S, et al.; The effect of SUV discretization in quantitative FDG-PET Radiomics: the need for standardized methodology in tumor texture analysis; Sci Rep. 2015;5(August):11075

Aside from the above reference, these are also some interesting reads on bin count / bin width usage.

- Tixier F, Hatt M, Cheze-Le Rest C, Le Pogam A, Corcos L, Visvikis D: Reproducibility of Tumor Uptake Heterogeneity Characterization Through Textural Feature Analysis in 18F-FDG PET. J Nucl Med 2012; 53:693–700.

- Desseroit M-C, Tixier F, Weber WA, et al.: Reliability of PET/CT Shape and Heterogeneity Features in Functional and Morphologic Components of Non–Small Cell Lung Cancer Tumors: A Repeatability Analysis in a Prospective Multicenter Cohort. J Nucl Med 2017; 58:406–411.

Regards,

Joost van Griethuysen

R

R

Op vrijdag 20 oktober 2017 18:38:13 UTC+2 schreef Daniel Gutmann:

[Daniel Gutmann]

Hi Joost,

many thanks for your prompt reply. I would also like to add that pyradiomics is really the most excellent tool 

in the field.

In my case the differences in intensities within the ROI reflect proton density fat fraction (PDFFs) and not scanner calibrations.

The sequences were acquired on strictly calibrated MRIs as part of a large national cohort study. Your argument makes perfect sense to me of course, in particular where “similar” lesions, such as glioblastoma or DCIS are concerned, since we would expect the tumors to have similar “ground truth” intensities. I wonder now whether PDFFs in a range from 0.5 to 30% can be treated in the same way? 

Sofar, my take was that I would first calculate the range of my rois and then set the bin width individually to be exactly the same for all of them ( somewhere between 40 to 120). Based on your arguments and Desseroits paper I begin to think that I should give fixed bin width a shot as well. 

Best regards,

Dan

--
You received this message because you are subscribed to a topic in the Google Groups "pyradiomics" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/pyradiomics/0Wx9c9UlUvM/unsubscribe.
To unsubscribe from this group and all its topics, send an email to pyradiomics...@googlegroups.com.
To post to this group, send email to pyrad...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/pyradiomics/78b1ccd8-7f16-40ba-9858-2f5202029792%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Joost van Griethuysen]

Hi Dan,

The short answer to your question about treating PDFF similarly is yes. In fact, I believe that the expectation of the "ground truth", i.e. that a gray value has the same meaning between images applies more to your PDFF scans (where a gray value has an absolute real world value) than a run-of-the mill MRI (where gray values are relative instead of absolute).Therefore, the need for comparable binning is even higher in your PDFF scans. If you have a lesion with a nearly homogenous lesion (which means the range will also be small most likely), you'd want that to be reflected in your feature values. Using a fixed bin count would mean you are 'creating' heterogeneity where there is none.

Regards,

Joost

Op zondag 22 oktober 2017 22:04:48 UTC+2 schreef Daniel Gutmann:

To unsubscribe from this group and all its topics, send an email to pyradiomics+unsubscribe@googlegroups.com.