50089_en-US

liliconcom

Image analysis for GC x GC data

Hi,

 

I'm at a conference on analytical chemistry. Someone asked if JMP could use image analysis to find differences between 2 images from GC x GC. The images look something like the one in this example.

 

Currently they just eyeball the images to look for differences - hotspots in one image that are not in another.

 

How could JMP help?

 

They have the raw data that produces the heat map. My feeling is that they should probably just use this in a wide format with MVA tools or others to find where there are big differences.

 

BTW, I have a couple of days to get back to them on this.

 

Thanks,

Phil

Re: Image analysis for GC x GC data

Thanks folks.

 

Yes, I agree that analysing the raw data makes more sense than using image analysis. Along with the dynamic linking to visual show where the differences are on the heat map. I will guide them in that direction. I can probably mock up something with a wafer map example.

 

Would be nice to build in FDE. I suspect FDE is going to replace Bootstrap Forest as my new answer-to-every-analysis-problem .

Re: Image analysis for GC x GC data

Hi Phil,

 

I agree with you on using MVA techniques to analyze the raw data.  Thinking PLS specifically and the example for octane rating from the Discovering Partial Least Squares book.

 

Martin's suggestion on comparing images sounds good too.  I have used the gray scale results from the image analysis add-in the past to compare images, but with John Ponte's add-in you also get RGB scale and Intensity values.  If the images are like the example you referenced I would bet you will find that the image analysis will work for you as well.

Re: Image analysis for GC x GC data

Hi Phil,

 

If they already have the raw data then analyse it with MVA as suggested. In addition to graphing the data I would consider PLS, FDE and Clustering with the actual method of choice depending on whether they have Y variables to predict.

 

Malcolm

Re: Image analysis for GC x GC data

Hi Phil,

 

that JMP is capable of doing Image recognition and comparison this success story of wildtrack should lead the way:

https://www.jmp.com/en_us/customer-stories/wildtrack.html

However, if you have the data for that image it would be possible to do this analytically. If you have image data in a matrix setup with values from 0 to 255 this should be an easy step to find the differences and locate as well visualize/highlight them.

 

I guess John Ponte could have already some material for that kind of image comparison based on coloring. May you want to reach out to him.

Martin