Endometrial cancer detection using Image processing

Hello Sir, I am working with endometrial cancer(Adenocarcinoma) images to detect cancerous cells using MATLAB R2008a. My images are 800px (800*533 Pixel). So shall I have to work with the whole image or have to crop it?

This image is a high mag and RGB image.shall I have to proceed in color based segmentation using K means clustering as in the demo of MATLAB?If so..then how can I extract the features from that.What should be the feature?cell size,cell density? or something else.

In a paper i saw that(lung cancer) at the final classification stage they have segmented R,G,B separately and fed it into neural Network for detection.Is it possible to feed the input into NNtool box,then how can I do that? Or after feature extration i can feed the feature as input in NNtoolbox?and how?

Please suggest.

Thanks in advance for your co-operation.

Regards, Shraboni Mondal.

Answer 1

[Jurgen]

Since you don't know where the cells are, cropping the images could mean deleting the areas with the cancer cells. So it is your decision to crop, since I can not see the images, nor am I specialized in endometrial cancer...

As to how to detect them: first you need to know how the cancer cells are distinct from normal cells (color? morphology?).

On a final note, there are many books on segmenting images, including the use of neural networks.

[Shraboni Mondal]

Thank you Mr. Jurgen,this is really helpful .I have one more doubt.Though I am also not specialized in Endometrial cancer,but I have gone through some medical and pathological book.There also I couldnt find out specifically how normal and affected cells are different.Can you please give me some idea about how to extract features from the images?

[Jurgen]

Are they microscope pictures of cells? Or macroscopic images from an endoscope? Or something else?