Automatic bone metastasis segmentation

I want to detect bone metastasis in bone scintigraphy. The best way I've found so far was to apply the canny edge function to my enhanced image, then apply the close morphological operation so that I can finally apply the imfill() function to obtain my regions. The initial image and the final result is attached: the metastasis are the black regions (in this case we have one in each shoulder and one in the head). My problem is that this only works if I manually select, for each scintigraphy, the threshold values of the canny edge function that can detect the edges of my tumor. What works for some images does not work for others. And I want to do this in a fully automatic way. I've also been trying the watershed funtion but so far I couldn't get any good results. Is there a way I can do this in a more automatic way?

Answer 1

[drummer]

Nuclear Medicine images are tricky.

Is normalization of the images a step of your pipeline? Wouldn't this work?

Another workaround would be parametrization of your images. This could take a while implementing as you're using whole-body images. This way, you could get a sample of healthy subjects' scintigraphy and perform statistical differences between them and a metastatic scintigraphy to find statistical differences of uptake. That would fully automatize your detection, let's by differences of p = 0.05.

Another approach, which is a trend in nuclear medicine is using deep learning to detect/segment medical images. Are you considering to use it as well?

[Laura Providência]

Yes I did normalize the images. And yes the next step is to use deep learning! I'm just new to the field of deep learning and I wasn't sure how I would apply that to this situation (I will have someone helping me with that). I thought I had to segment the metastasis previously, but now I see that is one of the jobs of machine learning? The parametrization also seems interesting. Thanks for your suggestions!