Hi,
The project describes the application of modern AI learning to medical segmentation. The segmentation framework heavily depends on the combination of tools we decide to leverage upon. Deep learning nets are a good fit for such segmentation problems but they may have to combined with Fuzzy or Evolutionary algorithms to achieve better accuracy. We can also investigate on the application of snake energy minimisation to come up with optimal contours for the complex images and maybe train ANNs on the contours.
I had good amount of experience with ANNs and evolutionary algorithms , was able to apply these techniques for problems including traffic network designs (implemented on-site research for the Lyon and Paris traffic networks), inventory management and recognition of diabetic eye. My research on DR detection in the eye is still in progress and have been using Genetic algorithm based snakes and ANN to classify the DR stage. I am happy to support this with examples taken from my implementations. With the confidence obtained from research experience , I want to share and build robust ideas for the segmentation problem you stated.
If interested, I am willing to document research article that can be spanned from our results, modified techniques, literature reviews and our discussions.
Please let me know if you want to discuss further on this.
Thanks,
Vamsi