Modern treatments for atrial fibrillation require ablation of problematic heart tissue. I have already created a new catheter-based vision system that allows inspection before, during and after treatment (image below). I have just started a further project that will use clinical data from many treatments to train a neural network to provide expert-quality inferences to assist surgeons during treatments.
I am using a C++ CPU-based implementation of a Convolutional Neural Network for ease of development and debugging. But the intention is to move towards a homogeneous implementation, e.g. with OpenCL, for best performance in the deployed product.
The Skull Canyon NUC looks built for the purpose - I hope I can get one soon.