Learning Doppler with deep neural networks and its application to intra-cardiac echography

Cardiac ablation therapy is an effective treatment for atrial fibrillation and ventricular tachycardia that relies on the creation of electrically isolating scars, e.g. through heat. The ability to reliably visualize and assess the formation of these lesions during the procedure would greatly enhance the therapy's success rate and safety. Tissue Doppler echography enables measurement of tissue strain, and could therefore be used to monitor and quantify the stiffening of developing lesions. In tissue Doppler, the tradeoff between spatiotemporal resolution and estimation accuracy/precision is balanced by manually tweaking the fast-and slow-time range gates, with the optimal settings varying across measurements and desired clinical objectives. Convolutional neural networks have shown remarkable performance at learning to execute a large variety of signal and image processing tasks. In this work, we show how a deep neural network can be trained to robustly fulfil Doppler imaging functionality, which we term DopplerNet.