Obtaining a 3D reconstruction of anatomical structures such as the left atria is a prerequisite to run physiological simulations to mimic its physical behaviour such as haemodynamics, mechanics or electrophysiology. Unfortunately,  some clinical applications are guided by 2D-based imaging acquisitions (e.g. 2D echocardiography), rather than full volumetric sequences such as from CT or MRI scanners. Nevertheless, the current development and progress of deep learning techniques is making possible the reconstruction of 3D geometries from multiple 2D acquisitions, once this relation has been learnt. The main goal of the project is to develop an algorithm to reconstruct/approximate the 3D anatomy of the left atria from multiple standard 2D acquisitions of echocardiography to subsequently be able to run fluid simulations.

Supervisor: Oscar Camara