Detecting Anatomy Openings in the Left Atrium Via a Triangular Ultrasonic Array Using Deep Learning

Cardiac arrhythmia is the clinical term for the set of diseases wherein the heart beats irregularly. A widespread treatment is ablating the arrhythmia maintaining regions, which requires electro-anatomical mapping. A proposed sparse ultrasonic/electrode array can potentially map the anatomy and activity in real time. However, a limited amount of elements causes a difficulty in mapping anatomical openings. We propose a deep learning model to increase the mapping capacity. We empirically show that our proposed method is able to accurately detect openings in a heart chamber anatomy simulation. We further improve the accuracy of the model by adding Fourier-based pre-processing steps. Finally, we demonstrate the robustness of the model to changes in the physical parameters of the problem.